Inter-compartment interaction in multi-impeller mixing
Part II. Experiments, sliding mesh and large Eddy simulations
Haringa, Cees; Vandewijer, Ruben; Mudde, Robert F.
DOI
10.1016/j.cherd.2018.06.007
Publication date
2018
Document Version
Final published version
Published in
Chemical Engineering Research and Design
Citation (APA)
Haringa, C., Vandewijer, R., & Mudde, R. F. (2018). Inter-compartment interaction in multi-impeller mixing:
Part II. Experiments, sliding mesh and large Eddy simulations. Chemical Engineering Research and Design,
136, 886-899. https://doi.org/10.1016/j.cherd.2018.06.007
Important note
To cite this publication, please use the final published version (if applicable).
Please check the document version above.
Copyright
Other than for strictly personal use, it is not permitted to download, forward or distribute the text or part of it, without the consent of the author(s) and/or copyright holder(s), unless the work is under an open content license such as Creative Commons. Takedown policy
Please contact us and provide details if you believe this document breaches copyrights. We will remove access to the work immediately and investigate your claim.
This work is downloaded from Delft University of Technology.
‘You share, we take care!’ – Taverne project
https://www.openaccess.nl/en/you-share-we-take-care
Otherwise as indicated in the copyright section: the publisher
is the copyright holder of this work and the author uses the
Dutch legislation to make this work public.
ContentslistsavailableatScienceDirect
Chemical
Engineering
Research
and
Design
j o u r n a l ho me p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c h e r d
Inter-compartment
interaction
in
multi-impeller
mixing.
Part
II.
Experiments,
sliding
mesh
and
large
Eddy
simulations
Cees
Haringa
∗,1,
Ruben
Vandewijer,
Robert
F.
Mudde
TransportPhenomena,DepartmentofChemicalEngineering,DelftUniversityofTechnology,vanderMaasweg9,
2629HZDelft,TheNetherlands
a
r
t
i
c
l
e
i
n
f
o
Articlehistory:
Received19November2017 Receivedinrevisedform24April 2018
Accepted1June2018 Availableonline13June2018
MSC: 00-01 99-00Keywords: Mixing Schmidtnumber CFD Multipleimpellers Rushton
a
b
s
t
r
a
c
t
Steady state multiplereference frame-RANS (MRF-RANS)simulations frequently show strongover-predictionsofthemixingtimeinsingle-phase,multi-impellermixingtanks, whichissometimespatchedbyadhoctuningoftheturbulentSchmidt-number.InPartI ofthiswork,weexperimentallyrevealedthepresenceofmacro-instabilitiesintheregion betweentheimpellers,aswellasapeakintheturbulentkineticenergyintheregionwhere theflowfromtheindividualimpellersconverges.TheMRF-RANSmethodwasfoundunable tocaptureboth.Inthissecondpaper,weshowthatthesliding-meshRANS(SM-RANS) approachdoescapturetheeffectofmacro-instabilities,whilestillunderestimatingthe tur-bulentkineticenergy.Consequently,theSM-RANSmethodmildlyover-estimatesthemixing time,whilebeinglesssensitivetotheexactmeshgeometry.Largeeddysimulationswith thedynamicSmagorinskymodelreasonablycapturethekineticenergycontainedin macro-instabilities,andproperlyassesstheturbulentkineticenergyintheregionbetweenthe impellers,evenforcrudemeshes.Consequently,themixingtimeisreasonablyassessed, andevenunder-predictedatthecrudestmeshes.However,theturbulentkineticenergyand energydissipationintheimpellerdischargestreamarepoorlyassessedbythedynamic Smagorinskymodel.
©2018InstitutionofChemicalEngineers.PublishedbyElsevierB.V.Allrightsreserved.
1.
Introduction
Computational fluid dynamics (CFD) has frequently been appliedtosimulatemixingprocessesinstirredtanks. Espe-ciallyReynoldsaveragedNavierStokes(RANS)modelsprovide thepossibility toevaluatethe mixingperformanceof vari-ousimpeller configurations without requiringan extensive experimentalcampaign.Anassessmentofmixingliterature focusing on Rushton turbines, conducted in Part I of this work,reveals thatRANSsimulationsarecapableof
reason-Abbreviations: MI, macro-instability;LDA,laserDoppleranemometry;PIV,particleimagevelocimetry; RANS,Reynoldsaveraged
NavierStokes;MRF,multiplereferenceframes;SM,slidingmesh;(S/R)KE,standard/realizablek−;RSM,Reynoldsstressmodel;(D/L)ES, detached/largeEddysimulation.
∗ Correspondingauthor.
E-mailaddress:cees.haringa@DSM.com(C.Haringa).
1 Currentaddress:DSMBiotechnologyCenter,AlexanderFleminglaan1,2613AXDelft,TheNetherlands.
ablypredictingthemixingbehaviorintanksstirredbyasingle impeller,albeitwiththerequirementofhighmeshdensities. Incontrast,thedimensionlessmixingtime95is
systemati-callyover-predictedinmulti-impellertankswithhighmutual impellerspacingwhenusingRANSmodels.Inparticular,for multiplereferenceframe(MRF)simulationsincreasingmesh densitiesresultsinanincreasingover-predictionof95
com-paredtoexperimentalresults.
Itiswellknownthatcompartmentsformaroundthe indi-vidualimpellersofamulti-Rushtontank.InPartIofthiswork,
https://doi.org/10.1016/j.cherd.2018.06.007
Nomenclature
Roman
Ct tracerconcentration,kg/m3
C off-bottomclearanceimpeller,m
CS Smagorniskyconstant,–
C inter-impellerclearance,m
D impellerdiameter,m D diffusioncoefficient,m2/s
Dt diffusioncoefficient,turbulent,m2/s
Ep/Et fractionofperiodicenergyinspectrumofu,–
f frequency,s−1 f1.1 basefrequency1,s−1 f2.1 basefrequency2,s−1 f1.2 harmonicfrequency1,s−1 f2.2 harmonicfrequency2,s−1 FQ pumpingnumber,– h axialcoordinate,m H tankheight,m
k slotnumber(autocorrelation),–
kt turbulentkineticenergy,m2/s2
kMI macro-instabilitykineticenergy,m2/s2
kt,* kt+kMI,m2/s2
kt,fit ktcomputedviaauto-correlationfit,m2/s2
kt,SI ktcomputedviaspectralintegration,m2/s2
kMI,fit kMIcomputedviaauto-correlationfit,m2/s2
kMI,SI kMIcomputedviaspectralintegration,m2/s2
N impellerrevolutions,1/s
P power,W
Po powernumber,W
Qax axialflowrate,L/s
r radialposition,m
R tankradius,m
Sij shearrate,s−1
Sct turbulentSchmidtnumber,–
T tankdiameter,m t timestepsize,s
u fluctuatingvelocity,m/s
U meanvelocitym/s
Utip impellertipspeed,m/s
V tankvolume,m3
Vi gridcellvolume,m3
w Tukey–Hanningwindow,–
Greek
laserwavelength,nm viscosity,dynamic,Pas
t viscosity,dynamic,turbulent,Pas
viscosity,kinematic,m2/s
t viscosity,kinematic,turbulent,m2/s
turbulentdissipationrate,m2/s3
ˆ auto-correlationcoefficient,– l density,kg/m3
noisecomponent,– lag lagtime,s
95 mixingtime,s
ij stresstensor,Pa
95 mixingnumberN·95,–
we posed the hypothesis that the over-predicted 95
origi-natesfrom anunder-predictioninmassexchangebetween thesecompartments,whilemixinginsidethecompartments is likelyproperly captured (Coroneo et al., 2011). Although stirredtankshavebeenstudiednumeroustimes,theregion betweenthe impellershadpreviouslyattractedlittle atten-tion.Wehencefocusedourstudyonthisregion,toprovide insightin theinter-compartment hydrodynamicsand their role in mixing. Using laser-Doppler anemometry (LDA),we observed thatthis transport isgoverned bytwo processes: macro-instabilities(MIs)atfrequenciesf/N=0.02–0.06(Nthe agitationrateins−1)inthehorizontalplaneseparatingthe compartments,andapeakinturbulentkineticenergy(kt)in
theplanesegregatingthecompartments,generatedbythe col-lisionoftheflow-loopsintheconvergingflowsectionnearthe wall.Boththeseeffectsareexpectedtoenhanceaxialmass exchange.
