This is an author-deposited version published in:
http://oatao.univ-toulouse.fr/
Eprints ID:5092
To link to this article
: DOI: 10. 1016/j.compchemeng.2011.09.016
http://dx.doi.org/10.1016/j.compchemeng.2011.09.016
To cite this version
:
Ouattara, Adama and Pibouleau, Luc and
Azzaro-Pantel, Catherine and Domenech, Serge and Baudet, Philippe and Yao
Kouassi, Benjamin Economic and environmental strategies for process
design. (2012) Computers & Chemical Engineering, vol. 36 . pp. 174-188.
ISSN 0098-1354
㩷
O
pen
A
rchive
T
oulouse
A
rchive
O
uverte (
OATAO
)
OATAO is an open access repository that collects the work of Toulouse researchers and
makes it freely available over the web where possible.
Any correspondence concerning this service should be sent to the repository
administrator:
staff-oatao@inp-toulouse.fr
㩷
Economic
and
environmental
strategies
for
process
design
Adama
Ouattara
a, Luc
Pibouleau
a, Catherine
Azzaro-Pantel
a,∗,
Serge
Domenech
a,
Philippe
Baudet
b,
Benjamin
Yao
caLGC-CNRS-INPT,UniversitédeToulouse,4,AlléeEmileMonso,BP84234,F-31432Toulouse1,France bProSim,StratègeBâtimentA,BP27210,F-31672LabègeCedex,France
cInstitutNationalPolytechniqueHouphouët-Boigny,Départementdegéniechimiqueetagro-alimentaire,BP1093Yamoussoukro,Coted’Ivoire
Keywords: Multiobjectiveoptimization Geneticalgorithm Eco-efficiency Economiccriterion Environmentalimpact
a
b
s
t
r
a
c
t
Thispaperfirstaddressesthedefinitionofvariousobjectivesinvolvedineco-efficientprocesses,
tak-ingsimultaneouslyintoaccountecologicalandeconomicconsiderations.Theenvironmentalaspectat
thepreliminarydesignphaseofchemicalprocessesisquantifiedbyusingasetofmetricsorindicators
followingtheguidelinesofsustainabilityconceptsproposedbyIChemE(2001).Theresulting
multiob-jectiveproblemissolvedbyageneticalgorithmfollowinganimprovedvariantoftheso-calledNSGA
IIalgorithm.AkeypointforevaluatingenvironmentalburdensistheuseofthepackageARIANETM,a
decisionsupporttooldedicatedtothemanagementofplantsutilities(steam,electricity,hotwater,etc.)
andpollutants(CO2,SO2,NO,etc.),implementedherebothtocomputetheprimaryenergy
require-mentsof theprocessandtoquantify itspollutant emissions.Thewell-knownbenchmark process
forhydrodealkylation(HDA)oftoluenetoproducebenzene,revisitedhereinamultiobjective
opti-mizationway,isusedtoillustratetheapproachforfindingeco-friendlyandcost-effectivedesigns.
Preliminarybiobjectivestudiesarecarriedoutforeliminatingredundantenvironmentalobjectives.
Thetrade-offbetweeneconomicandenvironmentalobjectivesisillustratedthroughParetocurves.In
ordertoaiddecisionmakingamongthevariousalternativesthatcanbegeneratedafterthisstep,a
syntheticevaluationmethod,basedontheso-calledTechniqueforOrderPreferencebySimilarityto
IdealSolution(TOPSIS)(Opricovic&Tzeng,2004),hasbeenfirstused.Anothersimpleprocedurenamed
FUCAhasalsobeenimplementedandshownitsefficiencyvs.TOPSIS.Twoscenariosarestudied;in
theformer,thegoalistofindthebesttrade-offbetweeneconomicandecologicalaspectswhilethe
lattercaseaimsatdefiningthebestcompromisebetweeneconomicandmorestrictenvironmental
impacts.
1. Introduction
In traditional chemical process design, attention has been focusedprimarilyupontheeconomicviability.Yet,chemicalplants can no longer be designed on the unique basis of technico-economicconcernsandtheotherdimensionsofsustainability– environmentaland social–leadingtotheso-called“Triple Bot-tomline”,mustbepartandparcelofthedesignphase.Thisstudy aimsatthedevelopmentofadesignframeworkforeco-efficient processes,following theguidelinesoftheenvironmentally con-sciousdesign(ECD)methodologyproposedbyAllenandShonnard (2002).
∗ Correspondingauthor.
E-mailaddresses:luc.pibouleau@ensiacet.fr(L.Pibouleau),
Catherine.AzzaroPantel@ensiacet.fr(C.Azzaro-Pantel),
Philippe.Baudet@ProSim.net(P.Baudet),beyao@yahoo.fr(B.Yao).
Amajordifficultytotackletheproblemisthattherearemany independentbutoftencompetingobjectivesthathavetobe consid-eredsimultaneously.Lotsofongoingresearchesaimatdeveloping asetofmetricsor indicatorsasproposedbyIChemE(2001).In thededicated literature,the amountof metricsmay vary from 10(AIChE,1998)to134(CSD,1996)todrawaquantitative pro-file of sustainability.The implementation ofa process-oriented sustainability metrics hasbeencarried out in parallelwith the developmentof adesign framework foreco-efficient processes, followingtheguidelinesoftheenvironmentallyconsciousdesign (ECD)methodologyproposedbyAllenandShonnard(2002).For thispurpose,severalindexesofenvironmentalimpactincluding ozone depletion,GlobalWarming Potential,humanandaquatic toxicity,photochemicaloxidation aswellasacidrainpotentials have tobetaken intoaccount. Such problems leadto multiple and mostoftenconflictinggoalsand must besolvedby means ofefficientmultiobjectiveoptimizationtools.Manyrecentworks usedthecombinationofmultiobjectiveoptimizationandlifecycle assessment approach for theeco-design of chemical processes.
Nomenclature
AP AtmosphericAcidificationPotential(eqtSO2/y)
Di distillateflowrateincolumnofdistillationi(kg/h)
EP EutrophicationPotential(eqtPO43−/y)
EBi ithenvironmentalburden
Fi alimentationflowrate(kg/h)
FUCA Frenchacronymfor“FaireUnChoixAdéquat” GWP GlobalWarmingPotential(eqtCO2/y)
HDA benzene production fromtoluene hydrodealkyla-tion
Hi,Ho enthalpiesofinputandoutputsteams(kJ/kg)
HP highpressure(Bar)
HTP HumanToxicityPotential(tC6H6/y)
hDi enthalpyofdistillateflowrateindevicei(kJ/kg)
hFi enthalpyofalimentationflowrateindevicei(kJ/kg)
hwi enthalpyofwasteflowrateindevicei(kJ/kg)
1Hstmi enthalpyofvapourizationofwaterinuniti(kJ/kg)
1Hstri enthalpy of vapourization of chemical stream i
(kJ/kg)
LP lowpressure(Bar)
MCDM multiple-criteriadecisionmaking MILP mixedintegerlinearprogramming MINLP mixedintegernonlinearprogramming
MP mediumpressure(Bar)
NLP nonlinearprogramming
˙mFO consumedflowrateoffueloil(t/h)
˙mstmi steamflowratedemandfordevicei(t/h)
˙m consumedflowrateofnaturalgasfuel(NM3/h)
NSGA non-sortedgeneticalgorithm
PCOP PhotoChemicalOxidationorsmogformation Poten-tial(eqtC2H4/y)
PEI potentialenvironmentalimpact Qbi heatsuppliedtoreboileri(kJ/h)
Qci heatextractedfromtheuniti(kJ/h)
R refluxratio
ratio fuelratio(%)
SBX simulatedbinarycrossover
TOPSIS TechniqueforOrderPreferencebySimilaritytoIdeal Solution
WAR WasteReductionAlgorithm
Wi wasteflowrate(kg/h)
furnaceorboileryield(%)
correlationcoefficient
Thisapproachisincreasinglybeingusedintheliterature(Azapagic &Clift,1999;Berhane,Gonzalo,Laureano,&Dieter,2009,2010; Gonzalo,2011).
Thisworkaimsatdevelopingadesignframework incorporat-ingenvironmentalissuesduringthepreliminarystagesofaprocess design.Theenvironmentcomponentisidentifiedasseveraldesign objectivefunctionsatthelevelofaprocessflowsheetsynthesis. First,a briefliterature surveydedicated toeco-friendlyprocess designviamodellingandoptimizationformulationisproposedin thispaper.Then,themethodologicalframeworkadoptedinthis workforeco-efficientchemicalprocessdesignisdeveloped.Inthis study,optimizationisperformedbya geneticalgorithm imple-mentedintheMultigenlibrary(Gomezetal.,2010)thatturnedout tobeparticularlywell-suitedformultiobjectiveprocess optimiza-tion.Thewell-known benchmarkprocessfor hydrodealkylation (HDA)of toluene toproduce benzene (Douglas,1988)is revis-itedinamultiobjectivemodeandillustratestheusefulnessofthe approach infinding environmentallyfriendly and cost-effective designs. A key point of the methodology is to capture in the
modellingapproachbothprocessandutilityproductionunits,since theenvironmentalimpactofachemicalprocessisnotonly embed-dedintheproductsinvolvedintheprocessbutisalsorelatedtothe energyconsumption,theeffectofflowrecycle,percentconversion andsoon.Thetrade-offbetweenthesevenconsideredobjectives (production,annualcostandfiveenvironmentalimpacts)begins byabiobjectiveeconomicoptimization(production,annualcost). Then, for a given production, the annual cost is deducedfrom theParetocurve.Biobjective optimizations areimplementedto establishmultilinearrelationsbetweensomeobjectivesinorder toreducethemultiobjectiveproblemsizetothreeantagonist cri-teria.Twocasesarefinallystudied.Intheformercasestudy,under afixedproductionlevel,theaimistofindtheoperatingconditions forgloballyimprovingtheenvironmentalimpacts.Inthelatter sce-nario,theobjectiveis tosatisfyenvironmentalrequirementsas thosethat might bedefinedby anEnvironmentalProtection in ordertohaveaGWP(GlobalWarningPotential)valuelessthan agiventhreshold;theotherenvironmentalimpactsarerequired tobeinferiortothevaluesobtainedinthefirstscenario.
