lable at ScienceDirect

Energy xxx (2014) 1e11

Contents lists avai

Energy

journal homepage: www.elsevier .com/locate/energy

Auto-ignition control in turbocharged internal combustion enginesoperating with gaseous fuels

Jorge Duarte, Germán Amador, Jesus Garcia, Armando Fontalvo, Ricardo Vasquez Padilla 1,Marco Sanjuan, Arturo Gonzalez Quiroga*

Department of Mechanical Engineering, Universidad del Norte, Km 5 Via Antigua Pto Colombia, Barranquilla, Colombia

a r t i c l e i n f o

Article history:Received 24 October 2013Received in revised form11 April 2014Accepted 14 April 2014Available online xxx

Keywords:Internal combustion engineAuto-ignition controlGaseous fuelsMethane numberRobust controlPID controller

* Corresponding author. Tel.: þ57 5 3509272; fax:E-mail addresses: jduartee@uninorte.edu.co (J. D

edu.co (G. Amador), jesusmg@uninorte.edu.co (J. Gedu.co (A. Fontalvo), Ricardo.Vasquezpadilla@csirmsanjuan@uninorte.edu.co (M. Sanjuan), arturogonzalez.quiroga@gmail.com (A. Gonzalez Quiroga).

1 CSIRO Energy Technologies, PO Box 330 NewcastlAustralia. Tel.: þ61 249606293.

http://dx.doi.org/10.1016/j.energy.2014.04.0400360-5442/� 2014 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Duarte J,fuels, Energy (2014), http://dx.doi.org/10.101

a b s t r a c t

Control strategies for auto-ignition control in turbocharged internal combustion engines operating withgaseous fuels are presented. Ambient temperature and ambient pressure are considered as the disturbingvariables. A thermodynamic model for predicting temperature at the ignition point is developed,adjusted and validated with a large experimental data-set from high power turbocharged engines. Basedon this model, the performance of feedback and feedforward auto-ignition control strategies is explored.A robustness and fragility analysis for the Feedback control strategies is presented. The feedforwardcontrol strategy showed the best performance however its implementation entails adding a sensor andnew control logic. The proposed control strategies and the proposed thermodynamic model are usefultools for increasing the range of application of gaseous fuels with low methane number while ensuring asafe running in internal combustion engines.

� 2014 Elsevier Ltd. All rights reserved.

1. Introduction

Knock phenomenon is a real barrier for increasing efficiency andlowering maintenance costs of spark ignition engines. Knockingcan cause severe damages in engines, especially in mechanicalparts like piston, valves and cylinder head [1]. Nowadays, it iswidely accepted that knock is due to auto-ignition in the end gasregion of the combustion chamber [2]. One important issue forknocking occurrence in turbocharged internal combustion enginesoperating with natural gas is the effect of ambient temperature onthe auto-ignition tendency of the fuel. Depending on local atmo-spheric conditions, manual derating of the engine is mandatory toavoid knocking [3]. Actions such as power reduction and sparkadvance have been implemented to avoid knocking condition [4].In this context, this paper presents a robust control strategy,implemented in a feedback control loop, for auto-ignition control in

þ57 5 3509255.uarte), gjamador@uninorte.arcia), aefontalvo@uninorte.o.au (R. Vasquez Padilla),q@uninorte.edu.co, arturo.

e, NSW 2300,

et al., Auto-ignition control in6/j.energy.2014.04.040

turbocharged internal combustion engines operating with gaseousfuels. The capabilities for adapting engine operation to changes inambient temperature while maximizing output power areexplored. The performance of the robust control strategy iscompared to a feedforward control strategy.

Some previous research in knocking prevention has beenfocused on the effect of fuel composition and engine operationmode. It was found that adding inert gases (N2 and CO2) causes asignificant increase in the knock-limited spark timing (KLST) [5]while thermal efficiency and emissions are slightly affected. It hasbeen reported that mixing fuels with different auto-ignition ten-dencies (natural gas and heptane) can be used for knocking control[6]. It has also been noted that auto-ignition is promoted by in-crements in residual gas temperatures and residual gas flow rates[7].

To prevent knocking, it has been proposed to adapt enginetuning according to fuel composition [2]. Authors propose aknocking protection map to set engine parameters based on theresults of a computational combustion simulator. It has been pro-posed using specific combustion chamber depending on fuel, todelay knocking appearance [8]. According to [8], a baseline-typecombustion chamber fueled with methanol allows operating withlower knocking intensity without significantly affecting thermalefficiency. Adding hydrogen to natural gas has been highlighted as away to get a lean combustion extension [9]. Adding oxygen to the

turbocharged internal combustion engines operating with gaseous

Nomenclature

BTDC before top dead centerC(s) transfer functionCp specific heat capacity [kJ/kgK]CO controller outputFFC feedforward controlFFTF feedforward transfer functionFI fragility indexFOPDT first order plus dead timeIAE integrated absolute errorKc controller gain [%CO/%TO]KLST knock-limited spark timingMs process transfer functionMs sensitivityMN methane numberMVM mean value modelnT polytropic coefficientNOx nitride oxidesp pressure [kPa]R crank-connection rod ratiorc compression ratio

T temperature [K]t time [s]TO transmitter outputu errorV volume [m3]

