• Aucun résultat trouvé

A simple and unified algorithm to solve fluid phase equilibria using either the gamma–phi or the phi–phi approach for binary and ternary mixtures

N/A
N/A
Protected

Academic year: 2022

Partager "A simple and unified algorithm to solve fluid phase equilibria using either the gamma–phi or the phi–phi approach for binary and ternary mixtures"

Copied!
13
0
0

Texte intégral

(1)

ContentslistsavailableatSciVerseScienceDirect

Computers and Chemical Engineering

j our na l h o me p ag e: w w w . e l s e v i e r . c o m / l o c a t e / c o m p c h e m e n g

A simple and unified algorithm to solve fluid phase equilibria using either the gamma–phi or the phi–phi approach for binary and ternary mixtures

Romain Privat

a,∗

, Jean-Noël Jaubert

a

, Yannick Privat

b

aÉcoleNationaleSupérieuredesIndustriesChimiques,UniversitédeLorraine,LaboratoireRéactionsetGéniedesProcédésUPR3349,1rueGrandville,BP20451,NancyCedex9, France

bENSCachan,AntennedeBretagne– AvenueRobertSchumann,35170Bruz,France

a r t i c l e i n f o

Articlehistory:

Received21September2012 Receivedinrevisedform 15November2012 Accepted16November2012 Available online 5 December 2012

Keywords:

Phase-equilibriumcalculation Algorithm

Binarymixtures Ternarymixtures Phi–phi Gamma–phi

a b s t r a c t

Anewalgorithmisproposedforcalculatingphaseequilibriainbinarysystemsatafixedtemperature andpressure.Thisalgorithmisthenextendedtoternarysystems(inwhichcase,themolefractionofone constituentinagivenphasemustbefixedinordertosatisfytheGibbs’phaserule).Thealgorithmhas theadvantageofbeingverysimpletoimplementandinsensitivetotheprocedureusedtoinitializethe unknowns.Mostsignificantly,thealgorithmallowsthesamesolutionproceduretobeusedregardless ofthethermodynamicapproachconsidered(–ϕorϕ–ϕ),thetypeofphaseequilibrium(VLE,LLE,etc.) andtheexistenceofsingularities(azeotropy,criticalityandsoon).

© 2012 Elsevier Ltd. All rights reserved.

1. Introduction

Severalindustrially importantchemical applicationssuchas single- and multi-stage processes (including flash, distillation, absorption,extraction,etc.)bringtwofluidphasesintocontact(e.g.

twoliquidphasesoraliquidphaseandavaporphase).Simulation andoptimizationof theseprocesses requirethattheproperties oftheequilibrium statesbecalculated,mainly,thetemperature T,thepressurePandthecompositionsofthecoexistingphases.

Robustandefficientalgorithmsforphase-equilibriumcalculations arethereforeneeded.TheGibbs’phaserulestatesthatcalculating two-phaseequilibriuminabinarysystemrequirestwointensive andindependentprocessvariablestobespecifiedwhicharetypi- callychosenfromT,P,xi(themolefractionofcomponentiinthe firstphase)andxi(themolefractionofcomponentiinthesecond phase).Thephase-equilibriumproblemthenconsistsofsolvinga setoftwoequilibriumequationswithrespecttotwounknowns:

1(T,P,x1)=1(T,P,x1)

2(T,P,x1)=2(T,P,x1)

(1)

Correspondingauthor.Tel.:+33383175128;fax:+33383175152.

E-mailaddress:romain.privat@univ-lorraine.fr(R.Privat).

whereidenotesthechemicalpotentialofcomponentiinagiven phase.Thesuperscripts anddenotethefirstandsecondequi- librium phases,respectively. Theprocedurefor solvingthetwo equilibriumequationsisstronglyrelatedto(i)thechoiceofthe twospecifiedvariables andto(ii)thenatureoftheequilibrium (liquid–liquid or liquid–vapor) involved. There are six possible combinations of two variables among thefour aforementioned ones(T, P,xi and xi).Four ofthem are oftenacknowledged to be of practical interest since theycan be easilygeneralized to p-componentsystemswithp>2.Thisiswhymanypapersandtext- booksproposesolutionproceduresforonlythefollowingfourcases (Michelsen,1985;Sandler,2006;Smith&VanNess,1975):

•Calculation of xi (or xi) and P, given xi (or xi) and T. For vapor–liquidequilibrium(VLE),thesekindsofcalculationsare calledbubble-pointpressureanddew-pointpressurecalculations.

•Calculationofxi (orxi)andT,givenxi (orxi)andP.ForVLE, thesekindsofcalculationsarecalledbubble-pointtemperature anddew-pointtemperaturecalculations.

Thetwo remainingcombinationswhich arescarcely– ifever– consideredare:

•CalculationofTandP,givenxiandxi.

•Calculationofxiandxi,givenTandP.

0098-1354/$seefrontmatter© 2012 Elsevier Ltd. All rights reserved.

http://dx.doi.org/10.1016/j.compchemeng.2012.11.006

(2)

Thefirstcalculationwhichrequiresthatthemolefractionsinthe twophasesbespecified,isclearlyofnopracticalinterest.Indeed, inthiscase,temperatureandpressurewhicharecertainlytheeas- iestprocessvariablestoworkwith,arenotspecifiedbutbecome accessibleaftersolvingthephase-equilibriumproblem.