2. Literaturereviewrelatedtoeco-friendlyprocessdesign In recent years, eco-design, which consists in taking into account environmental assessment at the preliminary stage of processdesign,hasbeenrecognizedasanefficient environmen-tallyfriendlyalternativetotraditionalprocessdesign.Currently, there is nostandardized methodology and almost no practical experienceinintegratingsustainablecriteriaintoprocessdesign (Azapagic,Millington,&Collett,2006).Generally,inthis kindof problem,theobjectiveistosimultaneouslymaximizeprofitwhile minimizingenvironmentalimpacts.Twoapproachesaregenerally consideredinchemicalsystemmodelling,thefirstoneisbasedon mathematicalprogrammingmethodsincludingeither determinis-ticalgorithmssuchasMINLP,NLP,MILPformulationsorstochastic optimizationtechniquestosolvethiskindofproblems.Some inter-estingreferencesinclude:SuryaandAlex(2002),Jia,Zhang,Wang, andHan(2006),TveitandFogelholm(2006),andVasiliosandShang (2008).Thesecondtechniquetomodelchemicalprocessesisthe useofsimulators.Flowsheetingprogrampackages,likeCHEMCAD, AspenPlus,HYSYS,PRO/II,and ProSimPlus,arecommonlyused inchemicalengineeringforprocessdesign.Theycanbeusedin anouter optimizationloop to optimizedifferent criteria.Some significantworks canbefoundinStanislav(2003),Lim,Dennis,
Murthy,andRangaiah(2005),Othman,Repke,Wozny,andHuang
(2010),andIskandarandRajagopalan(2011).Carvalho,Gani,and Matos(2008)developedagenericandsystematicmethodologyfor identifyingthefeasibleretrofitdesignalternativesofanychemical process.Thissystematicmethodologyhasbeenimplementedinto anEXCEL-basedsoftwarecalledSustain-pro.
More generally, sustainabledevelopment takesinto account theconceptoflifecycleassessment(LCA),whichisamethodfor analysingand assessing the environment impact of a material, productorservicethroughouttheentirelifecycle.Awholelifecycle includesallprocessesfromthecradletothegrave,i.e.rawmaterial, extraction,processing,transportation,manufacturing,distribution, use,reuse,maintenance,recyclingandwastetreatment.Björkand
Rasmuson(2002)developanewapproachtodesign
environmen-talimpactascumulativeformationofecopoints.Theyshowthat LCAisavaluabletool forenvironmentaloptimizationofenergy systems.Theapplicationofthewholelifecycleassessmentmay beconsideredas verytediousdue tothelackof informationat somestageofitsdevelopment.Tocircumventthisdifficulties,aset ofmetricsorindicatorsfollowingtheguidelinesofsustainability conceptshavebeendeveloped,IChemE(2001),AIChE(1998),CSD (1996)andsoon.Theseindicatorsusetheconceptofenvironmental
burdensdefinedasaquantitativemeasureofthepotential contri-butionofsubstancesreleased,toaparticularenvironmental poten-tialimpact.Itmustbeyethighlightedthattheyareoftenlimitedto acradle-to-gateorgate-to-gatestudy.Theenvironmentalburdens areusedtoevaluatetheenvironmentalimpacts insome strate-giesliketheWasteReductionAlgorithm(WAR)(Chen&Li,2008; Douglas&Heriberto,1999;Heriberto,Jane,&Subir,1999;Teresa, Raymond,Douglas,&Carlos,2003),theIChemEsustainability met-rics(DinizdaCosta&Pagan,2006;Labuschagne,Brent,&vanErck, 2005)theSustainableProcessIndex(SPI)(Ku-Pineda&Tan,2006; Narodoslawsky&Krotscheck,2004;Sandholzer&Narodoslawsky, 2007).Foragivenprocess,thepotentialenvironmentalimpacts arecalculatedfromstreammassflowrates,streamcomposition andemissionsfromutilitysystemsandarelativepotential envi-ronmentalscore(index)foreachchemicalcompoundisdeduced. TheSustainableProcessIndex(SPI)isanecologicalevaluation sys-temindex specially developed for the requirementsof process engineering.NarodoslawskyandKrotscheck(2004)usedthis envi-ronmentalevaluationmethodologytostudythesustainabilityof energyproduction systems. TheWARalgorithm is a methodol-ogyfordeterminingthepotentialenvironmentalimpact(PEI)of achemicalprocess.ThePEIbalanceisaquantitativeindicatorof theenvironmentalfriendlinessorunfriendlinessofa manufactur-ingprocess.TheWARalgorithmwasfirstintroducedbyHilalyand Sikdar(1994).Theyintroducedtheconceptofapollutionbalance whichwastheprecursortothePEIbalance.Douglas,Richard,and Heriberto(2000)usedtheWARalgorithmtoevaluatethe environ-mentimpactofanallylchlorideproductionfacility.Asystematic approachforsustainableassessmentofchemicalandenergy pro-ductionprocesswhichincorporatesexergyanalysistoquantifythe efficiencyofaprocessandanenhancedinherentsafetyindexto quantifythesocietalimpactofaprocess,hasbeenproposedbyLi, Zanwar,Jayswal,Lou,andHuang(2011).AccordingtoChenandLi (2008),thismethodgenerallydividedtheimpactcategoriesinto twogeneralareaswithfourcategories. Thefirstgeneralareais theglobalatmosphericlevelinvolvingGlobalWarmingPotential (GWP),Ozone DepletionPotential (ODP),Acidificationand acid-rainPotential(AP),PhotoChemicalOxidationorsmogformation Potential(PCOP).Thesecondgeneralareaisrelatedtothelocal toxi-cologicalimpactlevelandisassociatedwithHumanToxicity Poten-tialbyIngestion(HTPI),HumanToxicityPotentialbyeither inhala-tionordermalExposure(HTPE),AquaticToxicityPotential(ATP) andTerrestrialToxicityPotential(TTP).LiketheWARalgorithm,the methodologydevelopedbyIChemE(2001),alsousestheconceptof potencyfactorsfordifferenttypesofpollutants.Thetotal environ-mentalburdenrelatedtoanenvironmentalimpactcategoryis cal-culatedbysummingtheproductofdifferentpotencyfactorsof pol-lutantwiththemassflowrateofpollutantswhichcontributetothis environmentalburden.VasiliosandShang(2008)usedtheIChemE (2001)methodtocalculatetheenvironmentalimpactsofutility productionsystems.Thistechniqueisalsoadoptedinthisstudy fortheenvironmentalevaluation,sincethesecriteriahavebeen identifiedasrepresentativeoftheprocessindustries.Atthefinal stageoftheoptimizationstep,itmaybenecessarytohelpdecision makersindeterminingtrade-offsolutions,severaldecision anal-ysismethodscanbeimplemented.Pirdashti,Ghadi,Mohammadi, andShojatalab(2009)andZhouandPoh(2006)classifydecision analysismethodsintothreemaingroups:singleobjectivedecision making(SODM)methods,Multi-CriteriaDecisionMaking(MCDM) methods,anddecisionsupportsystems(DSS).Theiranalysis high-lightsthatMCDMmethodsthatarestructuralapproachtoanalyse problemswithseveralcriteriaandalternativesarethemostwidely usedstrategies.Theyhelpdecisionmakerstomakeconsistent deci-sionsbytakingalltheimportantobjectiveandsubjectivefactors intoaccount.AcomparisoncanbefoundinGoughandWard(1996). Amongtheavailabletechniques,theso-calledTechniqueforOrder
PreferencebySimilaritytoIdealSolution(TOPSIS)whichbelongsto MCDMgroupisusedforidentifyingthesetofoptimalparameters innumerousinvestigations.Li,Zhang,Zhang,andSuzuki(2009), Jiaetal.(2006),Ren,Zhang,Wang,andSun(2007)usedTOPSISas multicriteriadecisionmethodtotrade-offsolutions.
3. Eco-efficientchemicalprocessdesign 3.1. Generalframework
Thedesignofeco-efficientchemicalprocessesproposedinthis study(seeFig.1)integratesamathematicalmodelforthe consid-eredprocess,coupledwithanimpactassessment model,which is embedded in an outer multiobjective optimization loop. For processmodelling,commercialdesignandflowsheetingpackages couldbeclassicallyused.Theproposedframeworkisan alterna-tivedesignmethodologyforwasteminimizationtotheso-called WARalgorithm(Cabezas,Bare,&Mallick,1999;Young,Scharp,& Cabezas,2000),whichhasbeenextensivelyusedintheliterature. Letusrecallthatthismethodisbasedonapotential environmen-talimpact(PEI)balanceforchemicalprocesses.ThePEIisarelative measureofthepotentialforachemicaltohaveanadverseeffect onhumanhealthandtheenvironment(e.g.aquaticecotoxicolgy, globalwarming,etc.).TheresultofthePEIbalanceisanimpact (pol-lution)indexthatprovidesaquantitativemeasurefortheimpact ofthewastegeneratedintheprocess.
Recently,severalsystematicmethodologieshavebecome avail-ableforthedetailedcharacterizationoftheenvironmentalimpacts of chemicals,products, and processes, which include LifeCycle Assessment.The ideais tousetheirpotentialtodevelop a Life CycleAnalysismethoddedicatedtoprocess development.Since ourapproachfocusesondecreasingtheenvironmentalimpactsof themanufacturingstageandutilitysystems,onlya“cradle-to-gate” analysisisperformed.
Akeypointofthemethodologyimplementedhereistheuse of ARIANETM (http://www.prosim.net/en/energy/ARIANE.html,
2005) a decision supporttool dedicated tothemanagement of plants that produce energy under the form of utilities (steam, electricity, hot water, etc.) included in the PlessalaTM module
developedbyProSimS.A.(2005).ARIANETMisusedherebothto
computetheprimaryenergyrequirementsoftheprocessandto quantifythepollutantemissionsduetoenergyproduction. 3.2. Multiobjectiveoptimization
Like many real world examples,the problemunder consid-erationinvolvesseveralcompetingmeasuresofperformance,or objectives(Collette&Siarry,2002).Usingtheformulationof mul-tiobjectiveconstrainedproblemsofFonsecaandFleming(1998), ageneralmultiobjectiveproblemismadeupasetofncriteriafk,
k=1,...,n,tobeminimizedormaximized.Eachfkmaybe
nonlin-ear,butalsodiscontinuouswithrespecttosomecomponentsofthe generaldecisionvariablexinanm-dimensionaluniverseU.
f(x)=(f1(x),...,fn(x)) (1)
Thiskindofproblemhasnotauniquesolutioningeneral,but presentsasetofnon-dominatedsolutionsnamedPareto-optimal setorPareto-optimalfront.ThePareto-dominationconceptlieson twobasicrules:intheuniverseUagivenvectoru=(u1,...,un)
dominatesanothervector
v
=(v
1,...,v
n),ifandonlyif,∀
i∈{1,...,n}: ui≤v
i∧∃
i∈{1,...,n}: ui<v
i (2)Foraconcretemathematicalproblem,Eq.(2)givesthe follow-ingdefinitionoftheParetofront:forasetofncriteria,asolution f(x),relatedtoadecisionvariablevectorx=(x1,...,xm),dominates
Fig.1.Generaloptimizationandmodellingframework.
anothersolutionf(y),relatedtoy=(y1,...,ym)whenthefollowing
conditionischecked(foraminimizationproblem),
∀
i∈{1,...,n}: fi(x)≤fi(y)∧∃
i∈{1,...,n}: fi(x)<fi(y) (3)Onagiven setof solutions,itis possibletodistinguish non-dominatedsets.ThelastdefinitionconcernstheParetooptimality: asolutionxu∈UiscalledPareto-optimalifandonlyifthereisno
xv∈Uforwhich
v
=f(xv)=(v
1,...,v
n)dominatesu=f(xu)=(u1,..., un).These Pareto-optimal non-dominated individuals representthesolutionsofthemultiobjectiveproblem.