Greek symbolsb weight factorl lambda tuneu signal frequency [rad/s]r density [kg/m3]sd derivative time [s]si integral time [s]q crank angle [rad]

Subscripts0 ambient conditionsb burnedm air fuel blendR residual gass spark pointu unburned

J. Duarte et al. / Energy xxx (2014) 1e112

fuel also reduces the knock tendency; however it increases NOx

emissions [10].A considerable amount of literature has been published on

control strategies for knocking prevention. A knocking detectionscheme consisting of multi-feature extraction and neural classifi-cation has been proposed [11]. Authors developed a constructivelearning algorithm for the cycle-by-cycle knocking detection task. Afuzzy control system, in which different spark advances and timingsetting effects were used to determine knocking intensity, has beenreported [12]. It was also reported the development of an enginewhich operates with variable biogas/air ratios and performs a sta-ble operation without knocking [13]. The control algorithm for thisengine was designed to adjust variable biogas/air ratios to obtainhigh efficiency along with low NOx emissions.

A knocking control strategy based on modifying ratios of high-octane-fuel/low-octane-fuel fed into combustion chamberwithout changing ignition timing has been patented [14]. It has alsobeen patented an alternative knocking control strategy which usesa delivery system automatically configured to respond to variableoperating conditions by feeding a fluid like alcohol or water to atleast one of the engine cylinders of a vehicle [15]. Finally, it has beenpatented a knocking control strategy that uses a pre-combustionchamber with a first spark plug, followed by a chamber with anindependent spark plug; depending on operating conditions one orboth of the sparks work [16].

The most common knocking control strategies for modern en-gines have been based on the Feedback principle, with the draw-back that knocking is belatedly corrected. It is possible toimplement knocking control strategies based on the feedforwardprinciple, however it entails characterization of the disturbance,and for internal combustion engines there are many disturbancesthat lead to auto-ignition. This paper focuses on a robust PID con-trol strategy based on the feedback principle [17] but including asensitivity factor which allows faster responses to a disturbance.The first section of this paper presents the development and thevalidation of a model for predicting temperature at the ignitionpoint. Based on this model, the following section present Feedbackauto-ignition control strategies, testing its performance in terms ofrobustness and controller fragility. The next section presents a

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

feedforward auto-ignition control strategy focused on commondisturbances. The final section compares and discusses the per-formance of the developed auto-ignition control strategies.

2. Model for predicting temperature at the ignition point

2.1. Model description

To evaluate the performance of the control strategies addressedin this paper, a model for predicting temperature at the ignitionpoint was developed, fitted and validated. The model describes thecompression process of a four-stroke turbocharged engine takinginto account volume, pressure and temperature of air-fuel-unburned and residual-exhaust gases in the combustion chamber.It was considered simultaneous mass and heat transfer betweenair-fuel-unburned gases coming from the aftercooler and residual-exhaust gases remaining in the combustion chamber. A MVM(Mean ValueModel) approachwas chosen because of its reasonableprecision and low computational complexity [18]. MVM are controloriented models with time as the independent variable, where thediscrete nature of the engine is neglected and the evolution ofvariables are assumed to be continuous in an average sense over thecycle.

The process begins with an air-fuel mixture entering the mixingchamber. The mixture is fed into the turbocharger where an in-crease in pressure produces a rise in temperature. The valve in thesupply line regulates the amount of fuel fed into the mixingchamber and the aftercooler reduces air-fuel mixture temperatureto 40 �C. Then, the air-fuel mixture is fed into the combustionchamber where simultaneous mass and heat transfer between air-fuel-unburned and residual-exhaust gases mixture takes place. Themixture follows a polytropic compression process until reachingthe ignition point. At this point, the temperature of the mixtureshould be below auto-ignition temperature to avoid knocking. Aschematic diagram of the engine detailing the aforementionedprocesses is shown in Fig. 1.

The proposed model relates gases mixture temperature to itsexperimental polytropic coefficient. An experimental dataset froma large commercial engine was employed to determine the

turbocharged internal combustion engines operating with gaseous

MIXING

AIR

FUELTURBOCHARGER

AFTERCOOLER

MIXING OF FRESH AND REMAINING

GASES COMPRESSION

0 1 2 3

4

COMBUSTIONCHAMBER

Fig. 1. Schematic representation of the engine.

J. Duarte et al. / Energy xxx (2014) 1e11 3

polytropic coefficient of the compression processes within both,turbocharger and combustion chamber. Fig. 2 shows a picture ofthe engine and Table 1 summarizes its main characteristics.

Many gas compression and expansion processes may be usuallyapproximated by a polytropic process. In each case, the polytropiccoefficient must be determined experimentally. The most commonrelation used to describe polytropic processes is expressed in termsof pressure and volume. Likewise, polytropic processes can bedescribed in terms of temperature and pressure or temperature andvolume, as it is presented here. However, it is necessary to rely onan equation of state to represent the relation among variables.

In this work, the equation of state used is the ideal gas. Thischoice is justified because it was checked that compressibilityfactors for the gases mixtures along the process ranged from 0.997up to 1.01. Authors also checked that an equation of state like Peng-Robinson provides an accurate description of the processes withthe additional advantage that polytropic coefficients are not used.However, for auto-ignition control purposes, it was decided to usethe model with the lower complexity.

