+ All Categories
Home > Documents > Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling...

Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling...

Date post: 11-Mar-2018
Category:
Upload: truonghuong
View: 214 times
Download: 2 times
Share this document with a friend
52
Accepted Manuscript Modelling of organic Rankine cycle power systems in off-design conditions: An experimentally-validated comparative study Rémi Dickes, Olivier Dumont, Rémi Daccord, Sylvain Quoilin, Vincent Lemort PII: S0360-5442(17)30137-8 DOI: 10.1016/j.energy.2017.01.130 Reference: EGY 10265 To appear in: Energy Received Date: 29 August 2016 Revised Date: 5 December 2016 Accepted Date: 25 January 2017 Please cite this article as: Dickes R, Dumont O, Daccord R, Quoilin S, Lemort V, Modelling of organic Rankine cycle power systems in off-design conditions: An experimentally-validated comparative study, Energy (2017), doi: 10.1016/j.energy.2017.01.130. This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Transcript
Page 1: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

Accepted Manuscript

Modelling of organic Rankine cycle power systems in off-design conditions Anexperimentally-validated comparative study

Reacutemi Dickes Olivier Dumont Reacutemi Daccord Sylvain Quoilin Vincent Lemort

PII S0360-5442(17)30137-8

DOI 101016jenergy201701130

Reference EGY 10265

To appear in Energy

Received Date 29 August 2016

Revised Date 5 December 2016

Accepted Date 25 January 2017

Please cite this article as Dickes R Dumont O Daccord R Quoilin S Lemort V Modelling of organicRankine cycle power systems in off-design conditions An experimentally-validated comparative studyEnergy (2017) doi 101016jenergy201701130

This is a PDF file of an unedited manuscript that has been accepted for publication As a service toour customers we are providing this early version of the manuscript The manuscript will undergocopyediting typesetting and review of the resulting proof before it is published in its final form Pleasenote that during the production process errors may be discovered which could affect the content and alllegal disclaimers that apply to the journal pertain

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Modelling of organic Rankine cycle power systems inoff-design conditions an experimentally-validated

comparative study

Remi Dickesalowast Olivier Dumonta Remi Daccordb Sylvain Quoilina VincentLemorta

aThermodynamics Laboratory Faculty of Applied Sciences University of LiegeAllee de la Decouverte 17 B-4000 Liege Belgium

bExoes 6 Avenue de la Grande Lande F-33170 Gradignan France

Abstract

Because of environmental issues and the depletion of fossil fuels the world en-

ergy sector is undergoing many changes toward increased sustainability Among

the many fields of research and development power generation from low-grade

heat sources is gaining interest and the organic Rankine cycle (ORC) is seen

as one of the most promising technologies for such applications In this pa-

per it is proposed to perform an experimentally-validated comparison of dif-

ferent modelling methods for the off-design simulation of ORC-based power

systems To this end three types of modelling paradigms (namely a constant-

efficiency method a polynomial-based method and a semi-empirical method)

are compared both in terms of their fitting and extrapolation capabilities Post-

processed measurements gathered on two experimental ORC facilities are used

as reference for the models calibration and evaluation The study is first ap-

plied at a component level (ie each component is analysed individually) and

then extended to the characterization of the entire organic Rankine cycle power

systems Benefits and limitations of each modelling method are discussed The

results show that semi-empirical models are the most reliable for simulating the

lowastCorresponding authorEmail addresses rdickesulgacbe (Remi Dickes) olivierdumontulgacbe

(Olivier Dumont) remidaccordexoescom (Remi Daccord) squoilinulgacbe (SylvainQuoilin) vincentlemortulgacbe (Vincent Lemort)

Preprint submitted to Energy December 5 2016

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

off-design working conditions of ORC systems while constant-efficiency and

polynomial-based models are both demonstrating lack of accuracy andor ro-

bustness

Keywords Organic Rankine Cycle modelling off-design experimental data

simulation

1 Introduction

Among the many fields of research and development toward increased sus-

tainability power generation from low-grade heat sources (ie below 200C)

is gaining interest because of its enormous worldwide power potential In this

context the Organic Rankine Cycle (ORC) is acknowledged as one of the most5

suitable technologies for valorizing low-grade heat into electricity or mechanical

power [1] The working principle of an ORC is identical with that of a conven-

tional steam Rankine engine it constitutes a closed-loop thermodynamic cycle

into which a working fluid undergoes a series of processes (ie compression

evaporation expansion and condensation) aiming to partially convert thermal10

power from a heat source into mechanical power The distinction is related to

the nature of the working fluid instead of using water like in a conventional

steam Rankine cycle ORC systems employ organic compounds which are char-

acterized by lower boiling points and higher molecular mass By substituting

water for such organic fluids it is possible to perform efficiently the Rankine15

cycle at low power capacities and using heat from low-grade thermal sources [1]

The technology of the ORC is rather old and first experimental facilities

date from the late nineteenth century [2 3] Nowadays the total power ca-

pacity installed worldwide is estimated at 2 GWe [4] and ORC-based power20

systems have continuously been gaining in interest for more than a decade As

a figure of merit the number of papers yearly published about organic Rank-

ine cycles is illustrated in Figure 1 Most of these scientific works focus on

2

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

0

50

100

150

200

250

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Nu

mb

er o

f p

ub

licat

ion

s p

er y

ear

Year

ORC

ORC + modelling

ORC + control

Figure 1 Yearly number of publications related to ORC systems from 2001 to 2015 (source

advanced search with different keywords in ScienceDirect)

design optimization proper fluid selection exergyenergy analyses and various

techno-economic studies However a common feature of ORC-based systems is25

the versatile nature of the operating conditions In most of the fields of applica-

tion (eg solar thermal power combined heat and power geothermal or waste

heat recovery) the heat source (and eventually the heat sink) fluctuates in time

and the machine must adapt its working regime to ensure an optimal system

operation Despite of its importance the number of papers related to control30

aspects and off-design performance of ORC systems is comparatively low

A few steady-state performance analyses have been published for different

ORC architectures and applications For instance Gurgenci [5] proposed a sim-

ple semi-analytical model to assess the performance of ORC-based power plants35

The model aimed to easily derive the off-design behaviour of any ORC system

based on its design operating conditions The case of a 150 kWe solar pond

power plant was studied as an example and Gurgenci discussed the dependence

of the system efficiency in function of the turbine load and the hot and cold fluids

supply temperatures Another solar-driven ORC power plant was investigated40

3

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

in off-design operation by Wang et al [6] The system consisted of a 250 kWe

ORC module (R245fa as working fluid) coupled to a thermal energy storage and

compound parabolic collectors The off-design performance of the whole power

plant was assessed under variations in the ambient temperature and the heat

source mass flow rate Similarly Calise et al [7] studied a 230 kWe recupera-45

tive ORC power unit (n-butane as working fluid) coupled with solar parabolic

trough collectors After optimally sizing the different shell-and-tube heat ex-

changers (ie the recuperator economizer evaporator and superheater) the

authors evaluated the ORC off-design behaviour while varying the thermal heat

source both in terms of mass flow rate and supply temperature In the same50

power scale Fu et al [8] performed a theoretical study on a 250 kWe ORC

using R245fa as working fluid Only the influence of the heat source mass flow

rate on the power plant performance was considered The ORC was controlled

following a sliding pressure strategy the evaporation pressure was controlled

to ensure the working fluid to reach saturated liquid and vapour states at the55

outlet of the preheater and the evaporator respectively Hu et al [9] proposed

a more physical analysis and investigated three control schemes to operate a

70 kWe geothermal ORC unit namely a constant-pressure strategy a sliding-

pressure strategy and optimal-pressure strategy The system featured a radial

inflow turbine plate heat exchangers and used R245fa as working fluid Both60

the refrigerant mass flow rate and variable inlet guide vanes were used to adapt

the power plant behaviour in function of the operating conditions (variation of

the heat source supply temperature and mass flow rate) Manente et al [10]

studied a much larger geothermal power plant (gt 5 MWe) and performed a con-

strained optimization to maximize the system net power output Both R134a65

and Isobutane were considered as working fluid and three variables were used

to control the plant behaviour namely the pump speed the cooling air mass

flow rate in the condenser and the turbine capacity factor Both variations of

the ambiance and heat source supply temperature were considered in the study

Sun and Li [11] also analysed the off-design control of a 5 MWe ORC unit They70

demonstrated that the relationships between controlled variables (optimal work-

4

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ing fluid and air mass flow rates) and external perturbations (heat source and

ambient temperatures) are near linear function for maximizing the system net

power generation and quadratic function for maximizing the system thermal

efficiency Finally Quoilin [12] analysed the off-design performance of a micro-75

scale 15 kWe ORC prototype The system consisted of plate heat exchangers a

scroll expander and employed R123 as working fluid A control of the pump and

the expander speeds was proposed to maximize the ORC thermal efficiency All

the aforementioned studies were performed in steady-state conditions However

the transients affecting the boundary conditions of the ORCs are often faster80

than the response time of the system In such case proper control investigations

and off-design analyses require to account for the dynamic effects induced by

mass and energy accumulations in the various ORC components Such dynamic

performance assessment and control studies can also be found in the scientific

literature see for example [13 14 15 16 17 18 19 20]85

The works presented here above have one feature in common they all used

mathematical models to predict the behaviour of the ORCs and their compo-

nents in off-design conditions Indeed making measurements on existing power

units is costly and time-consuming and very few papers published experimental90

data characterizing ORC systems over their complete operating ranges (see one

example in [21]) In almost every case the experimental data (if there is any)

gathered on the facility only covers a narrow range of the feasible operating

conditions and they are not sufficient for a global empirical characterization of

the system Extrapolating the ORC performance in unknown working condi-95

tions can be performed by means of off-design modelling tools As shown in

the aforementioned papers there is a wide variety of modelling paradigms to

estimate the components state in an ORC system ranging from the simplest

method (eg to assume constant efficiencies for characterizing a turbine) to the

most complex one (eg CFD modelling of the same turbine) Each modelling100

method differs from the others in terms of complexity accuracy computational

speed calibration effort and domain of validity Commonly the most accurate

5

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 2: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Modelling of organic Rankine cycle power systems inoff-design conditions an experimentally-validated

comparative study

Remi Dickesalowast Olivier Dumonta Remi Daccordb Sylvain Quoilina VincentLemorta

aThermodynamics Laboratory Faculty of Applied Sciences University of LiegeAllee de la Decouverte 17 B-4000 Liege Belgium

bExoes 6 Avenue de la Grande Lande F-33170 Gradignan France

Abstract

Because of environmental issues and the depletion of fossil fuels the world en-

ergy sector is undergoing many changes toward increased sustainability Among

the many fields of research and development power generation from low-grade

heat sources is gaining interest and the organic Rankine cycle (ORC) is seen

as one of the most promising technologies for such applications In this pa-

per it is proposed to perform an experimentally-validated comparison of dif-

ferent modelling methods for the off-design simulation of ORC-based power

systems To this end three types of modelling paradigms (namely a constant-

efficiency method a polynomial-based method and a semi-empirical method)

are compared both in terms of their fitting and extrapolation capabilities Post-

processed measurements gathered on two experimental ORC facilities are used

as reference for the models calibration and evaluation The study is first ap-

plied at a component level (ie each component is analysed individually) and

then extended to the characterization of the entire organic Rankine cycle power

systems Benefits and limitations of each modelling method are discussed The

results show that semi-empirical models are the most reliable for simulating the

lowastCorresponding authorEmail addresses rdickesulgacbe (Remi Dickes) olivierdumontulgacbe

(Olivier Dumont) remidaccordexoescom (Remi Daccord) squoilinulgacbe (SylvainQuoilin) vincentlemortulgacbe (Vincent Lemort)

Preprint submitted to Energy December 5 2016

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

off-design working conditions of ORC systems while constant-efficiency and

polynomial-based models are both demonstrating lack of accuracy andor ro-

bustness

Keywords Organic Rankine Cycle modelling off-design experimental data

simulation

1 Introduction

Among the many fields of research and development toward increased sus-

tainability power generation from low-grade heat sources (ie below 200C)

is gaining interest because of its enormous worldwide power potential In this

context the Organic Rankine Cycle (ORC) is acknowledged as one of the most5

suitable technologies for valorizing low-grade heat into electricity or mechanical

power [1] The working principle of an ORC is identical with that of a conven-

tional steam Rankine engine it constitutes a closed-loop thermodynamic cycle

into which a working fluid undergoes a series of processes (ie compression

evaporation expansion and condensation) aiming to partially convert thermal10

power from a heat source into mechanical power The distinction is related to

the nature of the working fluid instead of using water like in a conventional

steam Rankine cycle ORC systems employ organic compounds which are char-

acterized by lower boiling points and higher molecular mass By substituting

water for such organic fluids it is possible to perform efficiently the Rankine15

cycle at low power capacities and using heat from low-grade thermal sources [1]