Facedtothediversity ofmathematicalproblems,itis recog-nized thatthere is not a unique and generalalgorithm ableto solvealltheproblems perfectly.Actually,methodefficiency for aparticularexampleishardlypredictable,andtheonlycertainty wehave is expressedbytheNo Free Lunchtheory(Wolpert & Macready,1997):there isnomethodthatoutdoesalltheother onesforanyconsideredproblem.Thisfeaturegeneratesacommon lackofexplanationconcerningtheuseofamethodforthesolution ofaparticularexample,andusually,norelevantjustificationfor itschoiceisgivenapriori.Apossiblesolutionwastodevelop sev-eralalgorithms,distinguishingthembytheirstructureandbytheir typeofvariables(continuous,integer,binary)andcollecttheminto adatabase:Multigen(Gomezetal.,2010)liesonthis principle, andcurrently,sixdifferentalgorithmsareavailable.Theaimwas totreatmultiobjectiveconstrained-optimizationproblems involv-ingvarioustypesofvariables(boolean,integer,real)andsomeof theseproblemscanbestructuraloptimizationones.Multigenhas beenimplementedinVBAandinterfacedwithMicrosoftExcel®
. ThealgorithmsmusthandleconstraintsaswellasPareto domi-nationprinciples.Inthatway,proceduresbasedonindependent objectivestocarryouttheselection(likeVEGA,Schaffer,1985)are notadaptedtotheconsideredproblems.Proceduresbasedofthe nichenotion(NPGA–Horn,Nafpliotis,&Goldberg,1994;MOGA –Fonseca&Fleming,1993)cannotguarantee acorrect conver-genceoftheParetofront,duetothelowdiversityofthegenerated populations.MethodslikeSPEA(Zitzler&Thiele,1999)andNSGA
II(Deb,Pratap, Agarwal, & Meyarivan,2002) favournot domi-natedisolatedindividuals.InSPEA,theprobabilityofselectionis afunction oftheindividualisolation, whichis quitedifficultto implement.InNSGA,individualsfromthemostcrowdedzonesare eliminatedaccordingtoacrowdingsorting.Takingintoaccountall thepreviousitems,NSGAIIwaschosenasthebasisof develop-mentoftheMultigenlibrary.InthispaperthepackageNSGAIIb, whichdiffersfromtheinitialNSGAIIbythecrossoveroperator,has beenretained.Notethat,NSGAIIbimplementsthesamealgorithm thanNSGAII,withcorrectionsoncrossoveroperatortoavoidthe creationofclonesinherentofSBXoriginalversion.Whenthe gen-eratedrandomnumberusedtoperformthecrossoverisgreater thanagivencrossoverprobability,thecrossovermayproducetwo childrenidenticaltotheparents:SBXcrossovercodedinNSGAIIb includesaforcedmutationofchildrenwhenthiseventoccurs.The objectiveistoavoidunnecessarycalculationsforclonesofexisting solutions:allsolutionsgeneratedbythereproductionprocedure arestatisticallydifferent.TheNSGAIIbalgorithmwasimplemented totreatvariablesofdifferentnature,eithercontinuous,orinteger. Sinceouranalysisisrestrictedtotreatcontinuousvariables,the waytohandleintegervariablesisnotdetailedinthispaper.The interestedreaderwillfinmoreinformationin(Gomezetal.,2010). Constrainedmultiobjectiveoptimizationisthemostcommon, kindofprobleminengineeringapplications.Ingeneral,threekinds ofconstraintsareconsidered:simpleinequality(≤),strict inequal-ity(<),andequality(=):
g(x)≤c1 r(x)<c2 h(x)=c3
⇔
constr1(x)=c1−g(x)≥0 constr2(x)=c2−r(x)>0 constr3(x)=c3−h(x)=0 (6)where(g,r,h)arereal-valuedfunctionsofadecisionvariablex=(x1,
c3)areconstantvalues.Inthemoregeneralcase,theseconstraints arewrittenasvectorsofthetype:
coEnstr1(x)=((c1−g(x))1,...,(c1−g(x))n1) =(constr1(x)1,...,constr1(x)n1)≥0 coEnstr2(x)=((c2−r(x))1,...,(c2−r(x))n2) =(constr2(x)1,...,constr2(x)n2)>0 coEnstr3(x)=(−|(c3−h(x))|1,... ,−|(c3−h(x))|n3) =(constr3(x)1,...,constr3(x)n3)=0 (7)
where n1, n2, and n3 arerespectively, the number or inequal-ity, strict inequality and equality constraints. This formulation impliesthateachconstrivaluewillbenegativeifandonlyifthis
constraint is violated. The conversionof Eq. (6), that is a clas-sical representationof constraintsset,toEq. (7) representation constitutesthefirststepofaunifiedformulationof constrained-optimization problems. In practice, due to round-off error on real numbers, the equality constraint constr3 was modified as
follows: coEnstr3(x)′=(−|(c3−h(x))|1+ε1,...,−|(c3−h(x))|n3+εn3) =(coEnstr3(x))+ Eε =0 E ε=(ε1,...,εn3),
∀
i∈{1,...,n3}, εi∈R (8) Eεiscalleda“precisionvector”oftheequalityvector,andtakes lowvalues(lessthan10−6forexample).Thisapproximationisnot
necessarywhenequalityconstraintinvolvesonlyintegerorbinary variables.
FromEqs.(7)and (8),theconstraintsatisfaction impliesthe maximization ofviolated constraintsin vectorsconstr1,constr2,
andconstr3.AccordingtoFonsecaandFleming(1998),the
satisfac-tionofanumberofviolatedinequalityconstraintsis,fromEq.(7), amultiobjectivemaximizationproblem.Fromatheoreticalpoint ofview,a constrainedmultiobjective optimizationproblemcan beformulatedasatwo-stepoptimizationproblem.Thefirststep impliesthecomparisonofconstraintsatisfactiondegreesbetween twosolutions,usingthePareto’sdominationdefinitionofEq.(3), butamoresimplesolutionconsistsincomparingthesumof val-uesofviolated constraintsonly,asin NSGAIIalgorithmofDeb etal.(2002),whichimplies therearenopriorityrulesbetween constraints.
3.3. Multiple-criteriadecisionmaking(MCDM) 3.3.1. Introduction
MCDMapproachesaremajorpartsofdecisiontheoryand anal-ysis.MCDMareanalyticmethodstoevaluatetheadvantagesand disadvantagesofalternativesbasedonmultiple-criteria.The objec-tiveistohelpdecision-makerstolearnabouttheproblemsthey face,andtoidentifyapreferredcourseofactionforagiven prob-lem.Huang,Poh,andAng(1995)mentionedthatdecisionanalysis (DA)wasfirstappliedtostudyproblemsinoilandgasexploration inthe1960sanditsapplicationwassubsequentlyextendedfrom industrytothepublicsector.Tillnow,MCDMmethodshavebeen widelyusedinmany researchfields.Differentapproaches have beenproposed by many researchers, includingsingle objective decision-making(SODM)methods,MCDMmethods,anddecision support systems (DSS). Literature shows that among MCDM methods,DAstrategiesarethemostcommonlyused(Zhou&Poh, 2006).TheselectionofasingleParetopointfromthePareto fron-tiermaybedifficultasthenumberofobjectivesincreases.Some intuitivemethodssuchastheso-called“knee-method”couldbe efficientforabinarycase.Thisiswhyothermethodsarenecessary
to tackle the multicriteria nature of the results generated by theGA.
3.3.2. Findingknees
Branke,Deb,Dierolf,andOsswald(2004),andTaboadaandCoit (2006)suggestpickingthekneesintheParetofront,thatistosay, solutionswhere asmallimprovement inoneobjectivefunction wouldleadtoalargedeteriorationinatleastoneotherobjective. 3.3.3. TOPSIS
Inpractice,theTechniqueforOrderingPreferencebySimilarity toIdealSolutions(TOPSISmethod)and othermultipleattribute decision making(OMADM)methods arethemostpopular(Ren et al., 2007; Yoon & Hwang, 1995). In this work, the TOPSIS methodologywasusedasdecisionmakingtool.Afterthe gener-ationofthePareto-optimalset,theTOPSISmethodisusedtoaid decision–makerintrade-offingthewholealternatives.Thebasic principleofTOPSISisthatthechosenalternativeshouldhavethe shortestdistancefromtheidealsolutionandthefarthestdistance fromthenegative-idealsolution(Opricovic&Tzeng,2004; Ren etal.,2007;Saaty,1980).
StepsusedinTOPSISasdescribedbyRenetal.(2007)arebriefly describedinwhatfollows:
Step1:Alltheoriginalcriteriareceivetendencytreatment.That consistintransformingthecostcriteriaintobenefitcriteria,which isshownindetailasfollows:
(i)Thereciprocalratiomethod(X′
ij=1/Xij)referstotheabsolute
criteria;
(ii)Thedifferencemethod(X′
ij=1−Xij),referstotherelative
cri-teria.
Aftertendencytreatment,constructamatrix X′[X′
ij]n×m, i=1,2,...,n; j=1,2,...,m. (9)
Step2:CalculatethenormalizeddecisionmatrixAasfollows: A=[aij]n×m aij= X′i j max(X′i j)
(j=1,2...,m.) foracriteriontobemaximized (10a) aij=1− X′i j max(X′i j)
(j=1,2...,m.)foracriteriontobeminimized (10b)
Step3:Determinethepositiveidealandnegativeidealsolution fromthematrixA
A+=(a+ i1,a + i2,...,a + im), a + ij =1≤i≤nmax(aij), j=1,2,...,m (11) A−=(a−i1,a−i2,...,a−im), a−ij = min 1≤i≤n(aij), j=1,2,...,m (12)
Step 4: Calculate the separation measures, using the n-dimensionalEuclideandistance.Theseparationofeachalternative fromthepositiveidealsolutionisgivenas
D+i =
v
u
u
t
mX
j=1 Wj(a+ij−aij) 2 (13)Similarly, theseparation from the negativeideal solution is givenas D−i =
v
u
u
t
mX
j=1 Wj(a−ij −aij) 2 (14)Step5:Foreachalternative,calculatetheratioRias
Ri=
D−i
D−i +D+i , i=1,2,...,n (15)
Step6:Rankalternativesinincreasingorderaccordingtothe ratiovalueofRiinstep5
3.3.4. FUCA
FUCAistheFrenchacronymforFaireUnChoixAdéquat(Make AnAdequateChoice).Thissimplemethodisbasedonindividual rankingsofobjectives;foragivencriterion,rankoneisassignedto itsbestvalueandrankn(nbeingthenumberofpointsofthePareto front)totheworstone.Then,foreachpointofthefront,aweighted summation(theweightsrepresentingthepreferences)ofranksis performed,andthechoiceiscarriedoutaccordingtothelowest valuesof thesum. Ina recent paper(Moralez-Mendoza, Perez-Escobedo,Aguilar-Lasserre,Azzaro-Pantel,&Pibouleau,2011)the FUCAmethodwascomparedwithclassicalMCDMproceduresona tricriteriaproblemrelatedtotheportfoliomanagementina phar-maceuticalindustry.ForeachsolutionfoundbyELECTRE(Teixeiro deAlmeida,deMiranda,&CabralSeixasCosta,2004),PROMETHEE (Zhaoxa&Min,2010)andTOPSIS,theFUCArankingisalsoreported. AverygoodagreementbetweenthethreeclassicalMCDM meth-odsandFUCAcanbeobserved,showingtheefficiencyoftheFUCA procedure,whichalwaysfindsthebestsolutionselectedbyoneof theothers.
