• Aucun résultat trouvé

On POD analysis of PIV measurements applied to mixing in a stirred vessel with a shear thinning fluid

N/A
N/A
Protected

Academic year: 2021

Partager "On POD analysis of PIV measurements applied to mixing in a stirred vessel with a shear thinning fluid"

Copied!
12
0
0

Texte intégral

(1)

To link to this article : DOI: 10.1016/j.cherd.2013.05.002

http://doi.wiley.com/10.1016/j.cherd.2013.05.002

To cite this version: Liné, Alain and Gabelle, Jean-Christophe and

Morchain, Jérôme and Anne-Archard, Dominique and Augier, Frédéric On

POD analysis of PIV measurements applied to mixing in a stirred vessel with

a shear thinning fluid. (2013) Chemical Engineering Research and Design,

vol. 91 (n° 11). pp. 2073-2083. ISSN 0263-8762

This is an author-deposited version published in :

http://oatao.univ-toulouse.fr/

Eprints ID : 10902

Any correspondance concerning this service should be sent to the repository

(2)

On

POD

analysis

of

PIV

measurements

applied

to

mixing

in

a

stirred

vessel

with

a

shear

thinning

fluid

A.

Liné

a,b,c,∗

,

J.-C.

Gabelle

a,b,c,d,e

,

J.

Morchain

a,b,c

,

D.

Anne-Archard

f

,

F.

Augier

e

aUniversitédeToulouse,INSA,LISBP,135AvenuedeRangueil,F-31077Toulouse,France bINRAUMR792IngénieriedesSystèmesBiologiquesetdesProcédés,Toulouse,France cCNRS,UMR5504,F-31400Toulouse,France

dADEMEFrenchEnvironmentandEnergyManagementAgency,20avenueduGrésillé,BP90406,49004AngersCedex01,France eIFPEnergiesnouvelles,Rond-pointdel’échangeurdeSolaize,BP3,69360Solaize,France

fUniversitédeToulouse,INPT,UPS,IMFT,AlléeC.Soula,F-31400Toulouse,France

a

b

s

t

r

a

c

t

P.O.D.techniqueisappliedto2DP.I.V.datainthefieldofhydrodynamicsinamixingtankwithaRushtonturbine andashearthinningfluid.Classicaleigen-valuespectrumispresentedandphaseportraitofP.O.D.coefficientsare plottedandanalyzedintermsoftrailingvortices.Aspectrumofdissipationrateofkineticenergyisintroducedand discussed.LengthscalesassociatedtoeachP.O.D.modesareproposed.

Keywords: Mixing;Particleimagevelocimetry;Properorthogonaldecomposition;Eigenvaluespectrum;Dissipation rate;Stirredtank

1.

Introduction

Advanced experimentaltechniques such as particle image velocimetry(P.I.V.)areavailablenowadays,andtheygeneratea hugeamountofdata,intermsofinstantaneousvelocityfields (inaplaneasfaras2DP.I.V.isconcerned).Indeed,advanced dataprocessingtoolsarerequiredtoextractmoreandmore informationfrom this sourceof experimentaldata. Proper orthogonaldecomposition(P.O.D.)orKarhunen–Loevemethod thatisknowntobeefficienttoisolatecoherentstructuresfrom aseriesofinstantaneousvelocityfields(Berkoozetal.,1993) willbeusedinthispaper.P.O.D.isamodaldecompositionof instantaneousvelocityfields,themodesbeingorthogonalto eachothers.Itcanbeexpressedas:

−→ Vk(x,y,z,t)= N

X

I=1 −→ V(I)k(x,y,z,t)= N

X

I=1 a(I)k(t)−→˚(I)(x,y,z) (1)

where −V→k is the kth instantaneous event of velocity field

measurement and −→

V(I)k is the Ith component of the P.O.D.

Correspondingauthorat:UniversitédeToulouse,INSA,LISBP,135AvenuedeRangueil,F-31077Toulouse,France.Tel.:+33561559786; fax:+33561559760.

E-mailaddress:alain.line@insa-toulouse.fr(A.Liné).

decomposition.ForeachmodeI,theP.O.D.methodgenerates temporal coefficientsa(I)k andspatialmodes−→˚(I).a(I)

k (t) isan

instantaneous(k)ortemporal(t)scalar(calledP.O.D.coefficient ofP.O.D.modeI);itisindependentofspace.−→˚(I)(x,y,z)isa spa-tialeigenfunctionofmodeIanditisindependentoftime(or instantaneouseventk).Modescanbederivedfromthe Fred-holmeigenvalueintegralequation,adaptedbySirovich(1987)

as:

Z

Z

˝

R(x,y,z,x′,y,z)−→˚(I)(x,y,z)dxdydz=(I)−→˚(I)(x,y,z) (2)

whereRisthecross-correlationtensorand(I)isthe

eigenval-ues.Followingthisapproach,theeigenvaluesareexpressed inm5/s2inafull3Danalysis.Indeed,onecanpointoutthat

thedimensioncorrespondstoakineticenergy(m2/s2)times

a volume(dxdydz). In2DPIV problem,the eigenvaluesare expressedinm4/s2,correspondingtoakineticenergy(m2/s2)

times a surface (dxdz). Usually eigenvalue sequences are rankedindecreasingorder.Clearly,thisorderingof eigenval-uesensuresthatthefirstP.O.D.modesarethemostenergetic.

(3)

Given its eigenvalue, (1) and its eigenfunction −→˚(I)(x,y,z),

it ispossibleto reconstructthe velocity field associatedto themode1(

−−→

V(1)k (x,y,z,t))whichwillbeconfirmedto repre-sentthemeanflow.Consideringmodes2and3,Oudheusden et al. (2005) have shown that, when their P.O.D. eigenval-ues are close to each other, the POD modes are coupled: usually,theycorrespondtoorthogonalcomponentsofa peri-odicprocess (coherentstructures induced bythe bladesof the impeller). Higher modes correspond to lower eigenval-ues that are associated tosmaller scale structures. Knight and Sirovich (1990) have shown that the eigenvalue spec-trum(plotofeigenvaluesversusmodenumberI)exhibitI−11/9

trendintheinertialsubrangeofturbulence.Thistrendwas confirmed by De Angelis et al. (2003) and Hoisiadas et al. (2005)(among others).Theanalysis ofthe differentmodes willbeaddressedinthis paper, inthe field ofmixingin a stirredtankwithaRushtonturbineandashearthinningfluid. Inaddition,aspectrumofdissipationrateofkineticenergy associatedtoeachmodeI,similartothespectrumof eigen-values,willbepresentedanddiscussed.Characteristiclength scales (wave length) associatedto each mode will be also proposed.