The compression process in the turbocharger is described byEquation (1):

Fig. 2. Turbocharged internal combustion engine. (Courtesy Porto Do Pecem, Brazil).

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

�T1T0

�nT

¼�p0p1

�1�nT

(1)

where T1 and p1 are temperature and pressure at the outlet of theturbocharger and T0 and p0 are ambient temperature and ambientpressure, respectively. The polytropic coefficient nT is described byEquation (2):

nT ¼ 1

1þln

����T1T0����

ln

����p0p1����

(2)

The compression process in the combustion chamber isdescribed by Equation (3):

T3T4

¼�V4

V3

�nC�1(3)

where V4 and V3 are given by Equations (4) and (5) [19]:

V4 ¼ V0

�1þ 1

2ðrC � 1Þ

�Rþ 1� cosq4 �

ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiR2 � sin2q4

q ��(4)

V3 ¼ Vd þ V0 (5)

where:

q4 ¼ 360� � qignition (6)

Table 1Turbocharged internal combustion engine parameters.

Engine model QSV91 e CumminsRated output 2000 Kw (100%) @ 1514 rpmMode operation Load governBore 180 mmStroke 200 mmCompression ratio 10.55:1, 11.4:1 and 12:1Displacement 91.6 lIgnition Individual coil on plugAir/fuel ratio 18.5:1, 18.2:1 and 18:1Cooling fluid WaterDesign Turbocharged with aftercooler

turbocharged internal combustion engines operating with gaseous

J. Duarte et al. / Energy xxx (2014) 1e114

The energy balance, assuming adiabatic conditions in thecombustion chamber, is described by Equation (7):

mRCpRðTe � T3Þ ¼ mmCpmðT3 � T2Þ (7)

where Te is the residual-exhaust gases exit temperature, mR themass of residual-exhaust gases that remains in the combustionchamber, and mm the mass of the air-fuel-unburned mix fed intothe combustion chamber. mR and mm can be expressed in terms ofvolume and density as described by Equations (8) and (9):

mR ¼ rRV0 (8)

mm ¼ rmVd (9)

where V0 and Vd are clearance (see Fig. 3) and displaced volume,respectively. The compression ratio (rc) is described by equation(10) as follow:

rc ¼ Vd þ V0

V0(10)

An experimental procedure was carried out to estimate thepolytropic coefficients of the compression processes in the turbo-charger and in the combustion chamber. It is important to bear inmind that temperature at ignition point depends on environmentalvariables such as pressure and temperature, and on operationalvariables of the engine such as ignition angle and temperature ofthe exhaust gases [20]. The temperature reached in the combustion

TDC

T3

Vo

Piston

Connectingrod

Crankshaft

Spark plug

Cylinder

T4 at ignition point

BDC

Fig. 3. Schematic diagram of the combustion chamber of a spark plug internal com-bustion engine.

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

chamber is calculated by means of Equation (11) which can bederived by combining Equations (3)e(10).

T4 ¼rRCpR

rmCpmðrc�1ÞTe þ T2

1þ rRCpR

rmCpmðrc�1Þ

�

2641rc

0B@1þ rc � 1

2

0B@Rþ 1� cos qþ

�R2 � sin2 q

12

1CA1CA3751�nc

(11)

where:

rc: compression ratio of the engine.q: crank angle (Engine operates with a constant ignition angle of14� before top dead center).Cpm: Air-fuel mixture specific heat capacity.rm: Air-fuel mixture density.CpR: Residual-exhaust gases specific heat capacity.R: Crank-connection rod ratio.T2: Temperature at the outlet of the aftercooler.T4: Temperature of air-fuel-unburned gases in the combustionchamber at BTDC.V0: Clearance volume.

2.2. Model assumptions

The following assumptions were taking into account for thedevelopment of the model:

� Homogenous distribution of mixture properties within thecombustion chamber.

� rm and Cpm of air-fuel-unburned and residual-exhaust gasesmixture are temperature dependent.

� Engine steady state operation, i.e. constant speed and constantair fuel rate supply.

� Polytropic processes during compression in the turbochargerand in the combustion chamber. The polytropic coefficients, nTand nC, were determined experimentally.

� Constant polytropic coefficients. It was assumed that the influ-ence of factors such as heat transfer at steady state on shirts,heat losses in the turbocharger and irreversibilities weredescribed by the polytropic coefficients.

� Engine data such as the compression ratio rc, turbochargeroutlet pressure and outlet temperature of exhaust gases weretaken from engine data sheet.

� The effect of relative humidity of intake air is negligible.� Heat transfer process only occurs between combustion productsremaining in the combustion chamber and the fresh charge ofair-fuel mixture.

� There are not chemical reactions, i.e. the simulation was carriedout before ignition takes place.

2.3. Model validation

The polytropic coefficients nT and nC, were calculated followinga DOE (Design of Experiments) approach for the same type of en-gine under different environmental conditions (ambient tempera-tures and ambient pressures) and compression ratios (10.6:1, 11.4:1and 12:1; see Table 2). Temperature and pressure data at the inletand at the outlet of the turbocharger, which vary as a function ofambient pressure and ambient temperature, were collected. Thepolytropic coefficient nT was calculated by means of Equation (2).An analogous procedure with temperature and volume data was

turbocharged internal combustion engines operating with gaseous

Table 2Ambient pressures and compression ratios.