3.4. Utilityproductionmodelling
ARIANETM software tool has been developed by ProSim
Company(French chemical engineeringSoftware Company) for designing assistance and optimal operation of power plants. ARIANETM makesitpossibletooptimizeandmodel anyenergy
combinedproduction plant (steam, hotwater, electricity, com-pressedair,cooling),whateveritssizeandcomplexity.ARIANETM
isadecisionsupporttooldedicatedtothemanagementofplants thatproduceenergyundertheformofutilities(steam, electric-ity,hotwater,etc.).ARIANETM,writteninVBA,presentsastandard
libraryofunitoperationsinvolvedinpowerplants:boilers (mono-fuel,bi-fuel,electrical),turboalternators (backpressureturbines, sidestreamturbines,condensating turbines),switchable or per-mutableturbines(inparallelwithelectricmotors),fuelturbines andthermalengines (withorwithoutrecoveryheatexchanger, withorwithoutpostcombustionboiler),valves,heatexchangers anddeaerators.Alltheseunitoperationscanbedescribed,from thesimplest(minimalistmodelling, withdefaultvalues) tothe mostcomplex(veryfinemodellingofthecharacteristicsand tech-nicalconstraints);thecomplexityofphysicalmodelsrequiringin manycasestousenonlinearequations.Theinherentmodel com-plexityanditsassociatedequationsaretotallytransparentforthe user.Intheframeworkofeco-designmodelling,theproblemlies inintegratingenvironmentalaspectsinsideARIANEsoftware,so thatemissionscanbetakenintoaccountintheglobalenergy bal-ance.Estimationmethods,toevaluatetheemissionsofallinvolved powerplantsequipmentitemshavethentobecompatiblewith existingroutinesthatnotrequirenumerousand complexinput data.InARIANE,classicalpollutantsasnitrogenoxides,carbonand sulphuroxides,andalsosolidparticleslikedusts,arewellknown. Theemissionsmodellingprocessstartsfromthecalculationof pollutantamountsinthesmokesofeachequipmentitem,global
emissionflowrateisfinallythesumofallflowrates,forallsite smokeemitters.Anotherproblemliesthenintheaccurate mod-ellingofthecombustionphenomenainallequipmentitemsofa powerplant.Wegivebelowsomeoftheequationsofunits oper-ationsimplemented in ARIANETM and usedat the eco-efficient
modellingstage.
4. ApplicationtoHDA(hydrodealkylationoftoluene) process
4.1. OverviewoftheHDAprocess
The traditional method of manufacturingbenzene fromthe distillationoflightoilsproducedduringthemanufacturingofcoke hasbeenovertaken bya number ofprocesses:The sourcesare now:catalyticreforming,hydrodealkylationoftoluene(HDA)and toluenedisproportionation.Theglobalworldbenzeneproduction capacitywillrisefromestimated41.8Mt/yin2009to55.8Mt/y in 2015: http://marketpublishers.com/lists/5873/news.html,
http://mcgroup.co.uk/researches/B/028500/Benzene.html. In
2005, the French production capacity was estimated at one Mt/yearonsevenindustrialsites,i.e.meanof143,000t/yearper site(www.ineris.fr/rsde/fiches/fichebenzene2005.pdf).
TheHydrodealkylation(Douglas,1988)processforproducing benzene,aclassicalbenchmarkinchemicalprocesssynthesis stud-ies,isusedinthispaper(HDAprocess).Thisprocessinvolvestwo reactions:theconversionoftoluenetobenzene(Eq.(16))andthe equilibriumbetweenbenzeneanddiphenyl(Eq.(17)).
Toluene+H2→Benzene+CH4 (16)
2Benzene↔ Diphenyl+H2 (17)
Douglas(1988) has firstextensively studied this process by usingahierarchicaldesign/synthesisapproach.Thehydrogenfeed streamhasapurityof95%andinvolves5%ofmethane;thisstream ismixedwithafreshinletstreamoftoluene,recycledtoluene,and recycledhydrogen.Thefeedmixtureisheatedinafurnacebefore beingfed toan adiabatic reactor.The reactoreffluent contains unreactedhydrogenandtoluene,benzene(thedesiredproduct), biphenyl,andmethane;itisquenchedandsubsequentlycooled inahigh-pressureflashseparatortocondensethearomaticsfrom thenon-condensablehydrogenandmethane.Thevapoursteam fromthehigh-pressureflashunitcontainshydrogenandmethane thatisrecycled.Theliquidstreamcontainstracesofhydrogenand methanethatareseparatedfromthearomaticsinalow-pressure flashdrum.Theliquidstreamfromthelow-pressureflashdrum consistingofbenzene,biphenyl andtolueneisseparatedintwo distillationcolumns.Thefirstcolumnseparatestheproduct, ben-zene,frombiphenylandtoluene,whilethesecondoneseparates thebiphenyl from toluene, which isrecycled backat the reac-torentrance.Energyissavedbyusingtheoutletstreamleaving thereactorasitstemperatureis intherangeof 620◦C,to
pre-heatthefeedstreamcomingfromthemixer,viaaheatexchanger (Fehe)someenergyintegrationisachieved(Cao,Rossiter,Edwards, Knechtel,&Owens,1998)(seeFig.2).
Asabovementioned,ARIANETMisusedherebothtocompute
theprimaryenergyrequirementsoftheprocessandtoquantify thepollutantemissions related toenergy production. Atypical steamgeneratingfacilityisconsidered:steamisproducedfrom aconventionalmono-fuelfiredboiler(40barand400◦C)andlet
downto lowerpressure levels (respectively,10bar,336◦C and
5bar,268◦C)throughturbineswhichproduceelectricityusedin
theplant.ARIANETMisalsousedtomodelthefiredheater
(fur-nace)oftheprocessasabi-fuelfiredheater(mixofnaturalgasand fuel).
Fig.2.HDAplantwithisutilityproductionunit.
TomodeltheHDAprocess,commercialdesignandflowsheeting packagescouldbeusedinordertocomputetheobjectivefunctions. However,ratherthanusingsuchpackages,theequationsproposed byDouglas(1988)havebeendirectlyimplementedandsolvedby theExcel®
solver.Biphenylhasbeenconsideredasapollutant,and theenvironmentalimpactcontributionsforthecomponentsinthe HDAprocesshavebeentakenfromSikdarandEl-Halwagi(2001).
4.2. InteractionbetweenHDAprocessandutilityproduction system
FromthestudiesofDouglas(1988)andTurton,Bailie,Whiting, &Shaeiwitz(1997,2009),sevenvariableswereselectedforHDA processbecauseoftheirinfluenceontheeconomicand environ-mentaloptimizationcriteria.Thesevariablesareinitiallyusedto computetheoverallmaterialbalanceofthemainchemical com-poundsandaswellastheassociatedthermodynamicproperties (enthalpy,density,heatcapacity, etc.)atvariousprocessnodes, withuseofSimulis® Thermodynamicsasacalculationserverof
thermodynamicproperties.Theyservefinallytocomputethesize ofallequipmentitemsinvolvedintheprocessforcarryingoutthe requiredunitoperations.Fig.3showstheinteractionbetweenthe variablesof theHDA processandtheutilityproductionsystem. HDAprocessoperationneedsthermalandelectricalutilities.The thermaldemandisformulatedintermsoffuelflowratetomeet theneedofheatingthemixtureinthefurnaceaswellasofsteam tomeettheheatingdemandsfortheexchangers,reboilerandother itemsoftheprocess.Eqs.(18)–(20)describethedemandofsteamin thecaseofareboiler.Theserequirementsobtainedfromtheenergy balancesoftheHDAsimulatoraretransmittedviaPlessalaTMas
variablestotheunitoperationsinvolved,whicharemodelledby useofArianeTM(boiler,furnace).Then,fueloilflowrateof,the
nat-uralgasflowrateandsteamflowratebecomesecondaryvariables forArianeTM.TheuseofArianeTMprovidestypicalresultsrelatedto
thethermalpowerplanti.e.thepowerproducedbytheturbineand theflowratesofallthepollutantsresultingfromfuelcombustion.
˙mstmi=
Qbi
1Hstmi
Fig.3.Interactionbetweenvariablesandsimulationtools.
with Qbi=DihDi+WihWi−FihFi+Qci (19)
and Qci=Di(R+1)1Hstri (20)
4.2.1. Furnacemodelling
Thefurnaceismodelledasabi-fuelboilerfedwithbothnatural gasandfueloil.Theirflowratesarelinkedbytheso-calledenergetic ratioandthefurnaceenergeticdemandasfollows:
Consumednaturalgasflowrateinthefurnace:QNG
˙m = QFurnace ·LHVNG·ratio
(21) Consumedfueloilflowrateinthefurnace:QFO
˙m = QFurnace ·LHVFO·(1−ratio)
(22) Energeticratio:
ratio= Energyprovidedbynaturalgas
Totalenergyofnaturalgasandfueloil (23)
Thefuelsconsideredarenaturalgas(fuel1)andfueloil(fuel2). Caseofamono-fuelboiler:
Theboilerisusedtoproducesuperheatedvapour,forthe oper-ationoftheturbineandhotwaterfortheotherunits
˙m= QBoiler ·LHVNG
. (24)
4.2.2. Sidestreamturbinemodelling
Aboilerisusedtoproductsteamathighlevelofpressureand temperature,and thenthis steamis expended througha back-pressureandcondensingturbinetogeneratepower.Theturbineis modelledwiththefollowingequations,implementedinARIANETM:
Si(Ti,Pi)=So(To∗,Po) (25)
Turbinemaximumwork
Wmax=Hi(Ti,Pi)−Ho(To∗,Po) (26)
Turbineisentropicyield = Weff
Wmax
(27)
5. Optimizationproblemformulation 5.1. Economicobjectives
AsmentionedbyCutlerandPerry(1983),amajorobjectiveof anyindustrial processis profitmaximization. In thisstudy,the benzeneproduction(ProdB)tobemaximizedandtheannualcost (Annualcost)havebeenchosenaseconomicfunctions.Thebenzene productionisadirectoutputoftheHDAprocess,whiletheannual costiscomputedaccordingtorelations(28)–(31).
