• Tidak ada hasil yang ditemukan

fluid method in axisymmetric coordinates Interface reconstruction and advection schemes for volume of Journal of Computational Physics

N/A
N/A
Protected

Academic year: 2024

Membagikan "fluid method in axisymmetric coordinates Interface reconstruction and advection schemes for volume of Journal of Computational Physics"

Copied!
13
0
0

Teks penuh

(1)

Contents lists available atScienceDirect

Journal of Computational Physics

www.elsevier.com/locate/jcp

Interface reconstruction and advection schemes for volume of fluid method in axisymmetric coordinates

Ananthan Mohan, Gaurav Tomar

DepartmentofMechanicalEngineering,IndianInstituteofScience,Bangalore,India

a rt i c l e i nf o a b s t ra c t

Articlehistory:

Availableonline31August2021 Keywords:

Volumeoffluid Axisymmetriccoordinates Multiphasesimulation

Volume of fluid (VOF) method is a sharp interface method employed for simulations of two phase flows.Interface inVOF is usually represented usingpiecewise linearline segments ineachcomputationalgridbasedonthe volumefractionfield.WhileVOF for cartesian coordinatesconserve mass exactly, existing algorithms do not show machine- precisionmassconservationforaxisymmetriccoordinatesystems.Inthiswork,wepropose analyticformulaeforinterfacereconstructioninaxisymmetriccoordinates,similartothose proposedbyScardovelliandZaleski(2000)[12] forcartesiancoordinates.Wealsopropose modificationsto theexisting advectionschemes inVOF foraxisymmetriccoordinatesto obtainhigheraccuracyinmassconservation.

©2021ElsevierInc.Allrightsreserved.

1. Introduction

Multiphaseflows areubiquitiousinseveralindustrial applications.Inthe lastthreedecades,therehasbeenasurgein thenumericalmethodsandalgorithmsforsimulationsofcomplexmultiphaseflows.Therehavebeenseveraldifferenttypes ofinterfacecapturingstrategiesthathavebeenproposedfortwo-phaseflows.ThemostpopularofthesearetheLevelset method,Volume ofFluid(VOF) method,andFronttrackingscheme [1,2]. VOFmethodswithgeometric advectionstrictly conservethevolumeofthetwophases.

Several improvementshave been madesince the inception ofthe method (see Hirt andNichols [3–8]).The simplest and earliest representation for interface reconstruction is simple line interface calculation (SLIC) in which the interface is approximatedby horizontal orvertical lines.Subsequently, piecewise lineinterface construction (PLIC) was introduced wheretheinterfaceisapproximatedasalinearlineatanangleinthecell[9].Higherorderinterfaceconstructionhasbeen proposed (suchasParabolic reconstruction by [4]),butconsidering theassociatedcomputational cost andcomplexity for geometricadvection,PLICisusuallypreferred.

In PLICbased method theinterface isapproximated asa line segment givenby m.x=a,where m=(mx,my) is the unit normal to the interface anda is the line constant. The reconstruction step in VOF methods consists of two steps, the computation ofm andthen the computation ofline constant a such that the volume cut by the interface matches thevolumefractionofthecell.Theinterfacenormaliscomputedusingthevolumefractiongradientbyusingan accurate gradient computing algorithm like Youngs method [9], a least-squares method (e.g.: (E)LVIRA [7]) or a height function method [10].Anotherpopularmethodforinterface reconstructionistheMoment-of-Fluid(MoF)method [11], whichuses the volume fractionvalue ofthe cell along withthe geometric centroids of the individual fluid volumeswithin the cell

*

Correspondingauthor.

E-mailaddress:[email protected](G. Tomar).

https://doi.org/10.1016/j.jcp.2021.110663

0021-9991/©2021ElsevierInc.Allrightsreserved.

(2)

to reconstructthe interface.The MoFmethods require one to tracktheadvectionof thegeometric centroids ofthefluid volumesalong withthevolumefraction. Inthe present,work we areprimarily interestedinthecomputation ofthe line constantgivenaninterfacenormalintheframeworkofVOFmethods.Neverthelessthereconstructionalgorithmpresented inthestudyisquite genericforstructuredgrids.

ScradovilliandZaleski [12] proposedanalyticalformulaeforthepiecewiselinearreconstructionoftheinterfaceincarte- sian coordinates that ledto a significantspeedupover the earlieriterativeschemes.However, forcurvilinear coordinates (suchasaxisymmetric coordinatesystem), theproposed analytical formulaecannot be employeddirectly. Diot etal.[13]

proposedasemi-analyticformulaeforarbitraryconvexcellswhichrequiredtheuseofNewtonsiterationforaxisymmetric coordinatesystem. Inthepresentwork,we derivesimilar analyticformulaeforaxisymmetriccoordinates,whichresultin aspeedupof∼28 overtheiterativecounterparts (Brent’srootfinding method).Further,we demonstratethattheexisting interfaceadvectionschemesinVOFforaxisymmetriccoordinatesarenotstrictlymassconserving.Inthisstudy,wepropose modifications tothecurrentoperatorsplitalgorithmsthat resultinmachine-precisionmassconservationinaxisymmetric coordinates.Weshowtheefficacyoftheproposedalgorithmsusingseveraltestcases.