Location Ambient pressure (kPa) Compression ratio

Barranquilla e Colombia 101.3 12.0Neuquen e Argentina 98.8 11.4Bogota e Colombia 74.0 10.6Fortaleza e Brazil 101.3 12.0

J. Duarte et al. / Energy xxx (2014) 1e11 5

applied to calculate the polytropic coefficient in the combustionchamber nC by means of Equations (3)e(6). The adjusted value ofthe polytropic coefficients nT and nC were 1.374 and 1.349, respec-tively. It is worth to highlight that these polytropic coefficients canbe used to predict temperature at the ignition point, i.e. in theabsent of combustion and at constant regime conditions.

Temperature at the ignition point was measured after com-pressing the mixture of air-fuel-unburned and residual-exhaustgases with the spark plug disconnected to avoid combustion.When the piston reached the ignition point, the exhaust valve wasforced by changing the angle of the camshaft, to allow the exit ofthe compressed mixture. Temperature at the ignition point wasmeasured by means of a sensor located immediately after the exitof the exhaust valve, at the entrance of the exhaust duct.

The model was validated by comparing predicted results with anew set of experimental data collected from the engine describedin Table 1, which operates at 100% load and at constant speed. Fig. 4illustrates the obtained results. To solve the mathematical model, asimulation programwas developed and implemented in Simulink�

under MATLAB� environment. Fig. 5 shows a simplified block dia-gram of the model. Natural gas was the fuel used in this study andits composition is shown in Table 3.

3. Feedback control strategy

The current control strategy is based on the one presented byAlfaro and Vilanova [17]. It begins with the regular form of a PIDcontroller as described by Equation (12). Errors are calculated bymeans of Equations (13)e(15).

uðtÞ ¼ Kc$

�epðtÞ þ 1

si$

ZeiðtÞ$dt þ sd$

dedðtÞdt

�(12)

epðtÞ ¼ b$rðtÞ � yðtÞ (13)

eiðtÞ ¼ rðtÞ � yðtÞ (14)

Fig. 4. Comparison between predicted and expe

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

edðtÞ ¼ g$rðtÞ � yðtÞ (15)

where Kc is controller gain, si is integral time, and sd is derivativetime, b and g are weight factors. The g parameter works as aselector (Possible values are 0 or 1), usually being zero, so the de-rivative kick off problem is avoided. Equation (16) describes thecontroller equation in the frequency domain:

uðsÞ ¼ Kc$

0B@epðsÞ þ 1

si$s$eiðsÞ þ

sd$ssdN$sþ 1

$edðsÞ

1CA (16)

By reordering Equation (16), it is possible to split it in twodifferent terms that are applied to the signals coming from the setpoint and the sensor-transmitter. The reordered expression isdescribed by Equation (17).

uðsÞ ¼ Kc$

�bþ 1

si$s

�$rðsÞ � Kc$

�1þ 1

si$sþ sd$s0:1$sd$sþ 1

�$yðsÞ

(17)

By introducing the definitions given by Equations (18) and (19),Equation (16) can be expressed as shown by Equation (20).

CrðsÞ ¼ Kc$

�bþ 1

si$s

�(18)

CyðsÞ ¼ Kc$

�1þ 1

si$sþ sd$s0:1$sd$sþ 1

�(19)

uðsÞ ¼ CrðsÞ$rðsÞ � CyðsÞ$yðsÞ (20)

Fig. 6 schematizes the final block diagram of the Feedbackcontrol strategy. The set point signal is fed to the Cr(s) block. On theother hand, the signal with information of the process is fed to theCy(s) block. Summing up, Cr(s) controls the process set point, andCy(s) controls feedback process performance.

The b coefficient of Equation (18) is used to reduce the propor-tional action of the controller, avoiding either a response with anexcessive overshoot or too oscillatory. This coefficient can vary from0 to 1. The other parameters are the commonly ones used in PIDcontrollers: controller gain (Kc), integral time (si), and derivativetime (sd). The signal represented as d(s) in Fig. 6, represents adisturbance that in the current paper accounts for changes inambient temperature. As previously stated, a robustness analysis ofthe control strategy is also carried out. This analysis takes into

rimental temperatures at the ignition point.

turbocharged internal combustion engines operating with gaseous

EXHAUSTGAS

ANALYZERExhaust Gas Temperature

Exhaust Gas Pressure

Exhaust Gas Composition EG Density

ENGINE2

FUELANALYZERFuel Composition

TurbochargerAftercoolerENGINE 1

Ambient Temp.Ambient Press.

Out P Turbo

Comb. Pressure

Comb.Temperature

AIR/FUELMIXTURE

AIRANALYZER

Comb. Pressure

Comb. Temperature

Air HeatCapacity

Air density

Comb. density

Comb. Heat Capacity

Comb. Molec. Mass

Air/Fuel Ratio

EG Heat Capacity

Exhaust Gas Temperature

Comb. Temperature

Air/FuelDensity

Air/FuelHeat Capacity

Ignition PointTemperature

ROBUST PIDCONTROLLER

SENSOR/TRANSMITTER

ACTUATOR(Retarder)

Sensorsignal

Controllersignal

Theta

Fig. 5. Simulation block diagram for predicting temperature at the ignition point.