Annual cost=0.1FCI+CRM+CUT (28)
FCI:Fixedcapitalinvestment($) 0.1FCI:depreciationcost($/y) CRM:costofrawmaterials($/y)
CUT:costofutilities($/y)
FCI=
X
i
(Purchase costi+Installedcosti) (29)
CRM=
X
i
Table1
Capitalcostestimationformainitems(M&S:Marshall&SwiftEquipmentsCostIndex=1,468.6(2009))fromChemicalEngineeringJanuary2010.
Equipmentinvestmentcost($) Nonlinearform Columncost
D:columndiameter(m) Purchasecost=9.201
!
M&S 280(101.9D1.066H0.802F c) H:columnheight(m)
Fc:materialpressure Installedcost=9.202
!
M&S280D1.066H0.802(2.18+Fc)+20.69!
M&S 280 4.7D1.55HF′ c F′c:material,trayspace,traytype
Exchangercost Purchasecost=
!
M&S 280(474.7A0.65F c) A:heatexchangerarea(m2) Installedcost=
!
M&S280
(474.7A0.65)(2.29+F c) Furnacecost Purchasecost=
!
M&S280
(5.52×103)Q0.85F c Q:furnaceabsorbed(293kW) Installedcost=
!
M&S280
(5.52×103)Q0.85(1.27+F c)
Table2
Capitalcostfortheutilitysystem.
Equipmentsinvestmentcost(1000$) Nonlinearform Fielderectedboiler 8.09F0.82f
p1
F:steamflowrate(kg/s);P:pressure(bar) With,fp1=0.6939+0.01241P−3.7984Exp(−3P2) Steamturbine Wst:power(MW) 25.79Wst0.41 Deaerator F:BFWflowrate(kg/s) 0.41F0.62 Condenser Q:heatdissipated(MW) 4.76Q0.68 CUT=
X
i ˙mUTiPUTi (31)˙mRMi:massflowrateofrawmateriali(kg/h)
PRMi:unitpriceofrawmateriali($/kg)
˙mUTi:flowrateofutilityi(kg/h,stdm3/h,m3/horkW)
PUTi:unitpriceofrawmateriali($/kg,$/stdm3,$/m3or$/kWh)
Forthemainequipmentitems,thepurchaseandinstalledcosts werecomputed fromtheclassical Guthrie’scorrelations (1969) (seeTable1).Fortheutilitysystem,capitalcostestimationwas performedbyusingexpressionsdevelopedinBruno,Fernandez, Castells,andGrossmann(1998)(seeTable2).Finally,thecostsof rawmaterialsandutilitiesfromTurtonetal.(2009)aredisplayed inTable3.Forconvenience,theAnnualcostwillbeexpressedin M$/yinwhatfollows.
5.2. Environmentalobjectives
The environmental impacts are quantified by the so-called environmental burdens listed by IChemE. From the classifica-tionproposedbyAzapagic,Emsley,andHamerton(2003),eight
Table3
CostofrawmaterialsandutilitiesusedinHDAprocess.
Rawmaterials Cost($/kg)
Toluene 0.648
Hydrogen 1
Utilities Cost($/commonunit)
Fueloil(n◦2) $549/m3
Naturalgas $0.42/stdm3
Electricity $0.06/kWh
SteamfromBoiler
Highpressure $29.97/1000kg
Mediumpressure $28.31/1000kg
Lowpressure $27.70/1000kg
Coolingtowerwater(30–40◦C) $14.8/1000m3
categoriesofimpactscanbeidentified,andenvironmentalburden EBiiscomputedas: EBi= n
X
j=1 ecijBj (32)whereecijistheestimatedclassificationofthejthparticipantinthe ithburdenandBjistheemissionofthejthgreenhousegas.Fora
givenapplication,thechoiceofparticularimpactscanbededuced from the flowsheet analysis, and they are quantified through massandenergybalances.Amongtheenvironmentalimpacts pro-posedbyAzapagicetal.(2003),fiverepresentativeitemsforthe consideredchemicalprocesshavebeenretainedforcomposingthe EnvironmentalBurdens(EBi).EachconsideredEnvironmental
Bur-dencanbebrieflydescribedasfollows(seethestudyofAzapagic etal.(2003)formoreinformation).
- GlobalWarmingPotential(GWPintCO2equivalent/y)whichis
ameasureofhowagivenmassofgreenhousegasisestimated tocontributetotheglobalwarming.Itisarelativescale,which comparesthemassofaconsideredgastothesamemassofcarbon dioxide.ItisequaltothesumofeightemissionsBjofgreenhouse
gasesmultipliedbytheirrespectiveGWPfactorsecjGWP.
-AcidificationPotential(APintSO2equivalent/y)isbasedonthe
contributionsofSO2andNOxtothepotentialaciddeposition,that
isontheirpotentialtoforH+protons.ThefiveAPfactorsecj APare
expressedrelativelytotheAPofSO2.
-Photochemical Ozone Creation Potential (POCP in
tC2H4equivalent/y) (or PCOP) very often defined as
sum-mer smog, is theresult of reactionsthat take place between nitrogenoxides NOx andVOCs exposed toUVradiations.The
POCPisrelatedtoethyleneasreferencesubstance,thatistosay thatthePOCPfactorsecjAPfortheeightconcernedproductsare expressedrelativelytothePOCPofC2H4.
- HumanToxicityPotential(HTPintC6H6-equivalent/y)expresses
thepotentialharmofchemicalsreleasedintotheenvironment. HTPincludesbothinherenttoxicityandgenericsource-to-dose relationships for pollutant emissions. It uses a margin-to-exposureratiotoevaluatethepotentialforhealthimpactfrom
exposuretoharmfulagents,bothcarcinogensandnon carcino-gens.It iscalculated byaddinghumantoxic releasestothree differentmedia,i.e.air,waterandsoil.Thetoxicologicalfactors arecalculatedusingtheacceptabledailyintakeorthetolerable dailyintakeofthetoxicsubstances.Thehumantoxicological fac-tors(26fromAzapagicetal.,2003)arestillatdevelopmentstage, sothat HTPcanonly betaken asanindicationandnot asan absolutemeasureofthetoxicitypotential.
-EutrophicationPotential(EPintPO43−equivalent/y)isdefinedas
thepotentialofnutrientstocauseover-fertilisationofwaterand soilwhichinturncanresultinincreasedgrowthofbiomass.
Itmustbeemphasizedatthislevelthattheapproachpresented heredoesnotaccountfortheentirelifecycleoftheprocesssince theenvironmentalburdensareassociatedwiththeemissions gen-eratedbytheprocessbothfromthematerialcomponentsandfrom theenergyvectors. Forinstance,theoutcomeof benzeneisnot consideredhereasitisviewedasthedesiredproductforthe pro-cess.Wearethusawarethattheapproachdevelopedheredoesnot embedthewholelifecycleoftheproduct.Moreover,the extrac-tionresourceassociatedwithrawmaterialssuchastolueneisnot included.Briefly,onlytheprocesscontributionisconsideredinthis study.
Ofcourse,alifecycleassessmentmethodologycouldbe attrac-tive to be more general and the abovementioned drawbacks could be partially overcome by using standard environmental databasesavailablein theliterature (e.g.EcoInvent forinstance http://www.ecoinvent.ch/)implementedinlifecycleassessment softwaretools.Thisissuewillconstituteaperspectiveofthiswork. 5.3. Multiobjectiveproblemformulation
Using the previous economic and environmental objective functionsdefinedabove,thefollowing multiobjective nonlinear optimizationproblemisformulatedasfollows:determinedecision variables(processoperatingconditions)inorderto:
Max(ProdB) (33)
Min(Annualcost) (34)
Min(EBi), i=1, 5 (35)
s.t.
Massandenergybalances(Excel®andARIANETM)
Boundsondecisionvariables
ProdBrepresentsthebenzeneproductionatthetopofcolumn 2(inkmol/h).
Theadditionalfollowingconstraintshavebeenconsideredinthe Multigeninterface,thenumericalvaluesvaluehavebeendeduced fromtheanalysisofDouglas(1988)andTurtonetal.(1997). -Thepurityofthebenzeneproductisatleast99.97% - Thehydrogenfeedmusthaveapurityof95% -Thereactoroutlettemperatureislessthan704.50◦C
-Thequencheroutlettemperatureislessthan621.16◦C
-TheconversionrateCisboundedas:0.5≤C≤0.9
-The hydrogen flow rate purged FPH (kmol/h) is bounded as:
30≤FPH≤300
-Allpollutants,CO2,NOx,CO,SO2,dustsflowrate(kg/h)musttake
onlypositivevalues. 5.4. GAparameters
Thetwoclassesofoptimizationproblemspresentedbelowhave beensolvedbyusingthecodeNSGAIIboftheMultigentoolbox, withthefollowing operatingparameters: populationsize=200,
Fig.4. Benzeneproductionvs.Annualcost.
numberofgenerations=200, Crossoverrate=0.75 andMutation rate=0.2.
Douglas(1988)thenTurtonetal.(1997)defineboundsonthe HDAvariablesasreportedinTable4.Theinitialpopulationwas randomlygeneratedbetweenthesebounds,anda particularset ofvariableshasbeenextractedfromtheinitialpopulation(column InitialvalueofTable4)forfurthercomparisonpurposes(seeTable5 line“Initial”).
6. Case1:economicfollowedbyecologicalstudy 6.1. Solutionstrategy
Fortheprobleminvolvingalltheobjectives,theParetofrontisa cloudofpointslocatedinahypercubeofℜ7.Toreducethis hyper-cubedimensioninordertointerpretasgoodaspossiblethegenetic algorithmresults,a twostepssolutionstrategyisimplemented. Inthefirststep,thebiobjectiveproblem((33)and(34))issolved toobtainanefficienttrade-off(ProdB*,Annualcost*)betweenthe maximalbenzene productionand theminimalinvestmentcost. Thenthebenzeneproduction,beingfixedatavalueProdB*obtained fromtheParetocurve,theproblem((33)–(35))issolvedforthe investmentcostlyinginalowrange[Annualcost*,Annualcost*+˛], withavalueof˛equalto15%forinstance.Adegradationofthe annualcostisalsoallowed,thusimprovingtheenvironmental per-formancesoftheprocessofbenzeneproduction.
6.2. Economicstudy:biobjectiveoptimization(ProdB,Annual cost)
TheParetofrontcorrespondingtotheoptimizationofthepair (ProdB,Annualcost)isdisplayedinFig.4.Thecurveiscomposedof twoportionsofstraightlines.Thekneelocatedonthecurvegives thebesttrade-off(Brankeetal.,2004),thatistosay(205.04M$, 299.96kmol/h).Withoutenvironmentalconsiderations,this solu-tion would be adopted for plant design. In the following, the benzeneproductionwillbelowerboundedat300kmol/handthe annualcostwillvaryintherange[205.04M$,299.96M$]inorder tocompetewithadditionalenvironmentalobjectives.Thevalues ofallconsideredobjectivesforthefirstpoint(204.5,299.9)noted “First”oftheflatlineofFig.4and theinitialpointdescribedin Table4areusedforcomparingobjectivesinTable5.