Besides experimental results, we reported steady state MRF simulationsinPartIofthis study.Due totheirsteady statenature,theyare inherentlyincapableofcapturing the influence of macro-instabilities. Combined with an under-estimated turbulent kinetic energy (kt), attributed to poor
capturingofturbulencegenerationinthecollidingflowregion byvirtueofReynoldsaveraging,thisleadstoahighly over-estimated95; 95wasfurthermore stronglyaffectedbythe
axial flowrate between the compartments, which in turn was highly mesh sensitive. This explained the increasing 95 with increasing mesh density. In this second part, we
assesswhethertransientsimulations,bothslidingmesh(SM) RANSandlargeEddysimulations(LES),docaptureallrelevant hydrodynamicsbetweenthe impellers,and properly assess themixingtimesinmulti-impellerstirredtanksystems.
2.
Literature
review:
Rushton
turbines,
LES
and
DES
Inthepreviouspartofthisworkwediscussedtheliterature regarding experimental assessment of flows in Rushton-stirredtanks,andtheuseofRANSmodelstomodelsuchflows. MixingresultsforLESwerealsoincludedinPartI,andarenot listedhere.Here,wediscussLESanddetachededdy simula-tions(DES)forflowmodeling.Toourknowledge,DEShasnot beenusedformixingstudiesinbaffledRushton-stirredtanks. 2.1. LES
Revstedt et al.(1998) usedfinitevolumeLES(FV-LES), with 2.12×105 gridcells,animplicitclosure modeland
momen-tumsourcetermsforimpellermodeling.DecentresultsforU
andktwerereportedinthebulk,withpooreragreementnear
theimpeller.Yeohetal.(2004,2005)appliedtheSmagorinsky (SGS)subgridmodelwithconstantCS=0.1(FV-LES,4.9×105
cellswithsliding-deforminggrid),reportinggoodresultsfor
U and kt, but providing no data on . Zhang et al. (2006)
reported similarresults. Alattice-Boltzmannapproach (LB-LES)isfrequentlyappliedinLESstudies(Eggels,1996).Eggels (LB-LES,SGS,CS=0.1)showedsomelocalunder-predictionof
Uax,butoverallagoodagreementwithdatabyBakker(1996)
wasreported.DerksenandVandenAkker(1999)(SGS-model,
CS=0.12,6×106gridnodes)reportedaccurateresultsforthe
dischargestreamvelocities,kt,andtrailingvortexbehavior.
Theyreportedaphase-averagedmaximumenergydissipation intheimpellerdischargeofmax¯ /N3D2≈4.6,over50%lower thanmeasuredexperimentally(DucciandYianneskis,2005).
Hartmannetal.(2004)reportedasimilarunder-estimationof withLB-LES,resultsthatarefurthersupportedbyMicheletti etal.(2004)(SGS,CS=0.1)withFV-LES.
Delafosse etal. (2008,2009)(sliding mesh,106 cells,
FV-LES,SGSmodel)explicitlynotedthatsettingCS=0.1leadsto
asignificant under-predictionof inthedischarge stream.
CS=0.1 was selected based on testing for awide range of
flows,withhighervaluesofCSleadingtoexcessiveturbulence
dampening.SettingCS=0.2stronglyimprovespredictionsfor
,withoutsignificantlyaffectingthepredictionsforvelocity andkt.Soosetal.(2013)(slidingmesh,1.6×106cells)alsoused
CS=0.2.ComparedtothedataofbothEscudiéetal.(2004),
and Wu and Patterson(1989), theyreported amild under-predictionofthevelocities andthe periodickinetic energy, whilektwaswellpredicted.Thevaluesforareinaccordance
withDelafosseetal.
TheworkofDelafosseetal.andSoosetal.indicatesa case-by-casetuningofCSmay berequired,whichisundesirable
fromtheperspectiveofpredictivecapabilities.Thedynamic SGS (DYN)model aims atsolving this issue by computing ratherthanprescribingCS.MurthyandJoshi(2008)(FV-LES,
DYN,1.3×106cells,slidingmesh)reportgoodresultsforthe
dissipationbasedpowernumberPo,butshownoprofilesof
orvaluesofCS.Jahodaetal.(2007)applythedynamicmodel
formixingin1and2impellergeometriesbutdidnotreport orPo.TheydidshowlocalvaluesofCS,whichwereinthe
rangeof0.05–0.1,belowthedefaultCS=0.1.
AdirectcomputationofCSusingdirectnumerical
simu-lation(DNS) byGillissen and Vanden Akker(2012) yielded
CS≈0.1,inagreementwiththeirowndynamicLES.This
indi-catesthattheCStuningconductedbyDelafosseetal.andSoos
etal.isnotinaccordancewithDNSobservations.Gillisenand vandenAkkernotedthattheunder-predictioninmaybethe resultofanunder-predictedkt-productionduetoinsufficient
meshresolutioninthevicinityofwalls.
2.1.1. DES
Detachededdysimulations(DES)blendaLESapproachinthe free-streamwithRANSinunder-resolved(wall)regions,and maytherebyreduceanydependenceofthebulkflowonwall effects,possiblyimprovingthepredictionsforifthe hypoth-esisbyGillisseniscorrect.Ofcourse,theaccuracyofthewall flowitselfwillstillbelimited,duetotheinherentassumptions oftheRANSmethodology.
Gimbunetal.(2012)presentedSpalart-Allmaras-DES sim-ulationsofaRushton-stirredtank,extensivelycomparingthe resultswith bothSKE-RANS and FV-LES (SGS, CS=0.1). The
bulkvelocitypredictionwasverysimilarbetweenthemodels. DESgenerallyyieldsthebestagreementwithexperimentalkt
data(Derksenetal.,1999),predictingslightlyhighervalues thanSKEandLES.LESandDESperformedsimilarinassessing thequalitativetrailingvortexbehavior,withtheSKEmodel predictingsignificantlylower radialspreadingofthevortex core,similartothestudyofSinghetal.(2011).DEScompared favorabletotheothermodelsinpredictingvelocitiesandkt
inthevortexcore.Overall,DESoutperformedRANS,and out-performedLESinregionswherewalleffectsaresignificant.
Charaetal.(2016)cametosimilarconclusions,observinggood agreementindischargevelocitiesandtrailingvortexbehavior. Theynotedthatthetangentialspreadofthetrailingvortexis slightlynarrowerthanexperimental(PIV)resultsshow.
Lane (2015) reported the energy dissipation behavior of variousturbulencemodelswithanA-310impeller(13.1×106
gridcells).Apowerrecoveryof69%wasobserved,i.e.PDES=
Table1–Meshesusedinthiswork.2IFrepresentsa 360◦domain.Thelastletter(s)representthemesh quality(C=crude,M=medium,F=fine,SF=super-fine).
Name Cells Domain Methods
2IF-C 648k 360◦ LES
2IF-M 1997k 360◦ SM-RKE,SM-RSM,LES
2IF-F 5884k 360◦ SM-RKE
2IF-SF 10584k 360◦ LES
(+t)S2ijdVis69%ofthepowerinputbasedontorque.For
various SST and KE formulations,theenergy recoverywas 68–91%andstronglymeshdependent,supportingthe obser-vationsbyCoroneoetal.(2011).Thelowenergyrecoveryfor DES,atthefinestmeshused,doeshintthatthewall treat-mentofDESdoesnotprovideasignificantimprovementover LESintermsofresolving.Acomprehensivecomparisonof LESandDESwithvariouslevelsofwallresolution,possibly supportedbyDNS,isrequiredtoprovidefurtherinsightin thereportedunder-predictionsof.Thatisoutofthecurrent scope,however.