As highlighted by many authors, P.O.D. eigenfunctions presentflow-field-likestructuresandtheiranalysismustbe drivenwithcare.However,thesignificanceofP.O.D.modescan becomeclearerbythedetailedanalysisofP.O.D.coefficients a(I)k.SomerelationbetweentwomodesIandJcanbetracked downbyplottingthetwocoefficientsseries[ak(I),a(J)k],k=1,N inthe(a(I),a(J))plane.Suchaplotiscalleda2Dprojectionof

thephaseportrait.Hydrodynamiccoherentstructurescanbe revealedwhen characteristicpatternsappearinthis plane, suchascircle,ellipseorLissajousfigures thatcanbe asso-ciated to periodic or intermittent structures. In particular, circularshapescansuggestcyclicvariationsofmodesIand

Jrelatedtovorticesorcoherentstructures.Inthiscase,(a(I),

a(J))seriescanbemodeledassine–cosinefunctionswithsame

period.Suchphaseportraitswillbeplottedinthispaperin order to emphasizethe couplingbetween modes 2 and 3, associatedtotrailingvorticesgeneratedbythebladesofthe impeller.

P.O.D. was appliedto hydrodynamicsin mixing tankby

MoreauandLiné(2006),Doulgerakis(2010),andGabelleetal. (2013). SnapshotP.O.D. studies werereviewedbyTabiband Joshi (2008) in the field ofChemical Engineering. Thegoal of this paper is to contribute to extract information from 2D P.I.V. data by P.O.D. technique in the field of mixing. Classical P.O.D. data processing is performed in terms of eigenvalueproblem,spectrumofeigenvaluesandphase por-trait analysis. New data processing is proposed in terms of dissipation rates and wavelengths associated to each modeI.

2.

Materials

and

methods

2.1. Mixingtank

2D-P.I.V.measurementswereperformedina70Lstirredtank equippedwithfourequallyspacedbaffles.ARushtonturbine (D/T=0.33) was used. Thevessel was standard (T=0.45m); theimpeller clearancewasC=T/3.Moreinformationabout the mixing tank dimensions is provided in Gabelle et al. (2013).

Fig.1–ApparentviscosityofZetag7587solutionsversus shearrate.

2.2. P.I.V.system

TheP.I.V.systemusedinthisstudyconsistedofalaserand animageacquisitionsystemprovidedbyLaVision(LaVision GmbH, Goettingen, Germany).Thesystem includeda laser Nd-Yag(Quantel,10Hz,200×2mJ),asynchronizationsystem andacharge-coupled-device(CCD)camera(ImagerIntense,12 bits,1376×1040pixels).FastFouriertransform(FFT)cross cor-relationwasusedtointerrogatethetwoimages,whichwere dividedintointerrogationarea(32×32or16×16pixels).For allexperiments,500–700imagepairswererecordedand sta-tisticalconvergenceofthevelocitywaschecked.Inorderto studytheinfluenceofspatialresolutiononthevicinityofthe impeller,twospatialresolutions1x(0.7–1mm)were tested. Thedelaybetweeneachframeinanimagepairwaschosen inrelationtotheimpellerrotationalspeed.Fourzonesinthe vesselwereinvestigated:oneintheimpellersteam,twoabove andonebelow.PODwasonlycarriedoutfortheimageatthe impellerlevel.

2.3. Workingfluid

The shear thinning fluid Zetag7587 (BASF, Ludvigshafen, Germany)wasused.Itoffersgoodtransparencyevenathigh concentration (0.4%). Rheological measurements were car-ried out using aHaake Mars III rheometer (ThermoHaake, Germany).FlowcurvesofZetag7587solutionsarepresented inFig.1forshearratesrangingfrom0.001to1000s−1.

Rhe-ologicaldatacanbefittedtothemodelofOstwalddeWaele (powerlaw)accordingto:

a=K

˙

n−1

(3)

withKthefluidconsistencyfactorandntheflowindex.The rheologicalparametersarereportedinTable1.Zetag7587is knowntobevisco-elastic(EscudierandSmith,2001).However, thechoiceofZetag7587atahighconcentration0.4%wasmade becauseathighconcentrationtheviscosityishigh;itisthus easiertomeasureturbulentflowcharacteristicsclosetothe impeller,theturbulentmicroscalesbeinglargeratrelatively

(4)

Table1–Rheologicalparametersfittedtothepowerlaw equation.

Fluid n K(Pasn) Shearrate

range(s−1)

Zetag75870.1% 0.422 0.30 1–350 Zetag75870.2% 0.376 0.74 1–600 Zetag75870.4% 0.349 1.65 1–1000

lowReynoldsnumberandthespatialresolutionofPIV mea-surementneededtocatchthephysicsbeinglarger.

2.4. Experimentalconditions

TheimpellerspeedNis205rpm.Theglobalshearrate˙ MOis

estimatedfromMetzner–Ottocorrelation:

˙ MO=kSN (4)

where the constant kS depends on the impeller type (kS=11.5foraRushtonturbine).Theshearrateisthusequal

to 40s−1. Given the rheological law of the shear thinning

fluid,theapparentviscosityis0.15Pas.TheReynolds num-beristhusequalto530,correspondingtothetransitionregion betweenlaminarandturbulentregime.Thevolumeaveraged valueofthedissipationrateofkineticenergyhεicanbe esti-mated;itisequaltohεi=0.2m2/s3 orW/kg.Theassociated

valueofglobalshearrateis h i =37s−1.Indeed,theflowis

turbulentclosetotheimpellerandlessandlessturbulentfar fromit.TheP.I.V.datawereacquiredinadomainclosetothe impellertip(6cm×6cm),ensuringturbulentflowconditions asitwillbeshownintheresultspresentationanddiscussion.