Thesecondcombination(wherethevariablesTandParespec- ified)appearstobeparticularlyconvenient.Togiveanexample, calculationofthecomplete isothermalorisobaricbinaryphase diagramsisstraightforwardusinganalgorithm withT andPas specifiedvariables,asopposedtoanalgorithmusingxiandxi as specifiedvariables.

Atthispoint, itisimportanttohighlightthatcalculatingthe compositionsoftwophasesinequilibriumatafixedTandPdiffers notablyfromtheclassical(P,T)-flashcalculation.Indeed,calculat- ingxiandxiatgivenTandPonlyinvolvesphasevariables(T,Pand themolefractionsinthetwophases)butdoesnotinvolveanyover- allvariable(theproportionofthephases,overallmolefractions, etc.).Incontrast,aflashalgorithminvolvesbothphasevariables andoverallvariables.Morespecifically:

•ina(P,T)-flashalgorithm,thevariablesTandParespecifiedalong withtheoverallcomposition(bycomparison,thecalculationof xiandxi atagivenTandPonlyrequiresTandPtobespecified),

•ina(P,T)-flashalgorithm,thetwophase-equilibriumequations arecoupledwithmaterialbalances(whereasthecalculationof xiandxi atagivenTandPonlyinvolvesthephase-equilibrium equations),

•ina(P,T)-flashalgorithm,theunknownsarexiandxiaswellas, themolarproportionofagivenphase(whereasforthecalculation ofxiandxi atagivenTandP,xiandxi arethesoleunknowns).

•AtagivenTandP,dependingontheoverallcomposition,a(P, T)-flashalgorithmcanindicatewhetherthesystemismadeupof asinglephase(molarproportionsofthephases<0or>1)ortwo phases(molarproportions∈[0;1]).Thecalculationofxiandxi is thenonlyperformedinthelattercase.

Insummary,a(P,T)-flashalgorithmandthecalculationofxiand xi atagivenTandPdonotinvolvethesameinputvariables,the sameunknownsandthesamesetofequationstosolve.Inaddition, contrarytoa(P,T)-flashalgorithm,thecalculationofxiandxi at agivenTandPallowsuncouplingthephase-equilibriumproblem andtheresolutionofthemass-balanceequationswhichmayhave twointerests:

–the dimension of the system of equations to solve becomes smallerwhichmayfacilitateitsresolution;

–thephase-equilibriumproblemtakestheadvantageousformof apseudo-linearsystem,aswillbeprovedinthenextsections.

Inthispaper,anewalgorithmisproposedtocalculatephase equilibria in binary systems at a specified T and P. As will be shown,thealgorithmcanbeusedeitherwiththe–ϕortheϕ–ϕ approachandallowsVLEorLLE(liquid–liquidequilibria)tobecal- culatedaswell.Thisprocedureisconsideredtobeuniversal,asit unifiesthevariousapproachesandkindsoffluid-phaseequilib- riumcalculations.Thegreatsimplicityofthealgorithmfacilitates easyimplementation.Thealgorithmcanbeusedtocalculatecom- plexphasediagramsexhibitingazeotropyorcriticality,andisthus intendedforaudiencesinacademicorindustrialenvironmentswho usephase-equilibriumthermodynamicsasatoolandwishtohave attheirdisposalauniquealgorithmtoperformeitherVLEorLLE calculations.

Therefore,thealgorithmcanbeusedtomodel,e.g.adistilla- tioncolumn,astrippingprocess,aliquid–liquidextractionorany separationunitinvolvingfluid–fluidphaseequilibria.

Thefinalsectionofthepapershowshowthisalgorithmcanbe simplyextendedtoternarysystemsbyspecifyingathirdvariable.

2. Generalformalismofthetwo-phaseequilibrium probleminbinarysystemsatspecifiedtemperatureand pressure

2.1. The–ϕapproach

In the present case, a liquid-activity-coefficient model (also knownasmolarexcessGibbsenergymodel)isusedtoevaluate thethermodynamicpropertiesoftheliquidphase.Anequationof state(EoS)isusedforthegaseousphase.

•Forabinarysysteminliquid–liquidequilibriumatafixedtem- peratureTandpressureP,Eq.(1)maybewrittenasfollows:

x1·1(x1)=x1·1(x1)

(1−x12(x1)=(1−x12(x1)

(2)

wherex1andx1arethemolefractionsofcomponent1ineach ofthetwoliquidphasesinequilibriumandidenotestheactiv- itycoefficientofcomponentiinaliquidphase.iisclassically derivedfromanexcessGibbsenergymodel,suchthat:RTlni= (∂GE/∂ni)T,P,nj

/

=iwhereGEisthetotalmolarGibbsenergyofthe liquidphase and ni, theamountofcomponent iin theliquid phase.Thesekindsofmodelarepressure-independentsothat, theliquid-activity coefficients dependonly onthemolefrac- tionsandthetemperature.Forthesakeofclarity,onlythetwo unknownsx1andx1inthephase-equilibriumproblematfixedT andPappearwithinparentheses,inEq.(2).