3.
Materials
and
methods
3.1. CFDsetup
3.1.1. Geometry
A2-impellerstirredtankwithH=2T,D=T/3,C=T/3andC=T
wasmodeled,withT=0.29m,andanagitationrateN=5s−1. Note that our own experiments were conducted after the CFD work, and because ofthe equipment availablein our labwereperformedwithT=0.26andN=5.78s−1.Thisgives
Re=4.34×104,comparedto4.67×104byJahodaetal.(2007)
and inour CFD work. We donot expect the results tobe affected;allresultsarepresentedindimensionlessform,and allworkwasconductedinthefullyturbulentregime.Detailed informationontheexperimentsispresentedinPartIofthis work.Allinternalsweremodeledassheetbodies(Gunyoland Mudde,2009;Coroneoetal.,2011).LESandSM-RANS Simula-tionswereconductedwitha360◦domain,asusingaperiodic meshwillconstrainthemotionofmacro-instabilities(seePart I ofthis study).In accordance withthe experimentalwork ofJahodaetal.,thetracerconcentrationistrackedwithtwo probes,thebottomprobeatheightT/4andtopprobeatheight 1.25T,placedbetweenthebaffles,atT/20fromthewall.
3.1.2. Numericalsetup
Thenumericalsettingswerelargelysimilartothoseusedin PartI.Spatialdiscretizationwassetto2ndupwind(Gunyoland Mudde,2009;Coroneoetal.,2011)withRANS,andbounded centraldifferenceswithLESsimulations.Standardwall func-tionswereemployedforallsimulations.Thefreesurfaceat thetopwasmimickedbyusingano-shearsurface,allwalls weresettono-slip.Convergencewasdeclaredwhenthe resid-ualswerebelow10−5withinatimestep.Themeshesusedin thiswork areequaltothoseinPartIofthe work,withcell countslistedinTable1.Duetothecomputationtimerequired
forSM-RANSandSM-LES,itwasnotpossibletotestall
mod-elswithallmeshes.Additionalinformationonthemeshesis providedinSupplementarymaterialA.
Tracerwasinjectedinasphericalvolumewitharadiusof 0.0125m;thevolumecenterwasaty=0.551m(fromthe bot-tom),atr=0.0725m,inthebaffleplane.Wesett=0.00333s withslidingmeshandthe2IF-C LESsimulation.The2IF-SF
LESsimulationwasconductedwitht=0.001667s.Mixingin mesh2IF-M was studiedin triplicate withLES; twice with t=0.00333s,andoncewitht=0.001667s,totesthow repro-duciblethe resultsare.Temporaldiscretizationwassecond orderimplicit.The tracerand bulkfluidhad equal proper-ties,=1000and=0.001,suchthatthetracerwillnotdisturb theflowfield.Subgridspeciesdiffusioniscoupledtothe sub-gridturbulentviscositytviatheturbulentSchmidtnumber,
Sct=t/(Dt)=0.7,givingaspeciesfluxJi=−(lD+t/Sct)∇Ci
withD=10−9themoleculardiffusioncoefficientandCi the
scalarconcentration. 3.2. Turbulencemodels
Therealizable k− (RKE)and Reynolds stressmodel (RSM) used in SM-RANS modeling are well established, we refer to,e.g.GunyolandMudde(2009)fortheirdescription.Based onpreviousliterature,wefocus onthedynamic Smagorin-sky(LES-DYN)modelforLES.Thismodelassumesisotropic subgrid turbulence, using an eddy viscosity formulation with t being the subgrid turbulent viscosity. The filtered
Navier–Stokesequationsforanincompressiblefluidread: ∂ ¯ui ∂t + ¯ui ∂ ¯ui ∂xj =− 1 ∂ ¯p ∂xi + ∂ ∂xj
[+t] ∂ ¯ui ∂xj (1)Here, t is the kinematic turbulent viscosity, t/. The
dynamicturbulentviscositytiscalculatedfromEq.(2)
ij−
1
3kkıij=−2tS¯ij (2) With Sij is the resolved-scalerate ofstrain tensor. The
subgrid-scalestresstensorijiscalculatedbytheapplied
sub-gridmodel.Thebasisofthedynamicsubgridmodel(LES-DYN) isformedbytheturbulentviscosityformulationofthe stan-dardSmagorinskymodel,Eq.(3):
t=L2s·
2 ¯SijS¯ij (3)
where Ls is the mixing length calculated as Ls=
min(d,CSV1/3c ) with the Von Karman constant, d the
nearest wall distance and CS the Smagorinsky constant.
Whereas the standard Smagorinsky model typically pre-scribesCS=0.1,thedynamicmodelcomputesCSbasedonthe
localresolvedmotionsusingatest-filterapproachbasedon theworkofGermanoetal.(1990)andLilly(1992).Wereferto
Kimetal.(2004)fordetailsontheimplementationinFLUENT. Forstability,CSiscappedbetween0and0.23.
3.3. Analysismethods
WerecordedvelocitytimeseriesinthebaffleplaneinourLES simulations.Intheimpelleroutflow,theimpellerfrequency andthefirstandsecondharmonicthereofwereremovedvia Eq.(6),similartotheexperimentalprocedurereportedinPart I.Dataintheregionbetweentheimpellers,referredtoasthe inter-compartmentregion(withthehorizontalplane segre-gatingthecompartmentsreferredtoasinter-compartment plane),wasanalyzedforthepresenceofMIs.Thevelocitytime seriesatthepointscoincidingwiththeexperimentallyused gridwereanalyzedbyfittingtheauto-correlationsignal(Eq.
(4))andbyanalysisofthespectraldensityfunction(Eq.(5)), equaltotheexperimentalassessment:
ˆfit()=b+c0e−˛0+˙ni=1cne−˛ncos(2·nf ) (4)
S(f)= t
12ˆ(klag)w(klag)cos(kflag)
(5) Forthesemethods,theinfluenceofperiodiccomponents canberemovedfromthefluctuatingvelocityviaEq.(6)incase oftheauto-correlationsignal,andEq.(7)incaseofthespectral densityfunction: ut=u·
1−˙(cn) (6) ut=u·1−Ep Et (7)
HereEp/Etisthefractionofenergycontainedinthe
macro-instability componentsofthe u-spectrum;weassumed all frequencies below f/N=0.1 contribute to Ep. The influence
ofmacro-instabilities inthe inter-compartmentregion was addressedwithbothmethodsforLESsimulations.Inthiscase, twobasefrequenciesandtheirfirstharmonicswereincluded inEq.(6).Intheinter-compartmentregion,ktistheturbulent
kineticenergybasedonut,kt*thetotalfluctuatingenergy,and
the macro-instability kineticenergykMI=kt*−kt. Subscripts
fitandSIareusedtodistinguishbetweenresultscalculated using Eqs. (6) and (7), respectively. For sliding mesh simu-lations, thereisnoturbulent componentinthe fluctuating velocity,hencekMIwasdirectlycomputedfromthefluctuating
Reynolds-averagedvelocity,whilektisprovidedbythe
turbu-lencemodel,andkt*followsfromtheirsum.Thelocationy=0
isusedtoindicatethehorizontalplaneexactlybetweenthe twoimpellers,ataheight5T/6.Adetaileddiscussionofthe experimentalanalysismethodsisprovidedinPart Iofthis paper.
4.