J. Duarte et al. / Energy xxx (2014) 1e116

account the controlled variable (Temperature) and themanipulatedvariable (air-fuel ratio). Then, the performance of closed controlloop, given by different tuning equations, is evaluated to obtain thebest IAE (integrated absolute error) index.

3.1. Robustness analysis

Relative stability or robustness is a very important parameter inclosed control loop. As a measure of stability, it has been used thegain margin in conjunction with the phase margin (Am, 4m) [21].Another proposed measure of robustness has been “the maximumsensibility parameter”, which is defined by Equation (21), and canvary from 1.2 up to 2 [17]. For robustness analysis, the maximumsensibility parameter has been used in the present paper.

Ms ¼ maxw

jSðiwÞj ¼ maxw

���� 11þ CyðiwÞ$PðiwÞ

���� (21)

where Cy(iw) is the controller transfer function and P(iw) is theprocess transfer function which are defined by Equations (19) and(23), respectively. Parameters in Equation (23) are calculated byprocess Fit 3 (method proposed for estimating dead time t0 andtime constant s) [22], thus a first order plus dead time transferfunction is obtained.

CyðiwÞ ¼ Kc$

�1þ 1

si$iwþ sd0:1$sd$iwþ 1

�(22)

PðiwÞ ¼ Kp$e�to$iw

sp$iwþ 1(23)

In addition to Equation (21), the IAE index (Integral AbsoluteError) defined by Equation (24), has been used to minimize theerror between set point and real time process behavior.

Table 3Natural gas composition [24].

Component Percentage

CH4 87.1C2H6 8.8C3H8 2.5C4H10 0.8N2 0.8

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

IAE ¼Z

ðrðtÞ � yðtÞÞdt (24)

To obtain the process transfer function of the SI engine modeldescribed by Equation (25), it was defined a step change of þ5%CO(Controller output).

PðiwÞ ¼ �0:31902$e�0:6045,iw

0:3855$iwþ 1(25)

The controller parameters are calculated using the tuning for-mulas of Table 4 [22]. Results in Table 5 show that the best set oftuning equations is given by Refs. l0, since Ms value is closer to 1.2.On the other hand, the worst tuning equations correspond to IAE,whose Ms value was 3.861. Figs. 7 and 8, illustrate the effect offrequency on Ms for different tuning equations.

Once determined that l0 tuning equations are the most robust,the effect of b coefficient on stability is studied. In this paper, it wascarried out a sensibility analysis by varying the b coefficient and theIAE index for a step change of 20 �C in ambient temperature. Resultsfor 0.1 � b � 1.4 are shown in Table 6. The optimum value of b,which leads to the lowest IAE index, is 1.3.

3.2. Controller fragility analysis

While robustness is related to the number of process charac-teristics that can change without making the process unstable,fragility is focused on how unstable a closed control loop maybecome due to changes on controller tuning parameters [17]. Tostudy this effect Alfaro [23] defined the fragility index as:

FID20 ¼ MsDε;max

MSO� 1 (26)

where MsDε,max represents the maximum extreme sensibility,which is the major robustness lost by the controlled system when

Cr(s)

Cy(s)

Process- Y(s)r(s)u(s)+ +

+

d(s)

Fig. 6. Block diagram of the Feedback control strategy.

turbocharged internal combustion engines operating with gaseous

Table 4Tuning equations.

Tuning set Kc si sd

l sp=1:2$t0$Kp sp t0=2l0

sp=2:2$t0$Kp sp t0=2Quarter decay ratio (QDR) 1:2=Kp$ðt0=spÞ�1 2$t0 1=2$t0IAE for disturbances 1:435=Kp$ðt0=sÞ�0:921 s=0:878$ðt0=sÞ0:749 0:482,s$ðt0=sÞ1:137IAE for set point changes 1:086=Kp$ðt0=sÞ�0:869 s=0:740� 0:130$ðt0=spÞ 0:308,s$ðt0=sÞ0:9292

J. Duarte et al. / Energy xxx (2014) 1e11 7

all its parameters change Dε, and MSO the maximum nominalsensibility. Alfaro et al. [17] defined a specific index to study the PIDcontroller fragility. This is called fragility index D20(FID20), whichmeans that all controller parameters change 20%. According to theresults obtained by Alfaro [23], the following conclusions can bedrawn:

� A PID controller is fragile if its index is higher than 0.5� A PID controller is not fragile if its index is lower than 0.5� A PID controller is elastic if its index is lower than 0.1

By applying the FID20 index to the SI internal combustion enginemodel, it was concluded that the proposed control loop is notfragile. After varying �20% all the parameter, Ms values from 1.397up to 1.421 were obtained. Table 7 and Fig. 9 show the results.

3.3. Process response improvements

According to the aforementioned results, it is possible to changecontroller parameters by around 20% without making it fragile. Athree level experimental factorial design was carried out to figureout if under any of these experimental conditions the IAE perfor-mance index could be minimized. The three level factorial design isshown in Table 8.

Results are shown in Table 9, where the minimum IAE indexcorresponds to the following controller parameters: Kc ¼ �1.090%CO/%TO, si ¼ 0.308 s, and sd ¼ 0.242 s. These new tuning param-eters allow reducing IAE index from 10.805 up to 7.958. In thisanalysis the b coefficient was set as 1.3, which led to minimize theIAE index as it was shown in the previous section. Finally, thisrobust and non-fragile control strategy will be compared in thenext section with the feedforward control strategy.