The paperisorganizedasfollows.Wefirst presentanalytical formulaefortheinterface reconstructionschemesin ax- isymmetriccoordinatesinsection2.Insection3,weproposemodificationsintheexistinginterfaceadvectionalgorithmfor axisymmetricVOFandpresenttestcasestoshowtheefficacyofthescheme.Finally,insection4,wediscusstheimportant conclusions.

2. Interfacereconstructionscheme

Interfacereconstructioninthevolumeoffluid(VOF)methodrequiresthevolumefractionfield.Usingthevolumefraction field,apiece-wiselinearorahigherorderinterfaceisconstructedinagivengrid-cell.Interfacereconstructionisanintegral partofthegeometricadvectionschemestoensuremassconservationpropertyoftheVOFmethod[2].Initialconditionfor a multiphaseflow simulationrequires theinitial distributionofthevolume fractionfield, usually providedasan implicit functionofthespatialcoordinates.VOFIlibrary [14] isanopensourcelibrarytoinitialize theliquidvolumefractionfieldin cartesiancoordinatesystemsaccurately.InVOFI,forcellscutbytheinterface(seeFig.1),PLICreconstructionmethod[9,6]

isemployedtoapproximatetheinterfaceasalinearlinesegment,

m

.

x

=

a

,

(1)

where mis thelocalnormal attheinterface, x is apoint on theplane, anda isthe normaldistanceofthe originfrom the plane.Analyticalrelation,givenby ScardovelliandZaleski [12],betweenthevolume fractionandthelineconstant is employedtodeterminethelineconstanta.Thus,fortwo-dimensionalandthree-dimensionalCartesiancoordinatesystems, VOFI library canbe directly employed foraccurate assignment of the initialvolume fraction field on a givendiscretized domain usingan implicitequation oftheinterface. However, incurvilinearcoordinatesystems,fora givenimplicitfunc- tion,thepiece-wiselinearinterface constructedfromVOFIwouldrequirecomputationofthevolumefractionfieldusinga formulaspecific tothecurvilinearcoordinates.Forinstance,foraxisymmetriccoordinatesystem, themodified Gaussarea (shoelace)formulaforcomputationoftheareaofaconvexpolygonisgivenby,

V

= π

3

n

i=1

(

xi

+

xi+1

) (

xiyi+1

xi+1yi

)

(2)

where(xi,yi)fori=1,...,n(withxn+1=x1andyn+1=y1)arethecoordinatesoftheverticesofaconvexpolygonordered counterclockwiseasshownintheFig.1.

Thus,forinitializationofthevolumefractionfield,C,onceweobtainthelinearinterfaceineachgridcellusingtheVOFI library,weusetheaboveformulatocomputethevolume,V,andassignvolumefractionineachgridcellas,

C

=

V

2

π

rc

x

y

.

(3)

Here rc isthedistanceofthecenterofthe cellfromtheaxisofsymmetry,andxandy arethe grid-cellsizes inthe radial (r) andaxial(y) directions,respectively.Wenote that theabove procedureis followedessentiallyto minimize the errorinthevolume-fractionduringinitialization.Toillustratethis,weinitialize atorusofminorradiusrt=0.25 andmajor radiusr=0.50 inthecenterofacomputationaldomainofsize1×1 asshownintheFig.2.Thevolumeofthetoruscanbe analyticallycomputedasVt=2

π

2rrt2,wherethemajorradius,r,isthedistancetothecenterofthetorusfromtheaxisof symmetry.Wecomparetheresultsforinitialized volumeforvariousgridsizesbycomputingtherelativeerrorwithrespect tovolumeobtainedusinganalyticalformula,giveninTable1.

Thus, we haveshownthat forcurvilinearcoordinatesvolume fractionfield can beinitialized up tomachine accuracy.

Now, we derive an analytic relation forPLICreconstruction for axisymmetric coordinateson the lines ofScardovelli and Zaleski[12].WeuseYoungsmethod [9] togettheinterfacenormal(minequation (1))fromfluid-1(C=1)tofluid-2C=0 andisgivenbym= −∇C/|∇C|.TocompletethePLICinterfacereconstruction foragivenC,inadditiontothenormalm,

(3)

Fig. 1.Coordinatesoftheverticesofasimplepolygoncutbytheinterfaceorderedcounterclockwise.Herex3,y3andx4,y4representstheendpointsof thePLIClinesegmentwithinterfacenormalmandlineconstanta.

Table 1

Resultsforrelativeerrorinvolumeduringinitializa- tionofatorusofradius,rt=0.25,fordifferentgrid sizes.