•Forliquid–vaporequilibriumatafixedtemperatureTandpres- sureP,Eq.(1)isequivalentto:

P·y1·C1(y1)=Psat1 ·x1·1(x1)

P·(1−y1)·C2(y1)=Psat2 ·(1−x12(x1) (3) wherex1andy1 arethemolefractionsofcomponent1inthe liquidandvaporphases,respectively;Psati isthevaporpressure ofthepurecomponenti;thecoefficientCiisdefinedas:

Ci(y1)=ϕˆgasi (y1)

ϕsati ·FP,i (4)

whereϕˆgasi isthefugacitycoefficientofcomponentiinthegas phase,ϕisatisthefugacitycoefficientofpureiatitsvaporpressure attemperatureTandFP,iistheclassicalPoyntingcorrectionfactor forcomponentiatTandP(oftensetto1atlow tomoderate pressures).Eq.(3)canbealternativelywrittenasfollows:

x1·1(x1)=y1·[P·C1(y1)/P1sat]

(1−x12(x1)=(1−y1)·[P·C2(y1)/P2sat]

(5)

Asbefore, only the two unknowns (x1 and y1)in the phase- equilibriumproblematafixedTandPappearwithinparentheses inEq.(5).

2.2. Theϕ–ϕapproach

This approach requires the use of a pressure-explicit EoS to model both phases in equilibrium. The method offers the undeniable advantage of a unique formalism for

(3)

liquid–liquid,liquid–vapor andmoregenerallyfluid–fluidphase equilibria.Eq.(1)isthenwrittenasfollows:

x1·ϕˆ1=x1·ϕˆ1

(1−x1)·ϕˆ2=(1−x1)·ϕˆ2 (6) where x1 and x1 are themolefractions ofcomponent 1 inthe two fluid phasesin equilibrium;ϕˆi andϕˆi denotethefugacity coefficientsofcomponentiineachfluidphase.

For a binarysystem,thepressure-explicitEoSexpressesthe pressureofaphaseasafunctionofthetemperatureTofthephase, themolefractionz1ofcomponent1andthemolarvolumevofthe phase.Henceforth,thepressure-explicitEoSwillbedenotedbythe function:PEoS(T,v,z1).Expressionsforthefugacitycoefficientsare derivedfromtheEoSandarethusnaturalfunctionsofthevariables T,vandz1:ϕˆi(T,v,z1).

Consequently,thefugacitycoefficientsϕˆiandϕˆi ofcomponent iinthetwophasesinequilibriumareaccessibleatafixedtemper- atureT0andpressureP0,providedthatthemolarvolumes(vand v)andthemolefractionsofcomponent1(x1andx1)areknown:

ϕˆi=ϕˆi(T0,v,x1)andϕˆi =ϕˆi(T0,v,x1).Themolarvolumeofafluid phasecan becalculated bysolvingtheEoS,giventhetempera- ture,thepressureandthecompositionofthephase.Therefore,at fixedT0 andP0,theunknownsv andvareobtainedbysolving thetwoequations:P0=PEoS(T0,v,x1)andP0=PEoS(T0,v,x1)(by assumingknownvaluesofmolefractionsx1 andx1).

Finally,thephase-equilibriumproblematafixedtemperature T0andpressureP0usingaϕ–ϕapproach,isreducedtosolvingaset offourequationswithrespecttofourunknowns(v,x1,v,x1):

⎧ ⎪

⎪ ⎪

⎪ ⎨

⎪ ⎪

⎪ ⎪

x1 ·ϕˆ1(v,x1)=x1·ϕˆ1(v,x1)

(1−x1)·ϕˆ2(v,x1)=(1−x1)·ϕˆ2(v,x1) P0=PEoS(v,x1)

P0=PEoS(v,x1)

(7)

Onceagain,onlytheunknownsintheproblemappearwithin parentheses.

2.3. Writingofthephase-equilibriumproblemusingaunique formalism

Eqs.(2),(5)and(6)canbewritteninthefollowingunifiedform:

x1·F1 =x1·F1

(1−x1)·F2 =(1−x1)·F2 (8) ExpressionsforthefunctionsF1,F1,F2andF2aregiveninTable1 andareselectedbasedonthechosenapproach andthekindof phase-equilibriumcalculation(VLE,LLE,etc.)beingconsidered.

Table1

ExpressionsoffunctionsF1andF2appearinginEq.(8).

Approach: –ϕapproach ϕ–ϕapproach

Typeofphase- equilibrium calculation:

VLE LLE Fluid–fluid

equilibrium

F1 1 1 ϕˆ1

F1 P·C1/Psat1 1 ϕˆ1

F2 2 2 ϕˆ2

F2 P·C2/Psat2 2 ϕˆ2

3. Derivationofauniquealgorithmforphase-equilibrium calculationinbinarysystemsatafixedtemperatureand pressure

3.1. Generalsolutionprocedure

Eq.(8)canbeequivalentlywrittenas:

x1 ·F1 −x1·F1=0

−x1·F2 +x1·F2=F2−F2 (9) Sucha systemis highlynonlinear andthecoefficients F1,F1,F2 andF2dependonx1andx1.Usingmatrixnotations,Eq.(9)canbe alternativelywrittenas:

F1 −F1

−F2 F2

A(X)

x1 x1

X

=

0 F2−F2

B(X)

(10)

orequivalently,providingmatrixAcanbeinverted:

X=[A(X)]−1B(X) (11)

with:

A−1= 1 det(A)

F2 F1 F2 F1

(12)

Consequently,thevaluesofthemolefractionsx1andx1atiter- ation(k+1)canbededucedfromthevaluesofthequantitiesF1,F1, F2 andF2atiterationkasfollows:

⎧ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎩

x1(k+1)=

F1 (F2−F2) F1 F2−F1 F2

(k)

x1(k+1)=

F1 (F2−F2) F1 F2−F1 F2

(k)

=x1(k+1)

F1 F1

(k) (13)

Thisiterativeschemerepresentsthemainresultofthispaper andisactuallyanapplicationoftheclassicaldirect-substitution methodtothetwo-phaseequilibriumproblem.Algorithmsbased onthefollowingprocedurecanbeusedtocalculatephaseequilib- ria:

1Provideinitialestimatesfor theunknownsx1 andx1.Setthe iterationcountertok=0.

2CalculatethecoefficientsF1,F2,F1andF2forthecurrentiteration k.

3Defineaconvergencecriterionıasfollows:

ı=[|x1F1−x1F1|+|(1−x1)F2−(1−x1)F2|](k) (14) Ifıislowenough,thenconvergenceisreachedandtheprocedure isterminated.

4Updatetheunknownsx1andx1usingEq.(13);updatetheitera- tioncounterto(k=k+1)andreturntostep2.

Thesubsequentsectionsillustratetheefficiencyoftheiterative schemeanddiscussitsrangeofapplicability.

3.2. Preliminaryremarksregardingtheconvergenceofthe procedureforthegeneralsolution

Nextsectionsareintendedtoconvincethereaderthatthepro- posedformulationofthetwo-phaseequilibriumproblemcoupled withthesolutionprocedurecandealwithcomplexscenarioseffi- cientlyandhasawiderangeofapplicability.However,itisclearly

(4)

notpossibletoguarantyconvergenceinallcases.Itisindeedwell knownthatfixed-pointmethods –includingdirect-substitution methods–canpotentiallyfailduetotheexistenceofdiverging sequencesthatdependonthenatureoftheequationtosolveand theinitialguessestothesolution.Inthisregard,Luciaetal.have pointedoutthatevenchaoticandperiodicbehaviorscanbeexhib- itedbyfixed-pointmethods(Lucia,Guo,Richey&Derebail,1990), thusemphasizingthecomplexityofaconvergencestudy.There- fore,itseemsverydifficult,ifnotimpossible,todetermineapriori forwhichcases,themethodwillconvergeordiverge.InAppendices AandB,somemathematicalresultsontheconvergenceoffixed- pointmethodsareappliedtotwo-phaseequilibriumproblems;a generalmethodologyforstudyingfailedcasesisalsoprovided.

3.3. Numericalexample1:calculationofliquid–liquidphase equilibriumusinganactivity-coefficientmodel

Inordertoillustratethesimplicityoftheiterativeschemegiven byEq.(13),letuscalculatethecompositionsx1andx1oftwoliq- uidphasesinequilibriumatT=300Kforafictitiousbinarysystem (1)+(2),using the–ϕapproach. Thenon-idealityof theliquid phaseisaccountedforbyaVan-Laartypeactivity-coefficientmodel (Smith&VanNess,1975):

gE(x1)= A12A21x1x2

A12x1+A21x2

, with

⎧ ⎪

⎪ ⎩

x2=1−x1

A12=9000Jmol−1 A21=7000Jmol−1

(15)

TheactivitycoefficientsofthetwospeciesarederivedfromtheVan LaargEmodel:

i=exp

Aij

RT

Ajixj

Aijxi+Ajixj

2

,with

i=/jand(i,j)∈{1;2}2

R=8.314472Jmol1K1 (16) ThealgorithmproposedfortheLLEcalculationatafixedtempera- tureusingthe–ϕapproachisasfollows:

1Provideinitialestimatesfortheunknowns,forinstance:x1(0)= 0.01andx1(0)=0.99.Settheiterationcounterto:k=0.

2Calculate

⎧ ⎪

⎪ ⎨

⎪ ⎪

1(k)=1(x1(k)) 1(k)=1(x1(k)) 2(k)=2(x1(k)) 2(k)=2(x1(k))

fromEq.(16).

3Defineı=[|x11−x11|+|(1−x1)2−(1−x1)2|](k).

Ifı<(whereistheprecisionafforded,e.g.=10−6),thena solutionisreachedandtheprocedureisterminated.

4Updateunknownsx1 andx1usingEq.(13)andTable1(calcula- tionofaLLEusinga–ϕapproach).Theiterativeschemecanbe writtenas:

⎧ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎩

x1(k+1)=

1·(22) 1·21·2

(k)

x1(k+1)=

1·(22) 1·21·2

(k)

=x1(k+1)

1 1

(k) (17)

5Setk=k+1.Returntostep2.