The technology of the ORC is rather old and first experimental facilities

date from the late nineteenth century [2 3] Nowadays the total power ca-

pacity installed worldwide is estimated at 2 GWe [4] and ORC-based power20

systems have continuously been gaining in interest for more than a decade As

a figure of merit the number of papers yearly published about organic Rank-

ine cycles is illustrated in Figure 1 Most of these scientific works focus on

2

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

0

50

100

150

200

250

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Nu

mb

er o

f p

ub

licat

ion

s p

er y

ear

Year

ORC

ORC + modelling

ORC + control

Figure 1 Yearly number of publications related to ORC systems from 2001 to 2015 (source

advanced search with different keywords in ScienceDirect)

design optimization proper fluid selection exergyenergy analyses and various

techno-economic studies However a common feature of ORC-based systems is25

the versatile nature of the operating conditions In most of the fields of applica-

tion (eg solar thermal power combined heat and power geothermal or waste

heat recovery) the heat source (and eventually the heat sink) fluctuates in time

and the machine must adapt its working regime to ensure an optimal system

operation Despite of its importance the number of papers related to control30

aspects and off-design performance of ORC systems is comparatively low

A few steady-state performance analyses have been published for different

ORC architectures and applications For instance Gurgenci [5] proposed a sim-

ple semi-analytical model to assess the performance of ORC-based power plants35

The model aimed to easily derive the off-design behaviour of any ORC system

based on its design operating conditions The case of a 150 kWe solar pond

power plant was studied as an example and Gurgenci discussed the dependence

of the system efficiency in function of the turbine load and the hot and cold fluids

supply temperatures Another solar-driven ORC power plant was investigated40

3

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

in off-design operation by Wang et al [6] The system consisted of a 250 kWe

ORC module (R245fa as working fluid) coupled to a thermal energy storage and

compound parabolic collectors The off-design performance of the whole power

plant was assessed under variations in the ambient temperature and the heat

source mass flow rate Similarly Calise et al [7] studied a 230 kWe recupera-45

tive ORC power unit (n-butane as working fluid) coupled with solar parabolic

trough collectors After optimally sizing the different shell-and-tube heat ex-

changers (ie the recuperator economizer evaporator and superheater) the

authors evaluated the ORC off-design behaviour while varying the thermal heat

source both in terms of mass flow rate and supply temperature In the same50

power scale Fu et al [8] performed a theoretical study on a 250 kWe ORC

using R245fa as working fluid Only the influence of the heat source mass flow

rate on the power plant performance was considered The ORC was controlled

following a sliding pressure strategy the evaporation pressure was controlled

to ensure the working fluid to reach saturated liquid and vapour states at the55

outlet of the preheater and the evaporator respectively Hu et al [9] proposed

a more physical analysis and investigated three control schemes to operate a

70 kWe geothermal ORC unit namely a constant-pressure strategy a sliding-

pressure strategy and optimal-pressure strategy The system featured a radial

inflow turbine plate heat exchangers and used R245fa as working fluid Both60

the refrigerant mass flow rate and variable inlet guide vanes were used to adapt

the power plant behaviour in function of the operating conditions (variation of

the heat source supply temperature and mass flow rate) Manente et al [10]

studied a much larger geothermal power plant (gt 5 MWe) and performed a con-

strained optimization to maximize the system net power output Both R134a65

and Isobutane were considered as working fluid and three variables were used

to control the plant behaviour namely the pump speed the cooling air mass

flow rate in the condenser and the turbine capacity factor Both variations of

the ambiance and heat source supply temperature were considered in the study

Sun and Li [11] also analysed the off-design control of a 5 MWe ORC unit They70

demonstrated that the relationships between controlled variables (optimal work-

4

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ing fluid and air mass flow rates) and external perturbations (heat source and

ambient temperatures) are near linear function for maximizing the system net

power generation and quadratic function for maximizing the system thermal

efficiency Finally Quoilin [12] analysed the off-design performance of a micro-75

scale 15 kWe ORC prototype The system consisted of plate heat exchangers a

scroll expander and employed R123 as working fluid A control of the pump and

the expander speeds was proposed to maximize the ORC thermal efficiency All

the aforementioned studies were performed in steady-state conditions However

the transients affecting the boundary conditions of the ORCs are often faster80

than the response time of the system In such case proper control investigations

and off-design analyses require to account for the dynamic effects induced by

mass and energy accumulations in the various ORC components Such dynamic

performance assessment and control studies can also be found in the scientific

literature see for example [13 14 15 16 17 18 19 20]85

The works presented here above have one feature in common they all used

mathematical models to predict the behaviour of the ORCs and their compo-

nents in off-design conditions Indeed making measurements on existing power

units is costly and time-consuming and very few papers published experimental90

data characterizing ORC systems over their complete operating ranges (see one

example in [21]) In almost every case the experimental data (if there is any)

gathered on the facility only covers a narrow range of the feasible operating

conditions and they are not sufficient for a global empirical characterization of

the system Extrapolating the ORC performance in unknown working condi-95

tions can be performed by means of off-design modelling tools As shown in

the aforementioned papers there is a wide variety of modelling paradigms to

estimate the components state in an ORC system ranging from the simplest

method (eg to assume constant efficiencies for characterizing a turbine) to the

most complex one (eg CFD modelling of the same turbine) Each modelling100

method differs from the others in terms of complexity accuracy computational

speed calibration effort and domain of validity Commonly the most accurate

5

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 3: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

off-design working conditions of ORC systems while constant-efficiency and

polynomial-based models are both demonstrating lack of accuracy andor ro-

bustness

Keywords Organic Rankine Cycle modelling off-design experimental data

simulation

1 Introduction

Among the many fields of research and development toward increased sus-

tainability power generation from low-grade heat sources (ie below 200C)

is gaining interest because of its enormous worldwide power potential In this

context the Organic Rankine Cycle (ORC) is acknowledged as one of the most5

suitable technologies for valorizing low-grade heat into electricity or mechanical

power [1] The working principle of an ORC is identical with that of a conven-

tional steam Rankine engine it constitutes a closed-loop thermodynamic cycle

into which a working fluid undergoes a series of processes (ie compression

evaporation expansion and condensation) aiming to partially convert thermal10

power from a heat source into mechanical power The distinction is related to

the nature of the working fluid instead of using water like in a conventional

steam Rankine cycle ORC systems employ organic compounds which are char-

acterized by lower boiling points and higher molecular mass By substituting

water for such organic fluids it is possible to perform efficiently the Rankine15

cycle at low power capacities and using heat from low-grade thermal sources [1]

The technology of the ORC is rather old and first experimental facilities

date from the late nineteenth century [2 3] Nowadays the total power ca-

pacity installed worldwide is estimated at 2 GWe [4] and ORC-based power20

systems have continuously been gaining in interest for more than a decade As

a figure of merit the number of papers yearly published about organic Rank-

ine cycles is illustrated in Figure 1 Most of these scientific works focus on

2

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

0

50

100

150

200

250

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Nu

mb

er o

f p

ub

licat

ion

s p

er y

ear

Year

ORC

ORC + modelling

ORC + control

Figure 1 Yearly number of publications related to ORC systems from 2001 to 2015 (source

advanced search with different keywords in ScienceDirect)

design optimization proper fluid selection exergyenergy analyses and various

techno-economic studies However a common feature of ORC-based systems is25

the versatile nature of the operating conditions In most of the fields of applica-

tion (eg solar thermal power combined heat and power geothermal or waste

heat recovery) the heat source (and eventually the heat sink) fluctuates in time

and the machine must adapt its working regime to ensure an optimal system

operation Despite of its importance the number of papers related to control30

aspects and off-design performance of ORC systems is comparatively low

A few steady-state performance analyses have been published for different

ORC architectures and applications For instance Gurgenci [5] proposed a sim-

ple semi-analytical model to assess the performance of ORC-based power plants35

The model aimed to easily derive the off-design behaviour of any ORC system

based on its design operating conditions The case of a 150 kWe solar pond

power plant was studied as an example and Gurgenci discussed the dependence

of the system efficiency in function of the turbine load and the hot and cold fluids

supply temperatures Another solar-driven ORC power plant was investigated40

3

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

in off-design operation by Wang et al [6] The system consisted of a 250 kWe

ORC module (R245fa as working fluid) coupled to a thermal energy storage and

compound parabolic collectors The off-design performance of the whole power

plant was assessed under variations in the ambient temperature and the heat

source mass flow rate Similarly Calise et al [7] studied a 230 kWe recupera-45

tive ORC power unit (n-butane as working fluid) coupled with solar parabolic

trough collectors After optimally sizing the different shell-and-tube heat ex-

changers (ie the recuperator economizer evaporator and superheater) the

authors evaluated the ORC off-design behaviour while varying the thermal heat

source both in terms of mass flow rate and supply temperature In the same50

power scale Fu et al [8] performed a theoretical study on a 250 kWe ORC

using R245fa as working fluid Only the influence of the heat source mass flow

rate on the power plant performance was considered The ORC was controlled

following a sliding pressure strategy the evaporation pressure was controlled

to ensure the working fluid to reach saturated liquid and vapour states at the55

outlet of the preheater and the evaporator respectively Hu et al [9] proposed

a more physical analysis and investigated three control schemes to operate a

70 kWe geothermal ORC unit namely a constant-pressure strategy a sliding-

pressure strategy and optimal-pressure strategy The system featured a radial

inflow turbine plate heat exchangers and used R245fa as working fluid Both60

the refrigerant mass flow rate and variable inlet guide vanes were used to adapt

the power plant behaviour in function of the operating conditions (variation of

the heat source supply temperature and mass flow rate) Manente et al [10]

studied a much larger geothermal power plant (gt 5 MWe) and performed a con-

strained optimization to maximize the system net power output Both R134a65

and Isobutane were considered as working fluid and three variables were used

to control the plant behaviour namely the pump speed the cooling air mass

flow rate in the condenser and the turbine capacity factor Both variations of

the ambiance and heat source supply temperature were considered in the study

Sun and Li [11] also analysed the off-design control of a 5 MWe ORC unit They70

demonstrated that the relationships between controlled variables (optimal work-

4

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ing fluid and air mass flow rates) and external perturbations (heat source and

ambient temperatures) are near linear function for maximizing the system net

power generation and quadratic function for maximizing the system thermal

efficiency Finally Quoilin [12] analysed the off-design performance of a micro-75

scale 15 kWe ORC prototype The system consisted of plate heat exchangers a

scroll expander and employed R123 as working fluid A control of the pump and

the expander speeds was proposed to maximize the ORC thermal efficiency All

the aforementioned studies were performed in steady-state conditions However

the transients affecting the boundary conditions of the ORCs are often faster80

than the response time of the system In such case proper control investigations

and off-design analyses require to account for the dynamic effects induced by

mass and energy accumulations in the various ORC components Such dynamic

performance assessment and control studies can also be found in the scientific

literature see for example [13 14 15 16 17 18 19 20]85

The works presented here above have one feature in common they all used

mathematical models to predict the behaviour of the ORCs and their compo-

nents in off-design conditions Indeed making measurements on existing power

units is costly and time-consuming and very few papers published experimental90

data characterizing ORC systems over their complete operating ranges (see one

example in [21]) In almost every case the experimental data (if there is any)

gathered on the facility only covers a narrow range of the feasible operating

conditions and they are not sufficient for a global empirical characterization of

the system Extrapolating the ORC performance in unknown working condi-95

tions can be performed by means of off-design modelling tools As shown in

the aforementioned papers there is a wide variety of modelling paradigms to

estimate the components state in an ORC system ranging from the simplest

method (eg to assume constant efficiencies for characterizing a turbine) to the

most complex one (eg CFD modelling of the same turbine) Each modelling100

method differs from the others in terms of complexity accuracy computational

speed calibration effort and domain of validity Commonly the most accurate

5

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 4: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

0

50

100

150

200

250

2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015

Nu

mb

er o

f p

ub

licat

ion

s p

er y

ear

Year

ORC

ORC + modelling

ORC + control

Figure 1 Yearly number of publications related to ORC systems from 2001 to 2015 (source

advanced search with different keywords in ScienceDirect)

design optimization proper fluid selection exergyenergy analyses and various

techno-economic studies However a common feature of ORC-based systems is25

the versatile nature of the operating conditions In most of the fields of applica-

tion (eg solar thermal power combined heat and power geothermal or waste

heat recovery) the heat source (and eventually the heat sink) fluctuates in time

and the machine must adapt its working regime to ensure an optimal system

operation Despite of its importance the number of papers related to control30

aspects and off-design performance of ORC systems is comparatively low

A few steady-state performance analyses have been published for different

ORC architectures and applications For instance Gurgenci [5] proposed a sim-

ple semi-analytical model to assess the performance of ORC-based power plants35

The model aimed to easily derive the off-design behaviour of any ORC system

based on its design operating conditions The case of a 150 kWe solar pond

power plant was studied as an example and Gurgenci discussed the dependence

of the system efficiency in function of the turbine load and the hot and cold fluids

supply temperatures Another solar-driven ORC power plant was investigated40

3

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

in off-design operation by Wang et al [6] The system consisted of a 250 kWe

ORC module (R245fa as working fluid) coupled to a thermal energy storage and

compound parabolic collectors The off-design performance of the whole power

plant was assessed under variations in the ambient temperature and the heat

source mass flow rate Similarly Calise et al [7] studied a 230 kWe recupera-45

tive ORC power unit (n-butane as working fluid) coupled with solar parabolic

trough collectors After optimally sizing the different shell-and-tube heat ex-

changers (ie the recuperator economizer evaporator and superheater) the

authors evaluated the ORC off-design behaviour while varying the thermal heat

source both in terms of mass flow rate and supply temperature In the same50

power scale Fu et al [8] performed a theoretical study on a 250 kWe ORC

using R245fa as working fluid Only the influence of the heat source mass flow

rate on the power plant performance was considered The ORC was controlled

following a sliding pressure strategy the evaporation pressure was controlled

to ensure the working fluid to reach saturated liquid and vapour states at the55

outlet of the preheater and the evaporator respectively Hu et al [9] proposed

a more physical analysis and investigated three control schemes to operate a

70 kWe geothermal ORC unit namely a constant-pressure strategy a sliding-

pressure strategy and optimal-pressure strategy The system featured a radial

inflow turbine plate heat exchangers and used R245fa as working fluid Both60

the refrigerant mass flow rate and variable inlet guide vanes were used to adapt

the power plant behaviour in function of the operating conditions (variation of

the heat source supply temperature and mass flow rate) Manente et al [10]

studied a much larger geothermal power plant (gt 5 MWe) and performed a con-

strained optimization to maximize the system net power output Both R134a65

and Isobutane were considered as working fluid and three variables were used

to control the plant behaviour namely the pump speed the cooling air mass

flow rate in the condenser and the turbine capacity factor Both variations of

the ambiance and heat source supply temperature were considered in the study

Sun and Li [11] also analysed the off-design control of a 5 MWe ORC unit They70

demonstrated that the relationships between controlled variables (optimal work-

4

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ing fluid and air mass flow rates) and external perturbations (heat source and

