Process Systems Engineering, 9. Domain Engineering IOANNIS G. ECONOMOU, The Petroleum Institute, Department of Chemical Engineering, Abu Dhabi, United Arab Emirates EFSTRATIOS N. PISTIKOPOULOS, Imperial College London, Department of Chemical Engineering, London, United Kingdom JEN-PEI LIU, Imperial College London, Department of Chemical Engineering, London, United Kingdom YOSHIAKI KAWAJIRI, Georgia Institute of Technology, School of Chemical & Biomo- lecular Engineering, Georgia I, Atlanta, United States KRIST V. GERNAEY, Center for Process Engineering and Technology, Department of Chemical and Biochemical Engineering, Technical University of Denmark, Lyngby, Denmark JOHN M. WOODLEY, Center for Process Engineering and Technology, Department of Chemical and Biochemical Engineering, Technical University of Denmark, Lyngby, Denmark CONCEPCIO ´ N JIME ´ NEZ-GONZA ´ LEZ, GlaxoSmithKline, Research Triangle ParkNorth Carolina, United States RENE ´ BAN ˜ ARES-ALCA ´ NTARA, University of Oxford, Department of Engineering Science, Oxford, United Kingdom 1. Molecular Modeling and Simulation for Chemical Product and Process Design ......................... 2 1.1. Introduction ................... 2 1.2. Elementary Statistical Mechanics . . 3 1.3. Major Molecular Simulation Methods 3 1.3.1. Molecular Dynamics (MD) ........ 3 1.3.2. Metropolis Monte Carlo Simulation . . 4 1.4. Applications ................... 4 1.4.1. Pharmaceuticals................. 4 1.4.2. Polymer Membranes for Gas Separation 6 1.4.3. Ionic Liquids for Sustainable Chemical Processes ..................... 8 1.5. Conclusions ................... 9 2. Energy Systems Engineering ....... 10 2.1. Introduction ................... 10 2.2. Methods/Tools/Algorithm ........ 10 2.2.1. Superstructure-Based Modeling ..... 10 2.2.2. Mixed-Integer Programming (MIP) . . . 11 2.2.3. Multiobjective Optimization........ 11 2.2.4. Optimization under Uncertainty ..... 11 2.2.5. Life-Cycle Assessment ........... 12 2.3. Energy Systems Examples ........ 12 2.3.1. Example 1–Polygeneration Energy Systems ...................... 12 2.3.2. Example 2–Hydrogen Infrastructure Planning ...................... 15 2.3.3. Example 3–Energy Systems in Commercial Buildings ............ 17 2.4. Conclusions and Future Directions . 18 3. Pharmaceutical Processes ......... 19 3.1. Introduction ................... 19 3.2. Pharmaceutical Process Development and Operation ................. 20 3.2.1. Crystallization .................. 21 3.2.2. Chromatography ................ 22 3.3. Conclusion .................... 25 4. Biochemical Engineering .......... 25 4.1. Introduction ................... 25 4.2. Industrial Biotechnology Processes . 26 4.2.1. Fermentation Processes ........... 26 4.2.2. Microbial Catalysis .............. 27 4.2.3. Enzyme Processes ............... 27 4.3. Modeling of Bioprocesses ......... 28 4.3.1. Modeling of Bioprocesses– Mechanistic Models .............. 28 4.3.2. Modeling of Bioprocesses–Data- Driven Models ................. 29 4.4. The Role of Process Systems Engineering ................... 30 Ó 2012 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim 10.1002/14356007.o22_o13
Transcript
Process Systems Engineering,9. Domain Engineering
IOANNIS G. ECONOMOU,The Petroleum Institute, Department of Chemical Engineering,
Abu Dhabi, United Arab Emirates
EFSTRATIOS N. PISTIKOPOULOS, Imperial College London, Department of Chemical
Engineering, London, United Kingdom
JEN-PEI LIU, Imperial College London, Department of Chemical Engineering, London,
United Kingdom
YOSHIAKI KAWAJIRI, Georgia Institute of Technology, School of Chemical & Biomo-
lecular Engineering, Georgia I, Atlanta, United States
KRIST V. GERNAEY, Center for Process Engineering and Technology, Department of
Chemical and Biochemical Engineering, Technical University of Denmark, Lyngby,
Denmark
JOHN M. WOODLEY, Center for Process Engineering and Technology, Department of
Chemical and Biochemical Engineering, Technical University of Denmark, Lyngby,
Denmark
CONCEPCION JIMENEZ-GONZALEZ, GlaxoSmithKline, Research Triangle ParkNorth
Carolina, United States
RENEBANARES-ALCANTARA,University ofOxford,Department of Engineering Science,
and Conceptual Process Design . . . . 375.5. The Nature of Policy Formulation . . 385.6. The Nature of Sociotechnical Systems 395.7. Challenges for Modelers of
1. Molecular Modeling andSimulation for Chemical Productand Process Design
1.1. Introduction
Major chemical process industries (CPI) haveexperienced a substantial transformation in re-cent years worldwide due to an increased com-petition at a global level and a significant pres-sure from national governments and interna-tional organizations to develop new sustainableprocesses that consume significantly smallerquantities of energy and other natural resourcesand operate under zero (or close to zero) wasteproduction.
In parallel, major multinational CPI shiftedfrom low value commodity products to special-ty products of high added value where theunderlined materials are of considerably highercomplexity in terms of:
. Chemical structure
. Molecular and supramolecular architecture
. Micro- and mesostructure
. Performance in the end-use environment
These are nontrivial changes that require con-certed effort at different levels: basic research todevelop fundamental knowledge of physical phe-nomena, applied research to develop physicalmodels and parameters, and development workfor the generation of new processes that meet therequirements stated above. Accurate simulationand optimization methodologies are necessary atall length and time scales, from the submolecularlevel all the way to the macroscopic level wherethermodynamic and computational fluid mechan-ics models together with advanced numericalmethods are used in a concerted way. A consistenthierarchical development of physical models is ofoutmost importance.
This chapter refers to the development ofmolecularmodeling and simulationmethods forthe design of new chemical products and theimprovement of existing and design of newprocesses. Molecular simulation was intro-duced in the 1950s [1, 2] (! Molecular Model-ing, Chapter 3) as an abstract physical applica-tion for the primitive computers of the time andit evolved to a powerful engineering tool morethan 50 years later. At the same time, the needfor further development of simulation methods
2 Process Systems Engineering, 9. Domain Engineering
and physically accurate models remains as itwill be seen later in this chapter.
1.2. Elementary StatisticalMechanics
In statistical mechanics (! Molecular Dynam-ics (MD) Simulation), the properties of a bulkchemical system are calculated based on thecollective interactions between the moleculesthat make up the system. Almost all of thesystems of interest to process systems engineer-ing (PSE) followBoltzmann statistics and so thepartition function (Q) of a system of constantnumber of molecules (N) in a specific volume(V) and temperature (T) is [3]:
Q ¼ c
ZdpNdrNexp½�HðrNpN Þ/kBT � ð1Þ
where rN and pN denote the coordinates andmomenta of all N molecules, HðrNpNÞ is theHamiltonian of the system and c is a proportion-ality constant. For a system of N identical (indis-tinguishable) molecules: c ¼ 1/ðh3NN!Þ where his the Planck’s constant. The Hamiltonian pro-vides the total energy of the system as a functionof the coordinates and the momenta of the mo-lecules and is given as the sum of the kineticenergy (K) and the potential energy (U), so that:
HðrNpNÞ ¼Xi
p2i /ð2miÞþUðrN Þ ð2Þ
The potential energy U depends strongly onthe nature (complexity) of molecular interac-tions [3]. Intermolecular potentials range fromprimitive potentials (such as hard sphere, squarewell, etc.) to potentials of moderate complexity(such as Lennard–Jones, Stockmayer, etc.) andall the way to complex potentials that accountfor intra- and intermolecular interactions, manybody effects (polarizable potentials), etc.
From the partition function, one may calcu-late macroscopic thermodynamic propertiesusing the so-called bridge function, which forthe case of the constant NVT system (canonicalstatistical ensemble) is [3]:
AðNVTÞ ¼ �kBT ln QðNVTÞ ð3Þ
where A is the Helmholtz free energy. Unfortu-nately, the partition functionQ can be calculatedanalytically only for a very few simple systemsand significant approximations are needed
along the way in order that Equation (3) canlead to meaningful results [4].
Alternatively, one may calculate a macro-scopic property P as a statistical average overall microstates of the system, that is:
hPi ¼
ZdpNdrNPðrNpN Þexp½�HðrNpN Þ/kBT �Z
dpNdrNexp½�HðrNpNÞ/kBT �ð4Þ
Even then, calculation of hPi using bruteforce numerical integration requires extraordi-nary computing power. For example, for a 100molecule system using Simpson’s rule with just5 equidistant points along each coordinate axisone needs to evaluate the integrand of Equa-tion (4) at 10210 points [5].
A much more efficient approach is based onthe observation that some configurations of themolecular system aremuchmore important thanothers, so one should focus on sampling theseimportant configurations rather than randomconfigurations. This has been the basis of theso-called Metropolis Monte Carlo simulationmethod discussed briefly below.
1.3. Major Molecular SimulationMethods
1.3.1. Molecular Dynamics (MD)
In classical (Newtonian) mechanics, the follow-ing set of equations describes the evolution ofthe system over time [6]:
mir€i ¼ f i
fi ¼ �rri UðrNÞ i ¼ l; . . . ;N ð5Þ
where mi is the mass of molecule i and fi is theforce exerted on it. MD consists of solving theseN second order differential equations numeri-cally using a number of different methods de-veloped for this purpose. In thisway,MDallowsmonitoring of the evolution of the system withtime, and thus, time-dependent structure (poly-mer chain relaxation, etc.) and physical proper-ties (such as viscosity, diffusion coefficient,etc.) can be calculated.MD is usually performedin the microcanonical (NVE) statistical ensem-ble; however, the method has been extended tocanonical (NVT), isobaric-isothermal (NPT)and other statistical ensembles [6]. An impor-tant parameter concerning the robustness of the
Process Systems Engineering, 9. Domain Engineering 3
MD simulation is the time step used for thenumerical integration of the equations of mo-tion. For systems characterized by a relativelystiff potential (e.g., the case of chainmolecules),a typical time step is in the order of 0.1–1 fs. Anumber of advanced simulation techniques al-low the use of different time steps for differenttypes of forces. For example, a short time step isused for fast varying forces, such as bondstretching and bond angle bending and a longertime step is used for slowing varying forces,such as nonbonded intra- and intermolecularinteractions. Using state of the art computingfacilities, one may simulate a real system today(April 2010) for up to a fewmicro seconds. Thisis sufficient for the calculation of propertiessuch as chemical potential and self-diffusioncoefficient in systems that consist of small- andmedium-size molecules. For the calculationof dynamic properties of long chain molecules(e.g., polymers with a molecular mass higherthan 10000), alternativemethods are needed [7].
1.3.2. Metropolis Monte CarloSimulation
Metropolis Monte Carlo (MMC) simulation is astochastic method that allows efficient samplingof the multidimensional phase space of thesystem. In other words, this method allows‘‘jumps’’ in the phase space and so, no real timemonitoring of the system is possible. In MMC,the different states of the system are visited witha probability proportional to the Boltzmannfactor of the energyof the system [5]. The systemgoes from one configuration (state) to the nextconfiguration (state) based on different types ofmoves that satisfy microscopic reversibility andpreserve the macroscopic properties of the sys-tem that are set constant. In this way, MCsimulations are performed in the NVT, grandcanonical (mVT),NPT andmanyother statisticalensembles, depending on the system (pure fluidor mixture) and conditions (one phase, two, ormore phases, etc.) examined. In a typical NVTMMC simulation, particles are displaced ran-domly one at a time within the simulation boxand the new configuration is accepted or rejectedaccording to the Boltzmann factor of the energydifference between the two states, that is:
pNVT ¼ min½1; expð�DU/ðkBTÞÞ� ð6Þ
where DU ¼ U(new) � U(old) is the energydifference between the old and the new config-uration. Thermodynamic properties are calcu-lated based on Equation (4). Additional movesin the NPT, mVT and other ensembles includevolume fluctuation, random insertion and dele-tion of particles and so on, and acceptancecriteria are modified accordingly.
Amajor breakthrough inmolecular simulationwas the development of the Gibbs ensemble MC(GEMC) method which allows the simultaneoussimulation of several phases in equilibrium (e.g.,vapor–liquid equilibrium) [8]. The method hasbeen successfully applied to pure components,binary and multicomponent mixtures, and differ-ent types of phase equilibria (vapor–liquid, liq-uid–liquid, vapor–liquid–liquid, etc.) [9].
Development of efficient elementary movesfor long chain molecules has also been a veryactive area of research over the last two decades.A broad range of moves has been proposed forthe efficient relaxation of chain tails, internalsegments, branch points, and even moves thatallow exchange of molecular segments betweentwo different chain molecules [10]. A combina-tion of these moves allows today accurate sim-ulation of polymer melts with a molecular massof the order of several thousand.
1.4. Applications
1.4.1. Pharmaceuticals
Hydration energy plays a significant role inbiological processes and is currently an impor-tant predictive index formolecule availability inthe pharmaceutical industry. During the com-plex process of driving a molecule from anaqueous phase to a target protein active site,the driving force is directly related to the differ-ence between the hydration energy of the drugand the drug–protein association energy. More-over, desolvation of both protein site and drugmolecule occurs during this binding process,and recently developed docking/scoring meth-ods estimate this desolvation correction basedon free energy calculations. For some drugmolecules, solvation free energies may be esti-mated experimentally from concentration mea-surements in two-phase systems. However, inmost cases this is not possible and so accuratetheoretical or computational approaches are
4 Process Systems Engineering, 9. Domain Engineering
needed. Molecular simulation using realisticpotential models is able to provide accurateestimate of the property of interest and at thesame time a quantitative insight regardingthe molecular mechanisms associated with thehydration.
Recently, a simple thermodynamic cyclewasproposed to calculate the hydration Gibbs freeenergy, DhydG (P,T), of complex solute mole-cules [11]:
Solute ðwaterÞ ������!Dwater G Dummy ðwaterÞDhydG" #DdummyG
Solute ðvacuumÞ ������!Dvacuum G Dummy ðvacuumÞ
where, DwaterG is the Gibbs energy associatedwith the mutation of the solute molecules intomolecules of dummy atoms (atoms that do notinteract with their environment) in water,DvacuumG is the Gibbs energy associated withthe same process in vacuum, and finallyDdummyG can be seen as the hypothetical hydra-tion Gibbs energy of dummy species. In prac-tice, these atoms have no intermolecular elec-trostatic or van der Waals interactions, buttheir intramolecular bonded interactions are thesame as in the solute atoms. As a consequence,DdummyG is equal to zero and one can writethe following equation for the thermodynamiccycle:
The term DvacuumG contains only contribu-tions from intramolecular nonbonded interac-tions (forces acting between atoms in the samemolecule separated by more than three bonds),which exist in the solute molecule but not in thedummy molecule. The thermodynamic integra-tion approach was used by [11] to calculateDhydG of barbituric acid and various substitutedbarbiturates at 298 K and 0.1 MPa. In Figure 1,DhydG and the Lennard–Jones contribution to it,DLJG, are presented as a function of molecularmass for various mono- and di-substitutedbarbiturates.