4. Feedforward control strategy

The feedforward control strategy was conceived to give thecontroller the possibility of changing at the same time as commondisturbances occur. This allows counteracting a disturbance withminimal effect on the controlled variable. As mentioned before, inthe proposed control strategy the controlled variable is tempera-ture at the ignition point, and the disturbance is ambient tem-perature. To design the control loop, it is necessary to perform anopen loop characterization between the disturbance and thecontrolled variable. This identification allows determining how

Table 5Process controller parameters.

Tuning set Kc si sd Ms

l �1.6658 0.3055 0.3022 1.5902l0 �0.9086 0.3855 0.3022 1.2931Quarter decay ratio (QDR) �2.3988 1.209 0.3022 2.7112IAE for disturbances �2.9723 0.6150 0.3099 3.8614IAE for set point changes �2.3027 0.7190 0.2024 1.8407

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

much the manipulated variable has to be changed to counteractthe effect of the disturbance. This was done by changing theambient temperature by 10% with respect to the steady statevalue. Table 10 summarizes the characterization of the engineprocess and the disturbance in a FOPDT (First-Order-Plus-Dead-Time) function.

Once the open loop characterizationwas accomplished, the FFTF(Feedforward transfer function) described by Equation (27) wasdeveloped following the methodology presented by Smith andCorripio [22]. Kp, Kd, and KT represent: process, disturbance, andsensor transmitter gains, respectively. The term that contains pro-cess and disturbance time constants (sp, sd), is referred as a lead/lagblock, and its function is to compensate differences betweenmanipulated variable/controlled variable and disturbance/controlled-variable paths.

FFC ¼�� KdKT$Kp

�$

�sp$sþ 1sd$sþ 1

�$e�ðt0d�t0p Þ$s (27)

Results indicated that the difference between dead times isnegative (t0d�t0p < 0), so, this term was not implemented in thefeedforward controller. The total gain used was 3.5%COFF/%TOD.Finally, the FFC (feedforward control) structure used in this paper isrepresented by Equation (28):

FFC ¼ 3:5$�2:5600$sþ 11:5750$sþ 1

�(28)

The feedforward control strategy works in combination witha regular PID controller. The PID transfer function is defined by

Fig. 7. Effect of w on Ms for different tuning equations.

turbocharged internal combustion engines operating with gaseous

Fig. 8. Ms changing as w vary, for IAE disturbances, and IAE set point tuning equationssets.

Table 7Fragility index �D20 for closed control loop.

Kc si sd Ms FID20

MS�D20 �0.7269 0.3084 0.2418 1.3970 0.1319MSþD20 �1.0903 0.4626 0.3626 1.4212 0.0991MSO �0.9086 0.3855 0.3022 1.2931 e

Fig. 9. Fragility index �D20 for closed control loop vs frequency.

J. Duarte et al. / Energy xxx (2014) 1e118

Equation (29). This expression is a modified version suggestedby Smith and Corripio [24]. The controller is tuned by l

equations.

uðsÞ ¼ Kc$rðsÞ þ 1si$sþ 1

$

�uðsÞ þ Kc$sd$s

a$sd þ 1$yðsÞ

�� Kc$sd$sa$sd þ 1

$yðsÞ

(29)

It is important to point out that the final control element in theproposed control strategy is a spark retarder. This element receivesthe signal from the controller, and operates in a range between0� and 30� (Crank angle degrees) of ignition delay. Besides, the setpoint of the controlled variable was established as close as possibleto the auto-ignition temperature of the fuel. The feedforwardimplementation scheme is depicted at Fig. 10.

5. Results and discussion

At this section, a comparison between three different controlstrategies (feedback, feedback with robust control and feedfor-ward) is carried out. Feedback and feedforward control strategieswere chosen because those are the most suitable to be imple-mented. The proposed strategies were implemented by usingSimulink� from Matlab�.

To analyze and compare the performance of the proposedcontrol strategies, changes in ambient temperature were simu-lated. This was established because this type of engine is used at

Table 6b coefficient effect on IAE performance index.

b IAE Response type

0.1 10.8074 Stable0.5 10.8066 Stable1.0 10.8055 Stable1.2 10.8053 Stable1.3 10.8052 Stable1.4 10.8053 Stable

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

different places subject to different ambient temperature changeswhich is the most important disturbance when atmosphericpressure is constant (fixed location). Set point changes are notstudied in this research because it entails changing fuel composi-tion, which is not usual for a fixed place operation. Two ambienttemperature profiles were set to study the proposed control stra-tegies. The difference between the profiles comes from atmo-spheric pressure, which are 11 and 14.7 psia. The simulatedambient temperature profiles are shown at Fig. 11. Although it mayseem that an 11 psia pressure is unusual it can be reached at placeslike Bogotá (Colombia).

Ambient temperature profiles were initially simulated as stepchanges of 10 �C. The auto-ignition temperature for a natural gas-air rich mixture is 538 �C. Then, by selecting a security factoraround �8 �C, a 530 �C operation temperature allows obtaininghigh efficiency and maximum power without knocking. Thefollowing operating conditions were kept constant for evaluatingthe performance of the control strategy: speed rate, ignition angleand air/fuel ratio. As the expected application is stationary gener-ation, ambient pressure and electric output were set constant andengine load was set at 100%.