ambient temperatures) are near linear function for maximizing the system net

power generation and quadratic function for maximizing the system thermal

efficiency Finally Quoilin [12] analysed the off-design performance of a micro-75

scale 15 kWe ORC prototype The system consisted of plate heat exchangers a

scroll expander and employed R123 as working fluid A control of the pump and

the expander speeds was proposed to maximize the ORC thermal efficiency All

the aforementioned studies were performed in steady-state conditions However

the transients affecting the boundary conditions of the ORCs are often faster80

than the response time of the system In such case proper control investigations

and off-design analyses require to account for the dynamic effects induced by

mass and energy accumulations in the various ORC components Such dynamic

performance assessment and control studies can also be found in the scientific

literature see for example [13 14 15 16 17 18 19 20]85

The works presented here above have one feature in common they all used

mathematical models to predict the behaviour of the ORCs and their compo-

nents in off-design conditions Indeed making measurements on existing power

units is costly and time-consuming and very few papers published experimental90

data characterizing ORC systems over their complete operating ranges (see one

example in [21]) In almost every case the experimental data (if there is any)

gathered on the facility only covers a narrow range of the feasible operating

conditions and they are not sufficient for a global empirical characterization of

the system Extrapolating the ORC performance in unknown working condi-95

tions can be performed by means of off-design modelling tools As shown in

the aforementioned papers there is a wide variety of modelling paradigms to

estimate the components state in an ORC system ranging from the simplest

method (eg to assume constant efficiencies for characterizing a turbine) to the

most complex one (eg CFD modelling of the same turbine) Each modelling100

method differs from the others in terms of complexity accuracy computational

speed calibration effort and domain of validity Commonly the most accurate

5

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 5: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

in off-design operation by Wang et al [6] The system consisted of a 250 kWe

ORC module (R245fa as working fluid) coupled to a thermal energy storage and

compound parabolic collectors The off-design performance of the whole power

plant was assessed under variations in the ambient temperature and the heat

source mass flow rate Similarly Calise et al [7] studied a 230 kWe recupera-45

tive ORC power unit (n-butane as working fluid) coupled with solar parabolic

trough collectors After optimally sizing the different shell-and-tube heat ex-

changers (ie the recuperator economizer evaporator and superheater) the

authors evaluated the ORC off-design behaviour while varying the thermal heat

source both in terms of mass flow rate and supply temperature In the same50

power scale Fu et al [8] performed a theoretical study on a 250 kWe ORC

using R245fa as working fluid Only the influence of the heat source mass flow

rate on the power plant performance was considered The ORC was controlled

following a sliding pressure strategy the evaporation pressure was controlled

to ensure the working fluid to reach saturated liquid and vapour states at the55

outlet of the preheater and the evaporator respectively Hu et al [9] proposed

a more physical analysis and investigated three control schemes to operate a

70 kWe geothermal ORC unit namely a constant-pressure strategy a sliding-

pressure strategy and optimal-pressure strategy The system featured a radial

inflow turbine plate heat exchangers and used R245fa as working fluid Both60

the refrigerant mass flow rate and variable inlet guide vanes were used to adapt

the power plant behaviour in function of the operating conditions (variation of

the heat source supply temperature and mass flow rate) Manente et al [10]

studied a much larger geothermal power plant (gt 5 MWe) and performed a con-

strained optimization to maximize the system net power output Both R134a65

and Isobutane were considered as working fluid and three variables were used

to control the plant behaviour namely the pump speed the cooling air mass

flow rate in the condenser and the turbine capacity factor Both variations of

the ambiance and heat source supply temperature were considered in the study

Sun and Li [11] also analysed the off-design control of a 5 MWe ORC unit They70

demonstrated that the relationships between controlled variables (optimal work-

4

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ing fluid and air mass flow rates) and external perturbations (heat source and

ambient temperatures) are near linear function for maximizing the system net

power generation and quadratic function for maximizing the system thermal

efficiency Finally Quoilin [12] analysed the off-design performance of a micro-75

scale 15 kWe ORC prototype The system consisted of plate heat exchangers a

scroll expander and employed R123 as working fluid A control of the pump and

the expander speeds was proposed to maximize the ORC thermal efficiency All

the aforementioned studies were performed in steady-state conditions However

the transients affecting the boundary conditions of the ORCs are often faster80

than the response time of the system In such case proper control investigations

and off-design analyses require to account for the dynamic effects induced by

mass and energy accumulations in the various ORC components Such dynamic

performance assessment and control studies can also be found in the scientific

literature see for example [13 14 15 16 17 18 19 20]85

The works presented here above have one feature in common they all used

mathematical models to predict the behaviour of the ORCs and their compo-

nents in off-design conditions Indeed making measurements on existing power

units is costly and time-consuming and very few papers published experimental90

data characterizing ORC systems over their complete operating ranges (see one

example in [21]) In almost every case the experimental data (if there is any)

gathered on the facility only covers a narrow range of the feasible operating

conditions and they are not sufficient for a global empirical characterization of

the system Extrapolating the ORC performance in unknown working condi-95

tions can be performed by means of off-design modelling tools As shown in

the aforementioned papers there is a wide variety of modelling paradigms to

estimate the components state in an ORC system ranging from the simplest

method (eg to assume constant efficiencies for characterizing a turbine) to the

most complex one (eg CFD modelling of the same turbine) Each modelling100

method differs from the others in terms of complexity accuracy computational

speed calibration effort and domain of validity Commonly the most accurate

5

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 6: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ing fluid and air mass flow rates) and external perturbations (heat source and

ambient temperatures) are near linear function for maximizing the system net

power generation and quadratic function for maximizing the system thermal

efficiency Finally Quoilin [12] analysed the off-design performance of a micro-75

scale 15 kWe ORC prototype The system consisted of plate heat exchangers a

scroll expander and employed R123 as working fluid A control of the pump and

the expander speeds was proposed to maximize the ORC thermal efficiency All

the aforementioned studies were performed in steady-state conditions However

the transients affecting the boundary conditions of the ORCs are often faster80

than the response time of the system In such case proper control investigations

and off-design analyses require to account for the dynamic effects induced by

mass and energy accumulations in the various ORC components Such dynamic

performance assessment and control studies can also be found in the scientific

literature see for example [13 14 15 16 17 18 19 20]85

The works presented here above have one feature in common they all used

mathematical models to predict the behaviour of the ORCs and their compo-

nents in off-design conditions Indeed making measurements on existing power

units is costly and time-consuming and very few papers published experimental90

data characterizing ORC systems over their complete operating ranges (see one

example in [21]) In almost every case the experimental data (if there is any)

gathered on the facility only covers a narrow range of the feasible operating

conditions and they are not sufficient for a global empirical characterization of

the system Extrapolating the ORC performance in unknown working condi-95

tions can be performed by means of off-design modelling tools As shown in

the aforementioned papers there is a wide variety of modelling paradigms to

estimate the components state in an ORC system ranging from the simplest

method (eg to assume constant efficiencies for characterizing a turbine) to the

most complex one (eg CFD modelling of the same turbine) Each modelling100

method differs from the others in terms of complexity accuracy computational

speed calibration effort and domain of validity Commonly the most accurate

5

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 7: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

and reliable models implement detailed physics-based equations which leads to

high simulation time However the calculation speed is a key parameter to

maximize in the case of computationally-intensive simulations like control op-105

timization A common way to meet this requirement is to decrease the models

complexity resulting often in a loss of accuracy Therefore there is a trade-off

between modelling complexity and simulation accuracy which deserves being

studied

110

In this paper it is proposed to perform an experimentally-validated analysis

of different modelling methods for the simulation of ORC systems in off-design

conditions More specifically this work aims at comparing three common mod-

elling paradigms (presented in section 3) both in terms of their fitting and

extrapolation abilities Measurements on two experimental ORC test rigs are115

used as reference (for the models calibration and evaluation) and the database

are presented in section 2 The study is first applied to the components level

(ie each component is analysed individually) in section 4 and then extended

to the characterization of the entire ORC systems in section 5 A particular

attention is given to the complete ORC system modelling In most of the works120

presented in the state of the art here above the off-design ORC models rely

on several intrinsic user-defined assumptions like imposed superheating refrig-

erant mass flow rate condensing or evaporating pressure In this work except

for the condenser subcooling which needs to be specified (the ORC model is not

charge sensitive) the ORC model is developed so that the system performance125

is deduced by only taking as inputs the boundary conditions of the system

The modelling tools and source codes developed to perform this work can be

found in the open-source ORCmKit modelling library [22] and thermo-physical

properties of the fluids are computed with CoolProp [23]130

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 8: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

~

ΔP

~

EV EVPRE

RECREC

CD CDSUB

PPPP

EXPEXP

P1 P2 P3 P4 ΔP Pressure sensors

T1 T2 T3 T4Thermocouples

MF1 MF2Mass flow meters

VF1 VF2Volumetric flow meters

W1 W2Wattmeters

P1

P2

P1

P1P1

P2

T2 T1

T2

T2

T2

T2T2

T2

T2

MF1

W1

W1

T4

VF1

VF2

VF2

MF2

T1

T1T1

T3

T3

T3

T3

T3

T3

T3T3

T3T3

T3

P4

P4

P4

P3

P3

P3

W2

W2

Figure 2 Experimental facilities ORC1 (left) and ORC2 (right) - details about the sensors

are provided in Table 1

2 Test rigs and experimental database

In this work two experimental facilities (depicted in Figure 2) are used

as case study for the derivation of different kinds of models The following

section describes the two test rigs and the experimental campaigns performed

to characterize the systems performance135

21 Test rigs description

The first system considered is the Sun2Power ORC module developed by the

University of Liege for a solar thermal application [24 25] It is a 3 kWe recu-

perative organic Rankine cycle using R245fa as working fluid It is constituted

of scroll expander with variable rotational speed and a diaphragm pump Both140

the recuperator and the evaporator are brazed plate heat exchangers (protected

with a 3cm-thick thermal insulation) while an air-cooled fin coil heat exchanger

is used for the condenser Variable-frequency drives are used to control both the

rotational speeds of the pump and the condenser fan On the other hand the

7

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 9: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 1 Sensors properties (FS = full scale)

Sensor type Range Absolute accuracy

T1 (thermocouple type T) [133C 350C] 1C

T2 (thermocouple type T) [minus40C 133C] 1C

T3 (thermocouple type T) [minus40C 133C] 075C

T4 (thermocouple type T) [minus40C 133C] 5C

P1 (absolute pressure) [0bar 10bar] 1 middot FS

P2 (absolute pressure) [0bar 40bar] 1 middot FS

P3 (absolute pressure) [0bar 10bar] 075 middot FS

P4 (absolute pressure) [0bar 40bar] 075 middot FS

∆P (differentiate pressure) [0bar 20bar] 1 middot FS

MF1 (coriolis flow meter) [0kgmin 20kgmin] 015 middot FS

MF2 (coriolis flow meter) [05kgmin 50kgmin] 025 middot FS

VF1 (volumetric flow meter) [03m3h 30m3h] 5 middot FS

VF2 (volumetric flow meter) [01m3h 12m3h] 05 middot FS

W1 (wattmeter) [0W 2000W ] 1 middot FS

W2 (wattmeter) [0W 10000W ] 075 middot FS

expander rotational speed is controlled by means of a variable electrical load145

The second system investigated is the Microsol 10 kWe ORC unit developed by

EXOES and integrated into a concentrated solar power (CSP) plant [26] It is

also a recuperative cycle running R245fa as working fluid and the same pump

technology is used A scroll expander (grid-connected with constant rotational

speed) performs the expansion and two additional heat exchangers are installed150

to ensure the fluid preheating and subcooling (in total the second system in-

cludes five thermally-insulated brazed plate heat exchangers)

In addition to the cycle components both test rigs are fully instrumented for

measuring the experimental performance of each subsystem As illustrated in

Figure 2 thermocouples pressure sensors flow meters and electric power me-155

ters are installed along the plants to ensure a proper characterization of the

systems Technical details regarding these sensors are given in Table1 For the

sake of simplicity the Sun2Power and the Microsol experimental facilities will

be further referred to as ORC1 and ORC2 and Table 2 summarizes their main

8

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 10: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 2 Main features of the two experimental facilities

Properties Facility ORC1 Facility ORC2

Nominal net power output 3 kWe 10 kWe

Working fluid R245fa R245fa

Heat source fluid Thermal oil (Pirobloc HTF-Basic) Pressurized water (sim 10 bar)

Heat sink fluid Ambient air Water-glycol mixture (30 vol)

Expander Scroll expander (variable speed) Scroll expander (constant speed)

Pump Diaphragm pump (variable speed) Diaphragm pump (variable speed)