3.

Data

processing

Properorthogonaldecomposition(P.O.D.)isalinearprocedure, whichdecomposesasetofinstantaneousvelocityfieldsinto amodalbase(seeSection 3.1). Weproposeinthis work to investigateP.O.D.intermsofreconstructionofinstantaneous velocity fields in a mixing tank. This methodology allows separatingthe organized and the turbulent motions with-outthenecessitytocollectangularphaseddata(Moreauand Liné,2006).PODiscarriedoutusingsnapshotmethodwith all the sets of instantaneous velocityfields directly issued from the measurements. The result is then an orthonor-malbasisofeigenfunctionsandassociatedeigenvalues.The instantaneous velocityfield can be projectedon each POD eigenfunction. From these projections it is possible to get P.O.D.coefficientsandtoreconstructthevelocityfield(see Sec-tion3.2).Dissipationrateofkineticenergycanbeestimated fromthe2Dvelocityfield(Section3.3).Finally,characteristic lengthscales mayalsobedeterminedfrom thePIV experi-ments.

3.1. Snapshotmethod

P.I.V.measurementsareperformedinthe(x,z)plane,ona regu-larmeshwithLrowsandCcolumns,leadingtoNP=LCpoints per plane. Each instantaneous velocity field measurement constitutesasnapshotoftheflow.Thestatisticalanalysisis performedonNsnapshotstakeninthesameplane(P.I.V.data).

ThenumberofgridpointsbeingLC,oneobtainsthematrixof instantaneousvelocityvectordataas:

−−→ V(1)k x=

− → Vk(x1,z1) −V→k(x1,z2) −→Vk(x1,zC) − → Vk(x2,z1) −V→k(x2,z2) −→Vk(x2,zC) −→ Vk(xL,z1) −V→k(xL,z2) −V→k(xL,zC)

(5)

wherekistheindexoftheinstantaneousevent(k=1,N).This matrixofvectorscanbereshapedtobuildavector−→Vkwith2LC

rowsasfollows: −→ Vk=

uk(x1,z1) uk(x2,z1) uk(xL,zC) wk(x1,z1) wk(x2,z1) wk(xL,zC)

(6)

Thesnapshotmethodadoptedinthiseigenvalueproblem was proposed by Sirovich (1987). This methodis based on thesnapshotmatrixMcorrespondingtotheNinstantaneous velocityfields.ThematrixMcanbeexpressedas

M=

u1(x1,z1) u2(x1,z1) ... uN(x1,z1) u1(x2,z1) u2(x2,z1) ... uN(x2,z1) ... u1(xL,zC) u2(xL,zC) ... uN(xL,zC) w1(x1,z1) w2(x1,z1) ... wN(x1,z1) w1(x2,z1) w2(x2,z1) ... wN(x2,z1) ... w1(xL,zC) w2(xL,zC) ... wN(xL,zC)

(7)

ThematrixMhas2LCrowsandNcolumns,eachcolumnof thematrixMrepresentingthektheventoftheinstantaneous velocityfield.Theauto-covariancetensorRcanbederivedas:

R= 1 NM.M T=

u2(x1,z1) ... w2(xL,zC)

(8)

TheFredholmintegraleigenvalueproblemisrepresented inthe2Ddomain˝ofmeasurementby:

Z

Z

˝

R(x,z,x′,z)−→˚(I)(x,z) dxdz=(I)−→˚(I)(x,z) (9)

where(I)istheItheigenvalueand−→˚(I)x(x,z)istheIth

asso-ciatedeigenfunction.Thecorrelationtensorcanbeexpressed in the orthonormalbasisas adiagonal matrix.The eigen-valuesofR,(I)R, areexpressedinm2/s2 andare definedas:

(I)R =(I)/(dxdz).

The dimension of M.MT matrix is (2LC)2 whereas the

(5)

are−→˚(I)(x,z);theycanbesimplyrelatedtotheeigenvaluesof

MT.M,noted−−→˚′(I)(x,z),as:

−−→

˚′(I)(x,z)=MT−→˚(I)(x,z) or −→˚(I)(x,z)=M−−→˚′(I)(x,z) (10)

Thismustbekeptinmindinordertosavecomputingtime sinceM.MTandMT.Mhaveidenticaleigenvalues,butmayhave

significantlydifferentsizesifN2(2LC)2.Inaddition,the

spa-tialmodesconstituteanorthonormalbasis,thus: −→

˚(I).−→˚(J)=ıIJ (11)

3.2. Modaldecompositionofthevelocityfield

Eachinstantaneousvelocityfieldorsnapshotoftheflowcan beexpandedinaseriesofP.O.D.modes.TheP.O.D.coefficients a(I)k areobtainedbyprojectingeachinstantaneousvelocityfield ontheItheigenfunction−→˚(I):

a(I)k =−V→k.

−→

˚(I) (12)

TheIthvelocityvector −→

V(I)k correspondingtotheIth decompo-sitionisgivenby:

−→

Vk(I)=a(I)k−→˚(I) (13)

Onecanthusreconstructeachinstantaneousvelocityfield givenbyEq.(1).Onecaneasilyderive:

1 N N

X

k=1 a(I)ka(J)k =(I)RıIJ (14)

Thisrelationshowsthateachmodemakesanindependent contributiontothetotalkineticenergy.

3.3. Modaldecompositionofthedissipationrateof kineticenergy

Therateofviscousdissipationofkineticenergyisdefinedas:

ε=2aS:S (15)

whereaistheapparentviscosityofthefluidandSisthestrain

rate(orstretching)tensor,definedas:

(I)∝−11/3 (16)

Consideringthedecompositionofthevelocityfieldinthe sumofIthcomponents,onecanassociatetoeachIth com-ponentofthevelocityV(I)i,kacomponentS(I)ij,kofthestrainrate tensor: S(I)ij,k= a (I) k 2 · ∂˚(I)i ∂xj + ∂˚(I)j ∂xi

!