Using the same methodology, DhydG of vari-ous n-alkanes [12] and polar compounds [13]were calculated by [11]. An extensive evalua-tion of three widely used molecular models(force fields) to describe the polar compounds,namely TraPPE, Gromos and OPLS-AA, wasperformed. In all cases, the MSCP/E model wasused for water. An overview of the predictionsobtained from the different force fields for thepolar compounds and a comparison with exper-imental data is shown in Figure 2. For therelatively simple polarmolecules, such asmeth-anol and propanol, all force field predictions arein good agreement with experimental data. Forthe case of more complex multifunctional mo-lecules, including acetylsalicylic acid (ASA)and ibuprofen (IBP) that are of interest to phar-
Figure 1. MD predictions of DhydG and of DLJG against molecular mass for various mono- and di-substituted barbiturates at298 KBA ¼ barbituric acid, MB ¼ methyl barbiturate, EB ¼ ethyl barbiturate, iPB ¼ isopropyl barbiturate, BAR ¼ barbital,PRO ¼ probarbital, BUT ¼ butethal, PEN ¼ pentobarbital
Process Systems Engineering, 9. Domain Engineering 5
maceutical industry, MD predictions from dif-ferent force fields deviate and the agreementwith experiments is less satisfactory.
1.4.2. Polymer Membranes for GasSeparation
Polymeric membranes (either glassy or elasto-meric) are used widely for separation of mix-tures in chemical industry, medical applica-tions, etc. A major physical property in sucha process is the permeability (P) of component iin the polymer membrane defined as the productof the solubility (S) and diffusivity (D), so that:
P ¼ SD ð8Þ
Separation of a binary mixture of compo-nents i and j (where i is typically the mostpermeable of the two components) by a givenpolymer membrane is characterized by the idealseparation factor, which is the ratio of perme-abilities for components i and j according to theexpression:
aidij ¼ Pi
Pj¼ Si
Sj
� �Di
Dj
� �ð9Þ
where the ratios aSij ¼ Si/Sj and aD
ij ¼ Di/Dj
represent the solubility selectivity and the dif-fusivity selectivity, respectively. In rubberypolymers aD
ij is less than unity, while aSij � 1,
so ideal separation factor is governed by sel-ectivity of sorption. Polydimethylsiloxane(PDMS) is a widely used polymer membrane
and so ideal separation factors for various binarygas and liquid mixtures have been measured.The separation factor for n-C4H10/CH4 mixtureis used widely as a benchmark for hydrocarbonmixture separation capability of a given mem-brane material. A new atomistic force field wasdeveloped for PDMS that accounts for bondstretching, bond angle bending, dihedral angletorsion, and nonbonded intra- and intermolecu-lar interactions [14]. For the nonbonded inter-actions, the Lennard–Jones potential for short-range van der Waals repulsive and attractiveinteractions togetherwith a long-rangeCoulom-bic potential were used. The model was shownto predict accurately the thermodynamic prop-erties of polymer melts over a wide temperatureand pressure range [14]. It was further used forpolymer–gas mixture simulations. In Figure 3,pure gas n-C4H10/CH4 solubility, diffusivity,and permeability selectivities in the range of273–400 K calculated from MD simulationstogether with experimental data from [15] areshown.
The solubility of hydrocarbons in the poly-mer matrix was based on the Widom’s testparticle insertion method that allows accuratecalculation of the excess chemical potential ofthe solute in the solvent. MD simulation of thepolymer matrix for 5–10 ns followed by severalhundred thousands of solutemolecule insertionsin each polymer configuration (this is a relative-ly fast post-processing calculation) provides anaccurate estimate of the solubility.
Figure 2. Experimental data and MD predictions of DhydG of various polar molecules using different force fields at 298 Ka) Methanol; b) Propanol; c) Ethylamine; d) Acetone; e) Acetic acid; f) IBP; g) ASA; h) Benzoic acid
6 Process Systems Engineering, 9. Domain Engineering
For the calculation of diffusion coefficient,significant longer MD simulations, on the orderof 100 ns, are needed in order to ensure that thehydrocarbon molecules diffusing through thepolymer matrix reach the normal diffusing(Fickian) regime [16]. In this case, the diffusioncoefficient is calculated from the mean squaredisplacement of the hydrocarbon molecules,based on Einstein equation.
Figure 3 reveals that solubility selectivitydecreases significantly as temperature increaseswhile diffusivity selectivity increases but with asmaller rate. Finally, the ideal separation factorfollows closely the trend exhibited by solubilityselectivity. In all cases, MD predictions are inexcellent agreement with experiments [15] overthe entire temperature range.
For the accurate design of a polymer mem-brane for the separation of a real mixture, mix-ture permeability data are needed. It is oftenassumed that in rubbery polymers penetrantspermeate independently of one another. How-ever, this behavior needs to be confirmed for agiven system. Recent experimental data for the
n-C4H10–CH4 mixture in PDMS showed anincrease in CH4 solubility in the presence ofn-C4H10 in the polymer. On the other hand, onlya weak influence of CH4 on n-C4H10 solubilitywas reported. In Figure 4, experimental data andMD predictions are shown for the infinite dilu-tion solubility coefficient of CH4 in the PDMS–n-C4H10 mixture at 300 and 450 K. Simulation
Figure 3. n-C4H10–CH4 mixture behavior in PDMS as a function of temperatureA) Solubility (S) selectivity; B) Diffusivity (D) selectivity; C) Permeability (P) selectivityOpen symbols are experimental data [15] and closed symbols are MD predictions
Figure 4. Mixed gas CH4 solubility in PDMS at 300 and450 K as a function of n-C4H10 weight fraction in PDMSa) Experimental data [15] at 300 K (open points); b) MDpredictions at 300 K (closed points); c) MD predictions at450 K (closed points)
Process Systems Engineering, 9. Domain Engineering 7
results are consistent and in good agreementwith experimental measurements [15].
Finally, the diffusion coefficients of a mix-ture of CH4 and n-C4H10 in PDMS at ambientconditions are shown in Figure 5 and are com-pared to pure gas diffusion calculations. Clear-ly, CH4moleculesmove faster in the presence ofn-C4H10molecules in PDMSmatrix than in purepolymer. The same behavior is observed forn-C4H10 in the presence of CH4 molecules. Thepresence of a second penetrant species swellsthe polymer matrix resulting in an increase inthe diffusion coefficient of the first penetrant.The swelling behavior of PDMS in the presenceof mixed gases and the consequent increase indiffusivity and permeability coefficients of thecorresponding gases has also been reportedexperimentally by many investigators [15, 17].
1.4.3. Ionic Liquids for SustainableChemical Processes
Ionic liquids (ILs) (! Ionic Liquids) have re-ceived much attention for use as environmen-tally benign reaction and separation media. ILsare molten salts with melting points close toroom temperature. Their most remarkable prop-erty is that their vapor pressure is negligiblysmall, so that ILs are nonvolatile, nonflammableand odorless. Other characteristics of ILs in-clude a wide liquid temperature range, a highthermal and electrochemical stability, a highionic conductivity and good solvency proper-ties. In principle, ILs can be tailored for a
specific application by the right choice of cationand anion.
It is expected that ILs may revolutionize thechemical process industry in the years tocome [18]. For example, they are increasinglyused as novel processing media in combinationwith supercritical CO2. Due to the negligiblevapor pressure, it is possible to extract organicproducts from ILs using supercritical CO2 with-out any contamination by the IL. Despite thewealth of experimental data available, moredata are needed for process design, and theirexperimental determination is often difficult,time-consuming and expensive. Therefore, it ishighly desirable to develop predictive methodsfor estimating the relevant thermodynamic,phase equilibrium, and transport properties. Atthe molecular level, early molecular simulationstudies focused on the development of accurateforce fields and validation towards the predic-tion of structure and thermodynamic propertiesof ILs inmelt. Recently, thesemodels were usedfor the calculation of thermodynamic and trans-port properties of IL melts and mixtures.
A powerful approach toward the develop-ment of accurate force fields is to start from thesubatomic level with quantum mechanics cal-culations. A recent example refers to ab initiodensity functional theory (DFT) calculations(! Process Intensification, 1. Fundamentalsand Molecular Level, Section 2.2.3.1) per-formed on isolated IL molecules ([bmimþ][Tf2N
-], [hmimþ][Tf2N-], and [omimþ][Tf2N
-])in order to evaluate the minimum energy struc-ture and calculate charge density distribution ofthe molecule [19]. In Figure 6, schematic repre-sentation of DFT results are shown. DFT resultswere used for the development of a realisticatomistic force field that was used subsequentlyfor MD simulations.
ILs have very long characteristic relaxationtimes and so longMD simulations are needed inorder to obtain accurate thermodynamic anddynamic predictions. MD simulations of up to50 ns on bulk ILs at various temperatures andpressures were performed by [19]. Volumetric,dynamic, and transport properties together withstructure properties were calculated. Molecularconformations from MD simulations were ingood agreement with DFT results. The IL con-figurations generated from MD were used sub-sequently for the calculation of the excess
Figure 5. Diffusion coefficient of pure andmixed n-alkanesin PDMS at ambient conditions. Solid circles correspond topure CH4 and n-C4H10 diffusion coefficient in PDMS. Opensymbols correspond to n-alkanes in mixturea) CH4 mixed with 2% n-C4H10 in PDMS; b) CH4 mixedwith 10% n-C4H10 in PDMS; c) n-C4H10 mixed with 1%CH4 in PDMS
8 Process Systems Engineering, 9. Domain Engineering
chemical potential, and thus solubility, of CO2
in the IL using the Widom’s test particle inser-tion method. In all cases, excellent agreementwith experimental data was obtained. Represen-tative results concerning IL self-diffusion coef-ficient and CO2 solubility in [bmimþ][Tf2N
-]are shown in Figure 7 and Figure 8, respectively.
1.5. Conclusions
Molecular simulation is a mature computationaltool that can be used reliably by material scien-tists and chemical engineers for industrial prod-uct and process design. Highly robust and effi-cient computer codes have been developed by
Figure 6. Relative electronic energies DE of the isolated ion pairs optimized at B3LYP/6-311þG* level Energies are given inkJ/mol. Conformer III has been arbitrarily chosen as reference, thus relative energies are calculated by DEi¼Ei� E3. Energieswith zero point energy corrections are given in parentheses
Figure 7. Experimental data (markers) and MD simulationresults (lines) for the self-diffusion coefficient of variouscations at 0.1 MPaa) C4 mim; b) C6 mim; c) C8 mim
Figure 8. Henry’s law constant of CO2 in [bmimþ][Tf2N�]
a) Experimental data (blue squares and line);b) Molecularsimulation predictions from TraPPE (red squares); c) EPM2model (open diamonds) for CO2
Process Systems Engineering, 9. Domain Engineering 9
major academic and government laboratoriesworldwide and are freely available for researchand development purposes [20–23]. In addition,specialized computational chemistry andmodel-ing software companies offer state-of-the-artuser-friendly interfaces to support the rathercomplex computer codes [24, 25]. In this respect,molecular simulations can be performed almostroutinely by nonexperts.
Despite the above, significant challenges stillexist: Accurate force-fields are of outmost im-portance in order that simulation results resem-ble the real systems. Although significant ad-vances have been made in recent years, there isstill need for developments of intra- and inter-molecular interaction models for highly com-plex chemical compounds.
As molecular simulation matures, the com-plexity of the problems where it is appliedincreases. Very often, the detailed atomisticrepresentation of interactions is not necessaryanymore and a more coarse-grained representa-tion becomes more suitable. In such cases, asystematic hierarchical approach is needed inorder to parameterize a model consistently.Finally, one should recognize the fact that com-parison of simulation predictions against exper-imental data at various time and length scales isalways necessary in order to validate the modeland the methodology used.
2. Energy Systems Engineering
2.1. Introduction
Excessive energy consumption and conse-quent greenhouse gas (GHG) emissions havebecome two major crucial global issues, andthis situation is most likely to continue in thenext couple of years to come. Driven by thisurgent situation, technologies which canfacilitate a smooth transition from existingenergy systems to more advanced ones arereceiving more and more serious attention.However, although there already exist manytechnical options, they usually differ greatlyfrom one another in many aspects, and theyare often treated separately by their own tech-nical or political groups.
The concept of energy systems engineeringas an integrated approach for the energy sys-
tems of the future is introduced by [26].Energy systems engineering provides a meth-odological framework to address the complexenergy and environmental problems by anintegrated systematic approach which ac-counts complexities of very different scales,ranging from technology, plant, to energysupply chain, and megasystem. Energy sys-tems engineering employs systems-basedrepresentations and methods, such as super-structure-based modeling, mixed-integer pro-gramming (MIP), multiobjective optimiza-tion, optimization under uncertainty (see !Process Systems Engineering, 3. Mathemati-cal Programming (Optmization) and! Math-ematics in Chemical Engineering, Chap. 10,and life-cycle assessment ! Waste, 2. Life-Cycle Assessment). These methodologieshave been applied in energy systems of verydifferent nature and scale, including polyge-neration energy systems, urban energy sys-tems, hydrogen infrastructure, oil and gasproduction, wind turbine, electric power in-dustry, carbon dioxide capture and sequestra-tion, and distillation columns [27, 28].
2.2. Methods/Tools/Algorithm
2.2.1. Superstructure-Based Modeling
Superstructure-based modeling is an approachto simultaneously determine the optimal con-figuration of a process and its optimal operatingconditions viamathematical programming [29].It was first proposed to address process synthesisissues in heat-exchanger networks (HEN) [30],and widely used in process design thereafter,and it is regarded as one of the most significantaccomplishments in PSE [31].
Superstructure-based modeling has beenwidely used in a broad range of fields. Thesefields include heat-exchange networks [30],separation and distillation [32–34], reactor net-works [35, 36], water usage and treatment net-work [37], and energy systems [38, 39].
Superstructure-based modeling usually in-volves discrete decision making, e.g., inclusionof a certain type of reactor or not. Simultaneousmodeling of discrete decisions and continuousterms is usually implemented via MIP.
10 Process Systems Engineering, 9. Domain Engineering
2.2.2. Mixed-IntegerProgramming (MIP)
An optimization model with both integer andcontinuous variables is denoted as a MIP prob-lem [40] see ! Process Systems Engineering,3. Mathematical Programming (Optmization)and ! Mathematics in Chemical Engineering,Chap. 10. Integer variables in MIP problemsusually refer to 0–1 variables, also known asbinary variables, only, due to the fact that anyinteger variable can be represented in terms of aset of binary variables.
MIP is widely used in PSE. Typical applica-tions are, e.g., superstructure-based modeling,facility location and allocation problems, sched-uling problems. A canonical form of a MIPproblem is presented as follows:
minx;y
f ðx; yÞs:t: hðx; yÞ ¼ 0
gðx; yÞ � 0
x � 0; x 2 X � Rn
y 2 f0; 1gqx;y ð10Þ
where x is a vector of n continuous variables,and y is a vector of q 0–1 variables.
Depending on specific forms of the objectivefunction f, equality constraints h, and inequalityconstraints g, MIP problems can be classified intotwocategories:mixed-integerlinearprogramming(MILP) problems, where the objective functionand all constraints are linear, and mixed-integernonlinear programming (MINLP) problems,where either the objective function or some con-straints are nonlinear. MINLP problems can befurther classified as convex MINLP problems,where the objective function is a convex functionand the feasible region is a convex region, andnonconvex MINLP problems, where either theobjective function is a nonconvex function or thefeasible region is a nonconvex region.
ThemostcommonlyusedalgorithmforsolvingMILPproblems is branch andboundmethod [41].It has a huge number of varieties. Two commonlyused algorithms for solvingMINLP problems aregeneralized benders decomposition (GBD) [42]andouter approximation (OA) [43], bothofwhichhave a large amount of varieties.