Relative error in volume:E=|VVtVt|

Grid Current Solver

16×16 5.4×1016

32×32 7.1×1016

64×64 1.8×1016

128×128 5.4×1016

Fig. 2.Reconstructedinterfacefor16meshpointsofatorusofradiusrt=0.25 andmajoraxisr=0.50 initialized atthecenterof1×1 domainwhere reconstructedPLIClinesegmentsforeachmixedcellisdepicted.

we alsoneedtoobtain thelineconstant,a,whichis thenormaldistanceofthe interfacefromoneofthe verticesofthe computationalcell.Inwhatfollows,wepresentamethodologytogetthelineconstant(a)analyticallyforagiveninterface normalvectorandthevolumefractionofamixedcell.

As discussed in[12],using an analytical relationbetween thevolume fraction(C),interface normal(m) andthe line constant (a), we can implement an i felsei fendi f constructto determine the line constant, a. Thisapproach iscomputationally muchmoreefficientcompared tothealternative iterativeapproach togetthelineconstant.Giventhe equationoftheinterface,m1x+m2y=a,allcombinationsofm1,m2(suchthatm21+m22=1)canbereducedtooneofthe cases showninFig.3, eitherby changingthe originorby changingthe referencefluidfromfluid-1 to fluid-2,such that bothm1 andm2arepositiveandtheleftbottomcornerofthemixedcellunderconsiderationiscontainedinfluid-1.Fig.3 showsallthepossibleconfigurationsforinterfacearrangementwithm10 andm20.

(4)

Fig. 3.Variouscaseswhichcanariseinthestandardconfigurationoftheinterfacewithm1,m20 andfluid1 isoccupyingthebottomleftcornerofthe cell.

WefirstdiscussCaseAshownintheFig.3andsimilarprocedurecanbefollowedtoobtainrelationsforothercases.For theaxisymmetriccoordinatesystem,theshadedareashownintheFig.3forCaseAisgivenby,

V

= π

3

(

x0

+

x2

)

3m1 m2

− π

3

(

x0

+

x1

)

3m1

m2

− π

x20y1

.

(4)

Using theequation oflinem1x+m2y=a we have x1=a/m1(m2y)/m1 and x2=a/m1.In thepresentstudy,we assumex=y,butthesameanalysiscanbeeasilyextendedforx=y.Substitutingx1andx2intheequation(4) and collectingthetermsinpowersofa,weobtain,

π

y m21

a2

+

− π

m2

y2

m21

+

2

π

yx0 m1

a

+

π

m22

y3

3m21

− π

y2m2x0 m1

=

V

.

(5)

Thus,wehaveananalyticalrelationbetweenthevolumeandthelineconstanta.Wenotethattheaboverelationholdstrue onlywhentheinterfacecutsthroughthetopandthebottomedgesofthecellshownforCaseAinFig.3:(x2x0)x, y1=y and y2=0.These conditions yield the boundson the values ofa: m2yam1x.Substituting the above boundsforaintheequation(5),yieldtheboundsonthelimitingvolumes:

V1

= π

y

3

x2m21

+

3

xm1

(

2m1x0

ym2

) +

ym2

(

ym2

3m1x0

)

3m21

,

(6)

and

V2

= π

y2m2

(

ym2

+

3m1x0

)

3m21 (7)

ForagivenvolumefractionC andinterfacenormal(m1,m2),thevolume occupiedby fluid-1 intheconfigurationsshown intheFig.3isgivenby:V =2

π

rxyCwherer=x0+x/2 isthedistancefromtheaxisofsymmetrytothecellcenter.

If V1VV2 then theanalytical relation givenby equation (5) can be used to determinethe lineconstant a.Forthe quadratic equation ina givenby equation (5), we note that onlyone of the rootswill satisfy the requiredboundson a forcaseA.ForcasesBandC,we obtaincubicequationsthatcanbe solvedfora usingCardano’sformulaorusingBrent’s methodtofindtheappropriaterootwithnecessaryboundsforthelineconstant.Followingthesameapproachwe canget boundsonvolumeforcaseDas:

V3

= π

y2

(−

2

ym1

+

3

ym2

3m1x0

+

6m2x0

)

3m2 (8)

and

V4

= π

y2m1

(

y

+

3x0

)

3m2

.

(9)

Fig. 4shows all the possibleconfigurations of theinterface and thevolume bounds whichseparate each caseasthe interface normalinradial directionvariesfromminimumtoamaximum.Fig.4clearlyshowsthatthevariousboundsfor thecasesshowninFig.3donotoverlapandprovideauniquecriterionforcomputingthelineconstanta.Thus,wecanuse thefollowingalgorithmtoclassifyeachcase.

Welistbelowtheanalyticalrelationbetweenthelineconstant(a)andthevolume(V)foreachcase:

CaseA

π

y m21

a2

+

− π

m2

y2

m21

+

2

π

yx0

m1

a

+

π

m22

y3

3m21

− π

y2m2x0

m1

=

V

.

(10)
(5)

Fig. 4.BoundsfordifferentcasesshowninFig.3asafunctionofthex-componentoftheinterfacenormal,m1.Theinterfacialcellisplacedatadistance of1 fromtheaxisofsymmetrywithx=y=1.ThetotalvolumeofthecellisVcell=2πgivenbythetopboundaryintheplot.

