ML111890413

South Texas Project, Units 1 and 2 - Licensee Slide Regarding Questions on LOCA Frequency Analysis and Responses to Comments(Tac Nos. ME5358 and ME5359)

ML111890413
Person / Time
Site:	South Texas
Issue date:	07/06/2011
From:	Wright S South Texas
To:	Singal B K Plant Licensing Branch IV
Singal, B K, NRR/DORL, 301-415-301
Shared Package
ML111890371	List: ML111890376 ML111890380 ML111890391 ML111890408 ML111890413 ML111890420
References
TAC ME5358, GSI-191, TAC ME5359
Download: ML111890413 (8)

Text

QuestionsonLOCAFrequencyAnalysisGeneral:1. Howarefailuresthatdon'toccuratweldsconsidered?Theseinclude,forexample,ControlRodDriveMechanism(CRDM)failures,pressurizerheatersleevefailures,SteamGenerator(SG)tuberuptures,bottommountedinstrumentation(BMI)nozzlefailures,thermalfatiguefailuresatnozzles,componenttees,andothermixinglocations?Response:Weplantoaddressnon piperelatedfailuresandnonweldrelatedfailuresin2012sothefocusin2011istoconsiderpipebreaks.BasedonourunderstandingoftheprevailingdamagemechanismswebelievethatLOCAfrequencieswillbedominatedbyfailuresatweldlocationsincludingpipetosafeendandnozzlelocations.Howeverconsideringthefactthattherearemanyweldsdistributedratheruniformlyoverthepiperunstheweldlocationswillprovideanopportunitytoevaluatemanydetailedpipebreaklocations.ItisnotclearthatmovingtheassumedbreaklocationsbetweenweldlocationswillhaveasignificantimpactbutifitdoesthenwemayneedtoconsidersuchbreaklocationsintheLOCAfrequencyanalysis.Thisisanissuethatcanbebetterexaminedonceanumberofbreaklocationshavebeenfullyanalyzed.Weacknowledgethatwewilleventuallyneedtoincludenon weldeffectsordemonstratewhyaddingthemwillnotberisksignificant.Asfarasnon weldandnon pipelocationssuchasCRDMfailuresourapproachtoaddressingthemhasnotyetbeendefined.SGTRaremodeledinthePRAbutitisnotclearhowtubefailureswouldgenerateanydebris.Oursubmittalwillincludeajustificationofwhichlocationswereconsideredandthosethatwerescreenedoutandwhy.2. HowareLoss coolantAccidents(LOCAs)causedbyoverpressurizationeventssuchaswaterhammerorotherP Toverpressurizationevents(whichmaycausevesselfailureduetoembrittlement)considered?Response:ThepotentialforwaterhammerintheClass1pipingwasassessedintheRI ISIevaluationanddeterminednotbeacrediblefailuremechanismforLOCAsensitivepiping.ThePRAmodelexplicitlyconsidersover pressurizationduringlossofmainFW+ATWSconditionsandcalculatesaprobabilityofvesselfailure.WhilesuchfailuresareinfactmodeledinthePRA,theywillnotcontributetotheCHANGEinCDForLERFfromdesignoroperationalchangestoaddressGSI191-coredamagewillbeassumedforthecurrentdesignaswellasthereviseddesignifthevesselfails.3. HowareLOCAscausedbypressurizedthermalshock(PTS)consideredintheanalysis?Response:ThePRAmodelincludesamodelforPTSinducedvesselfailurefromthermalshockbutagain,nocreditistakentojustifycoredamagepreventionfollowingvesselfailure.OurunderstandingoftheNRCresearchonPTSindicatesthatPTSinducedvesselfailureisnotrisksignificantbutevenifitwaswhenthevesselfails,theissueofdebrisformationismootasfarasthePRAmodelisconcerned.

NOTE:ThemodelconsidersvesselintegrityfollowingoneclassofPTSevents,namelyexcesssteamflowinresponsetoatransient.Forexample,themodelquerieswhetherthevesselintegrityismaintainedfollowingaplanttripandfailurebothoftheturbinetotripandtheMSIVstoclose.PTSorthermaltransientimpactonthereactorvesselisnotincludedinthemediumorlargeLOCAeventmodels.NUREG1806wouldsuggestthatafutureupdateofthemodelsshouldincludethermalshockchallengesinresponsetomediumandlargeLOCAsorequivalent.PleasenotethatvesselintegrityisalsoqueriedafteranATWSwithfailureoftheMSIVstoisolate,inwhichcasethefailuremechanismofconcernisoverpressurization.ThislogicisincludedinthegeneraltransientandsmallLOCAmodelsonly.