Condenser Fin coil HEX (fan with variable speed) Brazed plate HEX

Subcooler na Brazed plate HEX

Evaporator Brazed plate HEX Brazed plate HEX

Preheater na Brazed plate HEX

Recuperator Brazed plate HEX Brazed plate HEX

characteristics160

22 Database description

For both test rigs experiments are conducted to characterize the systems

performance under various steady-state operating conditions In these experi-

mental campaigns the ORC systems are not operated in accordance with any165

dedicated control strategy Instead the test rigs are evaluated over extended

ranges of conditions (including non-optimal points) in order to properly char-

acterize their behaviours in off-design and part-load operations Quasi steady-

state performance points are obtained by averaging the measurements over 2-

minute periods in stabilized regimes (ie conditions for which the deviations in170

all the temperatures are lower than 1C with non-sliding pressures and with

constant mass flow rates) Two initial datasets of 57 and 59 experimental points

are collected for the facilities ORC1 and ORC2 respectively Because the mea-

sured numerical values are subject to different uncertainties possible errors or

sensor malfunction a thorough data post-treatment is performed In a first175

step outliers resulting of sensor malfunction or noise in the acquisition chain

are detected and discarded from the database For these points the measure-

9

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 11: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ments of one or several sensors are out of any confidence interval and do not

represent the physics of the machine These outliers are automatically identified

using the open-source GPExp library Based on Gaussian processes theory this180

numerical tool proposes a methodology for quality assessment of steady-state

experimental data as extensively described in [27] Once the outliers are iden-

tified and discarded from the original datasets a second post-process is applied

to the remaining measurements Because the sensors present a limited accu-

racy (in the form of noise or of a systematic error) any measurement gathered185

during the experimental campaign is contaminated by an unknown error Al-

though limited locally the propagation of these measurements errors results in

systems conditions that violate theoretical postulates onto which the models

are developed For instance the heat transfer rate experimentally evaluated on

the cold side of a well-insulated heat exchanger almost never match the heat190

transfer evaluated on the hot side (cfr Figure 3) However by accounting for

the sensors inaccuracy an ideal heat balance can be retrieved as it is assumed

in the heat exchanger models (heat losses in the heat exchangers are neglected

because of the good thermal insulation) As shown with this example most of

the measured variables are interdependent to each other and there are redun-195

dancy constraints which must be verified for every steady-state point Among

others these constraints include to verify both mass and energy balances in each

component to verify the equality between sensors measuring a same quantity

and to ensure feasible temperature profiles in the heat exchangers (ie ensure

a pinch greater than zero) A reconciliation method is thus applied to define200

an experimental database that can be used as reference for the calibration of

predictive models [28] The goal of the reconciliation is to correct the measured

values as little as possible while accounting for the sensors accuracy in order

to satisfy the system constraints Mathematically it can be formulated as the

10

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 12: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Hot side heat transfer [W]0 2000 4000 6000 8000 10000 12000 14000 16000 18000

Col

d si

de h

eat t

rans

fer

[W]

0

2000

4000

6000

8000

10000

12000

14000

16000

18000

Figure 3 Heat balance of an evaporator evaluated on both hot and cold sides - the blue and

red brackets represent the confidence interval when accounting for the sensors accuracy (NB

the wider red intervals of the hot side heat transfer are the result of poorer sensor accuracies)

definition of corrected values ci which minimize a penalty f(ci) function ie205

minci

f(ci) =

Nsumi=1

(mi minus ci)2

σ2i

st energy balance verified in each component

mass balance verified in each component

measurements redundancy respected

pinch in heat exchangers gt 0

(1)

where mi are the original measurements ci are the corrected values and

σi are the sensor absolute accuracies This optimization is performed for ev-

ery steady-state point of both test rigs In order to ensure the viability of the

reconciliation results the difference between the corrected values and the origi-

nal measurements is checked to be within the sensors accuracies Steady-state210

points which do not respect this condition or those whom the optimization

failed to respect the constraints in Equation 1 are also eliminated

11

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 13: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 3 Operating ranges of the experimental measurements

Facility ORC1 Facility ORC2

mwf [gs] [152 684] [277 619]

Pev [bar] [647 143] [98 205]

Pcd [bar] [159 663] [267 425]

Thtfhsu [C] [88 119] [137 169]

Thtfcsu [C] [176 257] [196 346]

Wnet [W ] [16 1255] [875 6000]

εnetORC [] [031 85] [148 491]

As a result of this post-treatment process two experimental datasets of 45

and 44 performance points are obtained for the systems ORC1 and ORC2 re-215

spectively These datasets are used as reference to characterize the performance

of both facilities in off-design conditions Ranges of the experimental data are

summarized in Table 3 and detailed values of the reconciliated measurements

are provided in the appendix (see Appendix A)

3 Modelling methods220

The performance of the power ORC systems and their components varies

with the operating conditions In this work three modelling methods are in-

vestigated to simulate each heat exchanger and mechanical device constituting

the ORC systems namely a constant-efficiency method (CstEff ) a polynomial

regression method (PolEff ) and a semi-empirical method (SemiEmp) This list225

of modelling approach is not exhaustive and many other types of models can be

found in the literature For instance more complex simulation tools like CFD

or advanced deterministic models (ie models which account for all the phys-

ical and chemical phenomena in the processes) exist to simulate the different

components (eg [29 30]) However these models are often computationally230

intensive and can hardly be coupled for performing system-level simulations

Since the ultimate goal of this work is the characterisation of complete ORC

power plants in off-design conditions only common modelling approaches that

12

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 14: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 4 Models inputs outputs and parameters

Component Inputs Outputs CstEff parameters PolEff parameters SemiEmp parameters

Pump Npp Tsu Psu Pex m Wmec Tex εispp εvolpp AUloss aij bij Alk Wloss Kloss AUloss

AUloss with i and j isin 1 2

Expander Nexp Tsu Psu Pex m Wmec Tex εisexp εvolexp AUloss cijk dijk dsu AUsu AUex AUamb

AUloss with i j and k isin 1 2 Closs Alk Wloss

Heat exchanger mh Phsu Thsu Qth εth eij αconvij and nij

mc Pcsu Tcsu with i and j isin 1 2 with i isin liq tp vap and j isin h c

are convenient for system-level simulations are investigated The assumptions

used to perform the modelling are given as below235

bull all the components are in steady-state conditions

bull heat losses in the heat exchangers are neglected (good thermal insulation)

bull pressure drops in the pipelines and the heat exchangers are lumped at a

single place in both the high and low pressure lines

bull heat losses in the pipelines are lumped at a single place in both the high240

and low pressure lines

bull heat exchangers feature counter-flow patterns

bull a global electromechanical efficiency of the pumps and the expanders of

87 is set in all conditions

bull kinetic and gravitational terms are neglected in the energy balance245

The models are implemented so as to predict the performance of existing devices

based on the component supply conditions only Table 4 summarizes the inputs

independent outputs and parameters of each model For the sake of conciseness

the constitutive equations of the models are not provided in the text but are

available in Appendix B The following section describes the different models250

investigated in this work

13

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 15: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

31 Constant-efficiency models

The first type of models considered in this work assumes constant perfor-

mance parameters whatever the operating conditions In the case of the pump

and the expander both the isentropic efficiency εis and the volumetric efficiency255

εvol are imposed as constant values In order to account for the heat losses in

these mechanical components a third parameter AUloss (representing a global

heat transfer coefficient with the ambiance) is added to the models and is kept

constant Regarding the heat exchangers the maximum heat power transferable

between two media is the one leading to a pinch equal to zero In practice the260

effective heat transfer in a heat exchanger is always a fraction (referred to as the

thermal efficiency εth) of this maximum heat power In order to characterize

the different heat exchangers a constant value is assigned to their respective

thermal efficiencies

32 Polynomial-regression models265

This second type of models does not impose constant values to the perfor-

mance parameters (ie εis εvol εth) but uses instead polynomial regressions to

account for the effect of the operating conditions A second-order multivariate

polynomial is applied for every component to keep the methodology systematic

Quadratic functions (ie polynomials of degree two) are chosen to limit Rungersquos

phenomenon and over fitting effects The generic form of the polynomials is

ε =

2sumi=0

2sumj=0

2sumk=0

aijkXiY jZk (2)

where X Y and Z are the most representative independent input variables that

influence the component efficiency These variables are identified for each class

of component (ie the pump the expander and the heat exchangers) as detailed

in the Appendix (Eq B8 to B14)

33 Semi-empirical models270

Another way for characterizing the ORC components is to use semi-empirical

models which implement physics-based equations While the two previous types

14

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 16: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

of model are empirical ie they implement equations that do not represent the

physics of the processes the semi-empirical models presented here below rely on

a limited number of physically meaningful equations whose parameters can be275

tuned to fit a reference dataset For instance the volumetric expanders are sim-

ulated by means of the grey-box model proposed by Lemort et al [31] Besides

of under- and over-expansion losses due to the fixed built-in volumetric ratio of

the machine the model accounts for internal leakages mechanical losses pres-

sure drops and heat losses The pumps are simulated in a similar manner The280

effective mass flow delivered by the pump is calculated as an ideal mass flow

rate to which an internal recirculation leakage is deduced (Eq B15) The mass

flow rate characterizing these leakages is modelled by means of an incompress-

ible flow through an equivalent orifice Finally the mechanical consumption of

the pump is obtained by summing the mechanical losses to the isentropic power285

(Eq B16) Regarding to the heat exchangers a three-zone moving boundary

model with variable heat transfer coefficients is used The modelling is decom-

posed into the different zones of the heat exchanger Each zone is characterized

by a global heat transfer coefficient Ui and a heat transfer surface area Ai The

effective heat transfer occurring in the heat exchanger is calculated such as the290

total surface area occupied by the different zones corresponds to the geometrical

surface area of the component (Eq B20) In the case of a fin coil heat exchanger

(eg the condenser of the test-rig ORC1) the model also accounts for the fin

efficiency by implementing Schmidtrsquos theory [32] Finally a flow-dependent re-

lationship is used to account for the effect of the fluids mass flow rates on the295

convective heat transfer coefficients (Eq B21)

34 Pipeline losses

Besides of the active components constituting the closed-loop cycles (heat

exchangers pumps and expanders) it is also important to account for the losses

induced by the interconnecting pipelines in the systems When modelling the300

complete ORC facilities (cfr section 5) these losses are lumped in each line

(ie high pressure and low pressure) by means of a single artificial component

15

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 17: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

placed at the inlet of the pump and the expander respectively Pressure losses

are simulated as a linear function of the fluid kinetic energy (Eq B22) while

ambient heat losses are modelled with a single AUloss coefficient as in Eq B24305

4 Component-level analysis

The models described here above have varying capabilities to simulate the

performance of a same component In this section a comparison of the fitting

and the extrapolation ability of the different models is applied at the compo-

nent level ie each component is studied independently to the others The310

post-processed experimental measurements described in section 2 are used as

reference for both the calibration (ie as training set) and the evaluation (ie

as test set) of the models

41 Fitting performance

In a first step the fitting performance of the models (ie the ability of the

models to fit an experimental database after calibration) is considered To this

end each component of the two ORC units is simulated by means of the three

different models (constant-efficiency polynomial and semi-empirical) which are

each calibrated using every experimental point of the reference datasets The

calibration is performed by tuning the model parameters so as to minimize

the mean relative errors committed on the different model outputs over the

entire calibration domain The minimization is performed with a derivative-free

direct search optimization algorithm Once calibrated the residuals between

the simulation results (ie the models outputs) and the experimental values are

analysed For example the case of an expander is depicted in Figure 4 The

experimental points used for the calibration and the evaluation of the models are

illustrated on the left side while the model outputs (ie the expander mechanical

power the fluid mass flow rate and the fluid exhaust temperature) are compared

to the reference data by means of parity plots In order to compare numerically

the performance of the three types of model the Root Mean Square Error

16

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 18: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Operating conditions CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 238 - PolEff = 54 - SemiEmp = 49

RMSE (unit= kgs) - CstEff = 15e-2 - PolEff = 12e-3 - SemiEmp = 16e-3

RMSE (unit= K) - CstEff = 19 - PolEff = 13 - SemiEmp = 11

Figure 4 Fitting performance of the expander models (a) Operating domain of the exper-

imental points (b) parity plot of the mechanical power (c) parity plot of the mass flow rate

(d) parity plot of the exhaust temperature

(RMSE) is evaluated for each model output y ie

RMSEy =

radicradicradicradic Nsumi=1

(yi minus yi)2N

(3)

where yi and yi correspond respectively to the reference and the predicted out-

put values of the ith point for a given model Although widely used in the

literature the RMSE is a scale-dependent quantity which can only be used

to compare the performance of different models for the prediction of a single

variable Furthermore the RMSE is not a normalized factor and it does not

illustrate comprehensively the precision of the models individually Therefore

the Mean Absolute Percent Error (as defined in Equation 4) is also proposed as

figure of merit to characterize the different models

MAPEy =1

N

Nsumi=1

|yi minus yi|yi

(4)

This study illustrated in the case of an expander is applied for every component315

of the two test-rigs and the global results (in terms of RMSE) are given in

Figure 5 For the readerrsquos convenience detailed values of the root mean square

errors and the mean absolute percent errors are provided in the Appendix C

Based on the results it can be seen that the models have varying success in320

matching the experimental measurements In most cases the constant-efficiency

17

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 19: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

EV1

RM

SE

Qdo

t

0

20

40

60

80

100

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

SUB2

RM

SE

Qdo

t

0

100

200

300

400

500

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

002

PP2

RM

SE

Wdo

t

0

50

100

150

PP2

RM

SE

mdo

t

10-3

0

2

4

6

EXP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

Wdo

t

0

200

400

600

(b) RMSE of the models outputs for the mechanical devices

Figure 5 Fitting performance for the component-level analysis

models lead to the highest simulation residuals Although straightforward and

easy to use the assumption of invariable components efficiencies should be

avoided for off-design modelling The polynomial and semi-empirical models

fit the datasets better but a clear trend cannot be observed In some cases325

(eg EV 1 and CD1) the polynomial regressions fit the best the dataset while

with other components (eg EV 2 and PRE2) the semi-empirical model show

the lowest residuals On average the absolute percent errors committed while

fitting the heat transfer rate in the heat exchangers are 59 35 and 41

for the constant-efficiency the polynomial-based and the semi-empirical models330

respectively Regarding the mechanical devices these global percent errors are

respectively equal to 76 11 and 21 for the prediction of the mass flow

rate and 219 72 and 71 for the mechanical power

18

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 20: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

