comparison of non-standard simulation methods for

13
Materials, Structural Integrity and Reliability Solutions Through Innovative Engineering Comparison of Non-Standard Simulation Methods for Performing Extremely Low Probability Assessments Robert E. Kurth, Cédric J. Sallaberry Engineering Mechanics Corporation of Columbus (Emc 2 ) 3518 Riverside Drive - Suite 202 Columbus, OH 43221-1735 Phone: (614) 459-3200 Fax: (614) 459-6800

Upload: others

Post on 16-Oct-2021

4 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering

ComparisonofNon-StandardSimulationMethodsforPerformingExtremelyLowProbabilityAssessments

RobertE.Kurth,CédricJ.Sallaberry

EngineeringMechanicsCorporationofColumbus(Emc2)3518RiversideDrive-Suite202

Columbus,OH43221-1735Phone:(614)459-3200Fax:(614)459-6800

Page 2: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 2

PSAM 14 – Los Angeles (CA)

•  In1985,theNRCproposedaModificationofGeneralDesignCriterion(GDC)4RequirementsforProtectionofDynamicEffectsofPostulatedPipeRupturesin10CFRPart50.▫  TheproposedamendmentswouldmodifyGDC4toallowforthedemonstrationof

pipinganalysestoserveasabasisforexcludingconsiderationofdynamiceffectsassociatedwithcertainpiperuptures.

▫  Theseanalysesconstitutewhatiscommonlyreferredtoastheleak-before-break(LBB)concept.

•  Thefinalrulerequiresadeterministicfracturemechanicsanalysis•  Thesupportingsafetyanalysismustdemonstratethatasubstantialrangeofstablepipecracksizescanexistwithdetectableleaks•  Thattheprobabilityoffluidsystemspipingruptureisextremelylowunderconditionsconsistentwiththedesignbasisforthepiping.•  Thedefinitionof“extremelylowprobability”ofpiperuptureisgivenontheorderof10-6perreactoryearwhenallpiperupturelocationsareconsidered.•  Thisprobabilityincludestheprobabilityofaninitiatingeventoccurringsuchasanearthquakeoranaccident.

Page 3: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 3

PSAM 14 – Los Angeles (CA)

•  Presently,theStandardReviewPlan(SRP)3.6.3doesnotallowforassessmentofpipingsystemswithactivedegradationmechanisms,suchasPrimaryWaterStressCorrosionCracking(PWSCC),whichhasoccurredinsystemsthathavepreviouslybeengrantedLBBexemptions

•  Along-termgoalhasbeentodevelopamodular-based,probabilisticfracturemechanicscodecapableofdeterminingtheprobabilityofruptureforReactorCoolantSystem(RCS)components.

▫  Theneedforthismodular-basedcodeisstronglydrivenbytheneedtoquantitativelyassessanLBB-approvedpipingsystem’scompliancewithGDC-4onaninterval(time)basis.

▫  BasedontheterminologyofGDC-4,thisprogramandcodeisentitledExtremelyLowProbabilityofRupture(xLPR).

Page 4: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 4

•  xLPR2.0hasbeencompleted▫  ItemploystheGoldSim®platformfortheprobabilisticcalculations▫  ItisafullyQA’edcodeforuseinregulatoryassessments▫  Becauseofthisoverheadeachsimulation,whichcoverstheentirepipelinelife,

takesapproximately4seconds

•  BecausetheabilitytocomplywithGDC4implies,forasimplerandomsamplingmethod,5to10millionsimulationswouldbeneededalternativestrategieshavebeendeveloped▫  Simplerandomsampling▫  LatinHypercubeSampling(LHS)▫  ImportanceSampling▫  AdaptiveImportancesampling*

•  Thefocusofthispaperisontheadaptiveimportancesampling

PSAM 14 – Los Angeles (CA)

*DetailsofthesetopicscanbefoundinPSAM103:“Efficientsamplingstrategiestoestimateextremelylowprobabilities”,C.Sallaberry,R.Kurth,F.Brust,E.Twombly

Page 5: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 5

PSAM 14 – Los Angeles (CA)

SRS LHS DPD

Better multi-dimensional coverage for DPD. Helps in conjoint influence

Densely cover of each input with LHS. Helps if extreme value are needed

Page 6: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 6

•  Thedetailsofthemethodology,whichislabeledGRC(GoldenRuleClusteringsincethemethodisbasedontheFibonaccisequence)•  Insteadweillustratetheapplicationtotwoproblems:▫  AtheoreticalproblemtoillustratetheGRCmethod▫  Ananalysisoftheapipingproblemwhere

!  Timetofirstcrackinitiation!  Timetofirstleakage!  Timeofpipefailure

arecalculated

PSAM 14 – Los Angeles (CA)

Page 7: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 7

•  GRCmethodis▫  Variancereductiontechnique▫  Purposeistooversampletheregionofinterestandunder-sampletheregions

withnoevents.

