




版權說明:本文檔由用戶提供并上傳,收益歸屬內容提供方,若內容存在侵權,請進行舉報或認領
文檔簡介
TheOptimizationModule?Copyright2014COMSOL.Anyoftheimages,text,andequationsheremaybecopiedandmodifiedforyourowninternaluse.Alltrademarksarethepropertyoftheirrespectiveowners.See/trademarks.OverviewTheCOMSOL?4.4ProductSuiteTheOptimizationModuleAdd-ontoCOMSOLMultiphysicsGeneral-purposepackage,notapplication-specificOptimizationforElectrical,Mechanical,Fluid,andChemicaldevicesandprocessesCombineswithanyCOMSOLMultiphysicsModuleTheOptimizationModuleContains:UserinterfacesforsettingupoptimizationtasksOptimizationsolversExamplemodelsfromdifferentfieldsWhatisOptimization?TheimprovementofanobjectivefunctionbychangingcontrolvariableswhilemaintainingasetofconstraintsObjectivefunction(costfunction)Anyreal-valuedscalaroutputfromasolvedCOMSOLmodelMass,displacement,pressuredrop,resistance,power,etc...Controlvariables(optimizationvariables,designvariables,...)Anycontinuoussetofreal-valuedinputstothemodelDimensions,materialdistribution,materialproperties,etc...ConstraintsAnyequalityorinequalityconditionthatcanbeexpressedintermsofthesolvedmodel,orthecontrolvariablesPeakstress,minimumsize,maximumtemperature,etc...QuickIntroductiontoOptimizationintheCOMSOLDesktopHowdoweoptimize?StartwithasolvedCOMSOLmodelDefineyourobjectfunction:Whatdoyouwanttomakebetter?Choosethedesignvariables:Whatdoyouwanttochange?Specifyyourconstraints:Whatlimitscannotbeexceeded?Optimize!Tuningfork,Desiredresonanceat440HzStartwithanexistingmodelForclarity,theCOMSOLDesktopishereshownasitappearsonalow-resolutionscreen.AddtheOptimizationStudyfeatureSelectfromasuiteofoptimizationalgorithmsSettoleranceandthemaximumnumberofmodelevaluationsIfsolvingonacluster,candistributesolutionsandruninparallelSpecifyanobjectivefunctionfreq:theresonancecomputedbytheeigenfrequencysolverTheobjectiveistomaketheresonantfrequency440HzMinimize,maximum,maximizetheminimum,minimizethemaximum,andhandlesumsofobjectivefunctionsPickcontrolvariablesandconstraintsGlobalParametersetsthetuningforklength,specifyinitialvalueChooseupperandlowerbounds(constraints)onthecontrolvariablesCanaddotherconstraints,ifdesiredSolve,andthenextractsolutionOptimizationTutorialsandExamplesGettingStartedExample:TuningForkAdjustthelengthofthearmsofatuningforksuchthatthefirstresonanceisat440HzApproximategradient,sincethemodelisremeshedduringtheoptimizationBOBYQASolverCanalsobedoneviatheLiveLink?productsandcontrolledfromvariousCADsoftware,MATLAB?,andExcel?./model/tuning-fork-computing-the-eigenfrequency-and-eigenmode-8499SizeOptimization:
BanddispersioninamicrochannelMinimizethedifferenceistransittimebetweeninsideandoutsideSizeOptimization:3DFlywheelFindholeradiiwhichminimizeflywheelmassMaximumvonMisesstressmustnotexceedyieldlimit/blogs/using-gradient-free-optimization/SizeOptimizationofaFlywheel,
withAdvancedConstraintsMakestressdistributionasuniformalongtheradiusaspossibleConstrainthemassnottochangeConstrainthemomentofinertianottochangeGradient-basedapproach/model/optimizing-a-flywheel-profile-4356SizeOptimization:
MinimizingS11ofanAntenna/model/optimizing-a-bowtie-antenna-14435ChangeFlareAngleandArmHeightGradient-FreeApproach,becauseremeshingisrequiredShapeOptimization:
OptimizinganAcousticHornMaximizethesoundintensityalongtheaxisofthehornTheshapeofthehornisdescribedbyasumofsinewavesTruncatedFourierseriesTheMovingMeshfunctionalityisusedtoavoidremeshingthedomainGradient-Basedapproach/model/optimizing-the-shape-of-a-horn-4353ParameterEstimation:Findingthematerialpropertiesbaseduponexperimentaldatahttp:///model/transient-optimization-fitting-material-properties-of-a-wall-10905http:///model/curve-fitting-material-model-data-to-experimental-data-5886http:///model/determining-arrhenius-parameters-using-parameter-estimation-10305/model/degradation-of-dna-in-plasma-1391TransientanalysisMinimizeleastsquaresdifferencewithexperimentalresultsLevenberg-MarquardtInverseProblem:ImagingImagingofsoilpropertiesbasedonpumpingexperimentsFindthepermeabilitypatternmostconsistentwithexperimentaldataReconstructinteriordatabaseduponobservationsfromtheexterior/model/aquifer-characterization-through-inverse-modeling-from-pump-tests-4410TopologyOptimization:PorousCatalystObjectiveistomaximizethereactionratewhileminimizingtheamountofcatalystwithinthereactorInitialcatalystdistributionishomogeneousTopologyoptimizationadjustsamountofcatalystwithineachmeshelement/model/optimization-of-a-catalytic-microreactor-4401TopologyOptimization:TelsaMicrovalveMaximizeratioofflowlefttorightcomparedtoflowrighttoleftforsamepressuredrop./model/topological-optimization-of-a-tesla-microvalve-14513TopologyOptimization:
MinimizeBeamCompliance/model/topology-optimization-7428MinimizethecomplianceAddconstraintontotalmaterialSIMPmethodMathematicalOptimization:MinimizeaFunctionχ1χ2
NotableCOMSOLConferencePapersonTopologyOptimizationTopologyOptimizationinMultiplePhysicsProblems,O.Sigmund,DTUMechanicalEngineering,/papers/1790/MultiphysicsTopologyOptimizationofHeatTransferandFluidFlowSystems,E.Dede,ToyotaResearchInstituteofNorthAmerica,/papers/6282/SimulationofTopologyOptimizedElectrothermalMicrogrippers,O.Sardan,D.Petersen,O.Sigmund,&P.Boggild,DTUMechanicalEngineering,/papers/5346/ImplementationofStructuralTopologyOptimizationinCOMSOL,B.Lemke,Z.Liu,&J.G.Korvink,DepartmentofMicrosystemsEngineering,UniversityofFreiburg,/papers/1543/TopologyOptimizationofDielectricMetamaterialsBasedontheLevelSetMethodUsingCOMSOLMultiphysics,M.Otomori&S.Nishiwaki,KyotoUniversity,/papers/12519/IndustrySuccessStory:TopologyOptimizationLeadstoBetterCoolingExtractfromCOMSOLNews2012.?2012COMSOL.Allrightsreserved.SolidWorksisaregisteredtrademarkofDassaultSystèmesSolidWorksCorporationoritsparent,affiliates,orsubsidiaries.AluminumcoldplatewithouthierarchicalmicrochanneltopologyAluminumcoldplatewithhierarchicalmicrochanneltopologyModeledaluminumcoldplatewithhierarchicalmicrochanneltopologyAluminumcoldplatesaremountedincarstocombatheatproblems,requiringoptimalcoolingchanneltopologyforminimizedplatesizePerformedCFDandHeatTransferanalysesinconjunctionwithLiveLinkTM
forSolidWorks?tocreateaprototypeusingoptimizedtopologyCOMSOLNews2012:EricDede,ToyotaResearchInstitute,AnnArbor,MIOptimizationModuleTheory,IntroductionTheOptimizationinterfacecandefinecontrolvariablefields,integralobjectives,andlocalconstraintsChangethespatialdistributionofmaterial,subjecttolocalbounds,tominimizemasswhileconstrainingsystemcomplianceThefamilyofoptimizationsolversOptimizationModuleSearchwithoutfindingGradientsAnalyticGradientMethodsMonte-CarloCoordinateSearchMMALevenberg-MarquardtNelder-MeadBOBYQARandomSearchSearchoneaxisatatime1storderapproximategradient2ndorderapproximategradientSNOPTLinearconvergenceQuadraticconvergenceLeast-Squaresproblemsonly(veryfast)ApproximateGradientMethodsWhatisthegradient?ThegradientisthederivativeoftheobjectivefunctionwithrespecttothecontrolvariablesGradientsoftheobjectivecanbecomputedeither:ApproximatelyCoordinatesearch:Finite-differenceinonecontrolvariableatatimeNelder-Mead:Evaluates(N+1)pointsofanN-dimensionaldesignspaceandconstructsasimplex,improvesworstpointBOBYQA:MakesprogressivelocalquadraticapproximationsAnalyticallyMMAandSNOPTusetheAdjointmethodtocomputeexactgradientCancomputegradientswithrespecttoallcontrolvariablesatonceRequiressmoothanddifferentiableobjectiveandconstraintfunctionsNoremeshingGradientbasedmethodsstartatapointwithinthedesignspaceandimprovethedesignDesignSpaceχ1χ2ObjectiveFunctionf(χ)(1)Startatanexistingdesign&computegradientdirection(2)Searchalonggradient&findminimumalongline(3)Repeatuntilnomoreimprovementispossible(1)(2)(3)ComparisonofAlgorithmsApproximateGradientAnalyticGradientObjectiveFunctionAnyscalaroutputMustbebothsmoothanddifferentiableDesignVariablesAnything,includinggeometricdimensionsAnythingthatdoesnotresultinremeshingofthegeometryRemeshingYesNoConstraintsCanonlyconstrainscalaroutputsConstraintsmustbedifferentiableandsmooth,butcanbeateachpointinspacePossibleanalysesAnycombinationofalldifferentanalysistypesAnystudywithonlyoneof:
Stationary,Transient,orFrequency-DomainRelative
PerformanceIncreasesexponentially
withthenumberofdesignvariablesPerformanceisnotverysensitivetothenumberofdesignvariablesTheMonte-Carlomethodsharesalloftheseproperties,butwillhavetheslowestconvergenceWhichoptimizationsolvertouse?Docontrolvariableschangethemesh?Constraintsonthesolution?Arethereanyconstraints?TopologyOptimization?SNOPTMMALevenberg-MarquardtBOBYQASmoothlyvaryingobjective?Nelder-Mead(orCoordinateSearch)MonteCarloYesYesYesYesYesNoNoNoNoVeryrandomNoisyOptimizationUserInterfacesTheOptimizationstudystepCentralcontrolpanelforalloptimizationChooseandtunesolversSpecifyglobalobjectivefunctions,controlparametersandconstraintsEnable/disablecontributionsfrominterfacesTheOptimizationinterfaceSetupgeneralobjectivecontributions,
includingleast-squaresDefinecontrolvariablefieldsSpecifygeneralconstraintsOptimizationModuleTheory,AdvancedDerivative-freesolversDirectsearchNelder-MeadCoordinatesearchTrustRegionBOBYQADerivative-FreeOptimizationSolversRequireonlyobjectivefunctionvalues,
noderivativesControlanything,includingtheCADgeometryRobustbutexpensiveParallelonclustersGradient-basedGeneral-purposeSNOPTMMALeast-squaresLevenberg-MarquardtGradient-BasedOptimizationSolversUsegradientinformationtocontrolsearchdirectionStationaryandtransientsolverscomputegradientsefficientlyManycontrolvariables(fields)CommonOptimizationTasksOptimizationTasksPureOptimizationOptimalDesignParameterSelectionGeometricOptimizationSizingShapeOptimizationTopologyOptimizationTargetMatchingInverseProblemsParameterEstimationImagingObjectiveFunctionsinCOMSOLGlobalObjectivesAcceptsanyglobalexpression–ofteninvolvingcouplingoperatorsExample:classicaloptimizationIntegralObjectivesIntegratesanexpressionoveradomain,boundary,edgeorpointExample:minimizingtheweightofastructureProbeObjectivesEvaluatesanexpressionatgivencoordinatesLeast-SquaresObjectivesComparesanexpressiontomeasuredvaluesinafileExample:fittingreactionconstantstomatchmeasuredconcentrationsControlVariablesinCOMSOLGlobalControlVariablesChooseanyexistingmodelparametersintheOptimizationstudystepWorkswithallsolversUsedwithderivative-freesolverstocontrolCADgeometryparametersControlVariableFieldsControlvariablesareassociatedwithpositionsinthegeometryFiniteelementinterpolationgivesmanydegreesoffreedomRequireusingtheOptimizationInterfaceOnlyworkswithgradient-basedsolversConstraintsinCOMSOLBoundsSetlimitsdirectlyonthecontrolvariables,oftenrequiredbysolversDesignconstraintsSetrelationsbetweenthecontrolvariablesDonotrequireevaluationofanyPDEsolutionPerformanceconstraintsSetconditionsonthePDEsolutionvariablesSameformatasanobjectivefunctionandasexpensivetoevaluateSensitivityEvaluationinCOMSOLSymbolicmathmachineryallowsefficientgradientevaluationAslongasallexpressionsaredifferentiableAdjointsensitivityFastevaluationofthegradientofanyobjectivefunctionwithrespecttoacontrolvariablefieldForwardsensitivityFastevaluationofthederivativeofaPDEsolutionfieldwithrespecttoindividualcontrolvariablesGradientofobjectivefunctionsiscomputedusingthechainruleNumericalgradientFall-backwhenexpressionsarenotsymbolicallydifferentiableTheSolversoftheOptimizationModuleAbouttheSNOPTSolverSNOPT=SparseNonlinearOPTimizerDevelopedbyP.E.Gill,W.MurrayandM.A.SaundersatStanfordUniversitySequentialQuadraticProgramming(SQP)methodSolvesasequenceofapproximatingquadraticprogrammingproblemswithlinearizedconstraintsOuterloopusesaquasi-NewtonstrategywhereanapproximateHessianisupdatedusinggradientsevaluatedatconsecutivestepsAbouttheMMASolverMMA=MethodofMovingAssymptotesDevelopedbyKristerSvanbergatKTH(RoyalInstituteofTechnology,Sweden)COMSOLversionismorespecificallyGCMMA=GloballyConvergentMMASolvesasequenceofconvexapproximationsAllowsgeneralnonlinearconstraintsEachapproximationisgeneratedfromlineardataatcurrentpointAllintermediatepointsarefeasible,unlessthefeasiblesetisemptyAbouttheLevenberg-MarquardtSolverSolverforleast-squaresproblemsRequiresanobjectivefunctiononleast-squaresformDoesnotallowanyconstraintsTrust-regionGauss-NewtonmethodComputesthegradientofeachterminthesumofsquaresseparatelyApproximatestheHessianfromfirst-orderderivativesonlyOften
溫馨提示
- 1. 本站所有資源如無特殊說明,都需要本地電腦安裝OFFICE2007和PDF閱讀器。圖紙軟件為CAD,CAXA,PROE,UG,SolidWorks等.壓縮文件請下載最新的WinRAR軟件解壓。
- 2. 本站的文檔不包含任何第三方提供的附件圖紙等,如果需要附件,請聯(lián)系上傳者。文件的所有權益歸上傳用戶所有。
- 3. 本站RAR壓縮包中若帶圖紙,網(wǎng)頁內容里面會有圖紙預覽,若沒有圖紙預覽就沒有圖紙。
- 4. 未經(jīng)權益所有人同意不得將文件中的內容挪作商業(yè)或盈利用途。
- 5. 人人文庫網(wǎng)僅提供信息存儲空間,僅對用戶上傳內容的表現(xiàn)方式做保護處理,對用戶上傳分享的文檔內容本身不做任何修改或編輯,并不能對任何下載內容負責。
- 6. 下載文件中如有侵權或不適當內容,請與我們聯(lián)系,我們立即糾正。
- 7. 本站不保證下載資源的準確性、安全性和完整性, 同時也不承擔用戶因使用這些下載資源對自己和他人造成任何形式的傷害或損失。
最新文檔
- 水運面試題目及答案
- 2025-2030中國裁斷機行業(yè)融資渠道與資本運作模式分析報告
- 2025-2030中國裁斷機行業(yè)信貸風險評估與融資渠道優(yōu)化
- 2025-2030中國網(wǎng)絡信息安全威脅演化與防御體系構建報告
- 古代宗教與天文歷法合同
- 核反應堆堆芯結構優(yōu)化合同
- 銀行人才測評試題及答案
- 核反應堆堆芯核物理實驗教育合同
- 銀行評級面試題及答案
- 飲料員工的工作總結
- 2025年北京市房屋租賃合同(自行成交版)
- 跌倒護理PDCA循環(huán)實施模板
- 事業(yè)單位離崗退養(yǎng)政策
- 2025年紙杯機市場分析現(xiàn)狀
- 新疆烏魯木齊市名校2025屆八下數(shù)學期末經(jīng)典試題含解析
- 《初中畢業(yè)班家長會課件中學銜接》
- 2024年青海省門源回族自治縣事業(yè)單位公開招聘村務工作者筆試題帶答案
- 2025-2030中國縫紉機器人行業(yè)市場發(fā)展趨勢與前景展望戰(zhàn)略研究報告
- 油氣倉儲物流優(yōu)化-全面剖析
- 妊娠劇吐護理查房
- 機械設備維護手冊
評論
0/150
提交評論