2.2.3. Multiobjective Optimization
Multiobjective optimization, or multicriteriaoptimization, is to simultaneously optimize a
problem according to two or more (conflicting)criteria subject to certain constraints (! EnergyManagement in Chemical Industry, Section3.1). Multiobjective optimization is suitable tobe applied to a problem where trade-offs existamongst its objective functions and optimaldecisions should be made in the presence ofthese trade-offs. Multiobjective optimization iswidely used in various fields, including productand process design, supply chain design, andenergy systems engineering. A common multi-objective optimization problem involved withenergy system design is to maximize profitabil-ity and minimize environmental impactssimultaneously.
A generic mathematical from of a multiobjec-tive optimization problem is presented as follows:
minx;y
U
f1ðx; yÞf2ðx; yÞ� � �fnðx; yÞ
8>>>><>>>>:
s:t: hðx; yÞ ¼ 0
gðx; yÞ � 0
x � 0; x 2 X � Rn
y 2 f0; 1gq ð11Þ
where x is a vector of n continuous variables,and y is a vector of q 0–1 variables.
The target of solving a multiobjective opti-mization problem is to obtain the utility functionU, where n scalar objective functions are to beoptimized simultaneously [44]. Usually, someconflicts exist amongst the objective functions.However, if there are no conflicts, then a singlesolution can be obtained where every objectivefunction attains its optimum. In this case, opti-mizing the objective functions simultaneouslyor separately arrive at the same optimal solution.To avoid such simple cases, multiobjectiveoptimization problems discussed hereafter al-ways involve conflicting objective functions.
Typical algorithms for solving multiobjec-tive optimization problems are parametricprogramming [45] and the e-constraint method[46, 47].
2.2.4. Optimization under Uncertainty
Uncertainty is inevitable and unpredictable inprocess planning and design over a long-termhorizon. Because of the very nature of these
Process Systems Engineering, 9. Domain Engineering 11
tasks, many parameters obtained at the planningor design phase are subject to considerablevariability and cannot be predicted with a cer-tain degree of accuracy. Optimization underuncertainty takes the impact of uncertain para-meters into consideration at the planning anddesign stage thus improves a plan or design interms of both feasibility and operability (see! Process Systems Engineering, 3. Mathemat-ical Programming (Optmization)).
2.2.5. Life-Cycle Assessment
Life-cycle assessment (LCA), also known aslife-cycle analysis, is to evaluate and quantifythe environmental impacts of a certain productor production procedure caused by its existence.The definition and method of product LCA isdescribed in detail in (! Waste, 2. Life-CycleAssessment).
Depending on the boundaries of a systemwhere LCA is applied, LCA can be classifiedinto the following four categories:
. Cradle-to-gate. It accounts for the environ-mental impacts of a product produced at allstages before it is sent to the gate of a factory.These stages usually consist of mining, pre-processing, and transportation.
. Cradle-to-grave. It accounts for the environ-mental impacts of a product in its entire lifetime, from manufacture up to disposal phase.
. Cradle-to-cradle. It accounts for the environ-mental impacts of a product in a recyclingprocess, from the production of a product of acertain type of material to the production ofanother product of the same material.
. Well-to-wheel. It is a specific type of LCAwidely used in fuel and transportation LCA,accounting for the energy consumption andemissions production from exploration to fi-nal consumption. According to the particularresearch interest, it can be further divided intowell-to-tank and tank-to-wheel stages, orwell-to-station and station-to-wheel stages.
Depending on the means an LCA impactfactor is evaluated, LCA can be classified intothe following two categories:
. Inventory-based LCA. Most conventionalLCA methods belong to this category. These
methods start from a breakdown of a systemunder study into fundamental componentsand processes, then extract inventory data ofthese components and processes from a hugeinventory database which contains inventorydata of all primary products and processes,then multiply these inventory data with theircapacity within the system under study andsum them up to provide the LCA indicator.
. Economic input–output LCA. This methodestimates materials and energy requirementsand environmental emissions in activities ofan economy. It uses information of industrytransactions, i.e., purchases of materials byone industry from another industry, and infor-mation of direct environmental emissions ofindustries, to evaluate the entire environmen-tal impacts of a system or process under study.
2.3. Energy Systems Examples
The aforementioned energy systems engineer-ing methodologies can greatly facilitate theplanning or design of energy systems of differ-ent types and scales, at different levels, fromdifferent aspects, and according to differentcriteria. Some of thesemethodologies have beensuccessfully applied in energy systems of verydifferent nature and scale, and have been sum-marized as follows:
. Polygeneration energy systems [38, 48–51]
. Urban energy systems [52]
. Hydrogen energy systems [53–55]
. Energy systems in commercial buildings [56,57]
. Electric power industry [58]
. Pulp and paper industry [59–63]
. Oil and gas production [64]
. Wind turbines [65]
. Carbon dioxide capture and sequestration [66,67]
. Separation and distillation [68]
2.3.1. Example 1–Polygeneration EnergySystems
A polygeneration energy system is a multiinputand multioutput energy system that coproduceselectricity and synthetic liquid fuels. Processdesign of a polygeneration energy system in-
12 Process Systems Engineering, 9. Domain Engineering
volves several typical energy systems engineer-ing issues, as follows:
. A polygeneration energy system is a verycomplex system which comprises many unitsand pieces of equipment. For each of theseunits and pieces of equipment, there usuallyexist many alternative technologies or typesof equipment. Making the optimal selectionfrom the many alternatives remains achallenge.
. As public concern over fast increasing GHGemissions grows, environmental impact of anenergy system has become an important de-sign criterion. Designing a polygenerationenergy system according to multiple designcriteria (economic, environmental, etc.) posesanother challenge.
. A polygeneration energy system usually hasan operating horizon of several decades, overwhich there exist many inevitable and unpre-
dictable uncertainties. Design of a polyge-neration energy system under uncertaintymakes the task further complicated.
A modeling and optimization frameworkfor the optimal process design of polygenera-tion energy systems is proposed by [48–51],based on the energy systems engineering ap-proaches presented in the previous section.First, a superstructure representation of apolygeneration energy system is constructed,as shown in Figure 9, where a polygenerationenergy system is divided into many functionalblocks. For each functional block, all alterna-tive technologies and types of equipment areincluded in the superstructure representation,thus all possible types of process design arecaptured.
Based on the superstructure representation, aMINLP design problem is developed in thefollowing form:
Figure 9. Superstructure representation of a polygeneration energy system (CCS ¼ Combined combustion system; GCS ¼Gasification chamber and syngas scrubber; HRSG ¼ Heat recovery steam generator; ASU ¼ Air separation unit
Process Systems Engineering, 9. Domain Engineering 13
miny;d;x
f ðy; d; xÞ
s:t: hdcðy; dÞ ¼ 0
gdcðy; dÞ � 0
hocðy; d; xÞ ¼ 0
gocðy; d; xÞ � 0
d 2 Rm; x 2 Rn; y 2 f0; 1gq ð12Þ
where
. Binary design variables are denoted as y,which represent the selection (or not) of tech-nologies or types of equipment for each func-tional block.
. Continuous design variables are denoted as d,which represent the capacities of the func-tional blocks.
. Continuous operational variables are denotedas x, which represent quantitative decisions tobe made at the operational stage, e.g., flow-rates, stream compositions and the like.
. Equality design constraints are denoted as hdc,which involve design variables only, e.g.,evaluation of initial capital costs.
. Inequality design constraints are denoted asgdc, which involve design variables only, e.g.,logical relations between different functionalblocks equality operational constraints hoc,which involve design and operational vari-ables, e.g., mass and energy balances.
In theMINLP design problem, there could bemore than one objective function, i.e., designcriterion. Here, both the economic and environ-mental behavior of a polygeneration energysystem is evaluated. Net present value (NPV)is selected to be the economic design criterion,which comprises the initial capital costs and thediscounted profit obtained over the entire oper-ating horizon. A cradle-to-gate GHG emissionsindicator is selected to the environmental designcriterion, mainly comprising three parts:
. GHG emissions produced within the processduring operation
. GHG emissions produced throughout mining,extraction, and other processing stages offeedstocks
. GHG emissions produced during equipmentproduction and plant construction
On obtaining these two objective functions, amulti-objective MINLP problem is formed asfollows:
miny;d;x
U
(f1ðy; d; xÞ ¼ �NPV
f2ðy; d; xÞ ¼ GHG
s:t: hdcðy; dÞ ¼ 0
gdcðy; dÞ � 0
hocðy; d; xÞ ¼ 0
gocðy; d; xÞ � 0
d 2 Rm; x 2 Rn; y 2 f0; 1gq ð13Þ
where f1 is the objective function representingthe NPV, and f2 is the objective function repre-senting the GHG emissions.
Equation (13) is solved using the e-constraintmethod. Optimal results are presented on apareto curve, as shown in Figure 10. For thisexample, there exist 18 different combinationsof technologies, but only four of them appear onthe pareto curve, according to different econom-ic and environmental design criteria. Each pointon the curve represents a different process de-sign. A decision-maker can thus pick up anypoint from the curve according to their specificinterest or requirements.
In Equation (13), all time-variant parametersare considered as piecewise constant functionsover the operation horizon, which is discretizedinto several time intervals. However, due to thevery nature of the long-term operation horizon,uncertainty is almost inevitable at the designstage, e.g., due to external factors, such asmarketdemands for products, prices of feedstocks andproducts. Here, all uncertain parameters can bepresented as random variables following givenprobability distribution functions p(x).
By incorporating the uncertainty into theMINLP design problem, the following two-stage stochastic programming problem results:
miny;d
fdðy; dÞþEu 2 Q½fsðy; d; uÞ�s:t: hdcðy; dÞ ¼ 0
gdcðy; dÞ � 0
d 2 Rm; y 2 f0; 1gqwith :
fsðy; d; uÞ ¼ minx
fsðy; d; x; uÞs:t: hocðy; d; x; uÞ ¼ 0
gocðy; d; x; uÞ � 0
x 2 Rn; u 2 Q ð14Þ
where the objective function is split into adeterministic term fd representing decisions atthe design stage, and the expectation of a sto-
14 Process Systems Engineering, 9. Domain Engineering
chastic term fs which depends on the realizationof uncertain parameters u at the operation stage.Discrete variables y and continuous variables dare ‘‘here-and-now’’ (design) variables whichshould be decided at the first-stage problembefore the realizations of uncertain parametersu occur, and x is a vector of ‘‘wait-and-see’’(operational) variables which can be decided attime interval t of the second-stage problemwhere all uncertain parameters have been ob-served. In the second-stage problem, therecourse term based on a specific realization ofuncertain parameters is optimized and corre-sponding corrective actions in terms of values of
x are made. Equation (14) is solved using adecomposition-based solution strategy.
2.3.2. Example 2–Hydrogen Infrastruc-ture Planning
Energy systems engineering methodologieshave been applied in hydrogen infrastructureplanning [39, 69]. The problem under study isillustrated in Figure 11: given a specific regionwhere several potential production sites andmarkets (city as shown in the Figure 11) areavailable, obtain the optimal infrastructure
Figure 10. Pareto curve for polygeneration energy systems designa) H–CH–G–GTH; b) Q–CQ–L–GTH; c) RC–CRC–G–GTH; d) Q–CQ–G–GTH
Figure 11. Illustrative representation of a hydrogen infrastructure planning problem
Process Systems Engineering, 9. Domain Engineering 15
which connects the production sites to marketsvia a supply chain from primary feedstocks,central production, distribution, forecourt re-fueling, to the final product over a long-termplanning horizon.
This approach addresses the following issuesinvolved in hydrogen infrastructure planning:
. Planning over a long-term future horizon
. Geological site allocation
. Representing the state of existing infrastruc-ture, especially the natural gas distributionnetwork, electricity grid, and existing hydro-gen production facilities
. All types of available primary feedstocks,production, distribution, and forecourt refuel-ing technologies
. Trade-offs between large-scale centralizedproduction and small-scale distributedproduction
. Transitions from one type of supply chainstructure to another over time
. Planning according to both economic andenvironmental performance indicators
A superstructure representation of themodeling framework is shown in Figure 12. Itcaptures all possible types of primary feed-stocks, production sites, production technolo-gies, distribution technologies, forecourt refill-ing technologies, and potential markets, andgives the optimal planning scheme over theentire future planning horizon.
Based on this modeling framework, a multi-objective optimization was conducted whereNPV was selected as an economic objective anda LCA-based environmental impact factor as anenvironmental objective. A pareto frontier com-prising the full range of trade-offs between theeconomic and environmental objectives was
Figure 12. A superstructure representation of the modeling framework for hydrogen infrastructure planninga) Gasoline equivalent WTW (well-to-wheel) emissionsSMR-LIQ ¼ Manufacturing of liquid hydrogen via steam methane reformingSMR-GAS ¼ Manufacturing of gaseous hydrogen via steam methane reformingNG COMP ¼ compression of natural gas (NG)GAS-LIQ ¼ Manufacturing of liquid hydrogen via gasificationGAS-GAS ¼ Manufacturing of gaseous hydrogen via gasificationELC-LIQ ¼ Manufacturing of liquid hydrogen via electrolysis of waterELC-GAS ¼ Manufacturing of gaseous hydrogen via electrolysis of waterCNG ¼ Compressed natural gasLIQ ¼ Liquid hydrogenGAS ¼ Gaseous hydrogenNG pipe ¼ Natural gas pipelineH2 PIPE ¼ Hydrogen pipelineSMR ¼ Onsite hydrogen production via steam methane reformingELC-N ¼ Onsite hydrogen production via electrolysis of water using nonrenewable electricityELC-R ¼ Onsite hydrogen production via electrolysis of water using renewable electricityELC-U ¼ Onsite hydrogen production via electrolysis of water using nuclear electricity
16 Process Systems Engineering, 9. Domain Engineering
obtained, shown in Figure 13. Any point on thepareto frontier represents an infrastructure designwith specific economic and environmental per-formances, and decision-makers can pick up anypoint from this curve as the final design accordingto their own specific interest and preference.
2.3.3. Example 3–Energy Systems inCommercial Buildings
The applications of energy systems engineeringmethodologies in polygeneration energy sys-tems and hydrogen infrastructure planning fo-cus primarily on the energy production side.However, energy systems engineering is notconfined within the scope of energy production.It can also be applied to model and optimize theenergy consumption within a process or system.Next, its applications in design of the energysystems in commercial buildings [49] are pre-sented to illustrate its potential applications onthe energy consumption side.
The energy system in a commercial buildingusually comprises both an energy consumptionsection and an energy supply section. Energydemands usually come from requirements forlighting, HVAC (heating, ventilating, and airconditioning), and refrigeration. The energysupply is usually obtained from grid electricity,district heat, and on-site energy generation, e.g.,distributed power generation and boilers. Majorissues to be addressed at the design stage aresummarized as follows:
. Selection of technologies. For each type ofenergy demands, several types of technolo-gies or types of equipment are usually avail-able. Selecting the optimal combination ofthem may become a challenging problemwhen facing with too many choices. Thisissue could be further complicated when in-volved with other design issues, e.g., integra-tion between energy consumption and energyproduction sectors.
. Integration. Integration amongst differentenergy consumption sectors within a systemcan reduce the entire energy demand of thesystem. For example, heat produced in therefrigeration sector of a supermarket couldbe collected to heat the aisle space, other-wise an extra amount of energy is requiredto meet the heating demand. The integrationissue could become more complicated whenon-site production technologies are alsoinvolved.
