Fluent优化技术培训

ANSYSFLUENT优化技术培训姓名：宋述军公司：安世亚太科技（北京）有限公司Preface•ThereleaseoftheadjointsolverinFluent13istheculminationofseveralyearsofR&Deffort.•Thisprojectwasrisky,buttherewardsaregreat–bothforANSYSanditsclients.•Therewereanumberoffalsestartsanddead-ends.•Writinganadjointsolverthatmeetstheneedsoftheengineeringcommunityisnotatrivialtask.•Wearepleasedtohavecomesofar,andlookforwardtogoingmuchfurther.KeyIdeas-FundamentalsWhatdoesanadjointsolverdo?•Anadjointsolverprovidesspecificinformationaboutafluidsystemthatisverydifficulttogatherotherwise.•Anadjointsolvercanbeusedtocomputethederivativeofanengineeringquantitywithrespecttoalloftheinputsforthesystem.•ForexampleDerivativeofdragwithrespecttotheshapeofavehicle.Derivativeoftotalpressuredropwithrespecttotheshapeoftheflowpath.KeyIdeas-FundamentalsFLOWSOLVERInputs•Boundarymesh•Interiormesh•Materialproperties•Boundarycondition1•Flowangle•Inletvelocity•…•…Outputs•Fielddata•Contourplots•Vectorplots•xy-plots•Scalarvalues•Lift•Drag•TotalpressuredropHigh-level“system”viewofaconventionalflowsolverHOWARECHANGESTOKEYOUTPUTSDEPENDENTONCHANGESTOTHEINPUTS?KeyIdeas-FundamentalsFlowsolutionq,andtheinputstotheproblemc.);(cqJOutput0);(cqRResidualsoftheNavier-Stokesequations0ccRqqRLinearizedNavier-StokesChangetheinputscbycccJqqJJChangeinoutputEngineeraparticularlinearcombinationoflinearizedNSequations–introducemultipliers0~~ccRqqqRqTTq~qJqqRT~AdjointequationsAftersubstitutionsccRqcJJT~ChangeininputsChangeinoutput•Topleveladjointsolveralgorithm–quasi-Newtonscheme•SolveusingAMG•Involvesafamiliarprocess–Solutionadvancementcontrols–Courantnumber,under-relaxation–Residuals&iterations–RoughlythesameeffortasaconventionalflowsolutionKeyIdeas–AdjointEquationsThisinnocent-lookingsystemofequationsisattheheartoftheadjointsolver.ConstructandsolveanadjointproblemtogetahelpersolutionApproximatelinearizationwithpseudo-timestepintroduced.Similartooriginalflowsolver,buttransposed.UpdatetotheadjointsolutionSystemJacobian(transposed)DerivativeofachosenobservablewithrespecttotheflowsolutionqJqqRT~qqRqJqMT~~.KeyIdeas-WorkflowWorkflow•Solvetheflowequationsandpost-processtheresultsasusual.•Pickanobservationthatisofengineeringinterest.Lift,drag,totalpressuredrop?•SetupandsolvetheadjointproblemforthisobservationDefinesolutionadvancementcontrolsSetconvergencecriteriaInitializeIteratetoconvergence•Post-processtheadjointsolutiontogetShapesensitivitySensitivitytoboundaryconditionsettingsContour&vectorplots•ModifythemeshusingameshdeformationapproachMeshmorphingKeyIdeas–ShapeSensitivityShapesensitivity:Sensitivityoftheobservedvaluewithrespectto(boundary)gridnodelocationsmeshnnxwDrag.)(Shapesensitivitycoefficients:VectorfielddefinedonmeshnodesNodedisplacementVisualizationofshapesensitivity•Usesvectorfieldvisualization.•Identifiesregionsofhighandlowsensitivity.•Thesearetheplaceswherechangestotheshapecanhaveabigimpactonthequantityofinterest.•Theguidanceisspecifictothequantityofinterest,andthecurrentflowstate.DragsensitivityforNACA0012KeyIdeas–GradientAlgorithm&OptimizationHowtochoosethenodedisplacementfield?•Steepestdescent•Displacementproportionaltothelocalsensitivity•Providesafirst-orderestimateoftheeffectofthechangebeforethechangeisevermade.meshnnwwDrag.)(FirstorderestimateofthechangeScalingfactorforthedeformation(userchooses)ChosennodedisplacementsnnwxmeshnnxwDrag.)(KeyIdeas–MeshMorphingFlowCompletingthedesigncycleMeshMorphing•Sensitivityoflifttosurfaceshape•UseBernsteinpolynomial-basedmorphingscheme•Adjointtodeformationoperation•Surfaceshapesensitivitybecomescontrolpointsensitivity•Benefitofthisapproachistwo-foldSmoothsthesurfacesensitivityfieldProvidesasmoothinteriormeshdeformation•SelectportionsofthegeometrytobemodifiedKeyIdeas–MeshMoprhingActualchange3.1P=-213.8Totalimprovementof8%Constrainedmotion•Somewallswithinthecontrolvolumemaybeconstrainednottomove.•Aminimaladjustmentismadetothecontrol-pointsensitivityfieldsothatdeformationofthewalliseliminated.Castasaleast-squaresproblem.KeyIdeas–MeshAdaptationSolution-basedmeshadaptation•Detailsnotpresentedhere•Regionsintheflowdomainwheretheadjointsolutionislargeissusceptibletohavingastrongeffectofdiscretizationerrorsonthequantityofinterest.•AdaptinregionswheretheadjointsolutionislargeAdjointsolutionDragsensitivityBaselineMeshAdaptedMeshAdaptedMeshDetailKeyIdeasWhathavewelearnedsofar?•Anadjointsolvercanbeusedtocomputethederivativeofachosenobservationofengineeringinterestwithrespecttoalltheinputdataforthesystem.•Theadjointequationsformalinearsystem.•Solvinganadjointproblemisnottrivial–aboutasmucheffortasaflowsolution.•Theadjointsolutionprovidesguidanceontheoptimaladjustmentthatwillimproveasystem’sperformance.•Anadjointsolutioncanbeusedtoestimatetheeffectofachangepriortoactuallymakingthechange.•Shapesensitivitydatacanbecombinedwithmeshmorphingtoguidesmoothmeshdeformations.•Anadjointsolutioncanbeusedaspartofagradient-basedoptimizationalgorithm.•Anadjointsolutionc