(17)

AccountingforEqs.(11)and(14),onecanthusderivetheIth componentoftherateofviscousdissipationofkineticenergy:

ε(I)=a(˙ ) (I)R 2 3

X

i=1 3

X

j=1 ∂˚(I)i ∂xj + ∂˚(I)j ∂xi

!

2 (18)

Itisinterestingtorealizethatthetotaldissipationrateof kineticenergyisthesumoftheIthcomponentoftherateof viscousdissipationofkineticenergy:ε(x,z,t)=

P

NI=1ε(I)(x,z,t).

RecallthattheeigenvalueIisindependentofspace.It repre-sentsthecontributionofeachmodetothetotalkineticenergy intheplaneofmeasurement.Onecanalsoestimatethe dis-sipationrateofkineticenergy

ε(I)

averagedintheplane:

ε(I)

= 1 LC L

X

l=1 C

X

k=1 ε(I)(xl,zk) (19)

Itisthuspossibletodrawaplotsimilartothespectrumof eigenvalues; itisthespectrumofdissipationrateofkinetic energy (plot ofdissipation rate

ε(I)

associatedto mode I

versusmodenumberI).Thisspectrumwillbepresentedand discussed.

Inaddition,thestrainrateofsmallscalesofturbulenceis largeandcanbeexpressedass′

ij≈u/.RecallingEq.(17)and

consideringthattheP.O.D.coefficienta(I)k havethedimension ofavelocity(m/s),theinverseofalengthscalecouldbe asso-ciatedtothegradientsoftheeigen-function−→˚(I).Thiswillbe

analyzedindepthinthediscussion.

3.4. Reshapedeigenfunctions

Theeigenfunctionsarevectorswith2LCrows.Thesevectors canbereshapedtoforma(L,C)matrixoflocalvectors:

−→ ˚(I)k =

˚x,k(x1,z1) ˚x,k(x2,z1) ˚x,k(xL,zC) ˚z,k(x1,z1) ˚z,k(x2,z1) ˚z,k(xL,zC)

−→ ˚(I)k =

−→ ˚k(x1,z1) −˚→k(x1,z2) −˚→k(x1,zC) −→ ˚k(x2,z1) −˚→k(x2,z2) −˚→k(x2,zC) −→ ˚k(xL,z1) −˚→k(xL,z2) −˚→k(xL,zC)

(20)

Thereshaped eigenfunctioncan beplottedintheplane (x,z).Itcorrespondstoa2Dvectorfield,eachvector˚−→k(xi,zj)

havingahorizontal˚x,k(xi,zj)andavertical˚z,k(xi,zj)

compo-nent.

3.5. Commentson2DP.I.V.measurements

Theflowfieldis3Dinthemixingtank.However,theanalysis performedwith2D-P.I.V.islimitedtoa2Dplaneof measure-ment.Intermsofmeanflow,itisclearthatthetangentialflow isnotmeasurednortakenintoaccountinthepresent analy-sis.Indeed,thegoalofthestudyis2-folds:onetheonehand, itistoextractfromthemeasurementstheorganizedmotion induced bytheimpellerrotation. Itwas showninprevious studies(MoreauandLiné,2006,amongothers)thatPODcan enabletodecomposetheorganizedmotionfromthe instanta-neousmeasurement,withoutcollectingangularphaseddata. Itwasalsoshownbymanyauthorsthat2Dmeasurementin averticalplaneenabletoextracttrailingvorticesinducedby aRushtonturbine.Onetheotherhand,thegoalofthisstudy istoanalyzeturbulentscales.Intermsofturbulence,itwas alsoshownbymanyauthorsthatthemeasurementof instan-taneousvelocityfieldinasingleplaneofmeasurementwas

(6)

sufficienttoestimatethelocaldissipationrateofturbulent kineticenergyasfarasthespatialresolutionofthePIVsystem ishighenough.Thisisrelatedtotheisotropyofsmallestscales ofturbulencecoupledtotheefficiencyofthemixingtank pro-cess.Keepingthisinmind,wemustbeawareofthefactthat thetangentialmeanflowisnotmeasuredandthatassociated largescalesarenotmeasuredtoo.IntermsofPODanalysis, the2DlimitationofPIVmeasurementsmayaffectthelargest scalesofmotion(intermediatemodesofPODdecomposition).

4.

Results

Theeigenvaluescorrespondtothecontributionofeachmode tothetotalkineticenergyinthe2Ddomainofmeasurement (Section4.1).Mode1isstructurallyclosetothe mean (sta-tisticalaverage)velocityfield(Section4.2).Whentwoofthe followingmodesare closetoeach other intermsof eigen-values,theenergybroughtbythesemodesisthesame.Such modesmayreveallargescalecoherentstructures(Sections4.4 and4.5).Highermodesmayberelatedtoturbulence(Sections

4.5and4.6).

4.1. Eigen-valuespectrum

Theeigen-valuespectrumisplottedinFig.2.Theeigen-values arenormalizedbythesumoftheeigen-values;these normal-izedeigen-valuescorrespondtothecontributionofeachmode

Fig.2–Eigenvaluespectrum.

tothetotalkineticenergyintheplaneofmeasurement.One cansee thatthe firstmodeexplainsmorethan 80%ofthe totalkineticenergy;itischaracteristicofmeanflow contribu-tion.Onecanobservethatmodes2and3andmodes4and5 areclosetoeachother.Theyrespectivelycontributeto3–5% and0.6–0.8%ofthetotalkineticenergy.Theuppermodes con-tributetolowerlevelsofkineticenergy.Onecannoticethat

Fig.3–FirstPODmode(a)eigenfunction,(b)pdfofnormalizedPODcoefficient,and(c)comparisonoffirstmodetomean flow.