TheiterationsfortheLLE calculationatT=300Karedetailedin Table 2. The proposed algorithm can be used to constructthe entireLLEphasediagram,whichexhibitsanuppercriticalsolu- tiontemperature(UCST).Todoso,thealgorithmhastoberunat intermediatetemperaturesbetweenTmin=300K(forinstance)and Tmax=UCST.Thesameinitialestimatesforx1andx1canbeused foralltheLLEcalculations(followingthefirststepoftheproposed LLEcalculationalgorithm).Alternatively,inordertoincreasethe speedofconvergence,theinitialestimatesforagivenLLEcalcu- lationattemperatureTcanbemadeatatemperatureclosetothe temperatureunderconsideration.

For the binary system (1)+(2), the coordinates of the upper liquid–liquid critical point are analytically found to be:

UCST482.95K and x1,crit0.4073. The complete LLE diagram, fromlowtemperaturestotheUCSTandcalculatedusingthepro- posedalgorithm,isshowninFig.1(a).

Speedofconvergenceoftheproposedalgorithm:

Theproposedalgorithmisazeroth-ordermethodsincederiva- tivesofthemodel donot updatethemolefractions withinthe iterative process (see Eq. (13)). A majordrawback of this kind ofmethodistheslowconvergencethatisobserved.Thisfeature isillustratedinFig.1(b)showingthatfortemperaturesfarfrom theUCST,severaldozensofiterationsareneededforconvergence.

WhenapproachingtheUCST,convergenceiseventuallyreached afteraverylargenumberofiterations.Atthecriticaltemperature, thisnumberbecomesnearlyinfinite.

Inaddition,despiteofverypoorinitialestimates,theproposed algorithmisobservedtoberemarkablyrobustandfacilitatesaccu- ratedeterminationofthecompositionsofthetwoliquidphasesin equilibrium,evenattemperaturesclosetotheUCST.Furthermore, thecalculationprocedurerequires quiteareasonable computa- tiontime: theFortran 90 program used tocalculate theentire LLE diagramin Fig.1(a)runsin less than1/10s, using anIntel Xeon E5345© CPU. Note however that stronglyinadequate ini- tialestimatesmayleadinsomerarecases,totheso-calledtrivial solution,producingfluidphasesofidenticalcompositions.When constructingtheentirephasediagram,suchsituationscanbeeas- ilyavoidedbyperformingaseriesofLLEcalculationsatincreasing Table2

IterationdetailsoftheLLEcalculationperformedinSection3.3.

Iteration x1 x1 1 1 2 2 logı

0.01000 0.99000

1 0.02789 0.93862 28.67378 1.00849 1.00356 12.69554 −0.46

2 0.03248 0.92352 27.54304 1.01330 1.00482 11.90531 −0.99

3 0.03379 0.91845 27.23057 1.01517 1.00521 11.65221 −1.48

4 0.03417 0.91668 27.13949 1.01585 1.00533 11.56517 −1.95

5 0.03429 0.91605 27.11251 1.01610 1.00536 11.53453 −2.43

6 0.03432 0.91583 27.10447 1.01619 1.00537 11.52365 −2.89

7 0.03433 0.91575 27.10207 1.01622 1.00538 11.51977 −3.36

8 0.03434 0.91572 27.10135 1.01623 1.00538 11.51839 −3.82

9 0.03434 0.91571 27.10114 1.01623 1.00538 11.51790 −4.28

10 0.03434 0.91571 27.10107 1.01624 1.00538 11.51772 −4.74

11 0.03434 0.91571 27.10105 1.01624 1.00538 11.51766 −5.19

12 0.03434 0.91571 27.10105 1.01624 1.00538 11.51764 −5.65

13 0.03434 0.91571 27.10105 1.01624 1.00538 11.51763 −6.10

(5)

temperatureswheretheinitialestimatesforaLLEcalculationata certaintemperaturearegivenbythecompositionsx1andx1from theLLEcalculationatthetemperatureimmediatelybelowthepre- scribedtemperature.Theseefficientinitialestimates reducethe numberofiterationsrequiredfortheconvergence(seeFig.1(b)) andmoreimportant,generallypreventthetrivialsolutionfrom beingreached,eveninthevicinityofthecriticaltemperature.

3.4. Numericalexample2:calculationofliquid–vaporphase equilibriumusinganactivity-coefficientmodelfortheliquid phaseandtheidealgasequationofstateforthevaporphase

As an illustration of this case, we consider the binary sys- tem acetone (1)+chloroform (2) which is known to exhibit a negativeazeotrope.TheNRTL(Non-RandomTwo-Liquid)activity- coefficientmodel(Renon&Prausnitz,1968)isusedtorepresent thenon-idealityoftheliquidphase:

gE(x1) RT =x1x2

21exp(−˛21)

x1+x2exp(−˛21)+ 12exp(−˛12) x2+x1exp(−˛12)

,

with

⎧ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎩

x2=1−x1 12=b12/T 21=b21/T b12=209.38K b21=−431.47K

˛=0.1831

(18)

Theactivitycoefficientsofthetwospeciesintheliquidmixture arethengivenby:

i=exp

xj2

jiexp(−2˛ji)

[xi+xjexp(−˛ji)]2 + ijexp(−˛ij) [xj+xiexp(−˛ij)]2

,

with

i=/j

(i,j)∈{1;2}2 (19)

Thevaporpressuresofthepurecompoundsaregivenby:

log[Psat

1 (t)/bar]=4.2184−1197.01/[(t/C)+228.06]

log[P2sat(t)/bar]=3.9629−1106.90/[(t/C)+218.55] (20) ThepressureofthesystemissettoP=1atm.FromSection3.1, theproposedalgorithmcanbewrittenasfollows:

1Provideinitialestimates fortheunknowns x1 and y1. Setthe iterationcounterto:k=0.