Data:V,m1

Result:Identificationofthecaseinwhichthestandardinterfacebelongsto.

ifm112 then if VV1 then

CaseB elseif VV2 then

CaseC else

CaseA // V1>V>V2

end else

if VV3 then CaseB elseif VV4 then

CaseC else

CaseD // V3>V>V4

end end

Algorithm 1:Classification of the standard case of the reconstructed interface wherem1,m2≥0.

CaseB

− π

3m21m2

a3

+

π

y m21

− π

x0

m1m2

a2

+

π (

y

+

x0

)

2

m2

− π

m2

y2

m21

+

2

π

yx0

m1

− π

x20 m2

a

+

− π

y

(

y

+

x0

)

2m1

m2

+ π

m22

y3

3m21

− π

y2m2x0

m1

− π

m1x30

3m2

+ π

m1

(

y

+

x0

)

3 3m2

=

V

(11)

CaseC

π

3m21m2

a3

+ π

x0

m1m2 a2

=

V (12)

CaseD

a

=

2

π

y3m1

+

3

π

y2m1x0

+

3m2V

3

π

y

(

y

+

2x0

)

(13)

WenoteherethattheothercasescanbereadilytransformedintooneofthecaseslistedintheFig.3byeitherchanging thefluid(byusing(1−C)insteadofC tocomputethevolumeandinvertingtheinterfacenormalm) orby changingthe origin(keepingthelocationoftheaxis-of-symmetrysamebutinvertingitsdirection).

Wenowcomparetheabovedescribedanalyticalmethodwiththeiterativemethodforfindingthelineconstantforthe caseA giveninFig. 3.The relativeerrorin thelineconstant is giveninTable 2forthe analyticalanditerativemethods withdifferenttolerances.Wenotethattheiterativemethodtoreachanerrorwithatoleranceof108 isabout28 slower comparedtotheanalyticalmethod.

(6)

Table 2

RelativeerrorinlineconstantaandtheratioofCPUtimerequiredbyiterativemethod(Brent’salgorithm)to thatrequiredbyanalyticalmethodfor10000 repetitionsforcaseAshowninFig.3fordifferenttolerancesused intheiterativemethod.

Comparision between analytical and iterative reconstruction methods

Method Tolerance Relative Error titerative/tanalytical

Analytical0 1

Iterative 1.0×104 1.8×104 14.2

Iterative 1.0×106 1.1×106 21.2

Iterative 1.0×108 6.7×1011 28.3

Oncethelineconstant,a,isobtained,thepositionoftheendpointsofthelinearapproximation oftheinterfacecanbe computed thuscompleting the construction ofa planar interface ina givencomputational cell. As discussedearlier, this moreprecise descriptionoftheinterface withinthegrid cellallows geometricadvectionwhichgivestheVOFmethodits strictmassconservationpropertywhilemaintainingasharpinterface.

In what follows,we discuss an operator split algorithm forthe geometric advection ofthe interface in axisymmetric coordinates. Wenote that the straightforwardextension ofthe 2D cartesianalgorithmdoesnot yield accurate results,as alsoindicatedbytheresultsobtainedfromtheexistingopensourcecodes.

3. Advectionoftheinterface

We present herea scheme for accurate geometric advection ofthe volume fraction in theaxisymmetric coordinates.

We have used a uniformgrid to describe thevariables with volume fractionbeing storedat the cell centers (Ci,j). The incompressible fluid flow is determined by the velocity field which is defined atthe cell faces (ui+1/2,j,vi,j+1/2). Here, u denotes the radial directionvelocityand v is theaxial velocity. Thevelocity field satisfies thediscrete divergencefree conditiongivenby:

(

ru

)

i+1

2,j

− (

ru

)

i1 2,j

ri

x

+

vi,j+12

vi,j1

2

y

=

0

.

(14)

Motionoftheinterfaceisgovernedbytheadvectionequationforthevolumefractionfield,

C

t

+

u

· ∇

C

=

0 (15)

Forincompressiblefluids,conservationoftheindividualvolumesofthetwofluidsresultsintheconservationofmass.

In the volume offluid method,geometric advection ofthe volume fractionfield is employed instead of using a higher orderscalaradvectionschemetosolveequation(15) since itlimitsthediffusionofvolumefractionfield thusmaintaining a sharpinterface andatthe sametime ensuresthe conservationofmasstohighdegree ofaccuracy [2].Givenavolume fractionfield,reconstructedinterfaceandsolenoidalvelocityfield,wecansolvetheequation(15) usinganoperatorsplitting algorithm consistingofan r-sweep anda y-sweepfollowing[15].Inordertoemploy anoperator splittingalgorithm,the advectionoftheinterface(equation (15)),using∇ ·u=0,canbewrittenas:

C

t

+ ∇ · (

uC

) =

C

(∇ ·

u

).