Figs. 12 and 13 show the effect of atmospheric pressure on theperformance of the proposed control strategies. The feedforward

Table 8Three level factorial design general characteristics.

Level Kc si sd

Low �0.7269 0.3084 0.2418Medium �0.9086 0.3855 0.3022High �1.0903 0.4626 0.3626

turbocharged internal combustion engines operating with gaseous

Table 9Three factorial execution results.

Run Kc si sd IAE

1 �0.9086 0.4626 0.3022 12.35412 �0.9086 0.3084 0.2418 9.22633 �0.9086 0.4626 0.2418 12.30054 �0.9086 0.3855 0.3626 10.86355 �1.0903 0.4626 0.2418 10.50306 �0.9086 0.3084 0.3626 9.34197 �1.0903 0.3084 0.2418 7.95988 �1.0903 0.4626 0.3626 10.62719 �1.0903 0.3084 0.3022 8.021510 �1.0903 0.3855 0.3626 9.354311 �1.0903 0.4626 0.3022 10.564912 �1.0903 0.3855 0.3022 9.290813 �1.0903 0.3084 0.3626 8.082114 �0.7269 0.3855 0.2418 12.957115 �0.7269 0.3855 0.3022 13.006216 �0.7269 0.3084 0.2418 11.035217 �0.9086 0.3084 0.3022 9.284418 �0.7269 0.3084 0.3626 11.139219 �1.0903 0.3855 0.2418 9.22720 �0.9086 0.4626 0.3626 12.408521 �0.7269 0.3855 0.3626 13.055722 �0.9086 0.3855 0.3022 10.805223 �0.7269 0.4626 0.3022 15.005924 �0.7269 0.4626 0.3626 15.047225 �0.7269 0.3084 0.3022 11.087326 �0.9086 0.3855 0.2418 10.746927 �0.7269 0.4626 0.2418 14.9659

Table 10Process and disturbance FOPDT identification.

K s t0

Process �0.3190 0.3855 0.6045Disturbance 0.2809 1.0050 0.0050

Fig. 11. Ambient temperature profiles to be studied in the addressed control strategies.

J. Duarte et al. / Energy xxx (2014) 1e11 9

strategy exhibits the smallest deviations when compared withthe PID robust strategy, and the feedback loop using only themodified PID. When the robust PID strategy is compared with thePID modified strategy, results show that the first one exhibits thelower IAE index. Consequently, the feedforward strategy dem-onstrates its advantage so it can be considered as a real alter-native to be implemented in internal combustion enginesoperating with gaseous fuels. As a further advantage of the

AirFuel

Mixer

A

TurboCharger

FT104

FC104

LS

TT103

TC103

104

Set PointSet Point

Fig. 10. Feedforward contro

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

proposed strategy, its real implementation only requires an extratemperature sensor and a slight modification on the internalcontrol logic.

Figs. 12 and 13 show that magnitude changes in disturbance donot affect the feedforward control strategy, which is a desiredcharacteristic. On the other hand, magnitude changes in distur-bances have a significant effect on the performance of the robustPID scheme and the modified PID strategy. To find out the influ-ence of the step changes in ambient temperature, temperatureprofiles were set using step changes of 4 �C, 6 �C and 8 �C, whichare depicted at Fig. 14. Results at Fig. 15 show that control loopresponses are similar, however, maximum values are lower due toa smaller step change. To highlight the effect of atmosphericpressure on the proposed strategies, Figs. 16 and 17 illustrate theresponses of the feedforward control strategy and the robust PIDstrategy, respectively. Results show higher deviations in temper-ature at low pressure due to a higher knocking tendency at theseconditions.

ftercoolerTT101

TT102

TC102

TC101

Set Point Set Point

RFB

l scheme implemented.

turbocharged internal combustion engines operating with gaseous

Fig. 12. Control strategies performance at P ¼ 11 psia.

Fig. 13. Control strategies performance at P ¼ 14.7 psia.

Fig. 15. Control strategies performance at P ¼ 14.7 psia and ambient temperature stepchanges of 4 �C, 6 �C y 8 �C.

J. Duarte et al. / Energy xxx (2014) 1e1110

6. Conclusions

In this paper, a modified control strategy for auto-ignition pre-vention in turbocharged internal combustion engines was imple-mented. Despite the level of correction of a feedforward strategywas not achieved, the current strategy showed significant im-provements in the control loop response. The following conclusionscan be drawn from the present study:

� The model for predicting temperature at the ignition point,developed and validated in this paper, can be used to implement

Fig. 14. Simulated ambient temperature profiles with different step changes.

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

predictive auto-ignition control strategies using robust PID. Thepredictive robust PID can be used along with the correctiveclassical PID to significantly improve auto-ignition control.

� The feedforward control strategy showed the best performancefor auto-ignition prevention in turbocharged internal combus-tion engines, however its implementation entails adding asensor and new control logic.

� It was verified that ambient temperature is a key disturbance inauto-ignition control. As intake air temperature rises, the auto-ignition tendency rises as well.

� Atmospheric pressure arises as a relevant parameter on internalcombustion engines behavior. As atmospheric pressure drops,the auto-ignition tendency rises.