Results
and
discussion
4.1. Validation:theimpelleroutflow
InFig.1,theimpellerdischargestreamprofilesareshown,in comparisontoexperimentalresultsfromliteratureaswellas PartIofthiswork.Experimentaldataonwasnotcollected inthecurrentstudyduetolimitationsintheequipment.As fortheMRFsimulationsinPartI,theSMsimulationsshow goodagreementwithexperimentaldataintheimpeller out-flow.Again,theRSMsimulationpredictsadecreaseinktand
nearthebladetip,wheretheRKEmodeldoesnot.Inthebulk oftheoutflow,RKEandRSMareinexcellentagreement.LESis wellcapableofcapturingUrad,butperformspoorerforktand
especiallyfor.Powernumbersandadditionaldiscussionon meshdependencyareavailableinSupplementarymaterialA. For the LES simulations, kt is computed based on the
resolved scales, removingperiodiccomponents (blade pas-sages)byEq.(6).Subgridkineticenergyisnotincludedinthe figures,whichmeans2IF-SFisexpectedtoyieldhigherktthan
2IF-Cand2IF-M,asisindeedobserved.However,2IF-SF,which isstillinsufficientlyfinetocaptureallenergycarryingeddies, doesover-estimatetheexperimentalkt.Thedissipationrate
ontheotherhandisstronglyunderestimated.Overall,the powernumberonenergydissipationfluctuatesaroundPo≈8,
Fig.1–ProfilesofUrad,ktandinthetopimpelleroutflowcomparedwithLDAdata.Bottomimpellerresultsareomitteddue tosimilarity.Toprow:SMsimulations.Solidline:2IF-F,SM-RKE.Dashedline:2IF-M,SM-RKE.Dottedline:2IF-M,SM-RSM.
Bottomrow:LESsimulations.Solidline:2IF-SF.Dashedline:2IF-M.Dottedline:2IF-C.Symbolsrepresentexperimental data.Abbreviations:W.P.=WuandPatterson(1989),M.J.=MurthyandJoshi(2008),D.Y.=DucciandYianneskis(2005).The bluecrossesintheUradplotrepresenttheupper-andlowerboundofthestudiesreviewedbyRanadeandJoshi(1990).
(2004) and Derksenand Van den Akker (1999) reported an under-estimationofwiththestandardSmagorninskymodel usingconstantCS=0.1.Delafosseetal.andSoosetal.found
thatsettingCS=0.2increasedagreement;ishighlysensitive
toCS.InagreementwithJahodaetal.(2007),weobservethe
dynamicmodelyieldsCS inthe range0.01–0.06.Theselow
CSvaluesareconsistentwiththestrongunder-predictionof
(Delafosse etal.,2008).Unfortunately,Jahodaetal.donot reportktandforverification.Clearly,accuratepredictionof
instirredtanksusingLESrequiresfurtherattention. 4.2. Mixingtimes
Themixingtimeisreportedindimensionlessform,95=N95,
with95beingthetimeinsecondsbeyondwhichCt/ ¯Ctisbound
between0.95and1.05,withCtthetracerconcentration.The
CoMquantifiesmixingintheentiredomain,andisdefinedas Eq.(8): CoM=
˙i (Ct,i− ¯Ct)/ ¯Ct 2 Vi ˙iVi (8)
where95,CoMisachievedwhenCoM<0.0283(Hartmannetal.,
2006). The results for all SM-RANS and LES simulations are giveninTable2.SM simulationspredict aprobe-based 95≈110forRKEbased onthe bottomprobe, inagreement
with the SM simulation of Jahoda et al. (2007), and 15% abovetheexperimentalvalue,bothforthebottomandtop probe.Overall,95predictedwithSM-RKEisconsistentlylower
thanwithMRF-RKE(PartI)atthesamemesh,whichimplies
thateitherturbulentmassexchangebetweenthe compart-ments (t) ishigher inSM simulations,or MIs, which are
inherentlysuppressedinMRFsimulations,resultinahigher
Table2–Comparisonofdimensionlessmixingtimes95
fortheSM-RANSandSM-LESmethods,forthemeshes studiedinthiswork.botandtoprepresentthe
probe-basedresults,CoMthecoefficientofmixing.The workofHartmannisfollowedtosettheCoM-boundary.
Mesh bot/top CoM
2IF-M-SMRKE 112.7/89.5 126.5 2IF-F-SMRKE 110.1/84.9 122.0 2IF-M-SMRSM 133.3/107.9 147.0 2IF-C-LES 80.0/71.1 86.0 2IF-M-LES(t=1.67ms) 80.7/69.6 88.0 2IF-M-LES(t=3.33ms,run1) 81.6/69.6 n.m. 2IF-M-LES,(t=3.33ms,run2) 78.5/71.4 n.m. 2IF-SF-LES 94.0/79.1 99.0 Jahoda(LES) 81.5/n.r. n.m. Jahoda(EXP) 92.0/≈75 n.m.
n.m.=notmeasured,n.r.=notreported.
inter-compartmentmassexchange.Probeprofiles,reportedin
Fig.2AforMRFandSM,hintatthelatter:whereastheMRF
resultsshowaconstantincreaseindimensionless concentra-tionCt/ ¯Ct,wigglesintheprofilesforSMhintatthepresence
ofoscillatorymotionsintheinter-compartmentplane.
TheSM-RKEresultsaresimilarforbothmeshes,forprobe
dynamicsaswellas95.Reasonablesimilarityisobservedin
inter-compartmenthydrodynamicsbetweenthemeshes(see Section4.3),incontrasttoMRF-RKEinPartI.Thisgivessome confidencethatthenear-equal95isasignofmesh
indepen-denceforSM-RKE.TheresultsfromPartI,however,showthat
dataforSM-RKEonmesh2IF-SFisrequiredinordertomake
suchastatementwithconfidence;thecomputational require-mentsdidnotallowustoconductsuchasimulationwithin thisproject.Moregenerally,theMRFresultswerehighly
sen-Fig.2–Mixingprofilesfordifferentsimulations(bottomprobesignal).Black:typicalexperimentaldata,Jahodaetal.(2007). Top:RANSsimulations,including2MRFsimulations(reportedinPaperI).Bottom:LESsimulations.
sitivetodifferencesinQaxbetweenthemeshes,duetothe
flowfieldbeingfrozenduringmixing.Forslidingmesh sim-ulations,thedynamicratherthanfrozeninter-compartment flowfieldisexpectedtoleadtoalowermeshsensitivitythan wasobserved withMRF. OnlyoneSM-RSM simulation was conducted, yielding a higher 95, as well as shifted
inter-compartmentplane(seeSection4.3).Ingeneral,theinclusion ofturbulenceanisotropydoesnotseemtoimprovethe agree-ment withexperimentaldata, in the SM simulations,MRF simulationsofPartI,andsingle-impellerworkofGunyoland Mudde(2009).Hence,thereappearstobelittlereasontoopt forthemorecomputationallyexpensiveRSMmodel Rushton-stirredtankapplications,andwedidnotexploreSM-RSMat finermeshes.
TheLESsimulations with 2IF-C and 2IF-Myield95≈80
based on the bottom probe, in agreement with 95=81.5
observedbyJahodaetal.,usingasimilarmeshdensity.No effectofthetimestepsizeisobservedfor2IF-M.Simulation 2IF-SFyields95=94,inverygoodagreementwiththe
exper-imentalvalue 95=92 reported byJahoda et al. (2007). The
bottom-proberesponseprofilesforLESareshowninFig.2B. Itmustbekeptinmindthatthefiguresshowsingle realiza-tions,exceptforLESrunswithmesh2IF-M.Theresultsforthis caseshowlimitedvariationin95,whichgivesconfidencethe
differencein95betweenthismeshand2IF-SFisnotdueto
regularvariabilityintheLESmethod.Forthetopprobes,95
isinlinewiththeroughestimation madefromthemixing profilesofJahodaetal.Additionalproberesults,studyingthe sensitivityoftheresponsetoprobelocation,arereportedin SupplementarymaterialB.