. Building design. From an energy saving view-point, building design should also be involvedat the design phase. For example, sizing andpositioning of windows could be consideredtogether with the lighting requirement of abuild to minimize it.
. GHG emissions. From an LCA point of view,emissions from a commercial building comefrom two sources. One source is the emissionsproduced over the entire operation period, andthe other one is the emissions produced inmanufacturing and transporting equipmentand construction materials. Emissions from
Process Systems Engineering, 9. Domain Engineering 17
both categories should be considered at thedesign phase to give an overall environmentalimpact indicator.
To address these issues, a superstructurerepresentation of the energy system in a com-mercial building is firstly constructed, as shownin Figure 14. It comprises an energy supplysection, an energy conversion section, and anenergy savings section. The function of theenergy supply section is to provide electricityand heat for the entire energy system. Theenergy conversion section converts electricityand heat obtained from the energy supply sec-tion to all energy demand tasks, such as refrig-eration, lighting, ventilation, bakery, and spaceheating. The energy savings section furtherinvolves available types of energy savings tech-nologies, such as night blind and weir screen forthe refrigeration subsystem.
Based on the superstructure representation, amultiobjective MILP problem is formed andsolved to obtain the pareto curve, as shown inFigure 15. A decision-maker can pick up anypoint from the pareto curve according to theirspecific design criteria or interest. Once a designpoint is selected (e.g., A, B, C, or D), the systemconfiguration behind it can be obtained directlyfrom the model results.
2.4. Conclusions and FutureDirections
The introduced methodologies of energy sys-tems engineering cooperate with each other andprovide a systematic solution strategy for theplanning and design issues involved with anyenergy system. These methodologies are illus-trated via their applications in a simple exampleof polygeneration energy systems design. Itshows that energy systems engineering is oftremendous importance to guide the transitionfrom our existing generation of energy systemsto a more energy efficient and environmentallybenign one. It is certain that research in this fieldwill continue and prosper. Some recommenda-tions for future research directions are summa-rized as follows:
. The genericmodeling and optimizationmeth-odologies presented in this section can serveas a starting point, and more methodologieswhich are suitable for energy systems couldbe added into the scope of energy systemsengineering. This would certainly extend itsapplicable fields and enhance its capability.
. Modelingatamicrolevelcouldbeexplored.Themethodologies introduced here enable model-ing at strategic planning and process design
Figure 14. Superstructure representation of the energy system in a commercial buildingPE ¼ primary energy; GE ¼ grid electricity; DH ¼ district heat; OEG ¼ on-site energy generation; E ¼ electricity; h ¼ heat;R¼ refrigeration;L¼ lighting;V¼ ventilation;B¼ bakery; SH¼ space heating; P¼ production;ES¼ energy saving;D¼ demand
18 Process Systems Engineering, 9. Domain Engineering
levels, which can be regarded as modeling atmegalevel and macrolevel, respectively.Modeling at a microlevel, e.g., at the molecularlevel for biodiesel production, would give amuch better insight to these systems.
. The generic modeling and optimization meth-odologies introduced in this section could beused in the control field. Firstly, it canbe used inan integrated design and control scheme whereboth operational and control requirements aretaken into consideration at the design stage.Secondly, the frameworkcanbe alsoused in thecontext of model predictive control.
. Applications in energy value chain modelingand optimization. For example, bioenergy isexpected to play an important role in theongoing transition from conventional energysystem to amore sustainable and environmen-tally benign one. There also have been manycontroversies around bioenergy about its ca-pability to ameliorate energy security andclimate change, concerning its life-cyclegreen-house gas (GHG) emissions and com-petition on land use with food crops. Themodeling and optimization methodologiesdeveloped in this framework could be usedto guide the planning and design of a bioe-nergy value chain in terms of analyzing andquantifying net profit of bioenergy, producing
methodologies and tools for the optimal de-sign of bioenergy value chains with the righttechnologies at the right scale, and providingpolicy suggestions to direct the developmentof bioenergy.
3. Pharmaceutical Processes
3.1. Introduction
For pharmaceutical companies, drug develop-ment requires a huge investment (! Pharma-ceuticals, General Survey, Chap. 4). A studyestimates that the cost to bring a single new drugto the market costs over $800 million [70]. Theclinical period of a new drug is complex, anddivided into three phases. In phase I, normally asmall number of healthy volunteers are tested tofind safe dosages. In phase II, the drug candidateis given to a large number of patients. Phase IIItrials are more extensive, often carried out atmore than one clinical research centers. Figure 1shows the probability of entering the next phasefrom the previous phase. As can be seen, onlyone third of the drug candidates—which havealready passed the drug discovery stage andpreclinical trials—reaches the final stage, phaseIII. Furthermore, nearly 40% of the drugs that
Figure 15. Pareto frontier for the energy system design in a commercial building
Process Systems Engineering, 9. Domain Engineering 19
pass phase III do not get marketing approval.Therefore, wise decision-making at an earlierstage can reduce the development cost dramati-cally, and methods/tools that support such de-cisions play a crucial role in drug development.
The three clinical trial phases are closelyrelated to the development of drug manufactur-ing, and often proceed concurrently; the optimalproduction process design is explored during theclinical trials, and scale-up is performed simulta-neously. In phase I, a small-scale pilot facility–inthe order of one-hundredth of production scale–isusually sufficient, where as in Phase III a pilotplant of one-tenth production scale is often need-ed to supply a sufficient amount of the new drug,which is administered for this large-scale clinicaltrial [71]. Therefore, as the clinical tests proceed,the production volume goes up, and efficientmanufacturing that can supply such an amountof the drug candidate becomes more critical.
Process systems engineering (PSE) techni-ques have been applied to these unique issues inthe pharmaceutical industry to aid high-leveldecision-making. The complexity of theseproblems is recognized as one of the challengesby the PSE research community [72–74]. Someexamples of recent research activities includeportfolio management for drug research anddevelopment by multistage stochastic optimiza-tion [75], development of informatics infrastruc-ture for product development and manufactur-ing [76], supply chain optimization [77], model-based design/analysis of PAT systems [78],and resource investment and scheduling fornew drug product development [79].
3.2. Pharmaceutical ProcessDevelopment and Operation
Traditional pharmaceutical manufacturing most-ly relied on extensive laboratory testing for quali-
ty assurance. In this traditional approach, productquality was achieved predominantly by restrict-ing flexibility in the manufacturing process [80].Manufacturing procedures are treated as beingfrozen and changes in the process must be man-aged through regulatory submissions. As a result,many production processes are designed andoperated inefficiently, and relatively little efforthas been devoted to innovate and improve them.In some cases, the amount of product waste as aresult ofmistakes inmanufacturingwas as high as50%of the productmanufactured [81]. This led tosignificantly higher costs and even delays of newdrug development.
In 2002, the US Food and Drug Administra-tion (FDA) launched the process analyticaltechnology (PAT) initiative to challenge thehesitancy to innovate pharmaceutical manu-facturing [82]. This initiative, a paradigmchange of the FDA to inspect and approvepharmaceutical processes, promotes better un-derstanding of drug production processes. Here,the term analytical should be interpreted broad-ly; it includes ‘‘chemical, physical microbiolog-ical, mathematical, and risk analysis conductedin an integrated manner’’ [83]. For example,PAT encourages fundamental process under-standing for on-line or real time process controlto ensure product quality by reducing variabilityin the process [82]. Application reports of PATof specific processes include batch crystalliza-tion [84] (! Crystallization and Precipitation,Section 5.6), freeze drying [85], and fermenta-tion [86, 87]. A software for design of PATsystems has also been developed by [88].
Another change in pharmaceutical processdevelopment and operation was brought by theintroduction of quality by design (QbD) [89]. InQbD, the concept of Design Space has given asignificant impact to the pharmaceutical processdevelopment. The design space is defined as
Figure 16. Drug development phases [70, 71]a) Phase I; b) Phase II; c) Phase III; d) Registration launch
20 Process Systems Engineering, 9. Domain Engineering
‘‘the multidimensional combination and inter-action of input variables and process parametersthat have been demonstrated to provide assur-ance of quality’’ [89], which is initially deter-mined during product development and re-ported to the regulatory agency. Change ofprocess parameter values within the designspace usually does not require a regulatorypost-approval process [89]. This enables flexi-ble changes of manufacturing operations andwider applications of process automation,which had been severely limited previously inthis industry.
With such unprecedented changes, a largernumber of advanced PSE techniques are begin-ning to be applied to pharmaceutical processes.Some specific PSE technologies are mentionedin the documents from the regulatory agen-cies [83, 89], and studies in response to suchdemands have been carried out including pro-cess control [82, 86], process monitoring[78, 85, 88, 90], and development of multivari-ate design tools [91]. Furthermore, computer-aided process design and simulation tools arebeginning to be used in pharmaceutical processdevelopment [92].
Another important trend in the pharmaceuti-cal manufacturing is a revisit to continuousprocesses. Traditionally, pharmaceutical pro-duction relies on batch processing due to thelow production volume. However, understand-ing the process dynamics and dealing with thebatch-to-batch fluctuation often become thebottleneck for efficient production. Thus, thereliability of continuous processes, in additionto the higher productivity, is attracting pharma-ceutical manufacturers.
3.2.1. Crystallization
Although crystallization is widely used in thepharmaceutical industry, it remains one of themost poorly understood processes. In particular,controlling the size and shape of crystals is a bigchallenge, which requires substantial experi-mental, modeling, and computational efforts(! Crystallization and Precipitation).
The crystal size in a crystallizer can be char-acterized by the crystal size distribution (CSD).Estimating the CSD accurately is crucial forprocess development of crystallization. Among
the CSD estimation techniques, the sieve analy-sis is a primary offline technique, but this relieson good sampling which cannot be always real-ized. On the other hand, on-line measurementtechniques have been attracting attention in re-cent years. In particular, focused beam reflec-tance measurement (FBRM) has been employedinmany crystallization studies. FBRM is a probewhich can be installed directly in a crystallizereliminating the need for sampling (Fig. 17). Thison-line measurement technique obtains thechord length distribution (CLD). The challengehere is to find the relationship between the CLDand CSD. This problem has been recognized bythe PSE community, and techniques based onprojections onto convex sets [93], Monte Carlosimulation [94], and principal component anal-ysis [95] have been proposed.
Crystallization consists of two major me-chanisms, nucleation and growth (! Crystalli-zation and Precipitation, Section 4.1 and !Crystallization and Precipitation, Section 4.2).Nucleation is an event where the solute mole-cules in the solvent gather and form nuclei. Therate of nucleation can be modeled as a functionof the concentration of the API in the motherliquor C, and the the equilibrium concentrationat temperature T , C�ðTÞ:
Rb ¼ kbexp�16pg3 M
r
� �2
3k3T3 ln CC�ðTÞ
� �2
0B@
1CA ð15Þ
where g is the interfacial tension, and r thedensity, M the molecular mass, and k theBoltzmann’s constant. On the other hand,growth is the subsequent step where the size ofthe nuclei increases. The growth rate GðL; tÞ isusually expressed in terms of the degrees ofsupersaturation C�C�ðTÞ [96, 97]:
G ¼ kg;0exp � Ea
RT
� �ðC�C�ðTÞÞg ð16Þ
The mass balance can be described by thepopulation balance equation (PBE). If breakageand agglomeration is ignored, and it is assumedthat the crystal size can be represented by thecrystal length L, the PBE for a batch crystallizeris given by [98]:
qnðL; tÞqt
þ qðGðL; tÞnðL; tÞÞqL
¼ 0 ð17Þ
Process Systems Engineering, 9. Domain Engineering 21
where nðL; tÞ is the particle density. The bound-ary condition at L ¼ 0 is given by the nucleationrate:
nðt; 0Þ ¼ Rb
Gðt; 0Þ ; ð18Þ
It can be noted that this is a hyperbolic partialdifferential equation (PDE), which requires finediscretization. Therefore, an efficient solutiontechnique must be employed to reduce thecomputational cost to solve this equation, suchas a flux limiter and space-time conservationelement method [98, 99]. In addition to thecomputational difficulty associated with thePBE, there are many model parametersðg;Ea; kg;0; kbÞwhichmust be identified for eachapplication. These parameters can be estimatedby advanced computational techniques such asdesign of experiments (DOE) [100] and param-eter estimation techniques [101, 102]. In esti-mating these model parameters, it has beenreported that utilizing on-line measurement ofthe crystal size increases the reliability [102,103]. After obtaining the crystallization model,
model-based control and optimization [104,105] can be applied. These recent researchactivities on crystallization are influenced andencouraged in particular by the PAT initiative.A generic model-based frameowrk for crystal-lization modeling, control, and monitoring hasrecently been developed by [106].
3.2.2. Chromatography
Chromatography is often the only choice forseparation of thermally sensitive compoundssuch as proteins, or structurally similar com-pounds; it requires only a slight difference in theaffinity for adsorbent particles. Figure 3(a)shows a traditional batch chromatographic pro-cess, where the purified products A, B, andC arefractionated at the outlet ðx ¼ LÞ of the column.The feed mixture is supplied at the top of thecolumn which is packed with solid adsorbentparticles. Then the mixture is eluted with des-orbent, which can be water or organic solvent
Figure 17. Batch crystallization of paracetamol [84]A) Lab-scale crystallizer; B) Paracetamol crystals; C) Experimental setup of FBRMa) Computer; b) Temperature probe; c) Agitation; d) Cooling fluid (water); e) Batch crystallizer; f) FBRM
22 Process Systems Engineering, 9. Domain Engineering
(s). Because of the difference in the affinity forthe adsorbent, the migration speeds of the com-ponents are different and the components sepa-rate from each other as they move towards thebottom. Due to its batchwise operation, thethroughput, or the feed processing rate, is oftensmall. Furthermore, it consumes a large amountof desorbent which dilutes the product, and thusthe evaporation cost can be high.
To overcome these drawbacks, simulatedmoving bed (SMB) chromatography (! Chro-matographic Reactors, Chap. 2) has been devel-oped by the Universal Oil Products (UOP) in the1960’s, and applied to the separation of xyleneisomers. After their patent expired, its applica-tion areas expanded into sugar separations, inparticular for the production of high-fructosecorn syrup. In the past decade, applications inthe pharmaceutical industry have been gainingattention. In particular, separation of enantio-mers has been found to be one of the mosteffective applications. Figure 19 shows an SMB
process for active pharmaceutical ingredients(API) purification. An SMB system consists ofmultiple columns connected to each other mak-ing a circulation loop (Fig. 18B). Between everycolumn, there are inlet ports for feed and des-orbent streams, as well as outlet ports for extractand raffinate streams. The feed and desorbentare supplied continuously and at the same timeextract and raffinate are drawn continuouslythrough the ports. These four inlet/outlet portsare switched simultaneously at a regular intervalin the direction of the liquid flow. This systemdoes not reach a steady state but a cyclic steadystate (CSS), where the concentration profileschange dynamically, but the profiles of bothliquid and solid phase at the beginning of acycle are identical to those at the end of thecycle.
