Sie sind auf Seite 1von 38

2011ANSYS,Inc. May1,2012 1 Release14.

0
14.0Release
SolvingFSIApplicationsUsing
ANSYSMechanicalandANSYSFluent
Lecture6
CosimulationviaSystemCoupling
PartIV
2011ANSYS,Inc. May1,2012 2 Release14.0
Outline
Convergence
ThislecturefocusesonconvergingSystemCouplingsimulations.
Therearemanycontrolsthatcanimpactconvergence,sowellstart
byidentifyingthecontrolsavailable.
Obtaininggoodconvergencerequiresamethodicalworkflow.Well
discuss:
Usingseparatefluidandstructuralsimulationstofirstidentify
anynonFSIproblems
CorrectinitializationforFSIcases
Monitoringconvergence
Convergencecontrolsandtheireffect
IdentifyingunstableFSIcaseandhowtostabilizethesolution
Generalconvergenceadviceanddebugtools
2011ANSYS,Inc. May1,2012 3 Release14.0
Convergencerequiresanunderstandingofthesolution
processandthephysicsthatyoursetupdefines
Thismaybedifferentthantherealworldphysics
E.g.yourinitialconditionsmaybedifferent
Whenconvergencedifficultiesareencountered,follow
thesolutionprocess:
Usemonitorpointsandbackupfilestoanswersthese
questions
Convergence
How does
the flow field
respond?
What is the
force passed to
the structure?
How does
the structure
respond?
What are the
displacements
passed back
to Fluent?
2011ANSYS,Inc. May1,2012 4 Release14.0
Thereareanumberofcontrolsthatplayaroleingood
convergence,including:
Initialization
TimeStepSize
StabilityControls
NumberofFluentiterationsperstep
NumberofCouplingIterations
InterfaceUnderRelaxationFactors
Itisimportanttounderstandwhyacasehas
convergenceproblemsbeforeadjustingthesecontrols
Convergence
2011ANSYS,Inc. May1,2012 5 Release14.0
BeforestartinganFSIsimulation,youshouldrun
separatefluidsandstructuralsimulationsusing
representativeloads:
Checksthatthebasicsetupiscorrectforeachside
UsesettingsthatareascloseaspossibletotheplannedFSI
simulation(sametimestepsize,etc)
StructuralsimulationusingafixedPressureboundary
AFluentsimulationwithafixedboundaryorasimplymoving
boundary
Ifitsdifficulttoproviderepresentativefluidloads,
couldrunaSystemCouplingsimulationwithloads
passed1wayfromFluenttothestructure
Convergence
2011ANSYS,Inc. May1,2012 6 Release14.0
Next,considerhowtoinitializeyourFSIsimulation.Well
consideratransient2wayFSIsimulationhere:
Option1:Noprevioussteadystateresultsprovided
TheinitialFluentsolutioncomesfrompressureandvelocityvalues
setinFluent
Thestructureisnotprestressed(zeroinitialdisplacement)
Suitablewhentheinitialfluidpressurefieldisknownand
canbespecifiedaccurately(typicallyzero)
Noinitialforcepassedtothestructure
MakeuseoftheOperatingPressuresettinginFluent
Nootherloadsonthestructurethatrequireprestressing
Convergence Initialization
2011ANSYS,Inc. May1,2012 7 Release14.0
Convergence Initialization
Option2:SteadystateFluentresultsprovided
TheinitialFluentsolutioncomesfromaprevioussteadystaterun
Thestructureisnotprestressed(zeroinitialdisplacement)
SuitablewhentheinitialfluidforceontheFSIinterfacefromthe
steadystateflowresultswillnotcausesuddendeformationofthe
structureatthestartofthetransient,causingthesolverstofail
Nootherloadsonthe
structurethatrequire
prestressing
IftransientFluentresultsareprovided
asaninitialsolution,thetimemustbe
resettozerotobeconsistentwiththe
initialSystemCouplingtimeusing:
(rpsetvar flow-time 0)
2011ANSYS,Inc. May1,2012 8 Release14.0
Convergence Initialization
Option3:SteadystateFSIresultsprovided
TheinitialFluentandstructuralsolutionscomefromaconverged
steadystate2wayFSIsimulation
Thisprovidesthemostconsistentstartingpointfora
transient
Suitablewhenyouneedtoprestress/deformthe
structuretoproduceasmoothstarttothetransient
RequiresaSteady/StatictoTransientrestart,whichisnot
currentlysupported
ButcanrunaninitialtransientSystemCouplingcasethat
behaveslikeaSteady/Staticcase
2011ANSYS,Inc. May1,2012 9 Release14.0
TomimicaSteady/Staticcase:
SetTimeIntegrationoffintheTransientStructuralsystem
EquivalenttoaStaticStructuralcase
Fluentmustremainintransientmode
Canbehavelikeasteadystatesolverbyusingthepressurebased
coupledsolverwith1iterationpertimestep
UnderSolutionControlssetaFlowCourantNumberof1e6orhigher
andExplicitRelaxationFactorsof1forMomentumandPressure
SettheTimeStepSizeasapseudotransienttimestep
Solve,thenrestartafterturningonTimeIntegrationand
adjustingtheFluentsettingstoresolvethetruetransient
SeeRestartssection
Convergence Initialization
2011ANSYS,Inc. May1,2012 10 Release14.0
Option2(steadystateflowsolution)isthemostcommon
waytoinitializeatransientFSIcase
Typicallyprovidesanonzeroforcetothestructureatthe
startofthetransient
ThismaybeOK,producingasmallbumpatthestartofthe
transientthatcanbeignoredwhenpostprocessingtheresults
ItmaynotbeOKwhen:
Thefluidis(oralmostis)incompressible,particularlywhen
combinedwithasmalltimestep
Evensmallbumpscancauselargepressurechanges

Convergence Initialization
2011ANSYS,Inc. May1,2012 11 Release14.0
ItmaynotbeOKwhen:
Thestructureisveryflexible
TheinitialforceproduceslargedeformationsthatcauseFluentto
failorproducelargepressurechangesthatarepassedbacktothe
structure
Theinitialbumpcausesstructuralvibrationsthattakealong
timetodieout
Stiffnessproportionaldamping(BETAdamping)canbeusedto
dampouttheunwantedhighfrequencystructuralresponse
Forceanddisplacementmonitorpointshelpshow
whatshappeningatthestartoftherun
Remembertotrackmonitordataeveryiterationsoyoucansee
howtheforce/displacementrespondateachCoupling
Iteration
Convergence Initialization
2011ANSYS,Inc. May1,2012 12 Release14.0
Forstartupproblemsgobacktothequestionsoutlined
earlier:
Whataretheinitialforcespassedtothestructure?
Iftheresalargestepchangeintheforces,howwouldyouexpect
thestructuretorespond,givenyourtimestep?
Whatdisplacementscanyouexpect?
Howwouldyouexpectthefluidsystemtorespondtothose
displacements,givenyourtimestep?
Transientstartuprobustnessproblemsareoftenmade
worsebyreducingthetimestepsize
Foragivendeformation,thefluidhaslesstimetorespond>
largeracceleration,largerpressurechanges
Convergence Initialization
2011ANSYS,Inc. May1,2012 13 Release14.0
ThedefaultDataTransferconvergencechartprovidesagoodoverview
oftheforce/displacementconvergenceattheFSIinterface(s)
Thesawtoothpatternisdesirable,showingconvergencewithineach
timestep,thentypicallyajumpatthestartofthenexttimestep
Bydefault,thenormalizedchangeinadatatransfermustbelessthan
1%(i.e.2ontheLogscale)tobeconverged
MonitoringConvergence
Note:thischartshows
CouplingSteponthex
axis,soitisclearthat
convergenceisachieved
withineachstep.Coupling
Iterationisthedefault.