2CalculateP1sat(T)andP2sat(T)usingEq.(20).

3Calculate

1(k)=1(x(k)1 )

2(k)=2(x(k)1 ) from Eq. (19). Calculate the coefficients C1(k) and C2(k) using Eq. (4), the pure component liquid molar volume correlations, thepurecomponent vapor pressurescorrelationsandanappropriateEoSforthegasphase.

4Define ı=[|Py1C1/Psat1 −x11|+|P(1−y1)C2/Psat2 −(1−x1)2|](k). Ifı<(where,istheprecisionafforded,e.g.=106),thena solutionisreachedandtheprocedureisterminated.

5Updatetheunknownsx1andy1:

⎧ ⎪

⎪ ⎩

x(k1+1) =

C1 (PC2−P2sat2) Psat1 1C2−Psat2 2C1

(k)

y(k1+1) =x1(k+1)×[P1sat1/(PC1)](k)

(21)

6Setk=k+1.Returntostep3.

Fig.1.(a)Composition-temperatureLLEdiagramcalculatedwithVanLaar’sactivity coefficientmodel.(b)Representationofthenumberofiterationsoftheproposed LLEcalculationalgorithmthatarenecessarytoreachconvergence(convergence criterion:ı<10−6)withrespecttothetemperature.

Atlowpressures,thecoefficientsCidefinedinEq.(4)aregen- erallyassumedtobeequalto1(thePoyntingfactorcorrectionis assumedtobeequalto1andtheideal-gasEoSisusedtomodelthe gas-phasebehavior).Theiterativeschemethenreducesto:

⎧ ⎪

⎪ ⎩

x(k1+1)=

P−Psat2 2 P1sat1−P2sat2

(k)

y(k1+1)=x(k1+1)×

P1sat1/P

(k)

(22)

ItappearsthatthetwoequationsinthesystemgivenbyEq.(22) canbesolvedseparatelyandsuccessively(i.e.thetwoequations canbedecoupled).Theupperequationonlyinvolvestheunknown x1whereasthelowerequationdirectlyyieldsy1asafunctionofx1. Asaresult,theinitialestimatefory1doesnotaffecttheiterative scheme.

Suchasituation,whichisaparticularcaseofthegeneralsolution procedure previously presented, arises because the VLE calcu- lation in binary systems at a specified T and P using the –ϕ approachatlowpressures,reducestotheresolutionofaunique equation(P−P1satx11−P2sat(1−x1)2=0)withrespecttoasin- gleunknown(x1).Theiterativeschemetakestheformofaclassical 1-dimensionaldirect-substitutionmethod.

Thisalgorithmwasappliedatt=64C(T=337.15K).Atthistem- peratureandatmosphericpressure,theazeotropicsystemacetone (1)+chloroform(2)exhibitstwodistinctVLE.Thesolutionthatthe

(6)

Table3

IterationdetailsforthetwoVLEcalculationsperformedinSection3.4usingtwo differentinitialestimatesforunknownx1.

Iteration x1 y1 logı

0.01000

1 0.20795 0.00636 −0.67

2 0.23475 0.16184 −1.46

3 0.23003 0.18760 −2.20

4 0.23115 0.18298 −2.83

5 0.23090 0.18407 −3.47

6 0.23095 0.18382 −4.11

7 0.23094 0.18388 −4.76

8 0.23094 0.18386 −5.40

9 0.23094 0.18387 −6.04

0.99000

1 0.66050 1.28752 −0.14

2 0.60094 0.75855 −0.94

3 0.62039 0.66233 −1.44

4 0.61130 0.69334 −1.77

5 0.61513 0.67879 −2.15

6 0.61343 0.68491 −2.50

7 0.61417 0.68219 −2.86

8 0.61384 0.68337 −3.22

9 0.61398 0.68286 −3.58

10 0.61392 0.68308 −3.94

11 0.61395 0.68298 −4.30

12 0.61394 0.68303 −4.66

13 0.61394 0.68301 −5.02

14 0.61394 0.68301 −5.38

15 0.61394 0.68301 −5.74

16 0.61394 0.68301 −6.10

proposedalgorithm convergesto,dependsonthechoseninitial estimatefortheunknownx1.Asanillustration,Table3showsthe resultsrespectivelyobtainedusingx(0)1 =0.01andx(0)1 =0.99.