(7)

uppermodesfollowalineartrendinlog–logplot(−11/9);this trendwillbeexplainedlaterinthediscussion.

4.2. Firstmodeanalysis

Thereshapedeigen-function−˚−→(1)(x,z)correspondingtomode

1isplottedinFig. 3a. Itlookslike avelocityfield but it is not, sincethe eigen-functions have no dimension. Indeed, theinstantaneousvelocityfieldassociatedtothefirstmode isgivenby

−−→

Vk(1)(x,z,t)=ak(1)(t)−˚−→(1)(x,z) (21)

foreachkrealizationofthevelocity.OnecanplotinFig.3b theprobability densityfunctionofthecoefficientsa(1)k . The coefficientsa(1)k werenormalizedbythesquarerootofthefirst eigen-value(1)R expressedinm2/s2.Thefirstremarkisthat

thepdfiscenteredontheunityvalue;thesecondremarkis thedispersionoftheNrealizationsa(1)k isweak,meaningthat a(1)k isalmostconstant.Onecanthusreconstructthevelocity fieldassociatedtothefirstmodeas:

−−→ Vk(1)(x,z,t)=

q

(1)R −˚−→(1)(x,z) ∼=cst ∀k (22)

Onecan compare this velocityfield (reconstructed with thefirstmode)tothestatisticalaverageoftheinstantaneous velocityfield,classicallydefinedas:

− → ¯V (x,z,t)= 1 N N

X

k=1 −→ Vkx(x,z,t) (23)

Thevertical profileof the horizontal component ofthe velocityfieldreconstructedwiththefirstmode

−−→

Vk(1)(x,z,t)is comparedtotheverticalprofileofthehorizontalcomponent themeanflow−→¯V (x,z,t)inFig.3c,ataradialpositioncloseto theimpellertip.Thetwoprofilesareidentical.Onecanthus confirmthatthevelocityfieldassociatedtothefirstmodeisa steady-statefieldcorrespondingtothemeanflow.

4.3. Organizedmotionassociatedtomodes2and3 Asaforementioned,theeigen-values(2)R and(3)R have simi-larordersofmagnitude,suggestingthattheymaycorrespond toorganizedmotions.Thisresultisexpectedinmixingwith Rushton turbine and may correspond to trailing vortices induced bytheimpeller blades. Theeigen-functions corre-sponding to mode 2, −˚−→(2)(x,z), and 3,˚−→(3)(x,z), are plotted

inFig.4aand b. Thesetwomodes exhibitvortices, almost symmetricalwithrespecttotheplaneoftheimpeller(z=0), suggestingsomeorganizationinthestructureofthesemodes. Onceagain,theseplotsare notvelocityfields.Thevelocity fieldassociatedtothesumofthesemodescanbewritten:

−−−→ Vk(2,3)(x,z,t)= 3

X

I=2 −→ V(I)k(x,z,t)=a(2)k (t)−˚−→(2)(x,z)+a(3)k (t)−˚−→(3)(x,z) (24)

Inordertohighlightthelinkbetweenmodes2and3,thek

realizationsofP.O.D.coefficientsa(2)k wereplottedversusthek

realizationsofP.O.D.coefficientsa(3)k inFig.5a.A2Dprojection

Fig.4–PODeigenfunctionsof(a)mode2and(b)mode3.

ofthephaseportraitinthe(a(2)k ,a(3)k )planeisthusobtained. Thesetofdatashowanorganizedshape;forsakeofsimplicity, itwillbeconsideredasanellipse.Thecoefficientscanthusbe relatedbytheexpression:

a(2)2k 2(2)R + a (3)2 k 2(3)R =1 (25)

wheretheeigen-valuesareexpressedinm2/s2.Theperiodic

natureofthese coherentstructuresassociatedwithmodes 2and 3could beexpressedassinusoidal variationsa(2)

k)

and a(3)

k)asshown inpreviouspapers(Duccietal.,2007; Doulgerakisetal.,2011;Gabelleetal.,2013):

a(2)(ϕk)=

q

2(2)R cos(ϕk) and a(3)(ϕk)=

q

2(3)R sin(ϕk) (26)

whereϕkisthephaseangle.Theparametersa(2)(ϕk)anda(3)(ϕk)

havethedimensionofvelocities(m/s).Itisthuspossibleto reconstructthetemporalevolutionoftheorganizedmotion associatedtomodes2and3:

−−−→ V(2,3)k (x,z,t)=

q

2(2)R cos(ϕk) −−→ ˚(2)(x,z)+

q

2(3)R sin(ϕk) −−→ ˚(3)(x,z) (27)

(8)

Fig.5–Modes2and3:(a)portraitofphase,(b)pdfofnormalizedPODcoefficientofmode2,and(c)pdfofnormalizedPOD coefficientofmode3.

OnecanplotinFig.5canddtheprobabilitydensity func-tionsofthenormalizedcoefficientsa(2)k /

q

(2)R anda(3)k /

q

(3)R . Thedistributionsarecenteredontheoriginandtheirshape canberelatedtosinusoidal distributions,asexpectedfrom thepreviousanalysis.

4.4. Organizedmotionassociatedtomodes4and5 Asaforementioned,theeigen-values (4)R and (5)R havealso similarordersofmagnitudeandmaythuscontributeto orga-nizedmotions.Theeigen-functionscorrespondingtomode4, −−→

˚(4)(x,z),andmode5,−˚−→(5)(x,z),are plottedinFig.6aandb. Comparedtomodes2and3,modes4and5showagain vor-ticalstructuresbutthelength-scalesseemtoberoughlyhalf thelength-scalesofmodes2and3.Similarlytotheprevious analysisofmodes2and3,thekrealizationsofa(4)k wereplotted versusthekrealizationsofa(5)k inFig.7aandtheprobability densityfunctionsofthecoefficientsa(4)k /

q

(4)R anda(5)k /

q

(5)R wereplottedinFig.7bandc,theeigenvaluesbeingexpressed inm2/s2.Fig.7aexhibitsastrongrelationbetweenmodes4

and5whereasthepdfdistributionsreveal noisysinusoidal distributions.Inordertogoindeeperdepth,themodes4and 5havebeenplottedversus modes2and 3inFig.8a–d.2D projectionsofthephaseportraitarethusplottedfor differ-entbases.Theorganizedstructuresrevealedbytheseplots exhibittwolobes(contrarytothepreviousportraitofphases

plottedinFig.5,withonlyonelobe)andcanbeanalyzedas follows:

• the relationbetweenmode 2andmode 4(Fig. 8a) corre-spondsroughlytoaplotofsin(2ϕk−(/2))versuscos(ϕk),

thatis−cos(2ϕk)versuscos(ϕk);

• therelationbetweenmode 2andmode 5(Fig.8b) corre-spondsroughlytoaplotofsin(2ϕk)versuscos(ϕk);

• the relationbetweenmode3and mode4(Fig.8c) corre-spondsroughlytoaplotofsin(2ϕk+(/2))versuscos(ϕk),that

iscos(2ϕk)versuscos(ϕk);

• therelationbetweenmode 2andmode 5(Fig.8b) corre-spondsagaintoaplotofsin(2ϕk)versuscos(ϕk).

Modes4and5followthusasinusoidaltrend,withaperiod whichishalftheperiodofmodes2and3.Itisthuspossible toreconstructthetemporalevolutionoftheorganizedmotion associatedtomodes4and5:

−−−→ V(4,5)k (x,z,t)=−

q

2(4)R cos(2ϕk) −−→ ˚(4)(x,z) +

q

2(5)R sin(2ϕk) −−→ ˚(5)(x,z) (28)

Thestructuresrevealedbymodes2to5canbecompared tovortexsheddingbehindbluffbodiesthathavebeen investi-gatedbymanyworkerssincethepioneeringpaperofRoshko (1955).SimilarLissajousfigureswereobtainedintheirworks.

(9)

Fig.6–PODeigenfunctionsof(a)mode4and(b)mode5.

4.5. Turbulentmotion

Followingthedecompositionmethod,areconstructionofthe turbulenceisbasedonallthemodeshigherthan6:

−−−−→ Vk(turb)(x,z,t)= N

X

I=6 −→ Vk(I)(x,z,t)= N

X

I=6 a(I)k(t)−→˚(I)(x,z) (29)

Theprobabilitydensityfunctionofhorizontal(u′

k)component

oftheturbulentvelocityvectorgivenbyEq.(29)isplottedin

Fig.9a,inapointlocatedinthejetoftheimpeller.Thepdf aresignificantlydifferentfromthepreviousones;aGaussian functionbasedonthemeanandrmsvaluesofthisdistribution isplottedandfitsperfectlythedata.

4.6. Kineticenergydissipationratespectrum

Asafore-mentioned,onecanderivetheIthcomponentofthe rateofviscousdissipationofkineticenergyaveragedinthe plane,

ε(I)

,expressedbyEq.(19).Itisthuspossibletoplotthe distributionofrateofviscousdissipationassociatedtoeach

Imodeversusthemodenumber,similarlytotheeigen-value spectrum(Fig.2).Thiskineticenergydissipationratespectrum

isplottedinFig.9b.Onecannoticethatuppermodesfollowa lineartrendinlog–logplot(−5/9);thistrendisrelatedtothe −11/9slopeoftheeigen-valuespectrumplottedinFig.2,asit willbeexplainedlaterinthediscussion.

5.

Discussion

5.1. Eigenvaluespectrum

Theeigen-value spectrumexhibitsa−11/9slopeinlog–log plotof(I) versustheeigen-valuenumberI.Thisresultwas

explainedbyKnightand Sirovich(1990)arguing thatitisa characteristicofinertialrangeofturbulence.Asexposedby theseauthors,inturbulentflows,theturbulentkineticenergy (tke)canberelatedtotheenergydensityspectrumofvelocity

EV(K),pervectorwavenumberK,as:

tke= 1 ϑ

Z

Z

Z

u2dv=

Z

Z

Z

EV(K)dK (30)

Clearly,thedimensionoftheenergydensityspectrumof velocityEV(K)isdim[EV]=l5/t2orl2/t2timesl3.Itcorresponds

toakineticenergyinavolume.Intheinertialrangeof turbu-lence,assumingturbulenceisotropy,onecanwritetheenergy density spectrumofvelocityEV per vectorwavenumberin terms of the energydensity spectrum ofvelocity ES() per scalarwavenumber:EV=ES()/42.Thetrendoftheenergy

densityspectrumofvelocityES()perscalarwavenumberis

well knownES()∝ε2/3−5/3. Onecanthusderivetheenergy densityspectrumofvelocityEV()pervectorwavenumberas

EV∝ε2/3−11/3.

KnightandSirovich(1990)consideredthenthePOD anal-ysis.AsaforementionedinSection3.1,theeigenvalue(I)is

expressedinm5/s2inafull3Danalysis,similarlytotheenergy

densityspectrumofvelocityEVpervectorwavenumber.Thus,

KnightandSirovich(1990)consideredthateacheigenvalueis ageneralizationoftheenergydensityspectrumofvelocityEV, carryingthesamephysicaldimensions.Itfollows(I)−11/3.

In addition, Knight and Sirovich (1990) stipulated that the wave number maybe relatedtothe eigenvaluenumber as ∝I1/3.Thus,theyobtainedthe−11/9trend(I)I−11/9.This

trendisobservedinourdataprocessing(Fig.2).Thisresult issurprisingatfirstappearance,sincetheReynoldsnumber oftheflowinthetankislow(itisequalto530), correspond-ingtothe“transition”regionofturbulence.Itisknownthat localisotropyonlyexistifthelocalReynoldsnumberislarge enough.ThefirstargumentisthattheglobalReynolds num-ber islow but the hydrodynamicsis heterogeneousin the tank.Thelevelofturbulenceishigherclosetotheimpeller (wherethedatawhereacquired)thanfarfromit.Secondly, Knight and Sirovichhaveshownthat aninertial rangecan exist at very modestReynolds numbers.Such −11/9 trend was also observed inthe same mixing tank with another shearthinningfluid(withyieldstress,Carbopol,Gabelleetal., 2013) but this trend was limited to the higher Reynolds number.