6.3. Studyofenvironmentalburdensviabiobjectiveoptimizations Thebenzeneproductionbeingfixedat299.96kmol/handthe Annualcostlyingintherange[205.04M$,299.96M$],the crite-ria are optimized by pairs for identifying redundantobjectives expressedintermsofmultilinearfunctionsofthethreeremaining ones(Annualcost,EPandAP).Thesedependentimpacts(GWP,HTP andPOCP)willberemovedfromthemultiobjectiveoptimization problem,wheresomedegradationsoftheAnnualcostareallowed
Table4
DecisionvariablesfortheHDAprocess.
Decisionvariables Lowerbound Initialvalue Upperbound
Conversionrate(%) 0.5 0.75 0.9
Hydrogenpurgeflowrate(kmol/h) 31 198 308
Flashpressure(bar) 30 34 34
Stabilizerpressure(bar) 4 10 10
Column1pressure(bar) 2 2 4
Column2pressure(bar) 1 1 2
Ratio(bi-fuelfurnace)(%) 0.1 0.85 0.9
Table5
Valuesofobjectives.
Objective ProdBt/y AnnualcostM$/y EPtPO43−/y APtC2H4/y GWPtCO2/y HTPtC6H6/y POCPtC2H4/y Initial 305.00 277.42 9759.06 11,190.34 1,884,528.48 18,699.58 2472.32 First 299.96 205.04 13,831.12 4829.26 1,410,428.13 26,537.88 1944.11 Gainvs.initial(%) −1.62 26.00 −29.44 56.85 25.16 −41.91 21.36 1410500 1411000 1411500 1412000 1412500 1413000 1413500 205,34 205,32 205,3 205,28 205,26 205,24 205,22 205,2 205,18 205,16
Annual cost (M$/y)
GWP (t CO
2
/ y)
Fig.5.GWPvs.Annualcost.
0 5000 10000 15000 20000 25000 30000 235 230 225 220 215 210 205 200
Annual cost (M$/y)
HTP (t C
6
H6
/ y)
Fig.6.HTPvs.Annualcost.
forgloballyimprovingalltheenvironmentalimpacts(includingthe redundantones).
(GWP,Annualcost):Nondominatedpointsappearonlyinlow rangesforthetwoobjectives(seeFig.5),suggestingastrong rela-tionbetweenthem(seesection5.4).(HTP,Annualcost):ThePareto frontisdisplayedinFig.6,wherethetwoobjectiveshaveopposite effects.(EP,Annualcost):TheresultsareshowninFig.7.Thecurves displayedinFigs.6and7exhibitverysimilartrends,duetoastrong linkbetweenthem(seeSection5.4).(AP,Annualcost):ThePareto
0 2000 4000 6000 8000 10000 12000 14000 235 230 225 220 215 210 205 200
Annual cost (M$/y)
EP (t PO
4
3-/y)
BT2 BT1
Fig.7.EPvs.Annualcost.
4300 4400 4500 4600 4700 4800 4900 235 230 225 220 215 210 205 200
Annual cost (M$/y)
AP (t SO
4
/y)
BT1
BT2
Fig.8. APvs.Annualcost.
1910 1915 1920 1925 1930 1935 1940 1945 1950 222 220 218 216 214 212 210 208 206 204
Annual cost (M$/y)
PCOP (t C
2
H4
/y)
Fig.9. POCPvs.Annualcost.
frontisdisplayedinFig.8.(POCP,Annualcost):theresultsreported inFig.9showthatnondominatedpointsappearonlyinalowrange forthePOCPobjective,suggestingastrongrelationbetweenthese twofunctions.(AP,EP):concerningthepair(AP,EP),the biobjec-tiveoptimizationgivesthefrontdisplayedinFig.10,whereitcan bepointedoutthatthetwoimpactsexhibitantagonistbehaviours.
0 1000 2000 3000 4000 5000 6000 7000 8000 9000 16000 14000 12000 10000 8000 6000 4000 2000 0 EP (t PO43-/y) AP (t SO 2 /y) BT1 BT2 Fig.10. APvs.EP.
Table6
Resultsofthemultilinearregressions–Case1.
Objective AnnualcostM$/y EPtPO43−/y APtC2H4/y y-Intercept Coeff.corr. Errormax(%)
GWPtCO2/y 5445.29 −0.64 43.95 90,529.67 0.9988 0.51 HTPtC6H6/y −6.29×10−5 1.92 −3.70×10−5 9.25×10−3 1.000 10−6 POCPtC2H4/y 0.44 4.07×10−3 8.67×10−2 1377.95 0.9999 0.03 4000 5000 6000 7000 8000 0 5000 10000 15000 205 210 215 220 225 230 AP (t SO2 / y) EP (t PO4 (3-) / y) Annual cost (M$ / y) TT1 TT2 TF1 TF2
Fig.11. Triobjectiveoptimization(Annualcost,EP,AP)– Case1.
6.4. Redundantobjectives
Fromtheprevioussection,environmentalimpactsGWP,HTP andPOCPwerefoundtobelinkedwithotherobjectivesAnnual cost, EP and AP. From 200 randomly generated values for the independentvariables(definedinTable4),theobjectivesAnnual cost,EP,AP,GWP,HTPandPOCPwerecomputedandmultilinear regressionswereperformed ontheExcel platform.The experi-mentwascarriedoutseveraltimesandgivesthefollowingresults ofTable6.
Inallcases, thecoefficientofcorrelationis verygood. From thecoefficientsofmultilinearequations,itcanbeobservedthat GWPandPOCParemainlyincreasingfunctionsoftheAnnualcost, whileHTPdependsprincipallyonEP.SoimpactsGWP,POCPand HTPbeingexplicitfunctionsoftermsAnnualcost,EPandAP,they canbesuppressedfromthefollowingmultiobjectiveoptimization phase.
6.5. Triobjectiveoptimization
Fromthepreviousbiobjectivestudy,onlythreeindependent objectivesareremaining: Annualcost,EPand AP.Theyarenow simultaneouslyoptimized,andtheresultsaredisplayedinFig.11. TheflatportionofthecloudofpointsnearAnnualcost≈205M$/y, EP≈10,000tPO43−andAP≈5000tSO2/ysuggeststhatgood
solu-tionsmayexistinthiszoneforthethreeobjectives.
Table7
ResultsofTOPSISanalysisforbiobjectiveoptimizations.
Pair AnnualcostM$/y EPtPO43−/y APtSO2/y (EP,Cost)1 226.42 3590.12 (EP,Cost)2 226.64 3574.40 (AP,Cost)1 210.13 4499.36 (AP,Cost)2 210.18 4498.29 (AP,EP)1 5918.37 4763.31 (AP,EP)2 5886.94 4819.97 6.6. Selectingsolutions 6.6.1. Biobjectiveoptimizations
Fromthethreeindependentobjectives,Annualcost,EPandAP, aTOPSISanalysisiscarriedouttwiceforthethreepossiblepairs inordertodetectineachcasethetwobestpointscalledBT1and BT2(BTsignifyBiobjectiveTopsis)inFigs.7,8and10.Theresults areindicatedinTable7.NotethattheFUCAprocedurecannotbe implementedonbiobjectiveproblems,insofarastheParetofront beingperdefinitionasetofnondominatedpoints,thesumofranks isconstantandequaltothenumberofpointsinthefrontplusone (thebestpointforoneobjectivebeingtheworstonefortheother). 6.6.2. Triobjectiveoptimization
ATOPSISandaFUGAanalysisarecarriedouttwiceontheglobal setofobjectives(Annualcost,EP,AP,GWP,HTPandPOCP).Inthese studieseventhedependantobjectivesGWP,HTPandPOCP, com-putedbyusingTable6fromthethreeindependentones,haveto beconsidered,becausetheycan influencethedecisionmaking. However,theresults areonly displayedfor thethree indepen-dentcriteriain Fig.11.Let uspointoutthat therankingswere performedwithoutanypreferencefactor.Thetwobestsolutions obtainedfromTOPSIS(respectivelyFUCA)arecalledTT1andTT2 (respectivelyTF1andTF2)onthe3Dcurve.
Table8exhibitstheresultsobtainedbothforthethree inde-pendentobjectivesandforthethreedependentonesinthelast columns.Foreachobjectivethegainin%iscomputedvs.solution called“first”inTable5.Thevaluesofthecorrespondingvariables aregiveninTable9.Obviously,thesolutionsobtainedwhen pro-jecting3-Dsolutionsinthecorresponding2-Dspacesgloballyover estimatethesolutionsobtainedinthepure2-Doptimizations.As apartialconclusion,preliminary2-Doptimizationscannotbeused forestimatingthesolutionsofthe3-Dproblem.TheTOPSISand FUCArankingsprovideyetsurprisinglywithscatteredvalues.The FUCAproceduregivesresultsthataremuch moreinagreement
Table8
ResultsofTOPSISandFUCAanalysisforthetriobjectiveoptimization.
Case AnnualcostM$/y EPtPO43−/y APtSO2/y GWPtCO2/y HTPtC6H6/y POCPtC2H4/y
TT1 227.34 4404.41 6221.15 1,599,077.01 8439.0316 2038.64 Gain −10.88 67.08 −28.82 −13.36 68.20 −4.86 TT2 222.28 4735.06 6043.41 1,563,468.01 9072.59 2022.27 Gain −8.41 64.61 −25.14 −10.85 65.81 −4.02 TF1 206.99 9770.38 4930.16 1,428,118.43 18,720.83 1939.25 Gain −0.95 26.98 −2.09 −1.25 29.46 0.25 TF2 209.03 10,109.41 4782.53 1,432,472.39 19,370.44 1928.70 Gain −1.95 24.45 0.97 −1.56 27.01 0.79
Table9
Valuesofthedecisionvariablesforthetriobjectiveoptimization.
Decisionvariables TT1 TT2 TF1 TF2
Conversionrateoftoluene 0.59 0.60 0.70 0.75
Hydrogenpurgeflowrate(kmol/h) 300 300 300 300
Flashpressure(bar) 33.98 33.98 33.98 33.88
Stabilizerpressure(bar) 9.99 9.71 10 9.97
Column1pressure(bar) 3 3 3 3
Column2pressure(bar) 1.68 1.21 1.31 1.02
Ratio(bi-fuelfurnace) 0.29 0.32 0.24 0.34
witha simplegraphicalanalysis:besides,theglobalgaininthe setofobjectivesishigherthanthevaluecomputedbytheTOPSIS analysis.AcloserexaminationofTF1andTF2finallyleadstoselect thesolutionTF2,becauseitgivesonlyonenegativegainin envi-ronmentalimpacts.Inthefollowingstudiedcase,onlytheFUCA rankingwillbeused.