Established mathematical models for chro-matography can be found in literature. Compre-hensive reviews are given by [107, 108]. One ofthe modeling approaches is the linear driving
Figure 18. Chromatographic separation processesA) Batch process; a) Chromatographic columnB) Simulated moving bed (SMB) process; a) Direction of valve switching
Process Systems Engineering, 9. Domain Engineering 23
force (LDF) model, where the mass transferbetween the liquid and the adsorbent particlesis described by a linear relationship character-ized by themass-transfer coefficient. In the LDFmodel, the mass balance equations in the liquidand adsorbent phases are given by the followingtwo PDEs, respectively:
ebqCiðx; tÞ
qtþð1�ebÞ qqiðx; tÞqt
þuqCiðx; tÞ
qx¼ 0 ð19Þ
ð1�ebÞ qqiðx; tÞqt¼ Kapp iðCiðx; tÞ�Ceq
n ðx; tÞÞ ð20Þ
where eb is the void fraction, Ciðx; tÞ is theconcentration in the liquid phase of componenti, qiðx; tÞ is the concentration in the solid phase,u is the superficial liquid velocity,Ceq
i ðx; tÞ is theequilibrium concentration in the liquid phase,and Kapp i is the mass-transfer coefficient, re-spectively. In addition to the mass balanceequations, the isotherm that describes the equi-librium between the liquid and adsorbent con-centrations must be specified. One of the mostwidely employed isotherms is the Langmuirisotherm:
qi x; tð Þ ¼ aiCeqi ðx; tÞ
1þbiCeqi ðx; tÞ : ð21Þ
To find the optimal design and operation ofSMB and batch chromatography, several ap-
proaches have been developed based on PDE-constrained numerical optimization of the rig-orous dynamic chromatographic model.Stochastic optimization approaches as well asNewton-based approaches have been pro-posed [109–111]. Alternatively, this problemcan be formulated as a multiobjective optimiza-tion problem to evaluate more than one objec-tives, such as throughput, purity, and desorbentconsumption, and obtain the pareto optimalset [112, 113].
For SMB chromatography, many new im-proved operations have been proposed to en-hance the performance. Traditional SMB sys-tems keep the liquid velocities constant duringa step, and then switch the four inlet/outletstreams at the same time. In PowerFeed sys-tems, however, the velocities become time-variant. Optimization of PowerFeed can beformulated as an optimal control problem [79,114]. Moreover, VARICOL systems performasynchronous valve switching, where the fourinlet/outlet ports are switched independently,not simultaneously [115]. A comprehensivesummary of modifications to SMB can befound elsewhere [111, 116]. Since the numberof operating parameters is larger in suchsystems, finding the operating and designparameters in such improved SMBs relieson a systematic and efficient optimizationapproach [117].
Due to the complex dynamics of the chro-matography, model-based feedback controlof a batch process is very challenging. Fur-thermore, the poor observability of chro-matographic processes makes automatic con-trol more difficult; although the temporalconcentration profiles at the outlet of thecolumn can be observed by a detector, thespatial concentration profiles inside the col-umns cannot be monitored directly. A fewrecent control strategies that tackle thesechallenges for batch processes can be foundelsewhere [118–120]. For SMB chromatogra-phy, there has been considerable advance inrecent years in model-based control techni-ques including linear and nonlinear modelpredictive controllers [121–123]. Further-more, several experimental techniques forstable monitoring of the purity have beenproposed which improve the controllerperformance [124].
Figure 19. Simulated moving bed chromatographic pro-cess for API purification (courtesy of AMPAC FineChemicals, Rancho Cordova, CA, USA)
24 Process Systems Engineering, 9. Domain Engineering
3.3. Conclusion
There are many exciting research challenges inthe pharmaceutical industry where PSE techni-ques can contribute. This is being accelerated bythe recent changes in drug manufacturing initi-ated by the regulatory agencies. The introduc-tions of PAT and QbD have brought a signifi-cant impact to the PSE community, and current-ly many advanced PSE technologies such asprocess modeling, control, optimization, anddecision-making support which can meet theunique demands from this industry are beingdeveloped.
4. Biochemical Engineering
4.1. Introduction
Process systems engineering (PSE) offers manytools for the chemical engineer. Today, forexample, modeling, simulation, and processevaluation tools are routinely applied to designand optimization problems in the bulk chemi-cals and fuels sector, where small process im-provements yield significant economic returns.In recent years there have been an increasingnumber of bioprocesses implemented and theseprovide a different type of challenge for PSE.This chapter has a focus specifically on biopro-cesses, and especially on the use of PSE tools forthe design and development of bioprocesses.
Bioprocesses have found application in theproduction of high-value products such as phar-maceuticals (and their intermediates). The pro-cess engineering emphasis in these cases is onrapid process implementation, rather than opti-mized development [125, 126]. However, inrecent years bioprocesses have also been in-creasingly applied to bigger volume productssuch as fine chemicals, bulk chemicals, and
biofuels, which are the new sectors of industrial(also called ‘‘white’’) biotechnology. Todaythere are significant new opportunities in whitebiotechnology for processes based on renew-able resources such as biomass and clean pro-cesses with reduced solvent inventories, renew-able catalysts, and mild conditions for reactionand separation [127]. In addition to the directprocess improvements, bioprocesses have alsofrequently been justified on the basis that theyare processes with potentially lower environ-mental impact than their chemical syntheticcounterparts. The main synthetic operations inbioprocesses include fermentation, microbialcatalysis, and enzyme catalysis (see Table 1).Downstream options are dependent on thenature of the product (i.e., macromolecular orlow molecular mass compounds (‘‘small mole-cules’’). Small molecules are frequently pro-cessed in a similar way to other chemical pro-ducts, although dilute aqueous solutions bringspecific problems which need to be addressed,both from the viewpoint of process optimizationand the environmental footprint. For instance,the downstream processing of some small mol-ecule bioprocesses could include large amountsof organic solvents for extraction from aqueoussolutions. In these cases, the organic solventsrequire processing, recycle, control and ulti-mately safe disposal. Macromolecules requiremore specialist operations such as filtration orchromatography. However, in all cases the mo-lecules are frequently sensitive to extremes ofpH and temperature, placing specific restric-tions and constraints on processing methods.Biocatalyst recovery (frequently for recycle)also necessitates filtration and centrifugation.
It is clear from the above that a range ofquestions need to be addressed when imple-menting new processes, and specifically biopro-cesses. For example:When should a bioprocess,rather than a chemical process be implemented?
Table 1. Process features of the three major types of bioprocess for chemicals production (I) represents options for immobilized enzyme
Catalyst recovery and/or recycle potentially via a continuous process recycle via filtration via immobilization
Reactor options stirred tank
bubble column
stirred tank
bubble column
stirred tank
packed bed (I)
fluidized bed (I)
membrane
Process Systems Engineering, 9. Domain Engineering 25
If a bioprocess is to be implemented, can theexisting infrastructure (feedstock, utilities, andplant) be used? How can a process plant beadapted for different biomass sources availablein different geographical regions? What is theoptimum biorefinery? What options exist forprocess integration? What are the environmen-tal, health, and safety issues of bioprocesses incomparison with chemical processes? What isthe environmental footprint of a bioprocesscompared with its chemical counterpart? Howcan bioprocesses be designed to maximize pro-cess efficiency, minimize environmental im-pact, as well as maximize sustainability?Whichvariables can be measured on a bioprocess?How can control contribute to more efficientoperation of the bioprocess?
Many of these questions can currently beaddressed qualitatively, but to have real valueit is necessary to assess the questions on aquantitative basis. In order to achieve this ef-fectively therefore, computer-based tools arerequired. In addition, models are needed. Overthe last decades, PSE has already developedmany of the appropriate tools, and those toolsoften rely on models. Nevertheless, some fur-ther developments are required. For example, inthe case of bioprocesses an extra option avail-able to the engineer is the improvement of thecatalyst itself. This requires models which takeinto account catalyst properties. In addition, oneemerging consideration is measuring the rela-tive sustainability of processes, and one can seelife cycle inventory and assessment (LCI/A)modeling tools and methods as a logical exten-sion of PSE. LCI/Amethodologies allow for theestimation of environmental impact across theentire life cycle of a process or product. LCI/Aestimations rely heavily on the characterizationof the process and its unit operations usingmodeling and simulation techniques, which arekey competences within PSE.
This chapter will first introduce the threedifferent types of bioprocesses that are of in-dustrial relevance. PSE methods and tools canbe applied in the bioprocess design phase–usingprocess models and design PSE tools–as well asto improve the process operation–using processmonitoring and control methods and tools. Pro-cess monitoring, with focus on applicationson bioprocesses, is already summarized in !Process Systems Engineering, 5. Process Dy-
namics, Control,Monitoring, and Identification,Chap. 2 andwill therefore not be discussed here.Process control issues are highlightedin ! Process Systems Engineering, 5. ProcessDynamics, Control, Monitoring, and Identifica-tion, Chap. 2 and! Process Systems Engineer-ing, 5. Process Dynamics, Control, Monitoring,and Identification, Chap. 3 for continuous andbatch/fed-batch processes, respectively. Thechapter only pays limited attention to data-driv-en modeling, where applications of data-drivenmodels in the area of soft sensors is highlighted,since this is one of the areas where probablydata-driven models will become increasinglyimportant in the future. The main focus of thischapter, however, will be on mechanistic mod-els, and on the current and future use of thosemodels within the design of sustainable biopro-cesses. This is likely to be one of the mostpromising R&D areas at this moment, wherePSE methods and tools will contribute tremen-dously to embed design principles during theearly stages of process development and design.Future trends are highlighted as well, whererelevant.
4.2. Industrial BiotechnologyProcesses
Before describing the PSE tools and some oftheir applications in more detail, it is importantto highlight the most relevant industrial biopro-cesses. Three major types of bioprocess can beidentified dependent on the nature of the biocat-alyst. These are outlined in the following sec-tions and the key process features are summa-rized in Table 1.
4.2.1. Fermentation Processes
For a significant number of chemicals, the use offermentation has become a standard alternativeto fossil-based feedstocks and technology. Nev-ertheless the possibility of growing microbialcells on a variety of sugars (derived from re-newable biomass) has reinvigorated interest inthis area. The consequence is that fermentationat a large scalewill becomemore common in thefuture chemical industry. Many different typesof fermentation process (using different strains
26 Process Systems Engineering, 9. Domain Engineering
to produce different products) can take place inthe same process plant which is a significantadvantage. The plant is relatively simple and thechallenges lie in adequate mixing (sometimeswith materials having complex rheology), suit-able oxygen input (for aerobic processes), andprocess control. Downstream, the separationprocess depends on the product, but will nearlyalways need to avoid high temperatures andextremes of pH. The solvent in such a processis water, meaning that the dilute product streamcombined with the presence of many otherproducts presents a significant process engi-neering challenge. Both, large molecular massand low molecular mass products can be madeby fermentation. The processes either focus onlow molecular mass products which can subse-quently be used as platform chemicals or bio-fuels or high molecular mass compounds suchas enzymes (for application in a range of indus-tries, including detergents, textiles, and foodingredients) or therapeutic proteins.
4.2.2. Microbial Catalysis
In fermentation, by definition, the catalyst isgrowing during the process. This means thatsome of the reactant (or substrate)will inevitablybe diverted from the product towards the cata-lyst, lowering the yield. An alternative (fornongrowth associated products) is to grow thecells first and subsequently carry out the reactionto increase the yield. This also enables thepossibility of growth and reaction on differentsubstrates (reactants) or under different condi-tions (such as temperature or pH) in each stage.Likewise the optimal cell concentration for con-version can be selected [128] after growth and
suitable media for effective product recoverychosen. For processes requiring oxygen it canbe highly important to select the optimal cellconcentration in order to avoid oxygen-transferlimitations. Several tools are now available forevaluating the oxygen supply issues in suchreactions [129, 130]. The three potential routesusingmicrobial catalysis are shown in Figure 20.
4.2.3. Enzyme Processes
The presence of so many products at the end offermentation or microbial catalysis is a conse-quence of the complexity of cells, where manyenzymes catalyze reactions giving a spectrumofproducts as well as decreasing the yield of thedesired product on the reactant. An alternative,for short pathways, is to isolate the enzymes andthen immobilize them on a solid support orbehind a membrane or via aggregation, suchthat they are large enough and have the rightproperties to be recycled (like a heterogeneouscatalyst). In this way a yield of product oncatalyst of around 5–10 t/kg immobilized bio-catalyst can be achieved, which typically issufficient to enable commercial implementationat an industrial scale. Such an approach has beenwidely used to assist in the synthesis of high-value compounds such as pharmaceuticals anda more limited number of well-known lower-value products such as high fructose corn syrup(HFCS). Many of these processes have alsobeen modeled [131]. Enzymatic processes canalso be carried out using soluble enzymes (i.e.,nonimmobilized), although these present chal-lenges in terms of separating and recycling thecatalysts (enzymes) when compared with theimmobilized enzyme processes.
Figure 20. Alternative process scenarios for use of microbial cells for nongrowth associated biocatalysisa) Route 1: Combined fermentation and microbial catalysis; b) Route 2: Fermentation separated from microbial catalysis; c)Route 3: Fermentation separated from microbial catalysis with intermediate processing to change catalyst concentration
Process Systems Engineering, 9. Domain Engineering 27
Traditionally, single enzymes were used forcatalysis. However, multiple enzyme mixturesare nowadays becoming attractive for catalyz-ing the production of several compounds at anindustrial scale. A classification of multien-zyme-catalyzed processes was proposed recent-ly [132], including reaction and process con-siderations for mathematical modeling of mul-tienzyme processes operated in a single reactor.
4.3. Modeling of Bioprocesses
Mechanistic and empirical (data-driven)model-ing approaches complement each other. Biopro-cesses are usually represented by a combinationof both where mechanistic models graduallyreplace empirical models as more knowledgeabout a process or a unit operation becomesavailable [133]. Empirical models representinput–output relationships in a data set withoutrequiring detailed knowledge of an underlyingmechanism. Usually, an empirical model canonly accurately predict conditions representedby the data set that was used to build the model.Empirical models are useful in a process controlcontext, where software sensors often rely onempirical models for the prediction of variablesthat are not measured directly due to on-linemeasurement difficulty or excessive sensor costas will be explained in more detail in the fol-lowing sections.
4.3.1. Modeling of Bioprocesses–Mechanistic Models
Mechanistic process models (! Biotechnolo-gy, 5. Monitoring and Modeling of Biopro-cesses, Section 5.3) for fermentation and bio-catalytic processes are developed based onmass, heat, and momentum balances, supple-mented with appropriate mathematical formu-lation of mechanisms (e.g., kinetic expressionsto reflect process dynamics). Specifically for thedescription of bioprocesses, the kinetic expres-sions themselves are often empirical, providinga simplified and idealized view of a complexbiological mechanism. Unlike empirical mod-els, mechanistic models usually offer betterextrapolation capabilities, which is critical ina process modeling context, where one is often
interested in investigating the process perfor-mance under different operating conditions onthe basis of simulations with a dynamic model.
Sufficient process knowledge is a necessityin order to optimize the design or the operationof a bioprocess. A mechanistic model capturesthat process knowledge in a structuredway [134]. The model therefore has great valuein planning experiments, or in determiningwhich critical process variables necessitate tigh-ter control [135–137].
Within the fermentation field, developmentof mechanistic models has a long history. Earlywork involving quantitative descriptions of bac-terial growth dates back to the 19th century, anda broad spectrum of modeling techniquesare available today [138]. For more detailedreviews on models and model types, a numberof key publications can be suggested on fermen-tation models [139–142], on specific modelingapproaches for enzyme production kinet-ics [143], on mechanistic model studies forbiocatalytic processes [144], and on modelingcellulase kinetics [145]. Finally, the opinionarticle of [146] highlights both the history andthe future of mathematical modeling in bio-chemical engineering.