42 Extrapolation performance

Additionally to the fitting performance another key property of the models335

to be assessed is their capability to predict the components performance in

unseen operating conditions To this end it is proposed to perform a cross-

validation in which the test set is defined outside of the domain of the training

set The experimental points are therefore divided for each component into

two subgroups of equal size an internal training dataset (used to calibrate340

the models) and an extrapolation testing dataset (used to cross-validate the

models outside of the calibration domain) In order to automatically define

these internal training and external testing datasets the following method is

applied systematically for each component individually (an illustrative example

is given in Figure 6 for the case of a heat exchanger)345

1 The experimental points are reported as a point cloud in a 2D graph

according to two key variables which illustrate the best the operating con-

ditions of the component In the case of the heat exchangers the two

variables are the heat power and the pinch point (see Figure 6a) whereas

the machine rotational speed and the pressure ratio are used for the me-350

chanical devices

2 The operating conditions forming the convex envelope of the point cloud

are identified and defined as part of the external testing dataset (see Fig-

ure 6b) The remaining internal points are kept as potential insiders for355

further division

3 Iteratively this process is repeated to the remaining points until the num-

ber of points included in the external testing dataset is equal to half of

the points in the dataset (see Figure 6 c-d) Ultimately the point cloud360

is divided in two groups of equal size half of the points in the inner-

most area of the point cloud form the calibration dataset (blue triangles

in Figure 6e) while half of the points in the outermost regions are used

19

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 21: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18a Experimental data

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18b Iteration 1

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18c Iteration 2

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18d Iteration 3

Pinch [K]20 30 40

Hea

t fl

ux

[kW

]

6

8

10

12

14

16

18e Final division

Figure 6 Training and testing data set identification in the case of the recuperator of ORC2

a) Point cloud of the experimental data b) First convex envelope calculation c) Second

convex envelope calculation d) Third convex envelope calculation e) Final training and

testing dataset (with 22 points in each group)

as extrapolation testing dataset (red stars in Figure 6e)

Once these domains are identified the models are calibrated with data of365

the training set (using the same methodology as in section 41) and then are

simulated in the testing set The example of an expander is depicted in Fig-

ure 7 where cross and circle markers refer to the training set and the testing set

respectively In order to quantify the extrapolation performance of each model

the RMSE and the MAPE are calculated in reference to the extrapolation test370

set only The same study is applied for every component of both test rigs The

results are given in Figure 8 and detailed values of the RMSE and the mean

absolute percent errors are provided in Appendix C

As in the fitting performance analysis the constant-efficiency models still

demonstrate poor performance Also it can be seen that polynomial mod-375

els do not necessarily lead to the lowest residuals anymore which highlights a

key drawback of these models the shape of the polynomial laws cannot be con-

trolled out of their calibration domain On the other hand semi-empirical mod-

els (which implement physically meaningful equations) are much more robust

in extrapolation On average the percent errors while extrapolating the heat380

power in the heat exchangers are 75 52 and 51 for the constant-efficiency

20

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 22: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

21

Pressure ratio [-]0 2 4 6

Sp

eed

[rp

m]

0

2000

4000

6000

8000Experimental domain

Training set Test set CstEff model PolEff model SemiEmp model

Reference value0 500 1000 1500 2000

Sim

ula

ted

val

ue

0

500

1000

1500

2000Mechanical power [W]

Reference value0 004 008 012

Sim

ula

ted

val

ue

0

004

008

012Mass flow rate [kgs]

Reference value320 340 360 380

Sim

ula

ted

val

ue

320

330

340

350

360

370

380Exhaust temperature [K]

RMSE (unit= W) - CstEff = 248 - PolEff = 147 - SemiEmp = 48

RMSE (unit= kgs) - CstEff = 12e-2 - PolEff = 15e-3 - SemiEmp = 17e-3

RMSE (unit= K) - CstEff = 26 - PolEff = 42 - SemiEmp = 14

Figure 7 Extrapolation performance of the expander models in the case of ORC1 (a) Exper-

imental data divided in inner calibration points and outer evaluation points (b) Parity plot

of the mechanical power (c) Parity plot of the mass flow rate (d) Parity plot of the exhaust

temperature

EV1

RM

SE

Qdo

t

0

50

100

150

CstEff model PolEff model SemiEmp model

REC1

RM

SE

Qdo

t

0

50

100

150

200

250

CD1

RM

SE

Qdo

t

0

100

200

300

400

EV2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

REC2

RM

SE

Qdo

t

0

500

1000

1500

2000

2500

CD2

RM

SE

Qdo

t

0

1000

2000

3000

4000

PRE2

RM

SE

Qdo

t

0

2000

4000

6000

8000

SUB2

RM

SE

Qdo

t

0

200

400

600

800

(a) RMSE of the models outputs for the heat exchangers

PP1

RM

SE

Wdo

t

0

10

20

30

40

CstEff model PolEff model SemiEmp model

PP1

RM

SE

mdo

t

10-3

0

05

1

15

2

EXP1

RM

SE

Wdo

t

0

50

100

150

200

250

EXP1

RM

SE

mdo

t

0

0005

001

0015

PP2

RM

SE

Wdo

t

0

20

40

60

80

100

PP2

RM

SE

mdo

t

10-3

0

2

4

6

8

EXP2

RM

SE

mdo

t

0

0005

001

0015

002

EXP2

RM

SE

Wdo

t

0

200

400

600

800

(b) RMSE of the models outputs for the mechanical devices

Figure 8 Extrapolation performance for the component-level analysis - Global results

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 23: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the polynomials and the semi-empirical models respectively Regarding the me-

chanical devices the global percent errors committed on the mechanical powers

are equal to 291 146 and 84 for each model respectively while smaller

residuals are observed for the mass flow rates with values of 96 26 and385

25

5 Cycle-level analysis

In practice models of individual components are often interconnected to sim-

ulate larger power systems In this section the two ORC units are simulated

by coupling in series the models of each sub-component For each ORC system

(ORC1 and ORC2) three different models are built (ie constant-efficiency

polynomial and semi-empirical) by using the corresponding component models

In order to best replicate the physics of the system these off-design models are

developed in such a way that the complete thermodynamic state of the ORC

can be deduced from the boundary conditions only ie the heat source and the

heat sink supply conditions as well as the pump and the expander speeds The

usefulness of such ORC models is very high they can be used to evaluate the

ORC performance over extended range of conditions and ultimately to derive

the optimal speeds to be set to the different components (pump expander and

condenser fan) in order to maximize the systems power output or net thermal

efficiency Inputs outputs and parameters of the ORC models are illustrated

in Figure 9 The exact mass of refrigerant in the systems being unknown the

ORC models are not made charge sensitive and the subcooling at the condenser

outlet is imposed for the different simulations [33] Apart of the cycle subcool-

ing there is not any user-defined intrinsic assumption of the ORC state (eg

imposed superheating refrigerant mass flow rate condensing or evaporating

pressure etc) Since the off-design modelling of an ORC is an implicit problem

that cannot be formulated causally (because of the multiple interactions be-

tween the different components) the thermodynamic states along the cycle are

found through an iterative optimization process driving internal key residuals

22

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 24: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

to zero More specifically the ORC model iterates on the condensing pressure

the evaporating pressure and the evaporator outlet enthalpy in order to drive

the following residuals to a value lower than 10minus6

res1 = 1minus mppsim

mexpsim(5)

res3 = 1minus hcdexhcdex2

(6)

res3 = 1minus hevexhevex2

(7)

The solver architecture is depicted in Figure 9 and further information about

the ORC model can be found in the ORCmKit documentation [22] As in the

component-level analysis both the fitting and the extrapolation performance of390

the three modelling approaches are evaluated while simulating the entire power

systems

51 Fitting performance

The ability of the three ORC models to fit the experimental datasets is first

investigated To this end the models of the different components calibrated395

with the complete database (ie the ones presented in section 41) are coupled

together to form the three ORC models These ORC models are then evalu-

ated in the same operating conditions than the experimental points while only

accounting for the external boundary variables The system performance pre-

dicted by each modelling approach are finally compared with the experimental400

data For example experimental and predicted T-s diagrams are shown in Fig-

ure 10 for two different operating conditions In the first case (left) it can be

seen that the three models replicate well the experimental conditions in terms

of temperature and pressure The second example (right) on the other hand

demonstrates larger discrepancies between the simulation results despite of the405

identical operating conditions

In order to numerically quantify the performance of the different models

RMSEs and MAPEs are calculated for the various energy flows involved in the

23

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 25: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

24

mhtfc Thtfcsu

Phtfcsu

mhtfh Thtfhsu

PhtfhsuPexpsu

Pexpex

Pcdsu

hcdsu

mhtfc

mpp

Phtfcsu

Thtfcsu

hevsu

Pevsu

mpp

Preccsu

Qcd

Prechsu

hrechsu

mpp

Pexpex

hcdex

ΔTsc

Pppsu

Pppex

Npp

PPmodel

Nexp

Npp

ΔTsc

MODELS PARAMETERS(EV ndash EXP ndash PP ndash CD ndash REC ndash Piping)

mwf

Pwfi Twfi hwfI

along the cycle

Wi Qi

along the cycle

OUTPUTSINPUTS ORC model

EVmodel

HPLossesmodel

EXPLossesmodel

LPLossesmodel

RECLossesmodel

Pppex

hevex

mpp

hexpsu

Npp

Pppsu

ΔTsc

mpp mexp

Wexp

hcdex2

Pcdex

CDmodel

1

23 4

5

6

7

hreccsu

Wpp

mhtfh

Phtfhsu

Thtfhsu

mpp

mpp

Pevex

hevex2

Qev

Figure 9 Inputs (in blue) outputs (in brown) parameters (in green) iteration variables (in

red) and solver architecture of the cycle model (case of the facility ORC1) The circled number

in each component model informs the execution order of the model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130Experimental data CstEff model PolEff model SemiEmp model

Entropy [JkgK]1000 1100 1200 1300 1400 1500 1600 1700 1800 1900

Tem

per

atu

re [

degC]

20

30

40

50

60

70

80

90

100

110

120

130

Figure 10 T-s diagrams predicted by the three ORC models for the system ORC1 (left)

experimental point 32 (right) experimental point 36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 26: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

20

40

60

80

100

120

140

160

180

200

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

200

400

600

800

1000

1200

1400

1600

CstEff model PolEff model SemiEmp model

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

(b) RMSE of the models outputs - ORC 2

Figure 11 Fitting performance for the cycle-level analysis - Global results

two systems and detailed values of these performance indicators are provided410

in Appendix C In comparison to the results presented in section 41 it is ob-

served that the residuals when modelling the complete ORC power systems are

larger than in the component-level analysis (the RMSEs are 23 times higher on

average) Such increase is due to the propagation and addition of the sub-model

errors along the ORC Unlike the component-level analysis which compared each415

component individually with identical supply conditions here the models inputs

and outputs are interdependent

When considering a complete ORC system two common variables used to

evaluate the global machine performance are the net mechanical power Wnet

generated by the engine and the net cycle efficiency εORC ie

εORC =Wnet

Qin

(8)

where Qin is the total heat power supplied to the system RMSEs committed

by the three ORC models to replicate these performance outputs are depicted420

in Figure 11 In the case of the first ORC system (ORC1) similar conclusions

to the component-level analysis can be drawn The constant-efficiency ORC

model leads to the highest simulation residuals while the polynomial-based and

25

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 27: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Experimental thermal efficiency [ ]0 2 4 6 8 10

Sim

ulat

ed th

erm

al e

ffici

ency

[ ]

0

2

4

6

8

10 CstEff (MAPE = 466 )PolEff (MAPE = 172 )SemEmp (MAPE = 896 )

(a) Thermal efficiency of system ORC 1

Experimental thermal efficiency []-1 0 1 2 3 4 5

Sim

ulat

ed th

erm

al e

ffici

ency

[]

-1

0

1

2

3

4

5 CstEff (MAPE = 174)PolEff (MAPE = 134)SemEmp (MAPE = 757)

(b) Thermal efficiency of system ORC 2

Figure 12 Parity plots of the thermal efficiency predicted by the three types of model for

both ORC units

the semi-empirical models offer better simulation performance On the other

hand results related to the second system (ORC2) are different and highlight a425

major drawback of the polynomial-based ORC model In some cases the cycle

state into which the residuals (as given in Eqs 5 - 7) are driven to zero may be

out of the calibration domain of some the subcomponents model However as

it as been mentioned previously polynomial regressions do not ensure any reli-

able results in extrapolation Therefore the polynomial-based ORC model may430

commit significant deviations compared to the reference data even though it is

re-evaluated in the same operating conditions used for to calibrate the subcom-

ponents models The robustness of an ORC model built by the interconnection

of multiple polynomial regressions cannot be ensured in all cases Regarding

the semi-empirical ORC model much better robustness is observed and good435

fitting performance are demonstrated with both ORC facilities

Finally the net efficiency predicted by the three modelling approaches for the

two test rigs are compared to the experimental data in Figure 12 Although the

ORC system models are re-evaluated on the calibration conditions (ie the refer-440

ence conditions used to calibrate each subcomponent models) significant resid-

uals can be observed More specifically the average percent error committed on

26

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 28: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

the net thermal efficiency by the constant-efficiency the polynomial-based and

the semi-empirical ORC models are 32 153 and 83 respectively Such

high values result from the accumulation of errors which affect the different vari-445

ables involved in the calculation of the net efficiency In conclusion even though

the models of the different components are well calibrated independently the

net thermal efficiency predicted by the cycle model can present significant de-

viations the highest average error being stated for the constant-efficiency ORC

model450

52 Extrapolation performance

Finally the capability of the three ORC models to extrapolate the whole

system performance in unseen operating conditions is analysed The cross-

validation methodology used to perform this study is identical to the component-

level analysis discussed in section 42 For each ORC facility the experimental455

data are divided in two subgroups of equal size an internal training dataset

(used to calibrate the different component models) and an extrapolation test-

ing dataset (used to cross-validate the ORC models outside of the calibration

domain) However it must be noted that the models calibrated in the extrap-

olation analysis at the component-level cannot be coupled together to perform460

the same analysis at the cycle-level Indeed the training sets defined for each

component (as presented in section 42) and used to calibrate the various mod-

els are not identical For instance while considering the system ORC1 the

experimental point 4 is defined in the training set of the pump but it is con-

sidered as external from the evaporator point of view In order to make the465

study consistent a common training set must be defined for all the components

of a same ORC engine To this end the experimental points are first reported

in a 2D graph accordingly to the net power output and the cycle thermal effi-

ciency then the method based on an iterative evaluation of the convex envelope

is applied until groups of equal size are obtained (cfr section 42 for further ex-470

planations) As an example the final data division performed for the first ORC

facility (ORC1) is depicted in Figure 13 The different component models are

27

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 29: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Net ORC power output (W)-200 0 200 400 600 800 1000 1200 1400