(16)

The RHS in equation (16) is the dilatation term or the divergence correction term and is needed in an operator split approachsinceineachstepofanoperatorsplitalgorithmthevolumefractionC isadvectedfirstinonedirectionandthen inthe otherdirection(s). Thediscrete divergencetermorthedilatationtermisincludedinthe formulationsince ineach step ofan operatorsplitalgorithm thevolume fractionC isadvected ina one dimensionalflowwhich isnot divergence free.Givenavolumefraction(Cni,j)andvelocityfield(uni+1/2,j,vni,j+1/2) atthenth timestep,thediscretized equation(16) isgivenby,

Cni,+j1

=

Cni,j

+

t ri,j

x

δ

Vi1/2,j

− δ

Vi+1/2,j

+

t

y

δ

Vi,j1/2

− δ

Vi,j+1/2

+

Cni,j

t ri,j

x

ri+1/2,juni+1/2,j

ri1/2,juni1/2,j

+

t

y

vni,j+1/2

vni,j1/2

(17)

where δVi+1/2,j=(ruC)ni+1/2,j is the amount of volume fraction fluxed through the right cell face. Similarly, fluxes δVi1/2,j,δVi,j+1/2 andδVi,j1/2canbecomputedforothercellfaces.

(7)

Usingoperatorsplitting,wecansplittheaboveequationasfollows:

Ci,j

=

Cni,j

+

t ri,j

x

δ

Vi1/2,j

− δ

Vi+1/2,j

+

Cci,j

t ri,j

x

ri+1/2,juni+1/2,j

ri1/2,juni1/2,j (18)

Cni,+j1

=

Ci,j

+

t

y

δ

Vi,j1/2

− δ

Vi,j+1/2

+

Cic,j

t

y

vni,j+1/2

vni,j1/2 (19)

where Ci,j istheintermediate valueofthe volumefraction. Here, Cic,j isafunction basedonwhich wecan getdifferent operator splitting approaches. If Cci,j=Ci,j then we get Sussmanand Puckett’s operator split approach [15]. An implicit schemeisusedinthefirstdirectionandanexplicitschemeintheseconddirectiontomaintaintheconservationofvolume fraction [16].The orderofsweepofdirectionisalternatedevery timestep[17] (“Strangspliting”)to achievesecond order accuracyintime.

Ifweuse, Cci,j

=

1 Cn i,j

0

.

5

0 Cni,j

<

0

.

5 (20)

whereCic,jdependsonlyontheprevioustimestepvalueofvolumefractionwegettheaxisymmetricformofWeymouth andYue’soperatorsplit approach [18].Thismethodisexplicitinboththesweep directions.Foranoperatorsplit method tobestrictlyconservativeithastofulfill,followingWeymouthandYue [18],asetofconcurrentrequirements,

1. Thevolumefluxtermsareconservative,and 2. thedilatationtermsumstozero,and

3. noclippingorfillingofacellisneededtoimpose0≤C1.

Usinggeometricadvection(describedbelow)forvolumefluxingandadivergencefreevelocity field,bothSussmanand Puckett’sandWeymouthandYue’smethodssatisfythefirsttworequirements.SussmanandPuckett’smethoddoesrequire clipping orfillingofacell toimpose 0≤C1 andthereforeisnotstrictly conservative.Howeverthe massconservation errorsintroducedinthismethodaresmallandthusithasbeenwidelyadoptedpreviously.WeymouthandYue [18] proved that,byimposingagridCourantnumberrestrictionof,

t

N

d=1

uxdd

<

1

2 (21)

in their method one can satisfy thethird requirement forthe operator split approachto be conservative. Here N is the number ofsweep directions and ud and xd are the maximum velocity andgrid size in each of those directions.Thus by usingWeymouth andYue’soperator splittingapproachwe canpreserve themass exactlywhenadvecting thevolume fractionfield.

The volumefluxthrough cellfaces,δVcellf ace,iscomputedgeometrically.Consider theschematic inFig.5,wherethe shadedregionshowsthevolumeoffluid-1 inthecelltobefluxed throughtherightface(δVi+1/2,j).Consideringtheface velocity(ui+1/2,j)tobepositive,thefluxcanbecomputedas,

δ

Vi+1

2,j

= (

ru

)

i+1/2,jV

2

π

r

r

y (22)

where V isthevolume offluid 1 fluxedthroughtherightface(shownastheshaded regioninFig.5),r isthedistance in the radial directionwhich contains the volume advected in this timestep and r is the distance to the center of this volumefromtheaxisofsymmetry.HerethevolumeV dependsontheorientationoftheinterface.Wecancalculaterby considering theconservationofvolume fluxedthrough therightface andsolving theresultingquadratic equation,which yields,r=ri+1

2,jr2

i+12,j2ri+1

2,juet.Usingthesectionofthepiece-wisereconstructedinterfacelyinginthevolume tobefluxedthroughthecell-faceovert time-stepandemployingtheGaussareaformula,givenbyequation(2),wecan calculatethevolumecutbythisregion.