Onemustalsohighlightthatthe−11/9trendcorresponds to modeslarger than20. Thisrelativelylarge valuecanbe explainedbythelimitationof2Dmeasurementofa3D veloc-ityfield.Asaforementioned,thevelocitycomponentnormal tothe 2Dplaneofmeasurement isnotmeasured.In addi-tion,thechoiceofZetagfluidwasdictatedinordertolimit theturbulencelevelclosetotheimpeller;itwasthus possi-bletomeasurealmostalltheturbulentscales.Therefore,the

(10)

Fig.7–Modes4and5:(a)portraitofphase,(b)pdfofnormalizedPODcoefficientofmode4,and(c)pdfofnormalizedPOD coefficientofmode5.

mainissueofthisworkistoexhibitthe−11/9trendofthe eigen-valuespectrumforlargevaluesofmodes, correspond-ingtoturbulentmotion.

5.2. Dissipationratespectrum

Onecan extend the development proposed by Knight and Sirovich(1990)totheanalysisofthespectrumofkineticenergy dissipationrate,plottedinFig.9b.Thisspectrumexhibitsa lin-eartrendinlog–logplotwithaslope−5/9.Thisresultcanbe explainedasfollows.Thespectrumofthedissipationinscalar wavenumbercanbeexpressedas:

ε=

Z

D()d (31)

Thedissipation density spectrum can be relatedto the energydensityspectrumbyD()≈2E()−5/3.Considering

oncemorethatthescalarwavenumberisrelatedtothe eigen-value number by ∝I1/3, we obtain the trend−5/9that is observedintheplotof

ε(I)

,normalizedbythe total dissi-pationrate(summationforallmodes).Thistotaldissipation rateisequalto0.42W/kg,almosttwicethevolumeaveraged valueofthedissipationrateofkineticenergyhεiwhichisequal to0.2W/kg.

Hereagain,the −5/9trendcorresponds tomodeslarger than20.However,themainissueofthis workistoexplain

the−5/9trendofthedissipationspectrumforlargevaluesof modes,correspondingtoturbulentmotion.

5.3. WavenumberassociatedtothemodeI

Thelast questionisrelated totheassumptionstatingthat ∝I1/3. Recall first thatthe reconstruction process isgiven

byEq.(1).Thedimensionofthecoefficienta(I)k isl/twhereas theeigen-function−→˚(I)(x,z)hasnodimension.Inaddition,the coefficient a(I)k isconstantinthe domainand onlydepends onthetimewhereastheeigen-function−→˚(I)(x,z)isconstant withtimebutitvariesinthespatialcoordinates(x,z). Conse-quently,anylength-scaleassociatedtotheIthmodecanonly berelatedtotheeigen-function−→˚(I)(x,z)andcannotberelated totheP.O.D.coefficientwhichisconstantforeachmodeinthe plane.Itisthusforeseeablethatthewavenumberassociated toeachImodedependsontheeigen-function−→˚(I)(x,z)and/or onitsgradients.If(Vx,Vz)arethehorizontalandvertical

com-ponentsofthevelocityvector −→V,and[˚(I)x(x,z),˚(I)z (x,z)]are

thehorizontalandverticalcomponentsoftheeigen-function −→

˚(I)(x,z),onecanexpressthecross-correlationfunctionAxxof thehorizontalvelocityVxbetween(x,z)and(x+L,z)as:

Axx(L)=Vx(x,z)Vx(x+L,z)= 1 N N

X

k=1 Vx,k(x,z)Vx,k(x+L,z) (32)

(11)

Fig.8–Portraitofphase:(a)base2-4,(b)base2-5,(c)base3-4,and(d)base3-5.

Afterderivation,basedonthereconstructionprocessand onEq.(14),oneobtains:

Axx(L)=Vx(x,z)Vx(x+L,z)= N

X

I=1

(I)R˚(I)x(x,z)˚(I)x(x+L,z) (33)

Inotherwords,foreachmodeI,onecandefinea cross-correlationfunctionA(I)xxintermsofA(I)xx(L)=(I)˚(I)x(x,z)˚(I)x(x+

L,z)orinnon-dimensionalform:

R(I)xx(L)=

A(I)xx(L)

(I)R

=˚(I)x(x,z)˚(I)x(x+L,z) (34)

ATaylordevelopmentattheorigingives

R(I)xx(L)=1− x2 2



∂˚(I)x ∂x



2 =1−(I)2x x2 (35)

Thus,awavenumbercaneasilyberelatedtothespatial gradientoftheeigen-function−→˚(I)(x,z).

Indeed,thekinetic energydissipation rateassociatedto each modeIwasexpressed byEq.(18). Dimensionally,one

canconsiderawavenumberassociatedtoeachImodeand definedby: (I)(x l,zk)=

v

u

u

u

t

3

X

i=1 3

X

j=1 ∂˚(I)i (xl,zk) ∂xj + ∂˚(I)j (xl,zk) ∂xi

!

2 (36)

This wave numbers can easily be derived from the eigen-function gradients. In factthis wavenumber variesin the planeofmeasurement.Sincetheeigenvalueisanaverageof thekineticenergyintheplaneofmeasurementassociatedto modeIand

ε(I)

isalsoaveragedintheplaneofmeasurement, onecanestimateanaverageofthewavenumberintheplane

(I)

similarlytoEq.(19).

This averaged wave number

(I)

isnormalized by the wavenumberofthefirstmodeandplottedversusthe eigen-valuenumberIinFig.10.ThecurveI∝

(I)

3

isalsoplotted. The normalized averaged wave numbers issued from the eigen-functionsgradientsbehavesasthewavenumberpower 1/3,asstipulatedbyKnightandSirovich.Thedefinitionofthe wavenumberassociatedtotheIthmode,(I),isa

generaliza-tionofthissimplifieddevelopmentgivenbyEq.(35).Further analysisshouldbedoneonthepdfofgradientsofeigen func-tionsintheplaneofmeasurementforeachImodetobetter understandthephysicalmeaningofthisscale.