To plot the complete isobaric phase diagram presented in Fig.2(a),itwasthusnecessarytoproceedintwosteps:theportion ofthephasediagramlocatedonthelefthandside(lhs)oftheneg- ativeazeotropewasobtainedbyperformingasuccessionofVLE calculationsatincreasingtemperatures,usingtheinitialestimate x(0)1 =0.01systematically.Theportionofthephasediagramlocated ontherighthandside(rhs)wasobtainedusingasimilarproce- durewithx(0)1 =0.99.Alsonotethatanaccuratedeterminationof theazeotroperequiresthatthetemperaturesteptbeefficiently managedwhencalculatingeachpartofthephasediagram.Asan example,thetemperaturecanbeincreasedusingaconstanttem- peraturestep(e.g.t=0.1K),untila trivialsolution(i.e.x1=y1)

is reached(meaningthat thecurrent temperatureis abovethe azeotropictemperature).Thetemperatureisthusreturnedtoits previousvalueandthenincreasedbyusingasmallertemperature step(e.g.t=t/10).Theprocedureisrepeateduntilt<tmin (e.g.tmin=10−4K).

Itisinterestingtonotethattheproposedalgorithmcanalsobe easilyusedtocalculateVLEwhenthemolefractionsofthecom- pounds inthetwo phasesare greaterthan1 orless thanzero, althoughthisisnotrelevantforengineeringcalculations.Thisfea- tureshowsthatthealgorithmdoesnotnecessarilydivergeatthe specifiedtemperatureandpressurewhennophysicallymeaning- fulVLEexists.Fig.2(b)showsthatconvergenceisreachedafter quiteareasonablenumberofiterationsforeachVLEcalculation.

Whenthephasediagramdoesnotexhibitcriticality,afewdozens ofiterationsaregenerallyrequired.Notethatthisnumberreaches aminimumatthepure-componentVLEpoint.

Onthechoiceofinitialestimatesforx1:

SincetheVLEcalculationreducestosolveoneequationwith respecttox1whenusingthe–ϕapproachwithCi=1,aninitial valueisnotneededfortheunknowny1andaninitialestimatemay beeasilychosenforx1,asexplainedbelow.

Firstofall,notethatitissimpletochoosetheinitialestimate inmostofthecaseswheretheliquid–vaporphasediagramdoes notexhibitazeotropyorathree-phaseline(oranycombinationof both).InsuchcasesandifaVLEexistsatthespecifiedtemperature andpressure,thenanyvalueofx1∈]0;1[leadstoacorrectsolution (divergenceproblemsarequitescarce).

Aspreviouslymentioned,phasediagramsexhibitingazeotropy aretrickiertodealwithsincetwoVLEmayexistatafixedtem- perature and pressure. The case of the binary systemacetone (1)+chloroform(2)atatmosphericpressureisreconsideredasan illustration.Threedifferentsituationsmayoccur:

•thecalculationconvergestoasolutionsuchthatx1>y1whichis locatedinthelefthandside(lhs)ofthephasediagramshownin Fig.2(a)

•thecalculationconvergestoasolutionsuchthatx1<y1whichis locatedintherighthandside(rhs)ofthephasediagramshown inFig.2(a)

•orthecalculationdiverges.

Thethreedomainsassociatedwitheachofthesethreesitua- tionsareshowninFig.3.Thedomainofdivergenceislargeatlow temperatures,narrowswithincreasingtemperatureandreduces toasinglepointattheazeotropictemperature.Thisfigurealso

Fig.2. (a)Isobaricphasediagramoftheacetone(1)+chloroform(2)systemplottedatP=1atmusingthe–ϕapproach.(b)Representationofthenumberofiterationsfor theproposedVLEcalculationalgorithmthatarenecessarytoreachconvergence(convergencecriterion:ı<10−6)withrespecttothetemperature.Thefullanddottedlines arerespectivelyassociatedwiththecalculationofthelefthandside(lhs)andrighthandside(rhs)ofthephaseequilibriumdiagram.

(7)

Fig.3.Influenceontheconvergenceoftheproposedalgorithm,ofthechoseninitial estimateforx1.

illustrateswhytheinitialestimatesx(0)1 =0.01andx1(0)=0.99can beusedtoconstructthephasediagramatalltemperatureswithout difficulty.Sincethedomainofdivergencegenerallyoccursinthe middleofthecompositionrange,itisadvisabletosystematically chooseinitialx1estimatesclosetozeroandone.

Moredetailsonthefixedpointmethodandconvergence-related issuescanbefoundinAppendixA.

3.5. Numericalexample3:calculationofafluid–fluidphase equilibriumusinganequationofstate

Inthissection,thePeng–RobinsonEoS(Peng&Robinson,1976) isusedtomodeltheliquid–vaporphasebehaviorofthebinarysys- temcarbondioxide(1)+n-hexane(2)atT=393.15KandP=40bar.

ThePeng–RobinsonEoSisgivenby:

PEoS(T,v,zi)= RT

v−bm− am

v(v+bm)+bm(v−bm) (23) wherePEoSisthepressureofthemixturecalculatedfromtheEoS,v isthemolarvolumeofthefluid;ziisthemolefractionofcomponent iinthefluidphaseunderconsideration(zi=xiforaliquidphaseand zi=yiforagasphase);andamandbmarethetwoEoSparametersfor themixture,whichfollowtheclassicalmixingrulesgivenbelow:

am=

2 i=1

2 j=1

zizj

aiaj(1−kij) and bm=

2 i=1

zibi (24)

Expressionsforthepure-componentquantitiesaiandbiarepro- videdinTable4.Thecriticaltemperatures,criticalpressuresand acentricfactorsofthepurecomponentsthatareneededtoevaluate Table4

ParametrizationofthePeng–RobinsonEoS.