▫  InthetheoreticalproblemThreeareasofinterest,morethantwostandarddeviationsfromthecentroid,needtohavetheirprobabilityofoccurrenceestimated

PSAM 14 – Los Angeles (CA)

Page 8: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 8

•  Forthetheoreticalproblemwewishtofindthepeaks•  Adaptivesamplingfocusestheinputsontheareaswheretheresponseis“ofinterest”

PSAM 14 – Los Angeles (CA) 8

•  Inthepictureonthebottomleftthe3DsurfaceisprojectedontotheX-Yplane▫  RedpointsareSRS▫  Bluepointsareadaptive

•  Onthebottomrightthevaluesbelowalowerlimitareremoved

Page 9: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 9

PSAM 14 – Los Angeles (CA)

•  AcomparisonofthenormaldistributiontheoreticalCDF,theDPDrepresentation,andtheGRCrepresentationareshowntotheleft

•  BoththeDPDandGRClieonthetheoreticalline.Agoodstart.

•  TheDPDpoints,shownasanopenblackcircle,areuniformlyspacedalongthexaxisandthenmovedtotheconditionalmeanvalue

•  TheGRCpointsdependonthevalueoftheinputwhenaresponseofinterestisfound

•  Forthissampletheinputwasat~15%value

•  50%oftheGRCpointsareplacedbelowthisvalue,theremainderabove

9

Page 10: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 10

PSAM 14 – Los Angeles (CA)

•  TheCDFandCCDFareshownfortheyieldstrength

•  The“pivot”pointisnearthemean

•  ThepointsaretheDPDvaluesaftertheadaptivealgorithmhasbeenapplied

•  ClearlytherewillbemoresamplestakennearthepivotpointwhenthesamplingisbasedonthesimplerandomsamplingoftheintervalnotonthevalueoftheCDF

•  TheDPDPDFareusedtofindtheweightingvaluefortheresponse

10

Page 11: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 11

•  BecausethefocusofGRC4isonanextremelylowprobabilityofrupturethreequantitiesareexamined▫  Timetofirstcrackinitiation▫  Timetofirstleakage▫  Timeofrupture

•  BecauseoftheQArequirementsofxLPRtheGoldSim®platformcannotimplementtheGRCmethod

•  Therefore,apreprocessortoxLPRhasbeenwrittenwhichimplementstheGRCmethod

•  ThispreprocessoriscallPROMETHEUSandhasbeenreleasedasabetaversion,PROMETHEUS1.0

PSAM 14 – Los Angeles (CA)

Page 12: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 12

•  Acomparisonoftheresultsforaprobabilisticanalysisisshown

•  2,500simulationusingSRSwithLHSxLPR2.0(dottedline)

•  100,000SRSsample(dashedline)

•  2,500GRCsample(solidline)

•  All3methodsconvergetosample60yearresult

PSAM 14 – Los Angeles (CA) 12

•  GRCresultestimatesthe100,000SRSresultthroughtheentirerange•  Theuncertaintyislarger•  Theabilitytoestimatethe1in100,000pointallowsthexLPRcodetofocus

samplingusingstandardimportancesamplingmethods

Page 13: Comparison of Non-Standard Simulation Methods for

Materials,StructuralIntegrityandReliabilitySolutionsThroughInnovativeEngineering 13

•  Theuseofavarietyofsamplingmethodshavebeenstudiedtoallowefficientanalysisschemestobedevelopedto▫  Estimateextremelylowprobabilitiesofavarietyofevents▫  Identifytheareasoftheinputspacewhichleadtheeventsofinterest

•  Muchsmallersamplesizescanbeusedtoidentifyareasofinterestforfocusedsampling•  Comparisonofthemethodsshowsthat▫  TheSRS,LHS,andDPDmethodsallleadtostatisticallyequivalentmeanresults▫  TheGRCmethodfindsverylowprobabilityeventsmoreefficientlybutthecost

islargeruncertainty

•  TheuseoftheGRCmethodallowslessefficient(basedonthenecessarynumberofsimulations)tobefocusedontheregionsoftheinputspaceleadingtorareevents•  Thisfocusincreasestheefficiencyofthesecodes

PSAM 14 – Los Angeles (CA)