Assuming a homogeneous reactor environ-ment, a generally accepted classification ofmechanistic models of cell populations is pre-sented in Figure 21 [139, 146].Moving betweenmodels in Figure 21 is determined by the as-sumptions behind the mechanistic model. Forexample, if the assumption of a homogeneousreactor environment does not hold, then a dis-tributedmodel is needed (i.e., amodelwhere notonly time, but also space (1-, 2- or 3D), forms anindependent variable). Unsegregated modelsare common, and rely on an average cell de-scription. Unstructured unsegregated modelsare the simplest models. They use a singlevariable to describe the biomass [134, 142, 147].
Unsegregated structured models form animportant class. The distinguishing feature ofthese models is that they describe the biomassas consisting of several variables (such asNADH, precursors, metabolites, ATP, bio-mass), and have been used for modeling com-plex processes. An example is a structuredmodel of yeast intracellular metabolism [141].Morphologically structured models [141, 148]distinguish between different regions of the
28 Process Systems Engineering, 9. Domain Engineering
hyphal elements of filamentous fungi, and werespecifically developed to describe growth of thisimportant class of production organisms.
Segregated models consider individual cells.They were developed in recognition of the factthat cells in a population–a pure culture–aredifferent, and are most often formulated as apopulation balance model (PBM). An unstruc-tured segregated model characterizes cells byone distributed property (i.e., cell size or age ofindividual cells [149]) without considering in-tracellular composition. Obviously, structuredsegregated models are more complex, since thedistribution of one or more intracellular vari-ables is also considered. Solving the resultingmultidimensional PBM is difficult, unless theintracellular state can be captured with just afew variables [140]. One alternative to PBMs iscell ensemble modeling [140, 150], where theparameters of a single cell model are random-ized to generate a cell population. Despite thefact that segregated models are more complex,advances in data collectionmethodsmean that itbecomes more and more relevant to developsegregated models in order to increase ourunderstanding of the complex interactions be-tween individual cells [151].
4.3.2. Modeling of Bioprocesses–Data-Driven Models
Data-driven models are extremely useful in aprocess monitoring and control context (see! Process Systems Engineering, Process Sys-tems Engineering, 5. Process Dynamics,Control, Monitoring, and Identification, Chap.
3), especially for handling multivariate datawhich are increasingly becoming available on-line. In this contribution, the focus is on the useof data-driven models in software sensors. Theestimation of the concentration of analytes ofinterest using so-called ‘‘software sensors’’ is inmany cases a fruitful alternative to direct (oranalyte-specific)measurements using, for exam-ple, chromatographic and spectroscopic meth-ods, as illustrated in a recent review on methodsthat allow the on-line measurement of the cellmass concentration [152] ! Biotechnology, 5.Monitoring andModeling of Bioprocesses, Sec-tion 3.2: software sensorswere considered as onealternative approach that can compete withmethods such as dielectric spectroscopy, OD,IR spectroscopy, and fluorescence for in situmeasurement of the cell mass concentration.
Software sensors can in fact be divided intothree classes:
. Software sensors based on stoichiometry,elementary balances and first-principlesmodels
. Software sensors based on data-drivenmodel-ing methods, i.e. black-box approaches
. Hybrid software sensors,which arepartly basedon first-principles, but supplemented withblack-box approaches for parts of the systemthat are not sufficiently well understood
Only the second class of software sensorswill be highlighted here, and these seek toexploit correlations between the variables inthe process, without seeking any mechanisticexplanation for the observed correlations.Meth-ods such as artificial neural networks (ANN)
Figure 21. Classification of mechanistic bioprocess models
Process Systems Engineering, 9. Domain Engineering 29
and chemometric modeling techniques, forexample, partial least squares (PLS), belonghere [153–156]. In a recent publication, a mov-ing window principal component analysis(MW-PCA) method was successfully used toidentify phase changes in several different in-dustrially relevant batch processes [157]. TheMW-PCA method was solely based on changesin the statistical properties of on-line data, suchas pH, dissolved oxygen concentration, agitatorspeed and concentrations of CO2 and O2 in theexhaust gas, and appears to be useful for anycase where even slight changes in process prop-erties must be identified.
The main message is that black-box methodsare certainly useful, on condition that the user isawareof the limitationsof themethods.Moreover,some of these methods have the advantage thatthey can be applied rather easily in practice. Soft-ware for building chemometric models, to nameone example, is available from several softwarevendors or can be downloaded as freeware.
4.4. The Role of Process SystemsEngineering
4.4.1. Evaluation of Process Options
For some higher value products a bioprocessmay in some cases be the only route to a givenproduct (to ensure correct folding of a therapeu-tic protein, for example, or the synthesis of anoptically pure pharmaceutical intermediate).However, the more usual situation is that thereare other competing routes to the same product.Therefore, for now, biotechnology is just one ofa number of options for the production of che-micals and biofuels. The economic drivers forimplementation depend on existing infrastruc-ture, feedstock costs, feedstock availability aswell as the efficiency of the relevant (bio)cata-lyst and (bio)process technology. At the sametime, there are environmental drivers and, in thewider sense, sustainability drivers for the selec-tion of different process alternatives. The sus-tainability drivers are by no means simple, asthey require the balance of different sets of goalsand metrics that can present trade-offs in somecases. Objective functions to be optimizedshould not exclusively be based on economicsbut increasingly also on sustainability me-
trics [158] and integrated with life-cycle analy-sis. This will need to include evaluation offeedstocks and products as well as processes,including energy and mass integration. Thispresents a fascinating set of alternative routesand technologies from a given feedstock and/orto a given product(s). PSEhas a particular role toenable such evaluations on a quantitative basis,not only from the process perspective, but alsofrom the wider sustainability aspect. PSE alsobrings the advantages of rapid computationalmethods. Such simulations enable alternativesto be quickly evaluated. The answer in a specificcase to the problem formulated here will inaddition depend on regional factors. Feedstockavailability and cost is highly dependent ongeographical location. A parallel set of evalua-tions concerning the need to retrofit existingplant, or build new plant, is also required.
4.4.2. Evaluation of Platform Chemicals
While the increasing cost of oil is driving partic-ular interest in the production of new biofuelsfrom biomass there is little doubt that today ofequal importance is the production of chemicalsfrombiomass. Indeed for the supply of fuels in thefuture there aremany potential sources aside frombiomass. In a world with limited (or very expen-sive) oil it is less clear where the chemicals of thefuturewill originate.There is currently an existinginfrastructure basedon the use of the7 establishedplatform chemicals (toluene; benzene; xylene;1,3-butadiene; propene; ethylene; methane). Inthe short termone could consider ifwe can use thesame infrastructure and just create the7 chemicalsfrom alternative sources. However, in the longerterm it will be necessary to devise new processesbased on a different set of platform chemicals.One group will be based around glucose (thehydrolytic product of starch and cellulose andtherefore readily available from biomass). In abiorefinery it will be necessary to develop astructurewhich canmanage a rangeof feedstocks,a range of technologies, and a range of products.This presents a considerable challenge for designand optimization as well as process integration.An interesting examplewhich illustrates the com-plexity and the challenge that lies ahead is the useof glucose or fructose to produce 5-hydroxy-methylfurfural (HMF) or 2,5-furandicarboxylic
30 Process Systems Engineering, 9. Domain Engineering
acid (FDA) [159]. Greatest value is obtained bygoing the whole way from glucose to FDA.However even in this small reaction pathwaythere are many alternative technologies. Somecan be integrated together, some give the requiredyield and selectivity, some are difficult to imple-ment and others are untested at scale. This illus-trates very well the challenge that design engi-neers face.
4.4.3. Process Integration
The solvent for most bioprocesses, with a fewexceptions, is water. Consequently the down-stream process is frequently difficult and this isexacerbated by the need to carry out separationsat moderate temperatures. Given the dilute na-ture of the streams it is frequently the case thatthe majority of the costs and environmental,health, and safety impacts are therefore in thedownstream process. For instance, in some finechemical and pharmaceutical applications ofbiocatalysis large amounts of organic solventsmay be used in the purification of a biocatalyticreaction. The dilution of the streams has histori-cally also driven the need for energy-intensiveseparation. In the case of transport fuels removalof water becomes an essential requirement toreduce costs and avoid transporting significantamounts of water. For example, in the case ofethanol which forms an azeotrope, this can be asignificant cost. In other cases the product maybe integrated within a biorefinery although atsome point water will need to be removed.Consequently the integration of water use andreuse via recycle is an essential part of the designof industrial bioprocess facilities. In addition,bioprocesses need to be designed with processsynthesis and process integration approaches,thus avoiding a process that is efficient in onepart and inefficient in another. Existing tools ofmass and energy integration such as pinch tech-nology (! Pinch Technology) will have animportant role. The issue of water use in abiorefinery is in many ways analogous to theissue of heat use in a conventional refinery.
4.4.4. Biorefinery Design
Two major types of biorefinery have been iden-tified for the future, based on lignocellulose
biomass utilization to provide a range ofsugars (for subsequent (bio)catalysis or fermen-tation) and oil-based material (from biomass)! Biorefineries–Industrial Processes and Pro-ducts. In each case the current research empha-sis on biorefineries is to ensure that all thefractions of a particular biomass in a givensituation are fully exploited. Likewise the de-velopment of downstream products is now be-ing explored. For example, glycerol (as a by-product of biodiesel production) can be used as aplatform chemical (e.g., via fermentation toproduce 1,3-propanediol). Another interestingexample concerns the production of bioethanol.This is widely developed as a biofuel althoughthere is considerable economic incentive fordeveloping a range of other products (e.g.,acetic acid) from ethanol, in other words usingit as a platform chemical [160].
4.4.5. Biocatalyst Design
A particular feature of bioprocesses is the use ofbiocatalysts ! Biocatalysis, 1. General, whichmay exist in several forms as indicated earlierand where options exist for modification. At thesimplest level as a protein (isolated enzyme), theoptions for swapping amino acids via proteinengineering exist. New enzymes which havebeen modified may display new tolerance toreactor conditions such as temperature or pHand may also have improved selectivity orreactivity (activity) on a given (nonnatural)substrate or reactant. Order-of-magnitudeimprovements have been found in a number ofcases although understanding themost effectivemethod of making changes to the enzyme de-pends on past precedent and, to some extent,structural knowledge [161]. In the case of mi-crobial catalysts, individual enzymes can beover-expressed (increasing reaction rate of agiven cell) and the regulatory control schemefixed to direct the carbon to give improved ratesand yields (via metabolic engineering). Somestart has also been made to the development ofpathwayswhere enzymes coming from a varietyof sources are cloned into single host to make anew pathway via a combination of geneticengineering and de-novo pathway engineer-ing [162]. In all these areas it is clear that thoseinvolved in PSE need to inform the biological
Process Systems Engineering, 9. Domain Engineering 31
engineers about what is required in a given caseand set suitable targets. Philosophically it isinteresting to note that process implementationmay come via process improvements or alter-natively via catalyst improvements. In manycases both will be required. Understanding thenecessary balance between these areas, as wellas their integration with each other will beimportant for the future development of thefield. PSE is particularly powerful in its abilityto predict and can therefore be used to directdecision-making and process development.
4.5. Assessing the Sustainabilityof Bioprocesses
As discussed before, bioprocesses have fre-quently been highlighted as greener chemistryor engineering, since they address many of thegreen chemistry and green engineering princi-ples [163] by offering reactions that are poten-tially more atom economic, operate under mildconditions, use mostly nonhazardous chemi-cals, and have less protection/deprotectionsteps. However, this tends to be true mostlywhen looking at the reaction part of the process,in other words, the biocatalysis. However, itcannot be generalizedwhen analyzing the entireprocess that in some cases may include the useof a large amount of organic solvents for down-stream processing. This is one of the areas inwhich systems engineering, in conjunction witha transparent application of life-cycle inventoryand assessment (LCIA) methodologies can andmust play a pivotal role.
Determiningwhether a process is sustainableor green is by no means a simple feat. It is moreakin to a multivariable optimization that is veryfamiliar to systems engineers, and for whichseveral proposals and methods have been pre-sented [164–166]. For an objective assessmentof the sustainability of a process, there is theneed to utilize the tools that system engineershave developed during recent years and applythem with a life-cycle approach. It is necessaryto move from the basic analysis of the biocatal-ysis alone and discrete unit operations (separa-tions) and use a whole system engineeringapproach instead. This implies utilizing multi-variate optimization techniques coupled with
LCA methodologies for a more objective anal-ysis of their ‘greenness’ or sustainability. Thiswould allow to develop bioprocesses that aresustainable by design, in such a way that they:
. Optimize the use of material and energyresources
. Eliminate or minimize environment, health,and safety hazards by design
. Integrate life-cycle thinking in the design
Analyzing and comparing sustainability willrequire a comprehensive assessment that bal-ances the three different spheres of sustainabili-ty (see Fig. 22). This can only be achievedthrough a multivariate optimization that willaccount for environmental performance, eco-nomic viability, and social responsibility(which include health and safety aspects).
Another important concept when assessingthe efficiency and the sustainability of processesis the differences between new process perfor-mance and retrofit performance. For instance, incomparing the sustainability or performance ofa well established process with a new biopro-cess, a situation that one often encounters is thefact that initially, the new process may not havethe same level of performance as the establishedtechnology, mainly because they are at differentpoints in the development curve, and thereforethe new process is suboptimal. On the otherhand, the established process can be retrofittedto improve its performance. Retrofitting andnew process development is not a new conceptfrom the systems engineering viewpoint. How-ever, additional modeling work is needed toestimate the achievable performance limits ofa fully developed process and an establishedprocess that undergoes retrofit. This will allowmore meaningful comparisons without unnec-essarily penalizing the new process for its lackof development, nor the established process forthe lack of timely retrofitting.
In the recent past there have been manyattempts tomeasure the ‘greenness’ of syntheticroutes, and the approaches have generated aseries of ‘green metrics’. Most of the ap-proaches have searched for a simple metric inan attempt to provide a low resolution view ofhow green is a given process. ‘‘E-factor’’ wasone of the first measures of greenness proposedto highlight the amount of waste generated in
32 Process Systems Engineering, 9. Domain Engineering
order to produce 1 kg of product [167, 168].This metric, while simple to understand, hasseveral key drawbacks by focusing on wasteinstead of efficiency, neglecting a view of thetype of waste generated, not accounting for therelative impacts, and the lack of life-cyclingthinking. For instance, bioprocesses have ingeneral large amounts of wastewater produced,which would indicate an extremely large E-factor, but at the same time the actual impactof the effluent might not be as large as an E-factor would suggest, since the waste is rela-tively benign. Additional metrics have focusedon efficiency, especially on mass efficiency (orits inverse, mass intensity) which addresses thefirst part of the disadvantages of the E-fac-tor [169, 170]. However, it has been widelyrecognized that measuring the sustainability oreven the ‘greenness’ of a process is a multi-objective optimization problem that must takeinto consideration the efficiency of the entireprocess regarding the use of mass and energy,the environment, health and safety characteris-tics of materials, and the inclusion of life-cyclethinking amongst others [171].
Some efforts have been made to comprehen-sively address and compare the sustainability ofprocesses in general and bioprocesses in partic-ular. These methodologies have been employedin some instances to assess the sustainability,environmental, health and safety aspects of bio-processes [172–181]. These methodologies at-tempt to measure the sustainability of biopro-cesses and in some instances they comparebioprocesses with their chemical alternatives.