Net

OR

C th

erm

al e

ffici

ency

(-)

0

002

004

006

008

01

Training data setExtrapolation testing data set

Figure 13 Data division of the system ORC1 for the extrapolation analysis

then calibrated using data of the internal training set and the component models

are coupled together to form the complete ORC power unit The three types of

ORC models are finally simulated in the operating conditions of the test set (ie475

in extrapolation) only Similarly as before RMSEs and MAPEs committed on

the different energy flows in the two systems are provided in Appendix C Like

in the fitting performance analysis (see section 51) the residuals committed

on the net power output and the net cycle efficiency are investigated and the

related RMSEs committed by each modelling method are depicted in Figure 14480

It can be seen that for both facilities the constant-efficiency ORC model leads

to the highest residuals while the semi-empirical ORC model demonstrates the

best extrapolation capability The polynomial-based ORC model presents inter-

mediate performance but as it has been highlighted previously viable results

cannot be ensured out of the calibration range (although convergence issues485

are not observed with the current simulations) Quantitatively speaking the

average percent errors committed on the net thermal efficiency by the constant-

efficiency the polynomial-based and the semi-empirical ORC models are 512

19 and 142 respectively

28

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 30: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

_Wnet

RM

SE

Wdo

t (W

)

0

50

100

150

200

250

0ORCnet

RM

SE

eps

ilon

(-)

0

0002

0004

0006

0008

001

0012

0014

0016

0018

CstEff model PolEff model SemiEmp model

(a) RMSE of the models outputs - ORC 1

_Wnet

RM

SE

Qdo

t (W

)

0

100

200

300

400

500

600

700

0ORCnet

RM

SE

eps

ilon

(-)

10-3

0

1

2

3

4

5

6

7

8

CstEff model PolEff model SemiEmp model

(b) RMSE of the models outputs - ORC 2

Figure 14 Extrapolation performance for the cycle-level analysis - Global performance results

6 Computational efficiency490

This last section is dedicated to the computational performance of the differ-

ent modelling methods The model computational time can indeed be a crucial

parameter if the model is used eg for Monte Carlo simulations in control

optimization problems or integrated into a larger system model As a figure

of comparison average computational times of the different models presented495

through the text are summarized in Table 5 These values should be considered

as a qualitative indicator only since they depend on the equations implementa-

tion and the computer performance It can be seen that the higher the model

complexity the higher the simulation time Constant-efficiency and polynomial-

based models show very similar computational efforts because of the fast calcu-500

lation of the polynomial regressions On the other hand semi-empirical model

(which often require implicit iterations and additional call to the working fluid

thermodynamic properties) are characterized by longer running times (4 times

higher for the heat exchanger model and more than 100 times higher for the ex-

pander model) Regarding the ORC system models similar trends are observed505

at a greater magnitude Besides of implicitly solving the components models

the ORC models also require internal iterations in order to derive the system

steady-state performance based on the boundary conditions only

29

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 31: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Table 5 Mean simulation times of the different modelling methods to evaluate one operating

point (simulations performed with a laptop Dell Latitude E5450 CPU Intel Core i7-5600U

26GHz 8GB RAM)

Pump Expander Heat Exchanger ORC

CstEff model 95times 10minus4 sec 91times 10minus4 sec 99times 10minus3 sec 11times 101 sec

PolEff model 11times 10minus3 sec 11times 10minus3 sec 12times 10minus2 sec 14times 101 sec

SemiEmp model 12times 10minus3 sec 11times 10minus1 sec 41times 10minus2 sec 35times 101 sec

7 Conclusions

Among the many topics of research and development in the energy sector510

power generation from low-grade heat sources is gaining interest and the organic

Rankine cycle (ORC) is seen as one of the most suitable technology for such

applications Aside of proper fluid selection and system design the off-design

characterization and control of ORC power systems is important due to the ver-

satile nature of their operating conditions Because of the incompleteness of the515

experimental data mathematical modelling tools are often required to predict

the system performance as a function of the boundary working conditions To

this end a wide range of modelling paradigms can be chosen to simulate the

power plants and their sub-components In this work it is proposed to anal-

yse and compare three modelling methods to simulate in off-design conditions520

ORC-based power plants and their constitutive components (heat exchangers

pumps and expanders) namely

bull a constant-efficiency method which assumes constant components efficien-

cies whatever the operating conditions

bull a polynomial regressions method which adapt the components efficiencies525

to the operating conditions by means of quadratic functions (second-order

multivariate polynomials)

bull a semi-empirical method which simulate the components by means of a

limited number of physically-meaningful equations

30

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 32: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

These models are compared in terms of fitting and extrapolation performance530

To this end experimental measurements gathered on two ORC facilities are

post-processed and used as reference for the models calibration and evalua-

tion Both root mean square errors (RMSEs) and mean absolute percent errors

(MAPEs) are calculated for the sake of model comparison The analysis is first

performed at a component level (ie each pump heat exchanger and expander535

is studied individually) and then extended to the entire ORC power units Nu-

merical results drawn from the study can be summarized as follows

1 In the component-level analysis the absolute percent errors committed

while fitting the heat transfer in the heat exchangers are on average

59 35 and 41 for the constant-efficiency the polynomial and semi-540

empirical models respectively Regarding the mechanical devices these

global percent errors are respectively equal to 76 11 and 21 for

the prediction of the mass flow rate and 219 72 and 76 for the

mechanical power

545

2 In the component-level analysis again it is demonstrated that the mod-

elling residuals are increased when using the models outside of the calibra-

tion domain (ie in extrapolation) More specifically the percent errors

while extrapolating the thermal power in the heat exchangers are on aver-

age 75 52 and 51 for the constant-efficiency the polynomials and550

the semi-empirical models respectively Regarding the mechanical devices

the average percent errors committed on the mechanical powers are equal

to 291 146 and 84 for each model respectively while smaller resid-

uals are observed for the mass flow rates with values of 96 26 and

25555

3 Because of the propagation of the models uncertainties RMSEs of the

residuals are on average 23 higher when modelling the complete systems

in comparison to the results of the component-level analysis

31

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 33: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

560

4 When modelling the entire ORC power systems in the reference boundary

conditions the average percent error committed on the net thermal effi-

ciency is equal to 32 153 and 83 for the constant-performance the

polynomial and the semi-empirical ORC models respectively Such high

values result from the accumulation of errors which affect the different565

variables involved in the calculation of the net efficiency

5 Like in the component-level analysis it is seen that the simulation resid-

uals are increased while using the ORC models in extrapolation The

average percent errors committed on the net thermal efficiency rise to570

512 19 and 142 for the constant-performance the polynomial and

the semi-empirical ORC models respectively

Although they are fast to implement to calibrate and to compute it can be

seen that constant-efficiency models demonstrate poor performance for both

component- and cycle-level simulations In most cases they lead to the highest575

residuals and should only be considered for off-design simulation if the operating

conditions remain close to the nominal operating point Polynomial-based mod-

els are also fast to calibrate and to evaluate They reveal very good fitting per-

formance while considering the components individually However polynomial-

based models can be unreliable in extrapolation and when coupled together580

They should only be used for characterizing the components individually and

within their calibration ranges for interpolation modelling Semi-empirical mod-

els on the other hand show good and robust performance in both fitting and

extrapolation at both component- and cycle-level analysis

585

Based on the current study semi-empirical models demonstrate to be the

most suitable for the off-design simulation of ORC systems despite of the higher

calibration and simulation times The proper simulation for a particular appli-

cation results from the classic trade-off between accuracy and complexity The

32

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 34: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

selected model should be accurate enough for the purpose of the simulation but590

its limitations should be known from the modeller It is important to note that

the modelling approaches investigated in this work are not exhaustive Other

forms of correlations can be used to characterize the components efficiencies

(eg first- or third-order multivariate polynomials more complex regressions of

the expander efficiency [34] etc) and models of different class could be coupled595

together to simulate the closed-loop systems

Finally it must be noted that the system-level simulations are performed

by imposing the cycles sub-cooling in the ORC model Since the goal of this

work is only to compare different modelling paradigms such a simplification600

is considered acceptable as it does not biased the analysis However in order

to perform valuable off-design simulations the ORC model should be improved

to be charge sensitive ie it imposes the total mass of refrigerant enclosed in

the ORC systems instead of the condenser sub-cooling This particular point

highlights another limitation of the simplest modelling approaches presented in605

this paper neither the constant-efficiency nor the polynomial-efficiency models

permit to properly estimate the amount of refrigerant enclosed in the various

heat exchangers Since these models do not rely on any heat transfer coefficient

they do not calculate the volume fraction occupied by each fluid phase in the

heat exchangers therefore the refrigerant mass enclosed in the heat exchanger610

cannot be properly computed Only the semi-empirical model of the heat ex-

changers (the one based on convective heat transfer coefficients characterizing

both fluids) may be used to perform a reliable charge sensitive modelling of the

ORC systems Prospective works include the development and the experimental

validation of such a charge-sensitive ORC model615

Acknowledgements

R Dickes thanks the Fund for Scientific Research of Belgium (FRS - FNRS)

for its financial support (research fellowship) The authors also would like to

33

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 35: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

gratefully acknowledge the contributors to the Sun2Power project (CMI group

Enertime Emerson ACTE and Honeywell) as well as Exoes for sharing the620

experimental measurements gathered on the Microsol CSP-ORC power plant

34

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 36: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Nomenclature

Acronyms and abbreviations

CD Condenser

CFD Computational Fluid Dynamics625

CSP Concentrated Solar Power

CstEff Constant-Efficiency

EV Evaporator

EXP Expander

FS Full Scale630

HEX Heat Exchanger

HP High Pressure

HTF Heat Transfer Fluid

LP Low Pressure

MAPE Mean Absolute Percent Error635

ORC Organic Rankine Cycle

PolEff Polynomial-Efficiency

PP Pump

PRE Preheater

REC Recuperator640

RMSE Root Mean Square Error

SemiEmp Semi Emperical

SUB Root Mean Square Error

35

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 37: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

VFD Variable Frequency Drive

WF Working Fluid645

Subcripts and supercripts

amb ambient

c cold

cd condenser

conv convective650

dis displacement

ev evaporator

ex exhaust

exp expander

h hot655

htf heat transfer fluid

ijk index

in incoming

is isentropic

liq liquid660

lk leakage

log logarithmic

loss losses

max maximum

mec mechanical665

36

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 38: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

net net

nom nominal

pp pump

rec recuperator

sc subcooling670

sim simulated

su supply

th thermal

tp two-phase

vap vapour675

vol volumetric

wf working fluid

Variables

α Heat transfer coefficients Wm2K

∆ Differential minus680

m Mass flow kgs

Q Heat Power W

V Volume flow rate m3s

W Power W

ρ Density kgm3685

σ Sensor accuracy

ε Efficiency

37

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 39: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

ϕ Fluid kinetic energy kgm3s2

y reference output minus

A Surface area m2690

C Torque Nm

c Corrected measurements

d Diameter m

h Enthalpy Jkg

KB Model parameters minus695

m Raw measurements minus

N Rotational speed kgs

P Pressure Pa

rp Pressure ratio minus

s Entropy JK700

T Temperature K

U Heat transfer coefficient Wm2K

V Volume m3

XY Z Symbolic variables minus

y model output minus705

abcde Polynomial coefficients minus

n Exponent coefficient minus

38

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 40: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

References

[1] B F Tchanche G Lambrinos A Frangoudakis G Papadakis Low-grade

heat conversion into power using organic Rankine cycles - A review of710

various applications Renewable and Sustainable Energy Reviews 15 (8)

(2011) 3963ndash3979 doi101016jrser201107024

URL httpwwwsciencedirectcomsciencearticlepii

S1364032111002644

[2] F W Ofeldt rsquoEnginersquo - US Patent No 611792A (1898)715

URL httpwwwgooglechpatentsUS611792

[3] L Y Bronicki Short review of the long history of ORC power systems in

Keynote lecture of the 2nd International seminar on ORC power systems -

ASME-ORC 2013 Rotterdam (NL) 2013

[4] P Colonna E Casati C Trapp T Mathijssen J Larjola T Turunen-720

Saaresti A Uusitalo Organic Rankine Cycle Power Systems from the

Concept to Current Technology Applications and an Outlook to the

Future Journal of Engineering for Gas Turbines and Power 137 (October)

(2015) 1ndash19 doi10111514029884

URL httpgasturbinespowerasmedigitalcollectionasmeorg725

articleaspxdoi=10111514029884

[5] H Gurgenci Performance of power plants with organic Rankine cycles

under part-load and off-design conditions Solar amp Wind Technology 36 (1)

(1986) 45ndash51 doi1010160038-092X(86)90059-9

URL httpwwwsciencedirectcomsciencearticlepii730

0038092X86900599

[6] J Wang Z Yan P Zhao Y Dai Off-design performance analysis of a

solar-powered organic Rankine cycle Energy Conversion and Management

80 (2014) 150ndash157 doi101016jenconman201401032

URL httpdxdoiorg101016jenconman201401032735

39

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 41: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[7] F Calise C Capuozzo A Carotenuto L Vanoli Thermoeconomic anal-

ysis and off-design performance of an organic Rankine cycle powered

by medium-temperature heat sources Solar Energy 103 (2013) 595ndash609

doi101016jsolener201309031

URL httpdxdoiorg101016jsolener201309031740

[8] B R Fu S W Hsu Y R Lee J C Hsieh C M Chang C H Liu Perfor-

mance of a 250 kW organic rankine cycle system for off-design heat source

conditions Energies 7 (6) (2014) 3684ndash3694 doi103390en7063684

[9] D Hu Y Zheng Y Wu S Li Y Dai Off-design performance comparison

of an organic Rankine cycle under different control strategies Applied745

Energy 156 (2015) 268ndash279 doi101016japenergy201507029

URL httpwwwsciencedirectcomsciencearticlepii

S0306261915008582

[10] G Manente A Toffolo A Lazzaretto M Paci An Organic Rankine Cycle

off-design model for the search of the optimal control strategy Energy 58750

(2013) 97ndash106 doi101016jenergy201212035

URL httpdxdoiorg101016jenergy201212035

[11] J Sun W Li Operation optimization of an organic rankine cycle (ORC)

heat recovery power plant Applied Thermal Engineering 31 (11-12) (2011)