TheabovesmallcorrectionincomputingralongwiththeaccurateGaussformulaforaxisymmetricsimulationsallows us to improve upon the existing volume fraction advection schemes. Existing schemes [15] modify the velocity in 2D algorithms by using ru for velocity field and use 2D geometric advection scheme which results in a third order error (O(t2h)),wherehisthegridsizeandt isthetimestep)inmassconservation.Toillustratethisgeometricfluxingerror, consider the radial advection of the volume fraction field for a full cell (C=1) as shown in Fig. 6. The shaded region representsthevolumeadvectedthroughtherightfaceofthecellwitharadialcellfacevelocityue.

The existing schemes compute the volume as Vc=2

π

(reue2t)(uet)y,where re is the distanceof theeast cell facefromtheaxisofsymmetry,tisthetimestep,andy istheheightofthecell.Whereas,theproposedschemeyields
(8)

Fig. 5.The fluxed volume through the right face of the cell whenui+1/2,jis positive.

Fig. 6.The shaded region is the volume advected through the east cell face in a single timestep.

theexactvolume,V =2

π

(re2r)ry withr=re

r2e2reuet.Thus,theerrorinvolumecalculationisgivenby, E=

π

(uet)2y.

Wevalidate theproposedmodificationswiththefollowingtestcasesandcomparewiththeresultsobtainedusingthe opensourcemultiphaseflowsolverGerris [19] whichusesasimplerextensionofthecartesianoperatorsplitapproachfor geometricfluxing.

3.1. Advectionofatorus

Inthistestcase,atorusofradius0.25 isinitialized at(0.35,0.5)inacomputationaldomainofsize(2.0,1.0).Thetorus is advectedunderthe steadystate velocityof u=0.1/r forr>0.05 and v=0,where r isthedistancefromthe axisof symmetry. The fluid is advected forward in time t=7.8125 and then the velocity is reversed so asto advect the fluid backwardsintime tilltime t=15.625.Inordertoillustratethe effectofaccurate geometric fluxing,we advectthetorus using SussmanandPuckett’s method fora grid Courant numberof 1 andcompare theresults withthat obtained using Gerris flowsolver. For a grid sizeof x=y=1/128, thisresultsin a time step oft=0.0078125. Thus the fluid is advected a 1000 timestepsforward intime andthen backwards.As seen fromtheFig. 7,after1000 timesteps thetorus ishighlycompressedduringtheadvectionaslessarea(duetoaxisymmetry)occupiesthesamevolumeaswemoveaway fromtheaxisofsymmetry.Wenotethatthefinalinterface shapematchesverywellwiththeinitialpositionofthetorus, thusvalidatingouralgorithm.

Therelativeerrorinthevolumebetweentheinitialandfinaldistributionoffluid-1 forvariousnumberofforwardand backwardadvectiontimestepsisgiveninTable3.Thecorrespondingrelativechangeinthevolumeobtainedforthesame testcasesimulatedusingGerrisflowsolverisalso givenforcomparison.Wenotethattheerrorobtainedfromthepresent schemesishighly accurateincomparisontothoseobtainedfromGerris(Table3).

WerepeatthistestfordifferentCourantnumbersusingbothSussmanandPuckett’sandWeymouthandYue’sadvection schemes andcomparethe resultswiththoseobtainedusingGerrisflow solver.Table 4givestherelativeerrorinvolume

(9)

Fig. 7.Interfaceshapeafter1000 timestepsforwardandbackwardafteradvectingthetorusofradius0.25 withagridCourant(CFL)numberof1 advected usingSussmanandPuckett’smethodforgridCourantnumberof1.

Table 3

Resultsforrelativeerrorinthevolumeforatorusofradius0.25 advectedinradial directionforwardandbackwardintimefordifferentnumberoftimesteps.

Relative error in volume

Number of timesteps CurrentSolver (SussmanandPuckett)

Gerris Solver

1 5.2×1015 1.1×105

10 2.9×1014 7.2×105

100 4.8×1013 2.1×104

1000 3.4×1010 3.8×104

Table 4

Resultsforrelativeerrorinthevolumeforatorusofradius0.25 advectedinradialdirectionforwardandback- wardintimefordifferentgridCourantnumbers.

Effect of Courant number on relative error in volume

Grid Courant number Gerris Solver CurrentSolver (SussmanandPuckett)

CurrentSolver (WeymouthandYue)

0.1 9.5×105 1.4×109 1.9×1012

0.2 3.9×106 1.9×1010 2.4×1012

0.3 1.3×104 7.1×1011 2.6×1013

0.4 2.4×104 1.0×1010 2.5×1015

forvariousgridCourantnumbers.Weobservethat usingWeymouthandYue’soperatorsplittingapproachalongwiththe interface reconstruction algorithm presented we can obtain machine accurate volume conservation.We observe a small increaseinerrorasCourantnumberdecreaseswhichcanbeattributedtothe largenumberoftimestepswhichresultina largernumberofinterfacereconstructioneventsthusleadingtolargererroraccumulation.