Quantity Expression

Criticalcompacity c 1

3

−1+

3

6

2+8

3

6 28

Universalconstant˝a 8(5 c+1)/(4937 c) Universalconstant˝b c/( c+1)

Shapefunctionmi 0.37464+1.54226ωi0.26992ω2i

ac,i ˝aR2Tc,i2/Pc,i

bi ˝bRTc,i/Pc,i

˛i(T)

1+mi

1

T/Tc,i

2

ai(T) ac,i˛i(T)

Table5

Physicalpropertiesofpurecomponents.

Tc/K Pc/MPa ω

CO2 304.2 7.383 0.2236

n-Hexane 507.6 3.025 0.3013

N2 126.2 3.400 0.0377

Methane 190.6 4.599 0.0120

Ethane 305.3 4.872 0.0990

Propane 369.8 4.248 0.1520

thecoefficientsaiandbi,aregiveninTable5.Thebinaryinterac- tionparametersareestimatedusingthePPR78group-contribution method(Jaubert&Mutelet,2004;Jaubert,Privat&Mutelet,2010):

k12=k21=0.1178andk11=k22=0.Asexplained inSection 2.2,a pressure-explicitEoSformixturesthatcanrepresentbothaliq- uidandavaporphase,asopposedtoavolume-explicitEoS,uses thetemperatureT,themolarvolumevandthecompositionasvari- ables.Consequently,thefluid–fluidequilibriumproblematafixed TandPinvolvessolvingfourequations(seeEq.(7))withrespect tofourunknowns:themolarvolumesand compositionsof the twophasesinequilibrium. InthePeng–Robinson EoS,thefuga- citycoefficientofagivencompoundiinaphasecharacterizedby thetemperatureT,themolarvolumev,themolefractionsziofthe compoundsandthepressureP=PEoS(T,v,zi),isgivenby:

⎧ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎪

⎪ ⎩

lnϕˆi= bi

bm

Pv

RT −1

−ln

P(vbm)

R·T

− am

2√ 2RTbm

ıi− bi

bm

· ln

v+bm(1+√ 2) v+bm(1−√

2)

with : ıi=2

√ai am

nc

j=1

zj

aj(1−kij)

(25)

Thefollowingalgorithmisproposedforsolvingaliquid–vapor equilibriumataspecifiedTandP,modeledwithanEoS:

1Provideinitial estimatesfor the unknownmolefractions, for instance:[x1](0)=0.01and[y1](0)=0.99.Settheiterationcounter to:k=0.

2Calculatethemolarvolumesoftheliquidandgasphasesatthe currentiterationk(denotedbyv(k)L andv(k)G )bysolvingtheEoSat afixedT,Pandcomposition:

v(k)L isthesolutionoftheequation: P=PEoS(T,v(k)L ,x(k)1 )

v(k)G isthesolutionoftheequation: P=PEoS(T,v(k)G ,y(k)1 ) (26)

Notethat theresolution ofanEoSisa recurringthermody- namicissue,addressedmanytimes(Deiters,2002;Privat,Gani

&Jaubert,2010).ForacubicEoS,asimpleCardano-typemethod canbeused.Inaddition,thesolutionoftheEoSmayproduce eitheroneorthreeroots.WhentheEoShasthreeroots,theliq- uidroothasthesmallestmolarvolumewhereasthevaporroot hasthehighestmolarvolume.

3Calculatethefugacitycoefficientsofallcomponentsinallphases:

i∈{1;2}

ϕˆi,L(k)=ϕˆi(T,v(k)L ,x(k)1 )

ϕˆi,G(k)=ϕˆi(T,v(k)G ,y(k)1 )

(27)

4Defineı=[|x1ϕˆ1,L−y1ϕˆ1,G|+|(1−x1)ϕˆ2,L−(1−y1)ϕˆ2,G|](k). Ifı<(whereistheaffordedprecision,e.g.=10−6),thena solutionisreachedandtheprocedureisterminated.

Références

Documents relatifs

Figure 4 maps the distribution of topics for the 55 most-represented countries in the database. The number in each cell is the number of papers devoted to country x and related to

Soot aggregate size (number of primary par- ticles per aggregate) distributions are described by a lognormal function whose distribution width is largely independent of the

In December 1989, recognizing the need to coordinate environmental research activities in the United States, a committee of scientists, educators, and the general public

One mechanism is associated with the interaction of the pressure wave with the density discontinuities within the explosive [2] and the collapse of the available pore space

Evidemment, le problème est difficile pour la main-d'œuvre spécialisée, car nous sommes la première grande usine de notre branche dans le canton, si bien que les Fribour- geois

TABLE I: Summary of results; ε denotes the reconstruc- tion efficiency and ε tot the total efficiency including daugh- ter branching fractions, both in percent; N is the number

We prove by the entropy production method the exponential decay in time of the solution to the only steady state of the associated stationnary equation.. Keywords:

In constructing a model in which the possible worlds are maximal consistent sets of formulas we will have to specify when one world is accessible from another(that model, in this