2032ndash2041 doi101016japplthermaleng201103012755

URL httpwwwsciencedirectcomsciencearticlepii

S135943111100144X

[12] S Quoilin Sustainable Energy Conversion Through the Use of Organic

Rankine Cycles for Waste Heat Recovery and Solar Applications PhD

thesis University of Liege (2011)760

[13] D Wei X Lu Z Lu J Gu Dynamic modeling and simulation of

an Organic Rankine Cycle (ORC) system for waste heat recovery Ap-

plied Thermal Engineering 28 (10) (2008) 1216ndash1224 doi101016j

applthermaleng200707019

40

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 42: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[14] S Quoilin R Aumann A Grill A Schuster V Lemort H Spliethoff765

Dynamic modeling and optimal control strategy of waste heat recovery

Organic Rankine Cycles Applied Energy 88 (6) (2011) 2183ndash2190 doi

101016japenergy201101015

URL httpdxdoiorg101016japenergy201101015

[15] J Zhang W Zhang G Hou F Fang Dynamic modeling and multivari-770

able control of organic Rankine cycles in waste heat utilizing processes

Computers and Mathematics with Applications 64 (5) (2012) 908ndash921

doi101016jcamwa201201054

URL httpdxdoiorg101016jcamwa201201054

[16] H Xie C Yang Dynamic behavior of Rankine cycle system for waste heat775

recovery of heavy duty diesel engines under driving cycle Applied Energy

112 (2013) 130ndash141 doi101016japenergy201305071

URL httpdxdoiorg101016japenergy201305071

[17] M O Bamgbopa E Uzgoren Numerical analysis of an organic Rankine

cycle under steady and variable heat input Applied Energy 107 (2013)780

219ndash228 doi101016japenergy201302040

URL httpdxdoiorg101016japenergy201302040

[18] N Mazzi S Rech A Lazzaretto Off-design dynamic model of a real

Organic Rankine Cycle system fuelled by exhaust gases from industrial pro-

cesses Energy 90 (2015) 537ndash551 doi101016jenergy201507083785

URL httpwwwsciencedirectcomsciencearticlepii

S0360544215009780

[19] A Hernandez A Desideri C Ionescu S Quoilin V Lemort R De Keyser

Towards the optimal operation of an organic Rankine cycle unit by means

of model predictive control Proceedings of the 3rd International Seminar790

on ORC Power Systems (2015) 1ndash10

[20] P Tona J Peralez Control of Organic Rankine Cycle Systems on board

Heavy-Duty Vehicles a Survey IFAC-PapersOnLine 48 (15) (2015)

41

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 43: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

419ndash426 doi101016jifacol201510060

URL httplinkinghubelseviercomretrievepii795

S2405896315019369

[21] T Erhart U Eicker D Infield Part-load characteristics of Organic-

Rankine-Cycles in 2nd European Conference on Polygeneration Tarrag-

ona (Spain) 2011 pp 1ndash11

[22] R Dickes D Ziviani M de Paepe M van den Broek S Quoilin800

V Lemort ORCmKit an open-source library for organic Rankine cycle

modelling and analysis in Proceedings of ECOS 2016 Portoroz (Solvenia)

2016

[23] I H Bell J Wronski S Quoilin V Lemort Pure and pseudo-pure fluid

thermophysical property evaluation and the open-source thermophysical805

property library coolprop Industrial and Engineering Chemistry Research

53 (6) (2014) 2498ndash2508

[24] E Georges S Declaye O Dumont S Quoilin V Lemort Design of

a small-scale organic Rankine cycle engine used in a solar power plant

International Journal of Low-Carbon Technologies 8 (2013) 34ndash41 doi810

101093ijlctctt030

[25] R Dickes O Dumont S Declaye S Quoilin I Bell V Lemort Experi-

mental investigation of an ORC system for a micro-solar power plant in

Proceedings of the 22nd International Compressor Engineering at Purdue

Purdue (USA) 2014815

URL httphdlhandlenet2268170508

[26] V Rieu Microsol - A 10 kW solar power plant for rural electrification in

Presentation at the SolarPACES 2012 conference Marrakech 2012

[27] S Quoilin J Schrouff Assessing steady-state multivariate thermo-fluid

experimental data using Gaussian Processes the GPExp open-source li-820

brary Energies 9 (6) (2016) ndash doi103390en9060423

42

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 44: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

[28] O Dumont S Quoilin V Lemort Importance of the reconciliation

method to handle experimental data application to a reversible heat

pump organic Rankine cycle unit integrated in a positive energy build-

ing International Journal of Energy and Environmental Engineeringdoi825

101007s40095-016-0206-4

URL httpdxdoiorg101007s40095-016-0206-4

[29] M M Aslam Bhutta N Hayat M H Bashir A R Khan K N Ah-

mad S Khan CFD applications in various heat exchangers design A

review Applied Thermal Engineering 32 (1) (2012) 1ndash12 doi101016830

japplthermaleng201109001

URL httpdxdoiorg101016japplthermaleng201109001

[30] R Dickes Design and fabrication of a variable wall thickness two-stage

scroll expander to be integrated in a micro-solar power plant Master thesis

University of Liege (2013)835

URL httphdlhandlenet2268160458

[31] V Lemort S Quoilin C Cuevas J Lebrun Testing and modeling a scroll

expander integrated into an Organic Rankine Cycle Applied Thermal En-

gineering 29 (14-15) (2009) 3094ndash3102 doi101016japplthermaleng

200904013840

URL httpdxdoiorg101016japplthermaleng200904013

[32] T Schmidt Heat Transfer Calculations for Extended Surfaces ASHRAE

Journal (1949) 351ndash357

[33] D Ziviani B Woodland E Georges E Groll J Braun W Horton

M van den Broek M De Paepe Development and a Validation of a Charge845

Sensitive Organic Rankine Cycle (ORC) Simulation Tool Energies 9 (6)

(2016) 389 doi103390en9060389

URL httpwwwmdpicom1996-107396389

[34] S Declaye S Quoilin L Guillaume V Lemort Experimental study on

an open-drive scroll expander integrated into an ORC (Organic Rankine850

43

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 45: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Cycle) system with R245fa as working fluid Energy 55 (2013) 173ndash183

doi101016jenergy201304003

URL httpdxdoiorg101016jenergy201304003

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 46: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix A Experimental measurements

In this appendix the reference database obtained experimentally on the test855

rigs (see section 2) are provided The reconciliated experimental measurements

are summarized in Tables A6 and A7 for the first and the second ORC system

respectively

Appendix B Models constitutive equations

This appendix provides the constitutive equations of the models presented in860

section 3 Please refer to the nomenclature (see p 38) for any details regarding

the variables names

Appendix B1 Constant-efficiency models

minus Pump model

εispp =mpp(hexispp minus hsupp)

Wmecpp

= εispp (B1)

εvolpp =Vsupp

NppVdispp= εvolpp (B2)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B3)

minus Expander model

εisexp =Wmecexp

mexp(hsuexp minus hexisexp)= εisexp (B4)

εvolexp =Vsuexp

NexpVdisexp= εvolexp (B5)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B6)

minus Heat exchanger model

εth =Q

Qmax

= εth (B7)

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 47: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

6

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

1

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Tppsu

Tppex

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Nexp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[rpm]

[minus]

349

409

614

516

286

783

917

6179

192

382

966

910

597

329

973

934

176

226

70

652

240

3680

1

278

209

314

515

9102

499

816

7181

196

372

1054

965

640

395

1063

1029

181

222

63

475

200

1890

2278

209

405

120

5105

9103

720

6217

225

415

1048

965

660

394

1057

1025

188

300

67

420

200

1700

3

268

609

202

231

8107

4105

731

8327

310

502

1046

963

692

474

1054

1024

202

433

65

359

200

1620

4

263

009

201

151

2111

1109

951

2480

433

602

1055

963

755

636

1064

1036

213

619

56

255

200

1554

5

274

609

403

724

0107

2106

024

0248

251

441

1040

960

660

396

1048

1017

195

343

70

399

200

1640

6

599

509

414

417

6111

5106

623

0191

204

469

1025

991

670

374

1064

998

183

266

130

1041

400

7680

7

598

709

807

822

2110

9106

626

5229

238

498

997

965

662

417

1064

1002

188

338

128

939

400

7300

8

504

309

504

828

1103

299

630

2269

272

546

1041

999

737

458

1050

997

194

396

104

725

340

6320

9

448

009

501

759

8113

1110

260

2508

479

718

1042

975

825

695

1062

1021

228

668

82

263

320

3780

10

658

409

414

320

1137

9131

826

1223

236

508

1140

1114

733

413

1155

1083

209

302

150

1405

440

5660

11

659

309

414

220

3134

4128

426

3228

240

521

1151

1125

755

415

1161

1088

213

306

147

1399

440

6280

12

659

409

414

220

5129

5124

126

5229

242

536

1165

1138

781

418

1172

1099

214

308

146

1389

440

7100

13

657

609

508

625

4130

4124

930

3261

271

572

1153

1129

791

459

1160

1091

225

375

142

1283

440

6822

14

679

909

405

534

3137

0131

038

3324

329

613

1117

1092

774

535

1157

1088

234

465

148

1153

460

6334

15

669

909

503

747

3136

9131

250

1403

400

706

1143

1108

869

628

1164

1100

246

567

137

751

460

6340

16

654

209

402

666

3139

8135

368

2516

502

793

1126

1082

924

743

1168

1110

256

684

125

291

460

5640

17

673

409

504

343

3129

9124

246

2368

368

714

1179

1138

939

600

1187

1122

242

536

135

554

460

7960

18

676

809

605

236

5130

4124

540

6334

338

671

1174

1140

891

555

1182

1115

238

487

141

849

460

7820

19

683

709

607

129

2130

3124

134

1293

300

617

1160

1133

837

497

1170

1101

234

420

147

1107

460

7920

20

557

209

506

926

9133

0128

330

0279

288

566

1164

1131

781

455

1173

1114

233

391

129

1112

380

4380

21

556

809

505

232

3129

0124

634

9311

316

602

1157

1125

803

504

1166

1108

236

442

125

1039

380

4800

22

551

109

503

840

1131

9127

742

1368

365

651

1159

1120

829

567

1168

1113

239

508

121

883

380

4420

23

534

109

702

654

9143

1141

056

0473

460

721

1158

1116

857

669

1183

1133

255

617

113

660

380

3100

24

447

809

302

942

4134

4131

443

4403

397

644

1174

1123

824

578

1183

1137

248

530

102

730

320

2960

25

453

809

404

531

8134

9132

033

3330

332

581

1169

1126

785

489

1179

1131

238

434

109

852

320

2980

26

459

209

405

926

9133

0129

828

9294

300

546

1161

1120

762

443

1171

1121

234

387

111

917

320

3090

27

461

609

514

319

0134

8130

122

1236

249

474

1144

1098

718

403

1156

1105

222

288

108

944

320

2960

28

368

809

207

521

075

171

222

5229

238

452

946

884

672

374

953

911

211

310

70

446

250

6900

29

382

709

504

129

179

276

229

9291

289

515

928

870

690

454

934

894

220

402

70

352

260

6140

30

375

909

402

638

784

881

939

2378

363

582

935

871

724

543

945

907

227

494

68

244

260

5060

31

340

409

201

847

886

784

547

8447

422

623

938

861

744

611

953

919

241

568

59

150

240

3940

32

270

409

401

156

685

884

156

6521

479

634

939

844

761

673

948

921

244

656

50

66

200

2620

33

186

909

301

044

565

865

144

5451

413

505

888

774

692

587

897

877

241

577

39

63

140

2240

34

183

009

301

042

679

077

942

6444

408

501

901

795

628

572

910

891

243

566

40

141

140

1400

35

152

209

301

531

068

968

531

0384

369

432

899

793

593

506

909

892

234

429

43

176

120

1380

36

152

809

302

225

269

068

225

2315

308

380

904

796

566

466

914

896

227

363

46

204

120

1360

37

153

109

503

820

567

266

320

5259

263

341

905

794

548

440

914

896

219

301

45

216

120

1440

38

152

409

311

516

564

763

516

6235

245

314

901

792

524

427

912

893

212

239

44

246

120

1500

39

320

009

514

417

771

468

118

5213

225

387

912

854

586

357

919

883

210

255

57

540

220

5300

40

321

009

614

417

668

164

618

5214

225

392

903

845

603

367

910

874

209

255

57

481

220

6060

41

319

209

614

417

375

373

318

3214

226

374

879

826

537

327

897

861

209

254

60

554

220

4020

42

320

609

506

820

671

769

221

2233

242

410

877

823

575

350

885

849

216

311

61

479

220

4960

43

318

009

605

922

076

673

922

1249

255

407

828

774

517

362

880

845

217

321

62

452

220

3540

44

252

309

503

226

771

970

126

7302

295

441

883

812

589

419

891

863

225

376

50

331

180

2940

45

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 48: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPTT

ab

leA

7

Rec

on

ciliate

dm

easu

rem

ents

gath

ered

on

the

test

rigORC

2

mwf

mhtfh

mhtfc

Pppsu

Pppex

Pexpsu

Pexpex

Phtfh

Phtfc

Tppsu

Tppex

Tpresu

Tevsu

Tevex

Texpsu

Texpex

Tcdsu

Tsubsu

Thtfhsu

Thtfhex

Thtfcsu

Thtfcex

Wpp

Wexp

Npp

Point

[gs]