As suggestedby Kotheetal. [20],simplelinearadvectiontest casesdonotreveal theefficacyofadvectionalgorithms appropriately.Thus, we furthertest theefficacyofthealgorithm by subjectingitto amore severetest caseofadvection ofatorus inaHill’svortex [21] withasuperimposedradial velocityfield whichresultsina divergencefree velocityfield uptomachine-precision.Thisisaxisymmetricequivalentofthecircleinavortextestcasefor2Dcases[20].Forthisvelocity field,theinterfaceundergoesstrongtopologicalchangesincludingfragmentationandmergingduetostrongsheareffects.A torusofradius0.1 isinitialized at(0.2,0.8)inacomputationaldomainofsize(1.0,1.0)withL=0.5.Thefluidisadvected underhighlystrainedsteadystatevelocityfieldgivenby(forr0.05)

u

=

0

.

1

r

L (yL)

L

+

0.r05 (23) v

=

0

.

1

1

yL L

2

2

r

L

2

.

(24)

The fluidis advectedforwardintime t=4.0 and thenthevelocity isreversed soastoadvect thefluid backwardsin time tilltime t=8.0.Fora gridsizeofx=y=1/128,andgridCourantnumberof0.1 thisresultsina time stepof t=1.103. Thefluid is advected 4000 timestepsforwardin time andthenthe velocityis reversed toadvect 4000 timestepsbackwardsintime.

(10)

Fig. 8.Interfaceshapeafter4000 timestepsforwardandbackwardafteradvectingthetorusplacedinavortexusingWeymouthandYue’soperatorsplit approachforagridCourantnumberof0.1.

Table 5

Resultsforrelativeerrorinvolumeforatorusofradius0.1 placedinacomplexvelocityfieldadvectedforwardandbackwardintimefordifferentgrid Courantnumbers.

Effect of Courant number on relative error in volume

Grid Courant number Gerris Solver CurrentSolver

(SussmanandPuckett)

CurrentSolver (WeymouthandYue)

0.1 2.7×104 3.5×108 2.7×1012

0.2 1.6×104 3.4×107 2.2×1015

0.3 8.1×105 7.6×107 4.1×1015

0.4 5.4×104 4.2×106 1.9×1015

Fig. 9.Interfaceshapeafterevery1000 timestepscomparedagainstthetrajectoriesofLagrangianmarkerparticlesforsimulationusingagridCourant numberof0.1 usingWeymouthandYue’soperatorsplittingapproach.

As seenfromFig. 8after4000 timesteps theshape oftheinterface ishighly distorted duetohighly strainedvelocity field.Thefinalinterfaceshapematchesverywellwiththeinitialpositionofthetoroidthusvalidatingouralgorithm.

WerepeatthistestfordifferentCourantnumbersusingbothSussmanandPuckett’sandWeymouthandYue’sadvection schemes andcomparethe resultswiththoseobtainedusingGerrisflow solver.Table 5givestherelativeerrorinvolume forvariousgridCourantnumbers.Weobservethat relativeerrorusingSussmanandPuckett’soperatorsplittingapproach isanordersmallercomparedtotheresultsfromGerrisflowsolverevenforlargernumberoftimesteps.UsingWeymouth andYue’sschemeweobservemachineaccuratevolumeconservation.

(11)

Fig. 10.Interfaceshapeaftertimet=1.Theinterfaceisflattened againstthetopwallandstretcheddue totheunderlyingvelocity.Evenwhenthe cross-sectionalareachangesthevolumeofthetoroidalbubbleismaintainedveryaccurately.

Table 6

Resultsforrelativeerrorinthevolumeforabubbleinstagnationpointflowfortf inal=1.0 fordifferenttoleranceofPressure-Poissonsolverforboth SussmanandPuckett’sandWeymouthandYue’sadvectionmethods.

Toleranceof PoissonSolver

Divergenceof velocityfield

Relativeerror (SussmanandPuckett)

Relativeerror (WeymouthandYue)

1.0×104 1.1×104 1.1×104 2.3×105

1.0×106 5.1×105 1.1×106 1.5×106

1.0×108 2.6×107 2.5×106 2.3×109

1.0×1010 2.67×109 2.6×106 4.8×1010

1.0×1012 1.5×1011 2.6×106 1.7×1011

1.0×1014 7.5×1013 2.6×106 1.4×1011

InordertoillustratetheaccuracyofthereconstructedinterfacewepopulatetheinitialinterfacepositionwithLagrangian marker particles andadvect them withthe underlying velocity witha fifth order accurate time integrator [22] fortime t=4.0. Wecomparethe final positionsofthe markerparticleswiththat oftheadvected interface usingWeymouth and Yue’sadvectionschemeandfora gridCourantnumberof0.1.The Lagrangianmarkerparticles matchthe positionofthe reconstructedinterfaceverywellasshowninFig.9thusdemonstratingtheefficacyofthealgorithm.