� l0 tuning equations were the most robust equations for the caseof study. This was an unexpected result, since there are othertuning equations that exhibit a stable response for the pro-cesses. This result may be explained by the fact that dead timesin engines are short, which causes a significant increase ofprocess gain.

� The key parameter to measure fuel quality is the MethaneNumber (MN). A high-knocking-tendency fuel is obtained bycoal gasificationwith aMN around 30. On the other hand, a low-knocking-tendency fuel is a natural gas with a MN around 90.

Fig. 16. Feedforward control strategy performance at P ¼ 11 psia and P ¼ 14.7 psia.

turbocharged internal combustion engines operating with gaseous

Fig. 17. Robust PID strategy performance at P ¼ 11 psia and P ¼ 14.7 psia.

J. Duarte et al. / Energy xxx (2014) 1e11 11

Therefore, it is suggested to implement an additional controlloop by enriching low MN fuel with high MN fuel.

References

[1] Affleck W, Fish A. Knock: flame acceleration or spontaneous ignition?Combust Flame 1968;12:243e52.

[2] Saikaly K, Le Corre O, Rahmouni C, Truffet L. Preventive knock protectiontechnique for stationary si engines fuelled by natural gas. Fuel Process Technol2010;91:641e52.

[3] Bradley D, Kalghatgi G. Influence of autoignition delay time characteristics ofdifferent fuels on pressure waves and knock in reciprocating engines.Combust Flame 2009;156:2307e18.

[4] Bayraktar H. An experimental study on the performance parameters of anexperimental CI engine fueled with dieselemethanoledodecanol blends. Fuel2008;87:158e64.

Please cite this article in press as: Duarte J, et al., Auto-ignition control infuels, Energy (2014), http://dx.doi.org/10.1016/j.energy.2014.04.040

[5] Brecq G, Bellettre J, Tazerout M, Muller T. Knock prevention of CHP engines byaddition of N2 and CO2 to the natural gas fuel’. Appl Therm Eng 2003;23:1359e71.

[6] Amjad A, KhoshbakhiSaray R, Mahmoudi S, Rahimi A. Availability analysis ofn-heptane and natural gas blends combustion in HCCI engines. Energy2011;36:6900e9.

[7] Cho G, Jeong D, Moon G, Bae C. Controlled auto-ignition characteristics ofmethaneeair mixture in a rapid intake compression and expansion machine.Energy 2010;35:4184e91.

[8] Zhen X, Wang Y, Xu S, Zhu Y. Study of knock in a high compression ratiospark-ignition methanol engine by multi-dimensional simulation. Energy2012;50:150e9.

[9] Wang X, Zhang H, Yao B, Lei Y, Sun X, Wang D, et al. Experimental study onfactors affecting lean combustion limit of SI engine fueled with compressednatural gas and hydrogen blends. Energy 2012;38:58e65.

[10] Abdelaal MM, Rabee BA, Hegab AH. Effect of adding oxygen to the intake airon a dual-fuel engine performance, emissions, and knock tendency. Energy2013;61:612e20.

[11] Ortmann S, Glesner M. Development and implementation of a neural knockdetector using constructive learning methods. International Int J UncertainFuzz Knowl Based Syst 1998;6:127e37.

[12] Yue S, Li P. Automatic knock control system. In: Intelligent Control andAutomation, 2004. WCICA 2004. Fifth World Congress on, vol. 3. IEEE; 2004.pp. 2464e6.

[13] Yamasaki Y, Kanno M, Suzuki Y, Kaneko S. Development of an engine controlsystem using city gas and biogas fuel mixture. Appl Energy 2012;101:465e74.

[14] Oda T. Knocking control system and method for internal combustion engineusing multiple fuels; 2005. US Patent 6,951,202.

[15] Leone TG. Control for knock suppression fluid separator in a motor vehicle;2008. US Patent 7,389,751.

[16] Russell JD, Andri M. First and second spark plugs for improved combustioncontrol; 2009. US Patent 7,578,281.

[17] Alfaro VM, Vilanova R, Rica UDC, José S, Rica C. Sintonización de los con-troladores PID de 2GdL: desempeño, robustez y fragilidad. In: XIV CongresoLatinoamericano de Control Automatico (CLCA 2010), Santiago, Chile; 2010.pp. 267e72.

[18] Guzzella L, Onder C. Introduction to modeling and control of internal com-bustion engine systems. Springer; 2009.

[19] Heywood J. Internal combustion engine fundamentals, McGraw-Hill series inmechanical engineering. McGraw-Hill; 1988.

[20] Lapuerta M, Armas O, Molina S. Study of the compression cycle of a recipro-cating engine through the polytropic coefficient. Appl Therm Eng 2003;23:313e23.

[21] Åström K, Hägglund T. Advanced PID control. ISA-The Instrumentation, Sys-tems, and Automation Society; 2006.

[22] Smith C, Corripio A. Principles and practice of automatic process control.Wiley; 2006.

[23] Alfaro V. PID controllers fragility. ISA Trans 2007;46:555e9.[24] Saikaly K, Rousseau S, Rahmouni C, Le Corre O, Truffet L. Safe operating

conditions determination for stationary si gas engines. Fuel Process Technol2008;89:1169e79.

turbocharged internal combustion engines operating with gaseous