4.3. Inter-compartmentdynamics:slidingmesh
SMsimulationspredictalower95 thanMRF,althoughthey
stillover-estimatetheexperimentalvalue.ThewigglesinFig.2
hintthedifferencebetweenMRFandSMliesintheinclusionof MIs.Inter-compartmentdata(Fig.3)showskt*ishigherforSM
simulationsthanforMRFduetotheinclusionofkMI,whichis
0inMRFsimulationsbyconstruction.Comparedtothe exper-iments,kt*isstillunder-estimatedeverywhereexceptnearest
to theshaft; hence 95 inSM simulations stillexceeds the
experimentalobservation. For SM−RKE, kt* isquite similar
betweenthetwomeshes,butthecontributionbykMIdiffers:
approx.10–25%formesh2IF-F,and5–60%for2IF-M.Closeto thewall,kMIisindecentagreementwithexperimentaldata
for2IF-M,butduetoanunder-estimationofkt,kt*islower
overall.
UradiswellcapturedbySM-RKE,althoughtheMRF
simu-lationsperformedslightlybetterinquantitativelycapturing
Uax;still,thereversalfromconvergingtodivergingflowis
rea-sonably captured withSM-RKE.SM-RSM performspoorly:a shiftininter-compartmentplanepositionisclearlyobserved in theprofiles for Urad, Uax and kt*. In contrastto theMRF
simulations,reportedinPartI,thisdoesnotreduce95for
SM-RSM.IntheMRFsimulations,ashiftintheinter-compartment plane positionwas associatedwith ahigher axial flowrate
Qax between the compartments.Thefrozen planeposition
meantthishigherQaxledtoconsistentlyfastermixing.In
con-trast,inSMsimulationstheinter-compartmentplaneposition isdynamic,whichimpliesthateveniftheplanepositionis shiftedsomewhat,thisdoesnotresultinaconsistentincrease ofmassexchangebetweenthecompartments.
Probing Uax at several points inthe inter-compartment
region(SM-RKE)revealsstrong oscillationsinthe Reynolds-averagedvelocity,asshowninFig.4AandB.Sincethereare no turbulent oscillations inthe Reynolds-averagedvelocity profilebyconstruction,thereisnoneedforauto-correlation fittingtodeterminekMI.Therefore, theMIfrequencieswere
Fig.3–Axialprofiles(baffleplane)of(A):Urad,(B):Uax,(C):kt,comparingLDAresults(symbols)withSM-CFDdata(lines).In (C(,theblackrectanglesrepresentthetotalkineticenergykt*,thebluediamondstheturbulentkineticenergykt,andthered diamondstheMIenergykMI.Theblacklinesrepresentkt*,theredlineskMI.Lines:CFDresultsatdifferentmeshdensities (dotted:2IF-M,RSM,dashed:2IF-M,RKE,solid:2IF-F,RKE).(Forinterpretationofthereferencestocolorinthisfigurelegend, thereaderisreferredtothewebversionofthisarticle.)
theFourierspectrum.Thespectrumshowsvelocity oscilla-tionsarehighlyperiodicfor2IF-M(Fig.4C),withadominant frequency f/N=0.058 and its harmonics. This is in excel-lent agreement with the jet instability frequency reported byPagliantietal.(2008),andthevalueobserved experimen-tallyinPartIofthiswork.Experimentally,wealsoobserved aweakcontributionoff/N=0.02,andstrongcontributionof
f/N≈0.04inthe parallelflowregion. Averyweak f/N=0.02 canbeobservedwithSM,whilef/N=0.04isabsent.Case
2IF-Fshowsmorescatter(Fig.4D);thedominantpeaknowisat
f/N=0.045withastrongshoulderatf/N=0.06forr/R=0.512 andr/R=0.694.Thisisqualitativelymoreinlinewith exper-imentalobservations,althoughthecontributionoff/N=0.06 istoo low.Atr/R=0.694, anadditional peakis observedat
f/N≈0.01,whichrepresentstheveryslowoscillationvisible inFig.4BthatisabsentinFig.4A.
Thequalitativeexperimentalobservationthataxial oscil-lationsincrease withincreasing radialpositioniscaptured bySMsimulations.ThenextquestionishowmuchMIs con-tribute to mixing. Fig. 5 shows the velocity vectors in an axialcrosssection(baffle-plane)atseveralmomentsthough amacro-oscillation.Duringtheoscillatorymotion,cross-over flowfromthetoptobottomcompartmentconnectsthe down-wardnear-wall flowinthetopcompartment,toflowalong theshafttowardstheimpellerinthebottomcompartment (Fig.5Band D).Tracerconcentrationsnapshots(Fig.6) con-firmthattracertransportbetweenthecompartmentsoccurs
dominantlyalongtheshaft.Nearthewall,theparallel-flow planesegregatingthecompartmentsisdisplacedasawhole withnovisibleinter-compartmenttransport.Videosof inter-compartment mixing are availableonline, for anMRF-RKE,
SM-RKEandLESsimulation.
4.4. Inter-compartmentdynamics:largeEddy simulations
ThevelocitycomponentsresolvedinLESspanamuchwider rangeoffrequencies,wellintotheturbulentdomain.Hence, weanalyzethemwiththesamemethodsastheLDAdatain PartI:fittingofthevelocitysignalauto-correlationwith damp-ened cosines(Eq. (4)),and spectralintegration(Eq. (5)). For spectralintegration,thecut-off frequencybetweenMIsand regularturbulencewassetatf/N=0.1;theenergycontainedin thespectralrangef/N<0.1isfullyattributedtokMI.Thismay
beanover-estimationofthetrueMIenergy,butbecausethe samecut-offwasusedinLDA,thismethodprovidesthemost directcomparison.Duetothenegligibledifferencebetween simulations witht=0.0033andt=0.001667, thereported resultsfor2IF-Mareobtainedwitht=0.0033,forwhichmore datewasavailable.
Similar to the experimental procedure, the auto-correlation signal was fitted with2frequency components (f1.1 and f2.1) and their first harmonics (f1.2 and f2.2).Fig. 7
Fig.4–(A)Axialvelocityaty=0versustime,2IF-MSM-RKE.(B)Axialvelocityaty=0versustime,2IF-FSM-RKE.(C)Fourier transformofA;insetshowsthesamegraphonalogscale.(D)FouriertransformofB;insetshowsthesamegraphonalog scale.Blackline:r/R=0.2.Blueline:r/R=0.512.Lightblueline:r/R=0.694.(Forinterpretationofthereferencestocolorinthis figurelegend,thereaderisreferredtothewebversionofthisarticle.)