(12)

Fig.9–Turbulence(a)pdfofturbulentvelocityfluctuations and(b)spectrumofdissipationrateofkineticenergy.

Fig.10–Wavenumberversuseigenmodenumber.

6.

Conclusion

P.O.D.technique wasappliedto2DP.I.V. data.Each instan-taneousvelocityfieldcanbeexpandedinaseriesofP.O.D. modes. The first mode is associated to mean flow. The

followingpairsofmodesare associatedtotrailingvortices. Thehighermodesarerelatedtoturbulence.The−11/9trend oftheeigen-valuespectrumisverified.Thespectrumof dis-sipationrateofkineticenergyisplottedandthe−5/9trendis shownandexplained.Lengthscalesassociatedtoeachmode aredeterminedandplottedversusmodenumber.Thepower 3trendisshown.Thegoalofthispaperwastoshowthe dif-ferenttrends(−11/9trendoftheeigen-valuespectrum,−5/9 forthespectrumofdissipationrateofkineticenergy,and3for thelengthscalesassociatedtoeachmode)correspondingto turbulentmotions.Clearlythesetrendsarelimitedtomodes largerthan20.Theanalysisofthefull3Dflowshouldbe com-pletedtoimprovetheunderstandingandthequalityofboth measurementandprocessingoftheP.O.D.modeslowerthan 20.

Acknowledgement

This work was supported by the French Environment and EnergyManagementAgency(ADEME)andIFPEnergies Nou-velles.

References

Berkooz,G.,Holmes,P.,Lumley,J.L.,1993. Properorthogonal decompositionintheanalysisofturbulentflows.Annu.Rev. FluidMech.25,539–575.

DeAngelis,E.,Casciola,C.M.,L’vov,V.S.,Piva,R.,Procaccia,I., 2003. Dragreductionbypolymersinturbulentchannelflows: energyredistributionbetweeninvariantempiricalmodes. Phys.Rev.E67,050612.

Ducci,A.,Doulgerakis,Z.,Yianneskis,M.,2007. Decompositionof flowstructuresinstirredreactorsandimplicationsformixing enhancement.Ind.Eng.Chem.Res.47,3664–3676.

Doulgerakis,Z.,2010. LargeScaleVortexandStrainDynamicsin MixingVesselsandImplicationsforMacro-mixing

Enhancement.King’sCollege,UniversityofLondon,UK(PhD).

Doulgerakis,Z.,Yianneskis,M.,Ducci,A.,2011. Onthe manifestationandnatureofmacroinstabilitiesinstirred vessels.AIChEJ.57,2941–2954.

Escudier,M.P.,Smith,S.E.,2001. Fullydevelopedturbulentflowof Newtonianliquidsthroughasquareduct.Proc.R.Soc. London,Ser.A457,911.

Gabelle,J.-C.,Morchain,J.,Anne-Archard,D.,Augier,F.,Liné,A., 2013. Experimentaldeterminationoftheshearrateina stirredtankwithnon-newtonianfluid.AIChEJ.59(6), 2251–2266.

Hoisiadas,K.D.,Beris,A.N.,Handler,R.A.,2005. Viscoelastic effectsonhigherorderstatisticsandoncoherentstructures inturbulentchannelflow.Phys.Fluids17,035106.

Knight,B.,Sirovich,L.,1990. Kolmogorovinertialrangefor inhomogeneousturbulentflows.Phys.Rev.Lett.65,1356–1359.

Moreau,J.,Liné,A.,2006. Properorthogonaldecompositionfor thestudyofhydrodynamicsinamixingtank.AIChEJ.52, 2651–2655.

Oudheusden,B.W.,vanScarano,F.,Hinsberg,N.P.,vanWatt,D.W., 2005. Phase-resolvedcharacterizationofvortexsheddingin thenearwakeofasquare-sectioncylinderatincidence.Exp. Fluids39,86–98.

Roshko,A.,1955. Onthewakeanddragofbluffbodies.J.Aerosp. Sci.Technol.22,124–132.

Sirovich,L.,1987. Turbulenceandthedynamicsofcoherent structures.Q.Appl.Math.45,561–590.

Tabib,M.V.,Joshi,J.B.,2008. Analysisofdominantflowstructures andtheirflowdynamicsinchemicalprocessequipmentusing snapshotproperorthogonaldecompositiontechnique.Chem. Eng.Sci.63,3695–3715.

Figure

Fig. 1 – Apparent viscosity of Zetag7587 solutions versus shear rate.
Fig. 2 – Eigen value spectrum.
Fig. 4 – POD eigen functions of (a) mode 2 and (b) mode 3.
Fig. 5 – Modes 2 and 3: (a) portrait of phase, (b) pdf of normalized POD coefficient of mode 2, and (c) pdf of normalized POD coefficient of mode 3.
+5

Références

Documents relatifs

In all dimensions, we prove that the marked length spectrum of a Riemannian manifold (M, g) with Anosov geodesic flow and non-positive curvature locally determines the metric in

We compute the whole spectrum of the Dirichlet-to-Neumann operator acting on differen- tial p-forms on the unit Euclidean ball.. The Dirichlet-to-Neumann operator on

[r]

1) Case of three or less users: The SS-MC-MA system al- lows the allocation of a 528 MHz sub-band for each user. This system offers the same performance and advantages as MC- CDMA,

On a surface with a Finsler metric, we investigate the asymptotic growth of the number of closed geodesics of length less than L which minimize length among all geodesic multicurves

2014 The analysis of wavelength data of the Gd I spectrum has resulted in the detection of 201 odd and 173 even levels in addition to those reported by

The calculated values for the 4 f8(7P) 6 p levels are given in Table IX along with the predicted positions and percentage compositions.

Ensuite, en ajoutant la dernière partie, qui est celle du quiz, avec une fonctionnalité de partage des scores sur les réseaux sociaux (Facebook, Twitter et