2011ANSYS,Inc. May1,2012 14 Release14.0
Asalreadydiscussed,alsomonitor:
TheFluentsolveroutputtoassessflowfield
convergence
TheMechanicalsolveroutputtoassessstructural
convergenceandthereceivedforces
Fluentmonitorpointsforforceanddisplacementat
eachiteration
MonitoringConvergence
2011ANSYS,Inc. May1,2012 15 Release14.0
Thereareanumberofconvergencecontrols,soitseasy
toheaddownthewrongpath
Ingeneralstartwith:
Default#ofCouplingIterations:Min:1,Max5(fortransients)
DefaultDataTransferUnderRelaxationFactor:1
Perhaps5FluentiterationsperCouplingIteration(for
transients),butthisiscasedependent
ATimeStepSizebasedonthephysicsyouneedtoresolve
1/20
th
ofthehigheststructuralfrequencyofinterest
FlowfieldCourantnumber
Fromthisstartingpointyoucanadjustthecontrols
basedonthesolutionbehavior
ConvergenceControls
2011ANSYS,Inc. May1,2012 16 Release14.0
AdjustthenumberofCouplingIterationswhen:
Acaseisconvergingwell,withforcesand/or
displacementsmovingtowardsasteadyvaluewithina
stepwithoutoscillating
YoujustneedafewmoreCouplingIterationstoreachtheData
TransferConvergenceTarget
Butconsiderreducingthetimestepfirstfortransientcases
Steadystatecasestypicallyneedmorethan5iterations,
assumingyouareusing1step
Alternativelyusing1iterationandmanysteps
Usuallytheresnoneedtochangethiscontrolfortransient
cases(donotmakethisthefirstthingyouchange)
ConvergenceControls CouplingIterations
2011ANSYS,Inc. May1,2012 17 Release14.0
ConvergenceControls URF
AdjusttheDataTransferUnderRelaxationFactorwhen:
Youarerunningasteadystatecase
Shouldmovegraduallytowardsasteadystatesolution
AtypicalURFmightbe0.5
Fortransientcasesusethedefaultvalueof1ingeneral
Thephysicalinertiaofthesystemprovidesstabilitythatisnot
thereinsteadystatecases
Sometransientcasesareunstable,withforces/displacements
oscillatingwithinatimestep
IfyoucanachievestabilitybyreducingtheURFslightly,sayno
lowerthan0.75,thenthisisacceptable
Otherwiseusethestabilitycontrolsdiscussedlater
2011ANSYS,Inc. May1,2012 18 Release14.0
Keepthefollowingformulainmindifyouadjustthe
UnderRelaxationFactor:
% LoadReceived=(1 (1 URF)^
Max.CouplingIters
)*100%
Forexample,givenaURFof0.25andamaximumof5
CouplingIterations,wehave:
% LoadReceived=(1 (1 0.25)^
5
)*100%=76%
Sobytheendofthestep,only76%ofthechangeina
loadwillhavebeenreceivedbythereceivingsolver
Assumingastepchangeintheload,whichthenremains
constant
ConvergenceControls URF
2011ANSYS,Inc. May1,2012 19 Release14.0
JudgingthenumberofFluentiterationsto
useperCouplingIteration(notperTime
StepasshownintheGUI)requiresyouto
watchtheforcemonitorpointbehavior.
ConvergenceControls FluentIterations
ToomanyFluentiterationsper
CouplingStep wastedCPU.A
reasonableforceforthis
CouplingIterationcouldbe
obtainedinfeweriterations.
1Timestepshown
2011ANSYS,Inc. May1,2012 20 Release14.0
ToofewFluentiterations
perCouplingIteration will
needtousemoreCoupling
Iterationunnecessarily
Theforceisincreasing
steadilywithineach
CouplingIterationandfrom
oneCouplingIterationto
thenext allowitto
increasefurtherbefore
exchangingdatawiththe
MAPDLsolver
ConvergenceControls FluentIterations
NoticeitsnotobviouswhereeachCouplingIterationstarts/ends
5 CouplingIterationsperstepwereusedhere
Timestep1
Timestep2 Timestep3
2011ANSYS,Inc. May1,2012 21 Release14.0
Reasonablenumber
ofFluentiterations
Theforceisjust
startingtoflatten
outwithineach
CouplingIteration
ConvergenceControls FluentIterations
C.Iter 1 C.Iter 2 C.Iter 3 C.Iter 4 C.Iter 5
1TimeStep
TheresnopointinfullyconvergingwithineachCoupling
Iteration,sincetheforcewillchangeinthenextone
MoreFluentiterationsinthefirstCouplingIterationwouldhavecaused
theforcetomovefurtherawayfromtheconvergedvalue
Less Fluentiterationscanhelpstability keepstightercoupling
betweenMAPDLandFluentbynotlettingtheforcemovetoofarbefore
updatingthedisplacements
2011ANSYS,Inc. May1,2012 22 Release14.0
TimeStepSizeguidelinesaregenerallythesameasforthe
individualsolvers;afewthingstowatchfor:
Makesuretheindividualsolversconvergewellusingthe
selectedtimestep