Table3–AxialRMS-velocity,oscillationfrequenciesandtheircontributionsfor2IF-M-LESand2IF-SF-LESintheplane
y=0atdifferentradialpositions,usingtheperiodic-fittingapproach.Thecoefficientcrepresentthecontributiontothe Reynoldsstress ¯uu,withci.jthejthharmonicoffrequencycomponenti.Forbrevity,onlyuaxisreported,tablesforurad andutanareinsupplementarymaterialC.Toprows:2IF-M.Bottomrows:2IF-SF.
r/R 0.200 0.358 0.435 0.512 0.566 0.694 0.765 uax,M(m/s) 0.089 0.111 0.126 0.147 0.164 0.181 0.148 c1.1,ax,M 0 0.026 0 0.001 0.022 0 0 c1.2,ax,M 0.033 0.011 0.033 0.08 0.117 0.110 0.049 c2.1,ax,M 0.002 0.005 0 0.007 0 0.015 0.023 c2.2,ax,M 0.003 0.071 0.017 0.141 0.065 0.045 0.039 f1.1,ax,M/N 0.014 0.010 0.012 0.012 0.012 0.012 0.013 f2.1,ax,M/N 0.019 0.022 0.017 0.021 0.018 0.022 0.023 103·k t,fit,M/U2tip 9.35 8.98 9.59 9.36 9.84 9.91 7.00 103·k MI,fit,M/U2tip 1.21 2.32 1.42 1.92 1.89 1.41 0.63 uax,SF (m/s) 0.088 0.106 0.117 0.132 0.148 0.208 0.174 c1.1,ax,SF 0 0 0.042 0.068 0.086 0.135 0.088 c1.2,ax,SF 0.011 0.007 0 0.014 0 0 0.039 c2.1,ax,SF 0 0 0.005 0 0 0.013 0 c2.2,ax,SF 0.022 0.09 0.006 0.030 0.006 0.210 0.108 f1.1,ax,SF/N 0.034 0.026 0.010 0.024 0.020 0.022 0.023 f2.1,ax,SF/N 0.043 0.042 0.042 0.036 0.040 0.036 0.035 103·k t,fit,SF/U2tip 6.48 8.22 8.90 7.37 8.28 10.2 8.06 103·k MI,fit,SF/U2tip 1.80 1.01 0.60 1.89 1.62 4.00 2.10
positionsat axial positiony=0. Thedominant frequencies foraxialoscillationsinLESsimulationsareless sharpthan the experimentalfrequenciesreported inPart I, leading to a poorer fit quality. On average, 2IF-SF gives f1.1/N≈0.023
and f2.1/N≈0.039, albeit with significant variation between
radial locations. For f2, the harmonic f2.2/N≈0.078 has a
highermagnitude thatf2.1/N. Thesefrequencies liearound
the experimental value of f2.1/N≈0.06, which itself is not
observed in 2IF-SF. TheLES data furthermore hints at the presence of higher frequency oscillations, which are not captured in the current auto-correlation fit. For the axial oscillationsatmesh2IF-M,f1.2/N≈0.024hasarelativelyhigh
magnitude,andisnearlyequalinfrequencytof2.1.Harmonic f2.2/N≈0.04alsohasahighmagnitude.Whilethisfrequency
was relatively prominent experimentally, the strongest experimental component f/N≈0.06 is again not observed
Fig.5–Velocityvectorfields(2IM-RKE)duringfourstagesofamacro-oscillation.Thedashedredlinesindicatetheregions whereUax≈0,showingthatthroughouttheoscillationstheaxialseparationbetweenthecompartmentsislocallybroken, resultinginenhancedinter-compartmentmixing.
Fig.6–Snapshotsofmixingat6timepoints,2IF-M-SM,showingtracerexchangealongtheshaftisinfluencedbyMIs(video availableonline).
Fig.7–Fittedauto-correlationfunctionsfor2IF-M-LES(top)and2IF-SF-LES(bottom)aty=0,at3radialpositions.Blueline: rawdata.Redline:fitteddampedcosinefunction(seePartIfordetails).(Forinterpretationofthereferencestocolorinthis figurelegend,thereaderisreferredtothewebversionofthisarticle.)
in the simulation. As with the SM simulations, the radial oscillationshaveahighmagnitudeneartheshaft,dampening withincreasingradialposition,whiletheoppositeisobserved foraxialoscillations(Table3).
Thespectraldensity functions(Fig.8)show anf−1 scal-ingfor0.1<f/N<1,andanf−5/3scalingforf/N>1,inlinewith experimentalobservations.Themoderate spatialresolution ofthemeshesresultsinaquickdeviationfrom−5/3scaling atthehigh-frequencyendofthespectra.Forthe2IF-SF sim-ulations,thefilterlengthratiowasaround/=10−25with theKolmogorovlengthscale,thelowervalueneartheshaft andthehighernearthebaffle.Afinermeshmaybedesired, butforroutineusewithoutsuper-computingfacilities,2IF-SF isalreadymuchtoodemanding.
Inthelowfrequencyrange,thespectraldensityfunctions confirmthe observationsmadein theauto-correlation fits. Thespectrumfor2IF-Mdoesshowthebi-modalpeakobserved
experimentally,butatthelowerfrequenciesf1.2/N≈0.024and
f2.2/N≈0.04,whereasexperimentallyf/N=0.045and0.06were
observed. In2IF-SFthebimodal peakisnotobserved.Here
f1.1/N≈0.026andf2.2/N≈0.078contributemost,but,asinthe
auto-correlationdata,f/N≈0.045isabsent.
Boththeauto-correlationfitandspectraldensitymethod provideanestimateofthepercentageoffluctuatingkinetic energythatiscontainedinMIs.Firstwereportresultsforthe auto-correlationmethod.Usingthismethod,kMI,fit10–20%of
kt*atthemeasuredlocationsintheplaney=0for2IF-M,with
themaximumatr/R=0.358.2IF-SFyielskMI,fitis5–30%ofkt*,
withthemaximumatr/R=0.694.Forcomparison, experimen-taldatayieldedacontributionof8–20%withthemaximum atr/R=0.694whenusingtheauto-correlationmethod. Spec-tral integration gave anMI contribution iskMI,SI is25–30%
ofkt* for2IF-M,and 25–40%ofkt* for2IF-SF,comparedtoa
inte-Fig.8–Spectraldensityfunctionsofuax(darkblue)andurad (lightblue)at3radialpositions,intheplaney=0.(A)2IF-MLES.
(B)2IF-SFLES.(C)Experimentalresults.Dashedline:S(f/N)∝(f/N)−5/3.Dash-dotline:S(f/N)∝(f/N)−1.Dottedline:cut-off frequencybetweenMIandturbulence.(Forinterpretationofthereferencestocolorinthisfigurelegend,thereaderis referredtothewebversionofthisarticle.)
gration.Thedifferencebetweenauto-correlationfittingand spectralintegrationarisingduetotheinclusionoflimited fre-quencies intheformer, and alllow-frequencycomponents inthelatter. Next,wecomparethe profiles ofvelocityand kineticenergyinthe inter-compartmentregion.Becauseof themorestraightforwardcomparisonwithexperimentaldata, thespectralintegrationmethodisusedtodeterminetheMI contributioninthekineticenergydatapresentedinFig.9.
Aside from an offset at the baffle position, which is attributed tothe 2D baffle geometry,Urad is well captured
(Fig.9).ThechangefromconvergingtodivergingflowinUaxis
excellentlyassessedbyallbutthecrudestmesh.Overallgood agreementinktisobserved(despitethepoorerperformance
intheimpellerdischargestream),with2IF-Mover-estimative neartheshaft,whereas2IF-SFissonearthebaffle.Both simu-lationsreasonablycapturekMI,SIneartheshaft,butonlywith
mesh2IF-SFthepeakatr/R=0.694isproperlycaptured.Even thoughthefrequencydistributiondoesnotcompletelyagree withexperiments,theoverallkMI,SIandkt,SIarewellcaptured
by2IF-SF,inlinewiththegoodassessmentof95.Theslight
under-estimationof95bythecrudeandmediummeshmay
bebyvirtueoftheirhigherktneartheshaft:axialtransportof
tracerisstrongestatthislocation,aswasshowninthemixing snapshotsfortheSM-RANSsimulations.
5.
Concluding
remarks
Inthiswork,wereportslidingmesh(SM)andlargeeddy sim-ulations(LES)ofastirredtankwith2Rushtonimpellersat largemutualclearance.Earlierstudiesreported(1) experimen-talevidenceformacro-instabilities(MIs)ina2-Rushtonstirred tank,(2)anover-estimationofmixingtime95withthe
multi-plereferenceframe(MRF)simulationmodel,increasingwith meshdensity,(3)amilderover-estimationof95withSM,and
slightunder-predictionof95withLES.