For instance, a technology comparisonframework [181] that accounts for environment,health, safety, and life-cycle assessment im-pacts, was used to compare the establishedchemical process and a two-enzyme biocata-lytic process for the production of 7-ACA [158].The conclusion of this assessment was that thebioprocesswas greenerwhen comparedwith thechemical process. This was driven by the factthat the chemical process uses more hazardousmaterials, requires about 25% more processenergy than the enzymatic process and, has alarger life-cycle environmental impact: it usesapproximately 60% more energy, 16% moremass (excluding water), has double the green-house gas impact, amongst others.
However, although this type of assessment isuseful andmore common as time passes, there isan ongoing need for modeling methodologiesthat will seamlessly integrate sustainability fac-tors during bioprocess design and development.Systems thinking and systems engineering arethe skill and the discipline that will need to playan important role to make this happen, and aholistic view of bioprocesses and their interre-lations will be imperative.
4.5.1. Life-Cycle Inventory andAssessment
One of the tools to analyze systems holisticallyis LCIA (Fig. 23). LCIA is a methodology usedto evaluate the environmental profile of anactivity or process from the extraction of raw
Figure 22. The three spheres of sustainability through a triple bottom line assessment
Process Systems Engineering, 9. Domain Engineering 33
materials to its end-of-life. The resource con-sumption and emissions are inventoried andassessed from the extraction of raw materials,production, transportation, sales, distribution,use, and final fate. Depending on the goal andscope of the assessment, the boundaries can beset differently; for instance a ‘cradle-to-gate’assessment might be adequate when comparingtwo processes to the same product; or a ‘gate-to-grave’ boundary may suffice when comparingtwo different end-of-life technologies. Theresults of these assessments can be reported asdirect inventory data (for example life cycleenergy, life cycle mass, life cycle emissions),measures of individual potential impacts (suchas global warming or acidification), or as anaggregate score or index for high-level compar-ison (for example Eco-Indicator 99). LCIAmethodologies are described in detail in theliterature [182–188]. LCIA methodologies arein a way an extension of systems engineeringand provide a directly applicable framework toassess the sustainability of processes.
In the area of bioprocesses, the application ofLCIA is still not a widespread practice. Thereare however, examples on how several practi-tioners have applied LCA metrics primarilyusing case studies to better understand the widerenvironmental implications of bioprocesses andto compare them with chemical routes. Thistype of assessment has provided some key in-sights, such as the role of separations, a more
systematic and holistic method to evaluatingwaste impacts, and the nuances of renewabili-ty [173, 174, 178, 179]. For instance, a compar-ison of a process using metal catalysts and oneusing biocatalysts for the enantioselective re-duction of ketoesters in pharmaceutical synthe-sis was performed using a streamlined LCIAmethodology. The analysis identified someprocesses and reaction conditions that had thelargest significance on the impact of the synthe-sis. It was also concluded that whether themetal catalysts were better than bio-catalystsdepended mainly on the work-up from the useof organic solvents and energy-intensivesteps [172].
Developing life cycle inventories and asses-sing the LCIA impacts of bioprocesses is not asimple endeavor given the large amount of dataneeded from different sources. The more mate-rials are involved in the bioprocesses will re-quire more life cycle inventory data to be col-lected, verified and analyzed. On the other hand,the life cycle inventory data for biomaterials isnot always available. There have been efforts toincrease the body of knowledge of life cycleinventories and impacts of bioprocesses andmaterials either derived from biomass or neededin bioprocesses. These challenges have influ-enced the development and use of streamlinedlife cycle assessment methodologies and abinitio modeling approaches to estimatethe life-cycle impacts of bioprocesses and
Figure 23. Life-cycle assessment to evaluate the environmental profile of a process from the extraction ofa) Raw materials; b) Production; c) Transportation; d) Sales, distribution, and use; e) Final fate
34 Process Systems Engineering, 9. Domain Engineering
bio-derived materials. This is precisely the sortof opportunity where system engineering canadd value, as the development of reliable, con-sistent, transparent, accurate and easy-to-usemodeling and streamlined techniques for LCIAwill continue to be an important need to be ableto routinely assess the sustainability ofbioprocesses.
The development of true sustainabilityassessments, with an embedded LCIA ap-proach will be necessary aligned with theearly modeling needs highlighted in this arti-cle. In order to routinely assess sustainabilityof bioprocesses and to embed sustainabilityprinciples into the bioprocesses design anddevelopment, the following modeling needscan be highlighted:
. Better deterministic models of unit operationsthat are part bioprocesses, such as fermenta-tion, biocatalysis, etc. This would need toinclude fundamental design parameters todesign more resource efficient bioprocesses.
. Development and enhancement of propertyprediction packages that would facilitateestimations of resources (e.g., energy require-ment) and the utilization of optimizationtechniques.
. More extensive use of process integrationtechniques on bioprocesses, especially at thedevelopment phase.
. Increased use of software sensors on biopro-cesses, in order to maximize the informationcontent that is available on-line. Closelyrelated to that, increased use of advancedcontrol and monitoring methods such thatbioprocesses can be operated as close aspossible to the optimum.
. Better understanding of life cycle inventoryand impacts of bioprocesses and bio-derivedmaterials.
. Increased understanding of the uncertaintiesin modeling bioprocesses, both from the pro-cess design and the sustainability assessmentviewpoints.
. Improved consistency and transparency ofLCIA methodologies as applied tobioprocesses.
. Improved streamlined LCIA methodologiesthat are easy to use by academia and industryalike.
. More routine application of multiobjectiveoptimization techniques for sustainability as-sessments of bioprocesses.
. Enhanced understanding of the interactions ofthe environmental, social and economic as-pects of bioprocesses for a holistic sustain-ability view.
Addressing these modeling and process un-derstanding needs will make it possible to inte-grate sustainability principles into process de-sign and development in a far more rigorousmanner.
4.6. Future Outlook and Perspectives
The development of new bioprocesses as acomplement to existing chemical and fuelproduction is an exciting endeavor that willoccupy many process engineers in the future.There will be a particular role for processsystems engineers in this developing sectorwith the advantages of quantitative decision-making tools and rapid simulation that thisbrings, including process design and sustain-ability principles. In the future suitable modelswill inform developments at the infrastructur-al level (evaluation of biorefineries, feed-stocks and integration), the process level(evaluation of alternative technologies andprocess integration) and the catalyst level(alternatives for protein and metabolic engi-neering). In addition, these models will allowthe integration of sustainability principles intoprocess design and development.
The further development of PSE tools(including property prediction packages and thedevelopment of a database for bio-based mole-cules) will be required. To routinely assesssustainability of bioprocesses will require aswell more robust and transparent environmentallife cycle inventory databases of bio-derivedmaterials; as well as better modeling and under-standing of the social and economic aspects ofsustainability and their relationships. Finally, anincreasing dialogue amongst the biochemicalengineers, biologists and other related areas ofexpertise will be necessary to enable the visionof sustainable industrial biotechnology to befully exploited.
Process Systems Engineering, 9. Domain Engineering 35
5. Policies and Policy Making
5.1. Introduction
Process engineers are becoming increasinglyaware that there is a significant role for themto play in the arena of policy making. This is notto say that there have not been past contributionsof PSE in the policy area nor that engineers havebeen insensible to the effects on society of thesystems they design and operate. However, anideological divide between the technical andsocial sides of a given problem has tended toreinforce the perception that our exclusive roleas engineers is to improve product quality andproduction efficiency and to minimize costs andrisks, and that these contributions in themselveswill automatically result in a net benefit tosociety. Such a perception is a result of ahistorical tendency to compartmentalize pro-blems, i.e., to dissociate the technical and socialsides of a system, leaving the decision-makingand analysis of the social side to the socialscientists and politicians.
The split between the technical and socialcamps has deep and old roots, as discussed in [,190]. Yet, it is increasingly accepted that socio-technical systems, i.e., systemscomposedbytech-nical artefacts and social arrangements (agentswith a purpose), are closely interdependent and,as such,need tobedesignedandanalyzedasa totalsystem[191]. Inrecentyears the trendhasbeguntoreverse towards the closing of the gap betweentechnical and social systems, because only bytaking into account the characteristics of thesocialsubsystems it will be feasible to develop effectiveand sustainable engineering systems.
To investigate the relations–past, current,and possible–between process systems engi-neering (PSE) and policy making it is conve-nient to start by providing some definitions forpolicy and its components, and a description ofhow policies are conceived, developed, andimplemented.
5.2. Policies and Policy Measures
The following two complementary definitionsof policy are adopted:
. A policy is a set of effective and acceptablecourses of action to reach explicit goals [192]
. A policy is a purposeful connection of endswith means [193]
In the classical point of view a policy is theproduct of rational choice; this assumption isalso going to be adopted for the rest of thisdiscussion, although the adequacy of this modelis in dispute [194].
Policies are constituted by combinations ofpolicy measures (also known as policy instru-ments). According to [192] policy measures canbe of different types:
. Exhortation (e.g., education)
. Economic incentives/disincentives (e.g., sub-sidies and taxes)
. Government provision
. Legislation/regulation
However, in all cases policy measures have aset of properties such as a degree of effective-ness and a range of implementation costs andtimes. Other properties, more difficult to mea-sure and quantify, but not less important, arerelated to issues such as equity, legitimacy, andpublic support.
A policy cycle is a sequence of steps throughwhich a problem is defined; alternative policiesto address it are proposed, analyzed, and refined;and a proposed policy is selected, implemented,and constantly challenged and revised [195].
5.3. Policy Making and the SystemsApproach
The first author to use systems theory to explainpolitical processes was [196]. In his portrayal,political systems convert inputs, such as politi-cal demands and public support, into outputs,i.e. a group of resulting decisions and actions(Fig. 24 A).
Policies themselves can be understood asblocks having policy measures as inputs and aset of desiredoutcomes as outputs [197], (Fig. 24B). Thus, policies are systems composed ofinteracting parts that are the means to reach afinal objective (the ends), as a result they can berepresented through a causal model.
The concept of policy cycle, in particular, hasits foundations in systems theory and is analo-gous to the concept of life cycle in engineering
36 Process Systems Engineering, 9. Domain Engineering
systems. The schematic policy cycle in Fig-ure 25, based on a model proposed in [198],depicts the process of policy decisions, theirimplementation, enforcement, and evaluation.
The policy cycle in Figure 25 is an idealiza-tion, as some stages in the process are sometimesmerged or altogether eliminated. The feedbackarrows intend to express that the process iscyclical with backtracking and revision steps.
The rest of this analysiswill focus on the thirdstage in the cycle (policy formulation) becauseof its many similarities with conceptual processdesign in PSE. These similarities explain whymost of the past efforts and the potential futureadvances of PSE in policy making are to befound in the policy formulation area.
5.4. Similarities between PolicyFormulation and Conceptual ProcessDesign
Policy formulation is a procedure with twocomponents (Fig. 26 A):
1. Synthesis (generation) of alternative coursesof action (alternative policies)
2. Analysis of the alternative policies, i.e., theestimation of their consequences to help inthe selection of the best policy alternatives.This step is generally performed through the
application of formal analytical methodssuch as simulation and optimization
The output from the policy formulation stageis the selection of a policy considered to be themost appropriate and thus the recommendedone for implementation [190]. Policies are cre-ated during the policy formulation step as newcomponents of a sociotechnical system. For thisreason, policy formulation has been character-ized as a design activity [].
During the synthesis step of policy formula-tion a set of policy measures (the buildingblocks) are combined to configure alternativepolicies; thus the policy maker has to decidewhich policy measures to select taking intoaccount their intrinsic characteristics (such aseffectiveness, cost, etc.) and their interactionswith other policy measures, as in the left-handpart of Figure 26A. Process synthesis, in turn, isthe invention of a structure and its associatedoperating conditions for a new chemicalmanufacturing process [199]. Inventing thestructure involves finding the best process con-figuration (which building blocks to include andhow to interconnect them) among a very largenumber of possible alternatives, as in the left-hand side of Figure 26 B.
In both cases synthesis and simulation stepsare applied in tandem and iteratively: a synthe-
Figure 24. Systems theory view of political and policy systemsA) Political system; B) Policy system
Figure 25. Simplified policy cycle, based on [198]
Process Systems Engineering, 9. Domain Engineering 37
sis step generating alternative policies/flow-sheets, and a simulation step evaluating eachof the alternatives so that they can be compared,and also informing the application of the nextsynthesis step in the loop (Fig. 26).
Policy formulation and conceptual processdesign belong to the type of problems wherethere is neither an a priori agreement aboutwhich criteria to use nor a prior articulation ofpreferences [201]. Both are processes of coevo-lution between what is possible and what isdesirable and proceed by generating potentialsolutions and evaluating them in a generation–evaluation cycle. Thus the goals, criteria, andthe artefact being designed (policy or chemicalprocess) evolve in a single front according to theexploration model proposed by [202]. As aresult there are no unique optimum policies,just satisficing ones [203].
It has already been mentioned that alterna-tive policies are evaluated during the analysisstep of policy formulation (Fig. 26 A). Thisevaluation entails the exploration of their im-plications in terms of what they can accomplish,alongside any desired or undesired side effects.Such a task can only be done by means ofsimulation, whereby either a point prediction(the forecast of the state of the sociotechnicalsystem at a particular point in time in the future)or a set of event frequency distributions isproduced. The analysis of policies is oftenuseful for the insights it provides even when
precise predictions are not feasible [190]. Sim-ulation can also be used to discover the rela-tionship between states through time, i.e., thedynamics of the system [].
In any of the above cases, simulation requiresthe development of models that relate policyalternatives to their impacts and the applicationof such models to predict the impacts of thepolicies being considered [204], however this iseasier said than done given the complexity of thesociotechnical systems that have to be predictedas will be discussed in Chapter 6.
5.5. TheNature ofPolicyFormulation
It is widely recognized that a one-for-all andstatic policy is unlikely to achieve the desiredgoals. This is because [205]:
. A good policy groups together (packages) aset of policy measures such that synergies areachieved, negative impacts are mitigated, andconflicts are resolved, thus ensuring that thepolicy will address the problem effectivelyand equitably over the long term [204]. Alter-native policies are not equivalent in theireffectiveness, implementation costs, publicacceptance, risk, etc. All of these propertiesare determined by the properties of its con-stituent policy measures and theirinteractions.
Figure 26. Analogy betweenA) Policy formulation; B) Conceptual process design [200]
38 Process Systems Engineering, 9. Domain Engineering
. The precise nature and scope of policies aregeographically and culturally dependent giv-en the variability of resources, of access totechnology, and of political constraints atdifferent locations and times.
. Even for a fixed time and place, the identifi-cation of a suite of alternative policies (ratherthan a single ‘‘optimal’’ one), together withclear indications of their trade-offs, is crucialto accommodate the diversity of stakeholders’preferences because, after all, decisions aboutdesirable futures, and the policies to attainthem, are essentially a question of socialvalues and political choice [206]. It is ac-knowledged that no set of values or framingscan definitively be ruled more rational, wellinformed or better than all others [207].
5.6. The Nature of SociotechnicalSystems
Sociotechnical systems are systems that involvethe interaction of human beings with physicalinfrastructure and, as in the case of purelytechnical artefacts, can also be designed. Theyare characterized for their sensitivity to initialconditions and for the complexity of the inter-actions between human actors (possibly mil-lions of them) and with heterogeneous physicalinfrastructure. Because of their human compo-nent, it is either impossible or prohibitivelyexpensive (in terms of time and cost) to performin vivo experiments on sociotechnical systemsin order to inform the decisions taken during theformulation of policies. The sensitivity to initialconditions and the complexity of interactionare two of the reasons that render accuratepoint predictions impossible, thus requiringto resort to mapping out a space of possiblestates and estimate the relative frequency thatany particular state will occur in a future timewindow [].