Asnotedearlier,ifyouhaveabumpatthestartofa
transient,asmallertimestepcanmakethisworse,not
better
Iftheforces/displacementsarechangingalotfromone
timesteptothenext,considerreducingthetimestep
buttoosmallatimestepmaystarttoresolvehigh
frequencystructuralmodesthatarenotofinterest
ConvergenceControls TimeStepSize
2011ANSYS,Inc. May1,2012 23 Release14.0
SomeFSIcasesareunstablenomatterwhatsettingare
usedforthepreviousconvergencecontrols
Unstablecasestypicallyincludeoneormoreofthe
following:
Thinstructures
E.g.sheetmetal,diaphragms,membranes
StructuralmaterialswithalowYoungsModulus
E.g.hyperelasticmaterials,rubber
Gasesmodeledasconstantdensity(incompressible)
orliquids
ConvergenceStabilization
2011ANSYS,Inc. May1,2012 24 Release14.0
Toidentifyanunstablecaselookatthe
force/displacementmonitorpointsagain
Oscillationwithin atimestep,fromoneCouplingIteration
tothenext,isindicativeofFSIinterfaceinstability
ConvergenceStabilization
1TimeStep
Mechanical
solverfailed
Pressurevalues
veryhigh
2011ANSYS,Inc. May1,2012 25 Release14.0
Theoscillationswilldevelopinthefirsttimestep
ThecasemayfailwithininthefirstfewCoupling
Iterations,oritmaymakeittothesecondorthirdtime
step,withtheoscillationsgrowinginmagnitude
Itisimportanttodistinguishandunstablecasefroma
poorlyinitializedcase
Apoorlyinitializedcaseshowsasuddenjumpatthestart
oftherun
Improvetheinitializationifthiscausesfailure
Anunstablecasestartswithsmalloscillationsthatgrow
Butwithhighlyunstablecasesyoumayonlyreachthe2
nd
or
3
rd
CouplingIteration,soitcanresembleasuddenjump
ConvergenceStabilization
2011ANSYS,Inc. May1,2012 26 Release14.0
InstabilityMechanism:
ConvergenceStabilization
Fluent Mechanical
Initialpressure
onFSIinterface
Forcespassed
toMechanical
Fluent Mechanical
P
Theinterfacedeforms
inMechanicalduefluid
pressure
Fluent
Mechanical
P
Pressure
decreasesin
Fluentdueto
increasein
volume
Fluent Mechanical
Theinterface
deformsin
Mechanicalinthe
oppositedirection
Fluent
Mechanical
P
PressureinFluent
increasesmore
duetoreduced
volume
Displacements
passedtoFluent
Forcespassed
toMechanical
Displacements
passedtoFluent
2011ANSYS,Inc. May1,2012 27 Release14.0
Tostabilizethesetypesofcases,tworpvars areavailable
intheFluentTUI:
(rpsetvar dynamesh/sc-bc-compressibility-type <value>)
(rpsetvar dynamesh/sc-bc-compressibility <value>)
Userpgetvar toquerythecurrentvalues:
(rpgetvar dynamesh/sc-bc-compressibility-type)
(rpgetvar dynamesh/sc-bc-compressibility)
ConvergenceStabilization
2011ANSYS,Inc. May1,2012 28 Release14.0
Fromanumericalpointofview,theserpvars modify
thecoefficientsintheFluentlinearsolutionmatrixfor
thecontinuityequationforcellsattachedtoanFSI
interface
Fromaphysicalpointofview,thisslowsdownthe
pressureresponseinFluentattheFSIinterface
Thismeansthepressuredoesntincrease/decreaseasfastin
Fluentwhenanewdisplacementisreceived
Thisdampedpressureresponseallowsustomovetowardsa
convergedsolutioninacontrolledway,ratherthanoscillating
around
Therpvars havenoeffectonconverged results
ConvergenceStabilization
2011ANSYS,Inc. May1,2012 29 Release14.0
Thesc-bc-compressibility-type rpvar defaultsto0,
whichmeansoff
Valuesof1and2arevalidandrepresentdifferentwaysto
adjustthematrixcoefficients
Type2hasprovenmostreliableandisalwaysrecommended
Thesc-bc-compressibility rpvar isascalingfactor
forthematrixcoefficientsanddefaultsto0,which
meansnoadjustmentismade
Values>0willadddampingwhenthetypeissetto1or2
Thelargerthevaluethemoredamping
Theappropriatevalueiscasespecific,butisgenerallyinthe
range0.01 500
ConvergenceStabilization
2011ANSYS,Inc. May1,2012 30 Release14.0
Lookingathowthedampingvaluechangesthe
pressureresponse:
ConvergenceStabilization
(rpsetvar 'dynamesh/sc-bc-compressibility 0)
Baselinedivergentcase
(rpsetvar 'dynamesh/sc-bc-compressibility 1)
Stilldivergent,butcompletesmore
CouplingIterations
Noticethepressureresponseisno
longerastepchange
1TimeStep
1TimeStep
2011ANSYS,Inc. May1,2012 31 Release14.0
ConvergenceStabilization
(rpsetvar 'dynamesh/sc-bc-compressibility 10)
Nowstable,butoverdamped.