[kgs]

[kgs]

[bar]

[bar]

[bar]

[bar]

[bar]

[bar]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[C]

[W ]

[W ]

[rpm]

[minus]

2770

801

907

426

798

086

833

6108

922

6350

355

538

836

1029

1011

863

616

616

1505

759

306

504

258

1134

243

1

3099

302

408

029

0114

6100

836

2102

223

7366

372

579

926

1154

1134

950

671

671

1518

852

322

532

314

2182

274

2

3287

302

508

429

1116

2101

837

1111

324

2380

386

580

918

1131

1109

931

668

668

1514

845

332

539

335

2073

291

3

3717

403

108

928

1133

7115

836

3117

522

9363

369

578

963

1198

1172

953

671

671

1524

893

310

534

423

3557

331

4

3503

003

009

029

9129

5113

036

6117

523

7373

379

602

980

1252

1230

1013

713

713

1528

911

326

540

387

3454

311

5

4538

604

110

233

0162

1138

042

3110

625

4405

412

628

1037

1282

1247

1009

719

719

1545

977

349

587

581

4641

405

6

4541

503

910

229

5159

4134

939

395

223

4377

384

596

1005

1236

1198

956

672

672

1541

942

321

558

590

4588

407

7

4962

704

710

831

9179

6151

542

3111

124

5389

396

623

1067

1299

1257

982

677

677

1548

1007

340

582

649

5755

447

8

5592

905

512

132

2204

8169

544

4110

324

7403

411

644

1104

1352

1299

1000

689

689

1568

1049

353

595

902

6889

508

9

5390

005

112

133

9195

7163

443

1111

724

7399

407

636

1087

1334

1286

993

687

687

1570

1029

354

588

829

6614

486

10

3642

002

410

732

0128

4111

138

9127

825

3420

425

609

925

1170

1144

963

715

715

1607

851

375

556

373

2260

323

11

4165

903

011

734

3149

1127

442

091

926

6439

445

629

979

1220

1187

976

728

728

1612

908

395

581

440

3312

372

12

4322

403

610

931

6154

6133

640

1110

424

8402

410

611

1016

1270

1239

993

722

722

1570

950

357

568

541

4648

386

13

5290

003

813

035

7189

5158

846

6110

427

0454

460

658

1038

1337

1292

1028

763

763

1682

976

407

619

546

5831

479

14

5287

203

813

137

6188

6157

547

9102

627

7467

472

669

1040

1333

1287

1033

771

771

1678

978

419

629

545

5502

478

15

5788

704

013

537

1200

5164

450

1100

228

1474

479

666

1028

1288

1229

962

712

712

1673

967

423

638

581

5421

525

16

5789

504

014

036

7199

3163

649

599

228

1475

480

663

1025

1279

1219

950

706

706

1668

963

425

632

583

5415

525

17

5789

404

314

036

7205

2169

350

099

028

3474

480

674

1066

1335

1280

1003

745

745

1668

1005

425

637

671

6072

527

18

4979

803

812

637

4178

3150

348

499

428

6476

483

678

1046

1322

1282

1050

789

789

1649

984

429

634

587

4513

449

19

4464

403

311

339

5163

6139

047

6103

328

3469

476

677

1024

1307

1273

1063

794

794

1651

961

421

629

516

3609

399

20

4461

503

310

542

5163

4138

749

6103

828

8473

479

683

1025

1305

1270

1072

800

800

1648

963

420

644

521

3248

397

21

5476

503

812

237

7177

6160

247

8103

326

7436

444

621

1018

1302

1276

1014

779

779

1692

951

387

626

775

5641

488

22

5783

204

113

239

2187

1167

750

2101

927

3453

462

636

1037

1316

1287

1018

788

788

1692

972

405

637

854

5953

517

23

5784

604

113

139

2187

3167

549

0101

226

1444

452

627

1033

1315

1285

1013

781

781

1690

968

394

627

855

6047

516

24

5794

804

013

139

6189

0168

950

0100

626

4447

455

625

1027

1293

1261

979

754

754

1687

961

401

631

856

6225

517

25

5794

204

313

138

6193

1173

050

399

726

5444

452

630

1054

1325

1295

1009

774

774

1685

988

399

633

870

6431

519

26

5799

304

313

136

3192

5172

248

899

025

5429

438

617

1046

1320

1289

999

762

762

1680

980

384

620

871

6632

520

27

5794

904

513

138

0196

0175

549

798

426

1436

444

630

1073

1357

1327

1038

793

793

1678

1008

390

628

884

6820

519

28

5800

104

514

040

0195

5175

250

298

127

0451

459

640

1076

1358

1329

1043

804

804

1675

1012

407

628

878

6750

519

29

4982

603

711

239

1166

0151

348

497

427

1445

451

616

1026

1290

1269

1025

807

807

1652

960

400

637

543

4850

442

30

5280

205

712

740

3173

6157

348

868

526

4455

463

622

1093

1302

1278

1028

817

817

1521

1041

410

634

730

5245

470

31

3591

003

109

338

1122

8112

444

168

026

6439

445

598

907

1169

1154

991

787

787

1499

909

396

604

339

1787

315

32

3978

203

709

339

9134

1122

846

567

827

0445

451

612

948

1217

1200

1022

809

809

1500

951

395

625

392

2393

350

33

4373

204

410

839

4147

7135

046

967

727

1452

458

622

1042

1264

1246

1040

821

821

1499

983

410

628

461

3477

387

34

4682

605

211

739

0160

0145

546

767

626

9450

456

627

1081

1316

1296

1064

837

837

1504

1027

409

627

518

4549

416

35

5579

006

513

137

2180

6163

848

567

526

3449

455

616

1115

1315

1290

1027

813

813

1510

1064

401

629

608

5829

500

36

3001

602

707

931

9103

996

138

062

123

4378

383

537

838

1090

1078

925

719

719

1419

850

345

551

251

1364

262

37

3564

103

709

335

3120

3110

940

362

025

2409

414

575

903

1191

1177

997

783

783

1427

922

365

575

330

2358

312

38

4164

004

810

436

0140

5128

743

361

925

8422

428

589

971

1238

1221

1012

798

798

1436

982

381

600

462

3569

367

39

4763

906

210

835

8158

5144

544

461

825

2415

422

591

1078

1287

1267

1026

802

802

1440

1029

369

611

574

4890

422

40

5368

207

912

737

8175

3159

546

961

926

2437

444

610

1128

1326

1304

1043

823

823

1447

1086

391

621

657

5838

478

41

5896

009

414

036

8192

2174

448

861

825

7436

444

600

1153

1311

1283

989

783

783

1446

1120

392

617

882

6669

528

42

6186

510

214

136

7197

8179

050

561

726

1442

451

605

1164

1314

1284

986

782

782

1445

1133

396

629

939

6856

555

43

5191

210

712

737

1171

4156

546

553

626

2434

440

589

1134

1301

1279

1019

821

821

1367

1109

391

615

586

5617

462

44

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 49: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Appendix B2 Polynomial regression models

minus Pump model

εispp =

2sumi=0

2sumj=0

aij(Npp)i(rp)j (B8)

εvolpp =

2sumi=0

2sumj=0

bij(Npp)i(rp)j (B9)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B10)

minus Expander model

εisexp =

2sumi=0

2sumj=0

2sumk=0

cijk(ρsuexp)i(rp)j(Nexp)k (B11)

εvolexp =

2sumi=0

2sumj=0

2sumk=0

dijk(ρsuexp)i(rp)j(Nexp)k (B12)

mexp(hsuexp minus hexexp) = Wmecexp +AUloss(Texp minus Tamb) (B13)

minus Heat exchanger model

εth =

2sumi=0

2sumj=0

eij(mhsu)i(mcsu)j (B14)

Appendix B3 Semi-empirical models865

minus Pump model

mpp = (ρsuppNppVdispp)︸ ︷︷ ︸midealpp

minus (Alk

radic2ρsupp(Pexpp minus Psupp))︸ ︷︷ ︸

mlkpp

(B15)

Wmecpp = (Wloss +KlossVsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wlosspp

+ (Vsupp(Pppex minus Pppsu))︸ ︷︷ ︸Wispp

(B16)

Wmecpp = mpp(hexpp minus hsupp) +AUloss(Tpp minus Tamb) (B17)

48

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 50: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

minus Expander model please refer to [31] for a detailed description of the

expander model

minus Heat exchanger model

Qi = AiUi∆Tlogi (B18)

Ui =

(1

αconvhi+

1

αconvci

)minus1

(B19)

Nsumi=0

Ai = AHEX (B20)

αconv = αconvnom

(m

mnom

)n

(B21)

Appendix B4 Pipeline losses

minus Pressure losses

∆P = Kϕsu +B (B22)

ϕsu =m2

ρsu(B23)

minus Heat losses

Qloss = AUloss(Tsu minus Tamb) (B24)

Appendix C Detailed results of the study

Detailed values of RMSEs and MAPES computed for both the component-870

level and the cycle-level analyses are summarized in Table C8 and C9

49

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 51: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Tab

leC

8

Glo

bal

resu

lts

-R

MS

E(u

nit

s

sam

eth

an

the

vari

ab

les)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

87

219

347

29

238

15

9e-0

315

3e-0

21923

1561

3258

472

2187

102

565

44

0e-0

371

1e-0

3

PolE

ffm

odel

68

89

125

854

56

3e-0

411

5e-0

31541

1550

1885

451

1340

72

136

31

3e-0

446

7e-0

3

Sem

iEm

pm

odel

77

73

151

549

80

1e-0

416

1e-0

31062

1560

818

382

979

87

205

58

2e-0

372

7e-0

3

Extrap

CstEff

model

126

231

338

36

248

17

6e-0

312

2e-0

21754

1780

3980

681

3191

84

612

60

7e-0

388

9e-0

3

PolE

ffm

odel

129

142

97

12

147

12

7e-0

314

8e-0

32230

2053

2842

662

6833

97

466

58

0e-0

417

2e-0

2

Sem

iEm

pm

odel

119

74

158

748

98

0e-0

416

7e-0

31236

1883

1044

449

1384

85

229

74

7e-0

390

6e-0

3

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

320

133

260

23

190

16

4e-0

316

4e-0

34842

2322

3540

3419

4343

80

632

43

0e-0

343

0e-0

3

PolE

ffm

odel

147

93

136

975

57

2e-0

457

2e-0

45276

2189

2871

3202

3115

281

1311

51

3e-0

351

3e-0

3

Sem

iEm

pm

odel

150

83

129

657

69

9e-0

469

9e-0

42118

2194

3405

2915

1206

79

365

61

6e-0

361

6e-0

3

Extrap

CstEff

model

441

91

337

19

207

15

3e-0

315

3e-0

34116

2665

5236

4040

2877

88

666

62

4e-0

362

4e-0

3

PolE

ffm

odel

153

402

151

19

94

68

0e-0

468

0e-0

45821

2954

15334

14696

5601

201

422

26

3e-0

326

3e-0

3

Sem

iEm

pm

odel

155

137

127

767

83

6e-0

483

6e-0

41637

2870

6177

4136

1452

102

392

79

5e-0

379

5e-0

3

Tab

leC

9

Glo

bal

resu

lts

-M

AP

E(u

nit

s

)

ORC

1O

RC

2

Qev1

Qrec1

Qcd1

Wpp1

Wexp1

mpp1

mexp1

Qev2

Qrec2

Qcd2

Qsub2

Qpre2

Wpp2

Wexp2

mpp2

mexp2

Com

ponent-levelanaly

sis

Fittin

g

CstEff

model

05

2185

631

3267

1325

135

2246

721

0117

325

628

461

5132

3146

407

312

0

PolE

ffm

odel

04

459

314

281

893

413

020

017

0113

811

128

735

886

526

800

508

1

Sem

iEm

pm

odel

05

1112

418

448

993

623

039

011

9104

408

143

921

096

643

311

211

7

Extrapola

tio

n

CstEff

model

06

5262

830

0358

6404

548

7288

920

0132

827

736

983

4151

2217

611

015

1

PolE

ffm

odel

07

382

014

0128

3212

921

445

223

7142

519

240

191

2146

694

200

830

5

Sem

iEm

pm

odel

06

5161

116

955

1120

331

137

715

9130

708

249

840

0108

055

216

015

4

Cycle

-levelanaly

sis

Fittin

g

CstEff

model

29

0144

028

0251

4262

636

836

843

1166

134

4263

189

5120

3139

407

107

1

PolE

ffm

odel

15

369

115

494

0107

012

912

941

3151

126

6235

772

6200

6111

805

405

4

Sem

iEm

pm

odel

18

8138

917

769

981

220

320

321

0142

334

9252

939

297

370

911

611

6

Extrapola

tio

n

CstEff

model

38

1140

734

3210

6346

044

744

745

7200

552

8413

882

5171

8179

509

909

9

PolE

ffm

odel

19

1276

320

1228

8131

523

023

058

1227

0135

81657

1153

0365

281

204

504

5

Sem

iEm

pm

odel

22

3328

020

683

7113

230

130

117

2193

555

2406

344

7147

289

715

615

6

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach

Page 52: Modelling of organic Rankine cycle power systems in off ... (pre...T D ACCEPTED MANUSCRIPT Modelling of organic Rankine cycle power systems in o -design conditions: an experimentally-validated

MANUSCRIP

T

ACCEPTED

ACCEPTED MANUSCRIPT

Highlights

o Three methods are compared to simulate the off-design operation of ORC engines

o Post-processed experimental measurements are used as reference database

o Fitting and extrapolation capabilities of these 3 modelling paradigms are studied

o Both component-level and system-level analyses are performed

o Semi-empirical models demonstrate to be the best modelling approach


Recommended