Compartmentformationaroundtheimpellersleadstoa parallel radial flow in the inter-compartment plane, caus-ingpoormassexchangebetweenthecompartments,thereby formingaratelimiting stepinmixing. Ourhypothesiswas thatRANSsimulationsover-predict95,asReynoldsaveraging
leadstoanear-shearfreeflowintheinter-compartmentplane, whichfailstoproperlycapturethegenerationofturbulence by the colliding flow. This in turn leads to a local under-predicting turbulent viscosity t. Both SM simulations and
MRFsimulations(PartI)indeedunder-predictedktinthe
inter-compartmentregion,comparedtoLDAmeasurements,which supports this hypothesis. Additionally, macro-instabilities (MIs)were observedtocontainaround30% ofthe fluctuat-ingkineticenergykt*intheinter-compartmentplane;these
pre-Fig.9–Axialprofiles(baffleplane)of(A):Urad,(B):Uax,(C):kt,comparingLDAresults(symbols)withLES-CFDdata(lines).In (C),theblackrectanglesrepresentthetotalkineticenergykt*,thebluediamondstheturbulentkineticenergykt,andthered diamondstheMIenergykMI.Theblacklinesrepresentkt*,thebluelineskt,theredlineskMI.Lines:CFDresultsatdifferent meshdensities(dotted:2IF−C,dashed:2IF−M,solid:2IF−SF.(Forinterpretationofthereferencestocolorinthisfigure legend,thereaderisreferredtothewebversionofthisarticle.)
dictthepresenceofsuchMIs,albeitslightlyunder-estimating theirkineticenergy,whereasthe frozen-flowfieldMRF sim-ulations byconstruction donot predict them. Overall, the under-predictedkt causes 95 to beover-predicted by both
SM-RANSandMRF-RANS,buttheinclusionofMIsmeansthe
over-predictionbySM-RANSislesssevere;approximately20% bySM-RKEwithrealizablek−(whichoutperformedSM-RSM),
whileanover-estimationofnear60%wasobservedwithMRF atsamemesh.Inthiswork,wehaveusedstructured hexa-hedralmeshes.Theresultsmaybedifferentatunstructured meshes of similar resolution, but the over-predictions are expectedtoholdoncemeshindependenceisachieved.
TurbulentSchmidtnumber(Sct)tuningwassuggestedto
improveagreementin95onsomeoccasions(Montanteetal.,
2005;GunyolandMudde,2009).Thecurrentresultssuggests suchtuningisnotbasedonphysicalconsiderations,but as apatchwork solutiontorepairaninadequateestimationof
kt (hence t and the effect ofMIs on multi-impeller
mix-ing behavior. Whilesuch a patch bean adequate solution if approximatemixing dynamicssuffice, the limitations of multi-impellerRANSsimulationsmustbekeptinmind,and theRANSmodelsstudiedinthecurrentwork(realizablek− and RSM)seeminadequate forquantitativemixing assess-mentinmulti-impellersystems.
Themixingtimewasunder-estimatedapprox.13% with LESatcrudemeshes(below2milliongridcells),inaccordance
withJahodaetal.(2007).Withameshofapprox.10million cells,95wasinexcellentagreementwithexperiments.The
MIfrequenciesintheinter-compartmentregionaspredicted inLESsimulationsarenotassharplydistinguishableasinthe LDAdata, and differencesinthe dominantfrequenciesare observed.Atthecrude-andmediummesh,kMIwasstill
under-predictedingeneral,buthigherneartheshaft,highervalues wereobservedthanexperimentally.Theregionneartheshaft iswheremostmassexchangebetweenthecompartmentsis predictedtotakeplace,whichmayexplainthelower95with
acrudermesh.Despitethedifferenceinfrequencies,kMIisin
goodagreementbetweenthefine-meshLESandexperiments, asiskt.
TheappliedLESapproachdidover-predictkt and
under-predictintheimpellerdischargestream.Thisisattributed totheemployeddynamicSmagorinskymodel,whichpredicts lowvaluesofconstantCS.WhiletakingCSconstantand
fine-tuningitmay increaseagreement(Delafosse etal.,2014),a moreuniversalapproachisdesired.Assuch,thereisroom forfurtherexplorationofLESandothertransientsimulation approaches ((ID)DES, hybrid LES-RANS,etc.) instirredtank applications. Strictlyspeaking,the finestcurrent LESstudy wasunder-resolved,especiallyintheimpellerregion.Higher resolutionstudiesmaybedesirableforfurtherfundamental studies,butalreadythecurrentfinemeshisconsideredtoo demandingforroutinemixingstudies.Despitethein
predict-ing,LESperformedwellintermsofmixingassessment,even onthecrudestmesheswhichfarfromresolveallenergy car-ryingturbulencescales.AsbothSMand LESmispredict95
toasimilardegreeatthecrudemeshes,andbothhave simi-larcomputationaldemands,botharecurrentlyconsideredto bereasonablyoptionsformixingassessmentprovidedtheir degreeoferroriskeptinaccount;bothareclearlypreferred aboveMRF.However,ifcomputationalfacilitiesallowforfiner meshes,LESisthepreferredchoice.
Acknowledgements
Wewish tothank our colleagues at ECUST, Shanghai and DSM Sinochempharmaceuticals forourongoing collabora-tion.WearegratefultoProf.H.vandenAkker,Dr.S.Kenjeres, Dr. L. Portela and Prof. H.J. Noorman for their advice and insights.CHacknowledgesallstudentsfollowingM.Sc.course CH3421, who contributed to the simulation results aspart oftheircoursereports.EvertWagner,JosThieme,Christiaan Schinkel,StefantenHagen,YoupvanGoozenandRuudvan Tolaregratefullyacknowledgedfortheirtechnicalassistance. Thisworkhasbeenconductedwithinamulti-partyresearch project,betweenDSM-SinochemPharmaceuticals,TUDelft, East ChinaUniversityofScienceand Technologyand Guo-jia,subsidizedbyNWOandMoST(NWO-MoSTJointprogram 2013DFG32630).Allsponsorsaregratefullyacknowledged.
Appendix
A.
Supplementary
data
Supplementary data associated with this arti-cle can be found, in the online version, at
https://doi.org/10.1016/j.cherd.2018.06.007.
References
Bakker,R.,1996.Laser-DopplerMeasurementsinaBaffledStirred TankReactorWithaDiscTurbine(Ph.D.thesis).Delft UniversityofTechnology.
Chara,Z.,Kysela,B.,Konfrst,J.,Fort,I.,2016.Studyoffluidflowin baffledvesselsstirredbyaRushtonstandardimpeller.Appl. Math.Comput.272,614–628,
http://dx.doi.org/10.1016/j.amc.2015.06.044.
Coroneo,M.,Montante,G.,Paglianti,A.,Magelli,F.,2011.CFD predictionoffluidflowandmixinginstirredtanks:numerical issuesabouttheRANSsimulations.Comput.Chem.Eng.35 (10),1959–1968,
http://dx.doi.org/10.1016/j.compchemeng.2010.12.007. Delafosse,A.,Liné,A.,Morchain,J.,Guiraud,P.,2008.LESand
URANSsimulationsofhydrodynamicsinmixingtank: comparisontoPIVexperiments.Chem.Eng.Res.Des.86(12), 1322–1330,http://dx.doi.org/10.1016/j.cherd.2008.07.008. Delafosse,A.,Morchain,J.,Guiraud,P.,Liné,A.,2009.Trailing
vorticesgeneratedbyaRushtonturbine:assessmentof URANSandlargeEddysimulations.Chem.Eng.Res.Des.87 (4),401–411,http://dx.doi.org/10.1016/j.cherd.2008.12.018. Delafosse,A.,Collignon,M.-L.,Calvo,S.,Delvigne,F.,Crine,M.,
Thonart,P.,Toye,D.,2014.CFD-basedcompartmentmodelfor descriptionofmixinginbioreactors.Chem.Eng.Sci.106, 76–85http://www.sciencedirect.com/
science/article/pii/S0009250913007690.