7. Case2:simultaneouseconomicandecologicalstudies 7.1. Context
Let us consider now that according to the EPA’s more stringent regulations, the GWP must be below a threshold of 1,400,000tCO2/y, which may have some effects on production
level.Besides,incentivemeasuresforenvironmentprotectionlead toimprovesimultaneouslytheotherenvironmentalobjectivesEP, AP,HTPandPOCP(ascomparedwiththepreviouscase).
Afterrandomlygeneratingseveralthousandsofvaluesfor inde-pendentvariables,it appearsthatvalues ofGWPslightly lower than1,400,000tCO2/yarelocatedinabenzeneproductionrange
of[250,280]kmol/h.Sothemultiobjectiveoptimizationisnow car-riedoutinthisrangeofproduction.Thesamestrategyasinthe previouscaseconsistingindecouplingobjectivesindependentand independentsetsisimplementedagain.Inordertoavoid extrapo-lationmodelproblemsthatmayoccur,multilinearregressionsare performedagainfordependentobjectives.Finally,atriobjective optimizationiscarriedoutforidentifyingthebestsolutionunder thesenewenvironmentalconstraints.
7.2. Multilinearregressions
For200valuesofindependentvariables withina production range[250,280]kmol/hrandomlygenerated,thevaluesoftheother objectivesarecomputedandtheresultsofmultilinearregressions aredisplayedinTable10.
0 0.5 1 1.5 2 2.5 x 10 3 4 2000 4000 6000 8000 1000012000 14000 250 255 260 265 270 275 280 AP (t SO2/y) EP (t PO4 (3-)/y) ProdB (kmol /h) TF3 TF4 TF5
Fig.12. Triobjectiveoptimization(ProdB,EP,AP)–Case2.
7.3. Triobjectiveoptimization
Atriobjectiveoptimizationiscarriedoutonthethree indepen-dentobjectives(ProdB,EPandAP),theotherdependantobjectives GWP,HTP,POCPandAnnualcost,beingcomputedbyusingTable10 fromthethreeindependentones.TheParetocurveobtainedfrom thetricriteriaoptimizationof(ProdB,EP,AP)isdisplayedinFig.12, wherethepointscalledTF3,TF4andTF5correspondrespectively tothebestsolutionfoundbyperforminganon weighted(resp. weighted)sumofranksofthesevenobjectives.
7.4. Selectingsolutions
AFUCArankingisfirstcarriedoutwiththesameweightfor alltheobjectives.ThebestsolutionnotedTF3isidentifiedamong thefivefirstones.However,inthissituation,benzeneproduction isimprovedonlyby14%.Thegoalofanyindustrialprocessbeing firstlyprofitmaximization,asecondFUCArankingisperformed withdifferentweightsonobjectives:0.4forbenzeneproduction
Table10
Resultsofthemultilinearregressions–scenario2.
Objective ProdBkmol/h EPtPO43−/y APtC2H4/y y-Intercept Coeff.corr. Errormax(%)
GWPtCO2/y 4901.44 −11.61 23.36 10,945.95 0.9397 3.75
HTPtC6H6/y −2.14×10−5 1.92 −3.68×10−5 1.66×10−4 1.000 6×10−6 POCPtC2H4/y 5.08 2.04×10−3 7.950×10−2 5.46 0.9942 0.69
COSTM$/y 0.6959 0 0 −5.03502 0.9997 2.10
Table11
ResultsofFUCAanalysisforthetriobjectiveoptimization.
Case BenzeneProd(kmol/y) AnnualcostM$/y EPtPO43−/y APtSO2/y GWPtCO2/y HTPtC6H6/y POCPtC2H4/y
TF3 253.50 171.30 7901.62 3737.16 1,249,024.11 15,104.58 1599.81 Gain −16.34 19.08 21.84 21.86 12.81 22.02 17.05 TF4 276.73 188.21 7149.41 4446.35 1,388,198.24 13,666.63 1772.36 Gain −9.49 18.04 29.28 7.03 3.09 29.45 8.11 TF5 259.88 175.95 8982.89 3799.92 1,269,260.85 17,171.37 1639.400 Gain −14.23 15.82 11.15 21.86 11.39 11.35 15.00
Table12
Valuesofthedecisionvariablesforthetriobjectiveoptimization–Case2.
Decisionvariables TF3 TF4 TF5
Conversionrateoftoluene 0.74 0.71 0.76 Hydrogenpurgeflowrate(kmol/h) 299.30 299.92 299.92 Flashpressure(bar) 29.27 28.73 32.57 Stabilizerpressure(bar) 8.60 6.02 7.00 Column1pressure(bar) 2.13 2.03 2.21 Column2pressure(bar) 1.02 1.33 1.74 Ratio(bi-fuelfurnace) 0.16 0.24 0.17
and0.1forthesixcriteria.Asinthepreviouscase,thebestsolution notedTF4isidentifiedamongthefivefirstones.Finally,a compro-misesolutionTF5generatedwiththerespectiveweightsof0.35for thebenzeneproduction,0.15fortheGWPand0.1foralltheothers isalsodetermined.
Table11exhibitstheresultsobtainedbothforthethree inde-pendentobjectivesand forthe threedependentones. Foreach objectivethegainin%iscomputedvs.solutionTF2ofTable8.The valuesofthecorrespondingvariablesaregiveninTable12. 8. Conclusions
Thispaperhaspresented a methodologyfor eco-design and optimizationofachemicalprocesstakingintoaccountthe contri-butionofutilitygeneration,viatheindustrialsoftwareARIANETM.
The well-known benchmark HDA process first developed by Douglas(1988)illustratestheapproach,whichistotallydifferent withthetraditionalend-of-pipetreatmentmethods.Theprocess wasdesignedunderclassicalengineeringobjectiveslikebenzene production and total annual cost, by also considering classical environmentalburdensastheGlobalWarmingPotential,the Acid-ificationPotential,thePhotochemicalOzoneCreationPotential,the HumanToxicityPotentialandtheEutrophicationPotential.A vari-antoftheclassicalmultiobjectivegeneticalgorithmNSGAIIwas usedforsolvingthevariousmultiobjectivesproblems.
Inapreliminarystudy,agoodvalueofthebenzeneproduction wasidentifiedonalimitedrangeofcosts,thenapossible degra-dationoftheannualcostinthisrange,andthusofthebenzene productionwereallowedtoimprovetheenvironmental perfor-mancesoftheprocess.InsteadofsearchingforaParetofrontonthe wholesetofobjectives,apreliminarystudyofobjectiveswas car-riedoutfordeterminingasub-setofdependentcriteriaexpressed asmultilinearfunctionsoftheremainingindependentones.Sothe multiobjectiveproblemwasreducedtoatricriteriaone.The val-uesofcorrespondingdependentobjectiveswerecomputedfrom theindependentones,andaTOPSISandFUCAanalyseswere per-formedonthewholesetofobjectives,FUCAgivingmuchbetter resultsthanTOPSIS.Analternativewouldbetoidentifyanideal pointontheParetofrontbydefiningametricdistancetoanutopia point(i.e.minimizingallobjectivesforinstance)asanobjective functionusingasingleobjectiveGA.
Asecondscenarioisbasedonanotherformulationforwhich a more stringent environmental regulation has tobe satisfied, expressedastheGWPcriteriontobelowerthanagiven thresh-old.Theobjectivealsohaspositivegainsoverallenvironmental objectivesEP,AP,HTPandPOCP.Multilinearregressionswere per-formedagain,toobtainanewtricriteriaproblem,solvedagainby meansofNSGAII.Bysimplytuningtheweightingfactorsusedin theFUCAranking,severalsolutionsofferingagoodtrade-offcanbe deducedforobjectives.Herethreesolutionscorresponding respec-tivelytohigh,mediumandlowrangesofbenzeneproductionare generated.Thismethodologycanbeappliedtoawidespectrumof chemicalprocessdesignproblemswithmultipleand environmen-talobjectivesandmakesexplicitthetrade-offsbetweenthem.
References
AIChE(1998).Sustainabilitymetrics.http://www.aiche.org/cwrt.
Allen,D.T.&Shonnard,D.R.(2002).Greenengineeringenvironmentallyconscious designofchemicalprocesses.UpperSideRiver,NJ:PrenticeHallPTR.
ARIANETM.http://www.prosim.net/en/energy/ARIANE.html.
Azapagic,A.&Clift,R.(1999).Lifecycleassessmentandmultiobjectiveoptimisation. JournalofCleanerProduction,7(2),135–143.
Azapagic,A.,Emsley,A.&Hamerton,I.(2003).Polymers:Theenvironmentand sus-tainabledevelopment.JohnWiley&Sons.
Azapagic,A.,Millington,A.&Collett,A.(2006).Amethodologyforintegrating sus-tainabilityconsiderationsintoprocessdesign.ChemicalEngineeringResearchand Design,84(A6),439–452.
Berhane,H.G.,Gonzalo,G.-G.,Laureano,J.&Dieter,B.(2009).Designof environmen-tallyconsciousabsorptioncoolingsystemsviamulti-objectiveoptimizationand lifecycleassessment.AppliedEnergy,86(9),1712–1722.
Berhane,H.G.,Gonzalo,G.-G.,Laureano,J.&Dieter,B.(2010).Asystematictoolfor theminimizationofthelifecycleimpactofsolarassistedabsorptioncooling systems.Energy,35(9),3849–3862.
Björk,H.&Rasmuson,A.(2002).Amethodforlifecycleassessment environmen-taloptimisationofadynamicprocessexemplifiedbyananalysisofanenergy systemwithasuperheatedsteamdryerintegratedinalocaldistrictheatand powerplant.ChemicalEngineeringJournal,87(3),381–394.
Branke,J.,Deb,K.,Dierolf,H.&Osswald,M.(2004).Findingkneesinmultiobjective optimization,parallelproblemssolvingfromnature.LNCS,3242,722–731. Bruno,J.C.,Fernandez,F.,Castells,F.&Grossmann,I.E.(1998).RigorousMINLP
modelfortheoptimalsynthesisandoperationofutilityplants.Chemical Engi-neeringResearchandDesign,76,246–258.
Cabezas,H.,Bare,J.C.&Mallick,S.K.(1999).Pollutionpreventionwithchemical processsimulators:Thegeneralizedwastereduction(WAR)algorithm– Full version.ComputersandChemicalEngineering,23,623–634.
Cao,Y.,Rossiter,D.,Edwards,D.W.,Knechtel,J.&Owens,D.(1998).Modelling issuesforcontrolstructureselectioninchemicalplants.ComputersandChemical Engineering,22Suppl.,S411–S418.
Carvalho,A.,Gani,R.&Matos,H.(2008).Designofsustainablechemicalprocesses: Systematicretrofitanalysisgenerationandevaluationofalternatives.Process SafetyandEnvironmentalProtection,86(5),328–346.