Policies are courses of action to design, plan,manage, or control complex sociotechnical sys-tems. In this sense the implementation of apolicy is an experiment but with unknown out-comes; furthermore, and because of there isoften a lack of time and resources, policy out-comes are rarely formally monitored.
A policy is also influenced by how the humanactors respond to it. Individual actors, or the
society as a whole, may take actions to subvertor circumvent it, rendering the policyineffective [204].
5.7. Challenges for Modelers ofSociotechnical Systems
Traditional policy analysis assumes that socio-technical systems reach static equilibrium, andthat they can be properly characterized andcontrolled. It has been argued that each of theseassumptions is true only in special circum-stances [208]. As a result, there are a numberof challenges that modelers of sociotechnicalsystems have to face due to the unique nature ofsuch systems.
5.7.1. Multiple Stakeholders
Policies involve multiple stakeholders withtheir own preferences, objectives, expectations,and beliefs [209]. Policies also have implica-tions for groups with no or little influence in thedecision-making and even for people that do notcurrently exist, as is the case of futuregenerations.
Because stakeholders can be individual peo-ple and organizations, the issue of individual vs.institutional behavior must also be taken intoaccount and included in the overall model; forexample, in the case of policies to reduce trans-port emissions there are many actors such asvehicle users, vehicle and fuel manufacturers,government agencies, and environmentalgroups [204].
Furthermore, the costs and the benefits ofpolicies are not evenly distributed between thestakeholders, as a result
. Stakeholders have contradictory interests andtheir interaction will often result in conflict
. The impact of a policy on a stakeholder con-ditions how they react to the policy
5.7.2. Incommensurable Values
The reconciliation of multiple incommensura-ble values (values that are not measurable) ispresent in all public policy decisions [204].
Process Systems Engineering, 9. Domain Engineering 39
Values that are intangible have often beenmonetized by assuming a hypothetical marketand then stating how much society would bewilling to pay either (i) to secure an improve-ment, (ii) to prevent a loss, or (iii) how muchsociety would be willing to accept as compen-sation [210]. The monetized values can then bemerged during a cost-benefit analysis. Howev-er, despite the fact that there are ethical pro-blems involved in monetizing health, welfare,and environmental values [211], these issues arenot always suitably considered.
5.7.3. Externalities
Externalities are secondary or unintended con-sequences, generally a nonmonetary cost orbenefit, incurred by a stakeholder who did nothave a choice and whose interests were nottaken into account [209]. There are differenttypes of externalities that may have to be mod-eled: spatial (consequences in different re-gions), intertemporal (consequences in differentgenerations), and social (consequences for dif-ferent social groups).
5.7.4. Uncertainty
Uncertainties are due to the long-term nature ofpolicies and the existence of unexpected im-pacts, for example, when a sociotechnical sys-tem undergoes structural change. They are pres-ent in three different forms [209]:
. Uncertainty about the model parameters andinitial conditions (which is usually addressedthrough sensitivity analyses)
. Uncertainty about the model structure (rela-tions between variables)
. Uncertainty about the applicability of themodel, i.e., its level of granularity and timescale, and about the selection of variables
Identifying and managing uncertainty is im-portant because the risk of wide-range and long-term hazards is proportional to uncertainty. Atthe same time, uncertainties limit the applica-bility for long-term forecasting; looking fartherinto the future (a large horizon of analysis)increases uncertainties. Users of models gener-
ally expect them to reduce uncertainty, butpolicy analysis often increases uncertainty byidentifying and raising new issues [190].
5.7.5. Emergent Behavior
An emergent property is one that cannot bepredicted from the knowledge of the systemcomponents, i.e., it is the product of many localeffects [212]. In practical terms this means thatthe system is computationally irreducible (thereis no simple set of equations to represent it) andthat the only way to figure out its evolution is byrunning the system itself [213].
5.7.6. Complexity of Causation
The complexity of sociotechnical systems re-sults in difficulties to establish and representcausation. There are several reasons forthis [190]:
. There is a very large number of variables andrelationships at different levels of granularityand time dynamics [214]; in particular, thebehavior of complex systems is dominated byinterconnected positive and negative feed-back loops [212].
. The effects of a policy are also shaped by how itis implemented and not only by its substance.
. The effects of a policy are modified as itreceives feedback, i.e., its content has aninteractionwith its effects. For example, theremay be unintended effects due to the stake-holders reactions to a policy by circumventingor subverting it.
In fact, it has been argued that policies aremore often facilitative than causative, i.e., con-sequences are as likely to depend on actions andinfluences other than the policy itself [215]. Tothe extent that this is true, it can be said that theeffects of policies can only be influenced but notcontrolled [208].
5.7.7. Objectivity in Policy Analysis
Policymust be interpreted in order to be analyzed.For example, its objectives may need to be re-
40 Process Systems Engineering, 9. Domain Engineering
defined in more concrete terms, decisions aboutthe types of effects that will be examined have tobe made, etc. However, this analysis is not value-free, as the purposes and mechanisms of analysisare conditioned by ideology and its conductshaped by the analyst’s concerns about the con-sequences of the analysis for society [190].
These issues raise ethical questions becausethe idealism of analysis as a pure scientific activi-ty getsmixedwith the impulses of the analyst as acitizen, reflecting ideology and values.
5.8. Types of Models Used in theAnalysis of Policies
Models can be used to explore, describe, ex-plain, and predict the behavior of a system [190].Effective policy formulation, in particular, de-pends on the understanding and modeling ofsociotechnical systems to forecast their futurebehavior, evaluate the likely impacts (econom-ic, environmental, social) of alternative poli-cies, and inform the decision of which policymeasures to adopt. All of these analysis tasks arepart of the so-called policy assessment, i.e., thecomparison of alternative policies using policyinstruments as inputs and measurable indicators(such as CO2 emissions, cost, etc.) as outputs.Unfortunately, while it is relatively easy to fitmodels to historical data, models are not as goodin predicting the future.
Many types of models have been used inpolicy making due to two factors:
. The development of models used during poli-cy making has attracted the participation ofmodelers from many different disciplines,each one bringing their ownmethods and tools
. The systems to be modeled are complex andpresent a number of challenges, as have beendescribed in Chapter 7. These challenges canonly be reasonably overcome through theconcurrent application of many methods orby restricting the modeling to parts of theoverall system
Most modeling methodologies are quantita-tive, but qualitative analysis is also useful forcertain aspects of policy making such asscenario building.
Several studies on the application of modelsduring policy development have been pub-lished; the list below is mainly based on [209]and [216] unless stated otherwise. The follow-ing descriptors are common to many of thedifferent categories of models in the list:
. Mainstream: a well known and widely usedtype of method.
. Descriptive vs. normative: a descriptive mod-el is one that given some inputs will producesome outputs, i.e., models used for simula-tion. A normative model suggests how thingsought to be (as opposed to a descriptivemodel, which describes how things are).
. Aggregated: the behavior of a system is takenas an average of the individual componentbehavior; although it is possible to avoid it,mathematical representations tend to use ag-gregation. Averaging is inappropriate for therepresentation of emergent behavior.
. Mechanistic: mechanistic models use mathe-matical equations to simulate a system andpredict its future state; they are useful in theunderstanding of the workings of a system butnot very reliable for prediction. Mechanisticmodels tend to be aggregated.
The general equilibrium models advocated byneoclassical economic theory are the most pop-ular type of economic models in use. Theyconsist of systems of equations and are basedon two premises:
. The economy behaves as a linear mechanicalsystem that tends to a stable equilibrium
. Human behavior is assumed to be rational andindependent, and can be represented by aver-aging the behavior of consumers andproducers
This type of models are used to forecast theeconomic impact of policies on the equilibriumof the system; however, it has been argued thatthere is little evidence to suggest that they havemuch predictive value [217], perhaps becauseboth of the basic assumptions are unrealistic.
Process Systems Engineering, 9. Domain Engineering 41
The parameters in macroeconomic modelstend to be calibrated rather than validated em-pirically. While macroeconomic models are notwithin the direct interest of PSE, they require thesame sort of solution methods as the traditionalPSE applications.
Optimization models are used to help in theselection of alternatives by minimizing capitaland operating costs under constraints of avail-ability, prices, etc. System optimality assumesthat decisions are taken centrally; however,society is composed of individuals and groupsmaking their own decisions.
Optimization is oneof the prime interest areasin PSE, with several types of applications [218].
5.8.3. Control Models (Mainstream, No-rmative, Aggregated, Mechanistic)
Process control is one of the mainstream PSEareas. A specialized branch of control is modelpredictive control (MPC) (! Process SystemsEngineering, 5. Process Dynamics, Control,Monitoring, and Identification, Chap. 5).
There is, however, considerable uncertaintyassociated to the model parameters and possibledisturbances when the prediction of the effectsis done over a horizon of many years. Exten-sions to MPC are required to address this issue.For example, stochastic predictive control [219]and a reformulation introducing multiple objec-tives and dynamics [220] have also been pro-posed to model the effects of policy over ahorizon of many years.
5.8.4. Data-Based Models
These models fit equations to data withouttrying to simulate the system in a mechanisticfashion, e.g., in the discovery of statistical pat-terns in historical data. Their predictions assumethat the future will resemble the past, but oneproperty of sociotechnical systems is that theyevolve and change. As a result, this type ofmodels by itself is not adequate for policysupport [212].
5.8.5. Game Theory (Descriptive)
Game theory attempts to model the behavior ofindividuals when confronted with a choice thatdepends on the choices of others and as such, itis helpful for the development of a strategy insituations of competition and/or cooperation.The recommended strategy is based on theso-called Nash equilibrium, which assumes aself-interest behavior and perfectly rationalagents (players) [221]. However, agents are notalways fully rational.
5.8.6. System Dynamics (Aggregated,Mechanistic)
System dynamics is amethodology based on thegeneral systems theory.
It uses the concepts of stocks (levels), flows(rates), feedback relationships, and time delayin order to model dynamic behavior.
System dynamics models are appropriate forthe (qualitative) identification of the importantvariables and causal links in a system, and forthe representation of the nonlinearities arisingfrom feedback loops and time delays. Thismethodology is popular in the modeling ofenvironmental systems, and to a lesser extentfor economic systems (! Ecology and Envi-ronmental Chemistry).
5.8.7. Network Theory (Descriptive)
Network theory groups a family of methods thatenable the representation of nodes and theirinterdependencies, and its subsequent analysis.As a result these methods are suitable to modelthe behavior of networks, such as those resultingfrom transportation systems or electricity gen-eration and distribution. Networks have nonlin-ear feedback loops, exhibit dynamic behavior,and their local effects cannot be averaged.
Several specialized types of networks can bedefined according to the focus that the modelerwants the model to have. For example, if thefocus is on the management of uncertainty,Bayesian Networks are risk assessment modelsexpressed in terms of influence diagrams thatare used to infer through causal links the prob-ability of an effect from the probabilities of itsassociated causes.
42 Process Systems Engineering, 9. Domain Engineering
5.8.8. Agent-Based Approaches
An agent is an autonomous entity with aninternal state. In turn, a multiagent system isa collection of agents that interact with otheragents and the environment within which theyoperate. The interactions between agents, whichare governed by rules, modify the internal stateof the agents. This type of model has the abilityto capture individual heterogeneous actors asautonomous decision-makers, with bounded ra-tionality (i.e., limited information), attemptingto maximize their own utility function. Theapproach is catalogued as bottom-up and hasthe property of exhibiting emergent behavior.
As in the case of networks, it is possible toextend the basic multiagent representation in anumber of ways, for example:
. Incorporating a stochastic approach within anagent-based system to facilitate the manage-ment of uncertainties.
. Embedding the agents in a spatial setting andthus creating artificial societies [222] or arti-ficial ecosystems [223], for example, throughthe creation of synthetic micropopulations toexplore policy impacts [].
5.8.9. Some Conclusions on Models forthe Analysis of Policies
There are a few general conclusions that can bedrawn from the breadth of methods used forpolicy analysis:
1. It is possible to develop partial models tocapture specific aspects of the sociotechnicalsystem when only one type of modelingmethodology is used.
2. To develop more inclusive models it is nec-essary to combine information from diversesources and integratemodels.Hybridmodelsintegrate differentmethods, e.g., mainstreammathematical simulationmodels with differ-ent tools to account for location (geographi-cal information systems), economics (mac-roeconomic models), etc.
3. In either case, partial or hybrid, models mayhave to be multiscale, i.e., spanning indivi-duals and organizations. This is the nature ofsome of the models being developed forPSE [224].
4. Models may also have to be multilevel, i.e.,using various time spans (just ensuring thatthe time spans are long enough to allow thecomplete unfolding of the dynamics of thesystem).
5. Model development should ideally allow theinclusion of values and objectives of multi-ple stakeholders and facilitate the communi-cation of the results.
5.9. Synthesis of Policies
A large portion of the space of alternative poli-cies is currently left unexplored because thesynthesis of policies is performed manually. Asystematic approach that aims to automate thegeneration of alternative policies has the poten-tial to accelerate the policy making process andimprove the effectiveness of the resulting poli-cies; the development of such an approach maybenefit from the experience gained in the area ofprocess synthesis in the last three decades. How-ever, the differences between process synthesisand policy formulation, in particular the perva-siveness of qualitative factors in the latter, re-quire a substantial adaptation of the methodsused in process synthesis.
A good policy ‘‘packages’’ a collection ofpolicy measures aiming to ensure the effective-ness and acceptability of the policy while main-taining a reasonable cost and a tolerable risk.The properties of a policy are determined by theproperties of its constituent policymeasures andtheir interactions, thus policy makers try to takeadvantage of the synergies between policy mea-sures and also try to avoid, or at least mitigate,the negative interactions between them.
A systematic approach for the synthesis ofpolicies is being developed in the area of trans-port policy to reduce CO2 emissions [225]:
This systematic approach may constitute thefirst step towards the development of a family ofcomputer-based systems that support the designof policies for different sectors, such as trans-port, energy, food, and water aiming to achieveenvironmental, security, health, and safety tar-gets. The output from such systems is a set ofpromising alternative policies, each annotatedwith its associated advantages and disadvan-tages. The final decision on which policyto implement rests with the decision makerswho may decide to include additional policy
Process Systems Engineering, 9. Domain Engineering 43
measures or remove some of the recommendedones.
5.10. Future Directions
The integrated design and analysis of socio-technical systems remains a challenge for both,policy developers and engineers; there is still alarge gap between the objectives of both com-munities and reality.
Engineers cannot expect to be able to formu-late, analyze, and predict the behavior of poli-cies in the same manner as they have done it forindustrial processes because sociotechnical sys-tems are different and more complex than tech-nical systems. There is, however, a large scopefor the application and/or adaptation of main-stream PSEmethods and tools for the support ofpolicy making, particularly in the analysis andsynthesis of policies.
A word of warning: there is a risk in havingexcessive confidence on models that may beinaccurate or incomplete because theymay givea delusion of control on systems that may beintrinsically unpredictable. Yet even in the caseof systems that cannot be properly characterizedand controlled, models can be useful if we askthe right question, i.e., to provide insights andfurther understanding of the workings of asociotechnical system rather than to predict itsfuture behavior.
References
1 N. Metropolis, A.W. Rosenbluth, M.N. Rosenbluth, A.N. Tell-
er, E. Teller, J. Chem. Phys. 21 (1953) 1087–1092.