3.2. Bubbleinastagnationpointflow

Inthistest caseweimplementtheVOFalgorithmpresented inthispaperfora morecomplexflow. Wesolve Navier- Stokesequationsinonefluidformgivenby:

ρ (

C

) ∂

u

t

+

u

.∇

u

= −∇

p

+ ∇ · μ (

C

)

u

+ (∇

u

)

T

+ ρ (

C

)

g

+

fγv (25)

whereuandparethevelocityvectorandpressure,respectively,

ρ

(C)and

μ

(C)arethefluiddensityandviscositywhich areafunctionofvolumefractionfield,C.WeuseChorin’sprojectionmethod [23] tosolvetheaboveequation(25) where wediscretized theadvectiontermusingasecondorderENOscheme [24] andthediffusiontermsusingcentraldifferencing.

Surface tension forces, fγ

v are acting only at the interface and have been modeled as volumetric body force using the continuumsurfaceforcemodelofBrackbill,Kothe,andZemach [25].TheinterfaceiscapturedusingCLSVOFalgorithmgiven bySussmanandPuckett [15].Thisalgorithmismassconservingandcalculatesthecurvatureandsurfacenormalwithhigh accuracywhichisusedforsurfacetensionforcecalculation.Theinterfaceisadvectedbysolvingtheadvectionequationsfor level-setfunction, φ,andvolumefraction, C.We initializeatoroidalbubbleofradius 0.1 at (0.2,0.5)inacomputational domainofunitsize,1×1.Thebottomboundaryhasaninletvelocityofunityintheupwardaxialdirectionandtheright boundary hasoutflowboundaryconditions.The topboundary actsasa rigidwall withno-slipandimpermeable surface.

The densityratioandviscosityratiois10 withthe Laplacenumberofthebubble,La= ρμD2σ =0.048.Theincomingaxial velocitydragsthebubbleandflattens itagainstthetopwall,stretchingitintheaxialdirectionconsiderably(see Fig.10).

Even though the fluid interface undergoes drastic change in its shape, the volume is conserved with a high degree of accuracy.

AnaccuratelycomputeddivergencefreevelocityfieldisessentialforaccuratevolumeconservationinVOFadvection.In numericalsimulations ofincompressiblemultiphaseflows usingprojectionmethod,thisisachievedbysolving apressure Poissonequation toahighdegreeofaccuracyby settinga smalltolerancevalueforsolvingthe matrixsystem.Theorder

(12)

oferrorinthedivergencefree conditionforvelocityisoftheorderofthesettolerancevalue.Solving Poissonequation is the mostexpensivestep inthe entiresimulation [2], thus studyingthe effectoflimiting toleranceforPoisson solveron theconservationofvolumeinVOFmethodsisrelevanttodeterminetheaccuracy ofthe simulation.Using theabovetest caseforagridsizeof 1281 andtimestepof5.104,westudytheeffectofaccuracyofdiscretized velocityfieldinmass conservationofthetwoVOFadvectionschemespresentedinthispaper.Thedivergenceofthevelocityfieldisthesumof eachdiscretized valueofdivergenceofvelocityatthecellcentergivenby(14) foreachcomputationalcell.

Table 6gives theeffect ofdivergenceof discretized velocityfield in volume conservationforVOFadvectionschemes.

SussmanandPuckett’soperator splitapproachresults infairlyuniformerrorinvolume conservationforlowertolerances sincetheerrorduetoclippingandfillingofthecellstomaintaintheboundednessofvolumefractionfieldcontributesthe mosttothe error.Asthe divergencefreenature ofthevelocity field deterioratestheerrorinvolume conservationscales downwiththetolerance.DuetotheverysmallgridCourantnumberusedinthesimulationeven forahighly divergence freevelocity fieldthereissmallerrorinthevolumeconservationwhenWeymouthandYue’soperatorsplittingmethodis employed.TherelativeerrorscalesuniformlywiththetoleranceandisbetterthanSussmanandPuckett’smethod.

4. Conclusions

In the presentwork, we havepresented severalimprovements for the implementationof volume of fluid methodin axisymmetriccoordinates.Wehavepresentedanalyticalrelationsforthereconstructionofpiecewiselinearinterfaceinax- isymmetriccoordinatessimilartothosegivenbyScardovelliandZaleski [12] forcartesiancoordinates.Theproposedscheme substantially reducesthecomputationalcostin comparisonto theiterativeschemes usuallyemployed forthereconstruc- tion. Further, we showed that for axisymmetric coordinate system, machine-precision advectionof volume fraction field canbeachieved.Weillustratedtheimprovementsbycomparingtheresultswiththepopularopensourcemultiphaseflow solverGerris.Finally,wewouldliketonotethatsimilarmodificationsintheadvectionschemeforvolumeoffluidmethod inother curvillinearcoordinatesystems(suchasellipticcoordinates)canbederived usingtheapproachpresentedinthis work.

CRediTauthorshipcontributionstatement

AnanthanMohan:

Referensi

Dokumen terkait