Chen,Q.&Li,W.(2008).Anapproachtopotentialenvironmentalimpactreduction inchemicalreactionprocesses.CleanTechnologiesandEnvironmentalPolicy,10, 97–105.
Collette,Y.&Siarry,P.(2002).Optimisationmultiobjective.Eyrolles.
CSD(1996).IndicatorsofSustainableDevelopment,NewYork.www.un.org/esa/ dsd aofwind/indindex.shtml.
Cutler,C.R.&Perry,R.T.(1983).Realtimeoptimizationwithmultivariablecontrolis requiredtomaximizeprofits.ComputersandChemicalEngineering,7,663–667. Deb,K.,Pratap,A.,Agarwal,S.&Meyarivan,T.(2002).Afastandelitistmultiobjective
geneticalgorithm:NSGA-II.IEEETransactionsonEvolutionaryComputation,2, 182–197.
DinizdaCosta,J.C.&Pagan,R.J.(2006).Sustainabilitymetricsforcoalpower generation inAustralia. Process SafetyandEnvironmentalProtection, 84(2), 143–149.
Douglas,J.M.(1988).Conceptualdesignofchemicalprocess.NewYork:McGrawHill. Douglas,M.Y.&Heriberto,C.(1999).Designingsustainableprocesseswith simula-tion:Thewastereduction(WAR)algorithm.ComputersandChemicalEngineering, 23,1477–1491.
Douglas,Y.,Richard,S.&Heriberto,C.(2000).Thewastereduction(WAR)algorithm: Environment impacts,energy consumption,and engineering economicals. WasteManagement,20,605–615.
Fonseca,C.M.&Fleming,P.J.(1993).Geneticalgorithmsformultiobjective opti-mization:Formulation,discussionandgeneralization.InProceedingsofthe5th internationalconferenceongeneticalgorithms(pp.416–423).
Fonseca,C.M.&Fleming,P.J.(1998).Multiobjectiveoptimizationandmultiple constrainthandlingwithevolutionaryalgorithm:PartI.Aunifiedformulation. IEEETransactionsonSystemsManandCyberneticsPartA:SystemsandHumans,1, 26–37.
Gomez,A.,Pibouleau,L.,Azzaro-Pantel,C.,Domenech,S.,Latgé,C.&Haubensack, D.(2010).Multiobjectivegeneticalgorithmstrategiesforelectricityproduction fromgenerationIVnucleartechnology.EnergyConversionandManagement,51, 859–871.
Gonzalo,G.-G.(2011).AnovelMILP-basedobjectivereductionmethodfor multi-objectiveoptimization:Applicationtoenvironmentalproblems.Computersand ChemicalEngineering,35(8),1469–1477.
Gough,J.D.&Ward,J.C.(1996).Environmentaldecisionmakingandlake manage-ment.JournalofEnvironmentalManagement,48(1),1–15.
Guthrie,K.M.(1969).Capitalcostestimating.ChemicalEngineering,76,114–142. Heriberto,C.,Jane,C.B.&Subir,K.M.(1999).Pollutionpreventionwith
chemi-calprocesssimulators:Thegeneralizedwastereduction(WAR)algorithm-full version.ComputersandChemicalEngineering,23,623–634.
Hilaly,A.K.&Sikdar,S.K.(1994).Pollutionbalance:Anewmethodfor minimiz-ingwasteproductioninmanufacturingprocesses.JournaloftheAirandWaste ManagementAssociation,44,1303–1308.
Horn,J.,Nafpliotis,N.&Goldberg,D.(1994).AnicheParetogeneticalgorithmfor multiobjectiveoptimization.InProceedingsofthe1stIEEEconferenceon evolu-tionarycomputation(pp.82–87).
Huang,J.P.,Poh,K.L.&Ang,B.W.(1995).Decisionanalysisinenergyand environ-mentalmodelling.Energy,20,843–855.
IChemE(2001).Thesustainabilitymetrics.http://www.icheme.org/sustainability/. Iskandar,H.&Rajagopalan,S.(2011).Aknowledge-basedsimulation-optimization frameworkandsystemforsustainableprocessoperations.Computersand Chem-icalEngineering,35,92–105.
Jia,X.,Zhang,T.,Wang,F.&Han,F.(2006).Multi-objectivemodellingand opti-mizationforcleanerproductionprocesses.JournalofCleanerProduction,14, 146–151.
Ku-Pineda,V.&Tan,R.R.(2006).Environmentalperformanceoptimizationusing processwaterintegrationandSustainableProcessIndex.JournalofCleaner Pro-duction,14(18),1586–1592.
Labuschagne,C.,Brent,A.C.&vanErck,R.P.G.(2005).Assessingthesustainability performancesofindustries.JournalofCleanerProduction,13(4),373–385. Li,X.,Zanwar,A.,Jayswal,A.,Lou,H.H.&Huang,Y.L.(2011).Incorporatingexergy
analysisandinherentsafetyanalysisforsustainabilityassessmentofbiofuels. IndustrialandEngineeringChemistryResearch,50(5),2981–2993.
Li,C.,Zhang,X.,Zhang,S.&Suzuki,K.(2009).Environmentallyconsciousof chem-icalprocessandproducts:Multi-optimizationmethod.ChemicalEngineering ResearchandDesign,87,233–243.
Lim,K.H.,Dennis,N.V.S.N.,Murthy,K.&Rangaiah,G.P.(2005).Synthesisanddesign ofchemicalprocesses.JournaloftheInstitutionofEngineer,Singapore,45(6) Moralez-Mendoza,L.F.,Perez-Escobedo,J.L.,Aguilar-Lasserre,A.,Azzaro-Pantel,C.&
Pibouleau,L.(2011).Selectingthebestalternativebasedonahybrid multiobjec-tiveGA-MCDMapproachfornewproductdevelopmentinthepharmaceutical industry.InIEEESymposiumonSwarmIntelligenceParis,11–15May.
Narodoslawsky,M.&Krotscheck,Ch.(2004).Whatcanwelearnfromecological valuationofprocesseswiththeSustainableProcessIndex(SPI)–Thecasestudy ofenergyproductionsystems.JournalofCleanerProduction,12,111–115. Opricovic,S.&Tzeng,G.(2004).CompromisesolutionbyMCDMmethods:A
com-parativeanalysisofVIKORandTOPSIS.EuropeanJournalofOperationalResearch, 156,445–455.
Othman,M.R.,Repke,J.U.,Wozny,G.&Huang,Y.(2010).Amodularapproach tosustainabilityassessmentanddecisionsupportinchemicalprocessdesign. IndustrialandEngineeringChemistryResearch,49,7870–7881.
Pirdashti,M.,Ghadi,A.,Mohammadi,M.&Shojatalab,G.(2009).Multi-criteria decision-makingselectionmodelwith applicationtochemical engineering managementdecisions.InternationalJournalofHumanandSocialSciences,4(15). ProSim(2005). Plessala AutomationServer. http://www.prosim.net/en/energy/
ARIANE.html.
Ren,L.,Zhang,Y.,Wang,Y. &Sun,Z.(2007).ComparativeanalysisofaNovel M-TOPSIS Method and TOPSIS. Applied Mathematics Research Express, 10 doi:10.1093/amrx/abm005(articleIDabm005)
Saaty,T.L.(1980).Theanalytichierarchyprocess.NewYork:McGraw-Hill. Sandholzer,D.&Narodoslawsky,M.(2007).SPIonExcel–Fastandeasy
calcula-tionofthesustainableprocessindexviacomputer.Resources,Conservationand Recycling,50(2),130–142.
Schaffer,J.(1985).Multi-objectiveoptimizationwithvectorevaluatedgenetic algo-rithms.InProceedingsoftheInternationalConferenceonGeneticAlgorithmsand theirApplications(pp.93–100).
Sikdar,S.K.&El-Halwagi,M.M.(2001).Processdesigntoolsfortheenvironment.New York:TaylorandFrancis.
Stanislav,V.E.(2003).Anexaminationofamodifiedhierarchicaldesignstructureof chemicalprocesses,http://www.che.ttu.edu/classes/che5000/EmetsThesis.pdf. Surya,K.C.&Alex,Z.(2002).AdefinitionofSteady-stateplantwidecontrollability.
IndustrialandEngineeringChemistryResearch,41
Taboada, H. & Coit, D. (2006). Data mining techniques to facilitate the analysis of the Pareto-optimal set for multiple objective problems. In Proceedings of the Industrial Engineering Research Conference Orlando, FL, (pp.43–48).
TeixeirodeAlmeida,A.,deMiranda,C.M.G.&CabralSeixasCosta,A.P.(2004). Mul-ticriteriamodelforprioritizationofresearchanddevelopmentprojects.Journal oftheAcademyofBusinessandEconomicals,(March)
Teresa,M.M.,Raymond,L.S.,Douglas,M.Y.&Carlos,A.V.C.(2003).Evaluating theenvironmentalfriendliness,economicalsandenergyefficiencyof chemi-calprocesses:Heatintegration.CleanTechnologiesandEnvironmentalPolicy,5, 302–309.
Turton,R.,Bailie,R.C.,Whiting,W.B.&Shaeiwitz,J.A.(1997).Analysis,synthesisand designofchemicalprocesses.UpperSideRiver,NJ:PrenticeHall.
Turton,R.,Bailie,R.C.,Whiting,W.B.&Shaeiwitz,J.A.(2009).Analysis,synthesisand designofchemicalprocesses(3rded.).UpperSideRiver,NJ:PrenticeHall. Tveit,T.-M.&Fogelholm,C.-j.(2006).Multi-periodsteamturbinenetwork
optimiza-tion:PartII.Developmentofamulti-periodMINLPmodelofautilitysystem. AppliedThermalEngineering,26,1730–1736.
Vasilios,P.&Shang,Z.(2008).Amulti-criteriaoptimizationapproachforthe designofsustainableutilitysystems.ComputersandChemicalEngineering,32, 1589–1602.
Wolpert,D.H.&Macready,W.G.(1997).Nofreelunchtheoremsforoptimization. IEEE.Trans.Evolut.Comput.,1,67–82.
Yoon,K.P.&Hwang,C.L.(1995).Multipleattributedecisionmaking:Anintroduction. ThousandOaks,CA:Sage.
Young,D.,Scharp,R.&Cabezas,H.(2000).Thewastereduction(WAR)algorithm: Environmentalimpacts,energyconsumption,andengineeringeconomicals. ComputersandChemicalEngineering,20,605–615.
Zhaoxa,S.&Min,H.(2010).MulticriteriadecisionmakingbasedonPROMETHEE method.InComputing,ControlandIndustrialEngineering,2010International Con-ferenceWuhan,China.
Zhou,P.A.&Poh,B.W.(2006).Decisionanalysisinenergyandenvironmental modelling:Anupdate.Energy,31,2604–2622.
Zitzler,E.&Thiele,L.(1999).Multiobjectiveevolutionaryalgorithm:Acomparative casestudyandthestrengthParetoapproach.IEEETransactionsonEvolutionary Computation,3,257–271.