Derksen,J.,VandenAkker,H.E.A.,1999.Largeeddysimulations ontheflowdrivenbyaRushtonturbine.AIChEJ.45(2), 209–221,http://dx.doi.org/10.1002/aic.690450202. Derksen,J.,Doelman,M.,VandenAkker,H.,1999.
Three-dimensionalLDAmeasurementsintheimpellerregion ofaturbulentlystirredtank.Exp.Fluids27(6),522–532,
http://dx.doi.org/10.1007/s003480050376.
Ducci,A.,Yianneskis,M.,2005.Directdeterminationofenergy dissipationinstirredvesselswithtwo-pointLDA.AIChEJ.51 (8),2133–2149,http://dx.doi.org/10.1002/aic.10468.
Eggels,J.G.M.,1996.Directandlarge-eddysimulationofturbulent fluidflowusingthelattice-Boltzmannscheme.Int.J.Heat FluidFlow17(3),307–323,
http://dx.doi.org/10.1016/0142-727X(96)00044-6.
Escudié,R.,Bouyer,D.,Liné,A.,2004.Characterizationoftrailing vorticesgeneratedbyaRushtonturbine.AIChEJ.50(1),75–86,
http://dx.doi.org/10.1002/aic.10007.
Germano,M.,Piomelli,U.,Moin,P.,Cabot,W.H.,1990.Adynamic subgrid-scaleeddyviscositymodel.In:StudyingTurbulence UsingNumericalSimulationDatabases.3:Proceedingsofthe 1990SummerProgram,StanfordUniv,pp.5–17(SEEN92-30648 21-34).http://adsabs.harvard.edu/abs/1990stun.proc..5G. Gillissen,J.J.J.,VandenAkker,H.E.A.,2012.Directnumerical
simulationoftheturbulentflowinabaffledtankdrivenbya Rushtonturbine.AIChEJ.58(12),3878–3890,
http://dx.doi.org/10.1002/aic.13762.
Gimbun,J.,Rielly,C.D.,Nagy,Z.K.,Derksen,J.J.,2012.Detached eddysimulationontheturbulentflowinastirredtank.AIChE J.58(10),3224–3241,http://dx.doi.org/10.1002/aic.12807. Gunyol,O.,Mudde,R.F.,2009.Computationalstudyof
hydrodynamicsofastandardstirredtankreactoranda large-scalemulti-impellerfermenter.Int.J.Multisc.Comput. Eng.7(6),559–576.
Hartmann,H.,Derksen,J.,Montavon,C.,Pearson,J.,Hamill,I., vandenAkker,H.,2004.AssessmentoflargeeddyandRANS stirredtanksimulationsbymeansofLDA.Chem.Eng.Sci.59 (12),2419–2432,http://dx.doi.org/10.1016/j.ces.2004.01.065. Hartmann,H.,Derksen,J.J.,vandenAkker,H.E.A.,2006.Mixing
timesinaturbulentstirredtankbymeansofLES.AIChEJ.52 (11),3696–3706,http://dx.doi.org/10.1002/aic.10997.
Jahoda,M.,Moˇst ˇek,M.,Kukuková,A.,Machon,V.,2007.CFD modellingofliquidhomogenizationinstirredtankswithone andtwoimpellersusinglargeEddysimulation.Chem.Eng. Res.Des.85(5),616–625,http://dx.doi.org/10.1205/cherd06183. Kim,S.-E.,etal.,2004.Largeeddysimulationusingunstructured
meshesanddynamicsubgrid-scaleturbulencemodels.AIAA Paper,34thFluidDynamicsConferenceandExhibit2548.
Lane,G.,2015.Predictingtheenergydissipationrateina mechanicallystirredtank.In:EleventhInternational ConferenceonCFDintheMineralsandProcessIndustries. CSIROAustralia,Melbourne,Australia
http://www.cfd.com.au/cfdconf15/PDFs/129LAN.pdf. Lilly,D.K.,1992.AproposedmodificationoftheGermano
subgrid?Scaleclosuremethod.Phys.FluidsA:FluidDyn.4(3), 633–635,http://dx.doi.org/10.1063/1.858280.
Micheletti,M.,Baldi,S.,Yeoh,S.,Ducci,A.,Papadakis,G.,Lee,K., Yianneskis,M.,2004.Onspatialandtemporalvariationsand estimatesofenergydissipationinstirredreactors.Chem.Eng. Res.Des.82(9),1188–1198,
http://dx.doi.org/10.1205/cerd.82.9.1188.44172. Montante,G.,Moˇst ˇek,M.,Jahoda,M.,Magelli,F.,2005.CFD
simulationsandexperimentalvalidationofhomogenisation curvesandmixingtimeinstirredNewtonianand
pseudoplasticliquids.Chem.Eng.Sci.60(8-9),2427–2437,
http://dx.doi.org/10.1016/j.ces.2004.11.020.
Murthy,B.,Joshi,J.,2008.Assessmentofstandard,RSMandLES turbulencemodelsinabaffledstirredvesselagitatedby variousimpellerdesigns.Chem.Eng.Sci.63(22),5468–5495,
http://dx.doi.org/10.1016/j.ces.2008.06.019.
Paglianti,A.,Liu,Z.,Montante,G.,Magelli,F.,2008.Effectof macroinstabilitiesinsingle-andmultiple-impellerstirred tanks.Ind.Eng.Chem.Res.47(14),4944–4952,
http://dx.doi.org/10.1021/ie800253u.
Ranade,V.V.,Joshi,J.B.,1990.Flowgeneratedbyadiscturbine.II. Mathematicalmodellingandcomparisonwithexperimental data.Chem.Eng.Res.Des.68(1),34–50
http://cat.inist.fr/?aModele=afficheN&cpsidt=6817059. Revstedt,J.,Fuchs,L.,Trägårdh,C.,1998.LargeEddysimulations
(24),4041–4053,
http://dx.doi.org/10.1016/S0009-2509(98)00203-6. Singh,H.,Fletcher,D.F.,Nijdam,J.J.,2011.Anassessmentof
differentturbulencemodelsforpredictingflowinabaffled tankstirredwithaRushtonturbine.Chem.Eng.Sci.66(23), 5976–5988,http://dx.doi.org/10.1016/j.ces.2011.08.018. Soos,M.,Kaufmann,R.,Winteler,R.,Kroupa,M.,Lüthi,B.,2013.
Determinationofmaximumturbulentenergydissipationrate generatedbyaRushtonimpellerthroughlargeeddy
simulation.AIChEJ.59(10),3642–3658,
http://dx.doi.org/10.1002/aic.14206.
Wu,H.,Patterson,G.,1989.Laser-Dopplermeasurementsof turbulent-flowparametersinastirredmixer.Chem.Eng.Sci. 44(10),2207–2221,
http://dx.doi.org/10.1016/0009-2509(89)85155-3.
Yeoh,S.,Papadakis,G.,Yianneskis,M.,2004.Numerical
simulationofturbulentflowcharacteristicsinastirredvessel usingtheLESandRANSapproacheswiththe
sliding/deformingmeshmethodology.Chem.Eng.Res.Des. 82(7),834–848,http://dx.doi.org/10.1205/0263876041596751. Yeoh,S.,Papadakis,G.,Yianneskis,M.,2005.Determinationof
mixingtimeanddegreeofhomogeneityinstirredvessels withlargeeddysimulation.Chem.Eng.Sci.60(8-9), 2293–2302,http://dx.doi.org/10.1016/j.ces.2004.10.048. Zhang,Y.,Yang,C.,Mao,Z.,2006.LargeEddysimulationofliquid
flowinastirredtankwithimprovedinner-outeriterative algorithm.Chin.J.Chem.Eng.14(3),321–329,