Notquitefullyconvergedwithinthe
timestepafter7CouplingIterations.
1TimeStep
(rpsetvar 'dynamesh/sc-bc-compressibility 2)
Nolongerdiverges,butstillunder
damped.Didnotfullyconvergeafter
7CouplingIterations.
1TimeStep
2011ANSYS,Inc. May1,2012 32 Release14.0
Forthecaseshownavalueof3wasthemagicnumber,
butthiswillbedifferentforothercases
Alwaysstartwithsmallvaluesthenincrease
Thefirstvalueyoutryshouldstilldiverge
Avoidsfallingintoatrapwheretheresponseissoover
dampedthatitappearsflat(i.e.converged!)
ConvergenceStabilization
(rpsetvar 'dynamesh/sc-bc-compressibility 3)
Goodresponse.Wellconvergedafter
5CouplingIterations.
1TimeStep
2011ANSYS,Inc. May1,2012 33 Release14.0
Notethattheresponseisafunctionoftherpvar value
ANDthenumberofFluentiterations
Considerthecaseshown ifyouperformmoreFluent
iterationsperCouplingIterationeachoscillationwouldextend
further(moreunstable)
Inpracticesetthenum.ofFluentiterationstoasmallnumber
andthenadjusttherpvar togettheoptimalresponse
TheresnopointinusinglotsofFluentiterations
ConvergenceStabilization
1TimeStep
2011ANSYS,Inc. May1,2012 34 Release14.0
See2022119intheKnowledgeResourcesontheANSYSCustomerPortal
forarecordeddemoofstabilizinganunstableFSIcase(justsearchFSI)
ConvergenceStabilization
2011ANSYS,Inc. May1,2012 35 Release14.0
Anumberofredflagsthatmayindicateyouareonthe
wrongpathtoimprovingconvergence:
UseofalargenumberofCouplingIterationspertimestep
Useofsmallinterfaceunderrelaxationfactorsintransient
cases
Atimestepsizethatismuchsmallerthanthatrequiredbythe
fluidonlyorstructuralonlysimulations
AlargenumberofFluentiterationsperstep,whencompared
withafluidonlysimulation
ConvergenceAdvice
2011ANSYS,Inc. May1,2012 36 Release14.0
ConvergenceAdvice
Iftherunfails:
Findoutwhichsolverfailedfirst lookatall theoutput
filestodeterminewheretheerrorstarted
Checkthemapping
IfMAPDL,werereasonableforcespassed?
IfFluent,werereasonabledisplacementsreceived?
Weretheindividualfieldsolversconverging?
Ifarunfailsinthefirstfewtimesteps,rerunforalimited
numberoftimestepsand/orwithalimitednumberof
CouplingIterationssothatyoucangenerateresultsfiles
andexaminetheresults
2011ANSYS,Inc. May1,2012 37 Release14.0
Canrequestadditiondebugoutput
inSCunderExecutionControl
Usefulfordebugging:
Mappingproblems
Toseetheforce/displacementvalues
thataretransferred
Simulationsthatcrash
Findoutwhichsolverencountered
problemsfirst,whattheerrorswere
DebugOutput
2011ANSYS,Inc. May1,2012 38 Release14.0
Summary
Obtaininggoodconvergencerequiresagoodunderstandingof
thedatatransferprocessbetweenparticipantsolversandhow
eachsolverwillreacttothenewboundarydatareceived.
ThoughtneedstobegiventohowSystemCouplingcasesare
initialized,particularlywhentheinitialfluidfieldwillcause
significantstructuraldeformations
Relyingonerrormessagesoradjustingcontrolswithout
understandingwhyasolverfailedisunlikelytoresolve
problems
Examiningmonitordataandbackupfilesareimportanttoolsto
guideparameterchangestoimproveconvergence
MonitordataisalsousedtoidentifyunstableFSIcases.rpvars
inFluentcanbeusedtostabilizethesecases

Das könnte Ihnen auch gefallen