4. Howarenon passivesystemLOCAfrequencies(e.g.,InterfacingSystemLOCAs(ISLOCAs),sealLOCAs,activesystemLOCAs)considered?Response:InterfacingLOCAsareexplicitlymodeledinthePRA.Whentheyoccurthepressureboundaryfailuresareoutsidethecontainmentandnotrelevanttocausingdebrisinducedcoredamageinsidethecontainment.ForISLOCAsthecoredamageisduetothediversionofcoolantoutsidethecontainment,bypassingthesumps,andinabilitytoestablishrecirculationflow.SealLOCAsduetoactivefailuresareincludedinthePRAmodelandtheGSI 191evaluationwillevaluatethedegreetowhichsuchfailurescangeneratedebris.Iftheyareshowntohavethecapabilitytogeneratedebris,theywillbeexplicitlyincludedintheanalysis.5. Whenandhowisprobabilisticfracturemechanism(PFM)beingusedtohelpdetermineLOCAfrequencyestimates(slide6)?Response:WehavenoplanstoperformanynewPFManalysisin2011butareleavingopentoperformsuchcalculationsifneededbasedontheresultsthisyear.Thefocusin2011istoevaluateawidespectrumofcasesthatspantheentireClass1pressureboundary.PFManalysesaremoreappropriateformorefocusedevaluationsatspecificandlimitednumberoflocations.AstheprojectproceedsandwhenandifitisdeterminedthatsomespecificlocationsareespeciallyimportantandwouldbenefitfromaPFMwewillincludethatbutatthisstageitistooearlytotellwhetherthiswillbenecessary.Uncertainties:1. Howarealeatoryandepistemicuncertaintiesconsideredandseparated(slide5)?Response:ThealeatoryuncertaintiesarereflectedintheassumptionthatLOCAsoccurasaPoissionprocesssothefrequencyofaLOCAisametricforthealeatoryuncertaintyaboutwhetheraLOCAwilloccurornot.TheuncertaintydistributionswedeveloparoundtheLOCAfrequenciesrepresentprimarilyepistemicuncertainties.Howeverwerecognizethatseparatingsourcesofuncertaintyintothesebinsissubjecttoitsownformofuncertaintyandcanbedebated.OurLOCAfrequencymodelofuncertaintyassumesthattheLOCAfrequencyisametricofaleatoryuncertaintyandthe uncertaintydistributionwedevelopthatisaroundthatistheepistemictype.Werealizethisisjustamodel.2. luncertaintiesdiscussedinslidesarecharacterizedasepistemicuncertainties.Whatisthebasisforthischaracterization

?Response:Seeaboveresponsetoitem1.3. HowisPIPExpdatabaseusedtoresolveuncertainties(slide6)?It'sunclearfromtheslides.Response:WemeantosaythatinformationcontainedinthedatabaseandinsightsdevelopedovermanyyearsincollectingandanalyzingthedatahelpstoreducethelevelofuncertaintythatweexpertsinputsinNUREG1829.InadditionanearlierversionofthesamedatabaseprovidedusefulinputtothelastNRCsponsoredprojectonLOCAfrequencies,NUREG/CR5750.Beforesuchdatawascollected,forexamplebackinWash 1400,verylittledataonnuclearpipingsystemswaseithergeneratedoranalyzed.TheentireefforttoimplementRIISIprogramswassupportedbyinsightsfromtheservicedata.Anearlyexampleofthekindofapplicationthatthisdatabasehasinreducinguncertaintiesistoguidethejudgmentsonhowtodefinehomogeneousweldpopulationsintogroupsforfailurerateestimation.Knowledgeofthedamagemechanismresponsiblefortheexperiencedpipefailuresisessentialforguidingthesejudgments(e.g.needtoseparatebi metallicweldssubjecttoPWSCCfromotherweldsnotsubjecttothisdamagemechanism)4. HowareNUREGs1829andCR 5750beingusedtoquantifyepistemicuncertainties(slide6)?WearepreparingaslidepresentationtouseduringourJuly7conferencecalltowalkthroughexamplesofhowweplantousethesereferencestoincorporateepistemicuncertaintiesintotheSTPspecificLOCAfrequencies.TobrieflysummarizeweplantoincorporateinformationfromNUREG 1829toestablishtheuncertaintydistributionparametersofourmodelfortheconditionalprobabilityofLOCAvs.LOCAcategory.5. WhydoesN ihaveuncertainty?Isn'tthenumberofweldsknown?Therearetworeasonsforthisuncertainty.Oneisplanttoplantvariability.Eachplanthasadifferentnumberofweldsforagivencomponent,therearedifferentnumbersofcoolantloopsinthePWRpopulation(2,3,and4),differentnumberofinterfacingsystemconnectionslikeECCS,etc.etc.Thesecondreasonisthateventhoughthesenumbersareknownwitheachplant'sorganization,thereareonlypublicallyavailablecountsforsomespecificplants.Basedonalimitedsamplewhichwillbedocumentedinoursubmittal,planttoplantvariabilityisresponsibleforafactorof2aboveandbelowthe"bestestimate"formanypipeweldcategories.6. Whyistherelittleuncertaintyassociatedwiththenumberoffailures(n ikinslide9)?Doesn'tlittleuncertaintyinthisparameterassumethatdatabasehascompletecoverageofalleventsandthatnoother"failures"haveoccurred?

Basedonourexperience,thereisverylittleuncertaintyinpipefailurecountsfortheClass1pressureboundarybasedonthePIPExpdataandcertainlymuchlessthantheuncertaintyweareassigningtothecomponentexposure.AlsothewholeideaofusingaBayes'methodforestimatingfailureratesisbasedontheideaofstartingwithapriordistributionthatmodelsaverylargeuncertainty.Inourapproachthepriorsareassumedtobelognormalwithrangefactorsof100.CalculationofLOCAestimates:1. Howistheintegritymanagementfactor(I ik)calculated(slide8)?IstheMarkovmodelusedtodetermineI ik?Response:Yes,theMarkovmodelisusedtocalculatethisfactor.ThiscalculationapproachwasworkedoutfortheEPRIRI ISIandisextensivelydocumentedintheattachedreferences.ThefirstreportontheMarkovmodelisReference[1]andtheinitialpipefailuredatadevelopedforuseofthismodelinReference[2].EPRIsponsoredreviewsofthisworkaredocumentedinReference[3]whichisincludedasanappendixtoReference[1].ThismodelanddataweredevelopedinitiallyinordertosupportestimatesinthechangeinCDFandLERFduetochangesinweldselectionsforNDEaspartoftheEPRIRI ISIprogram.TheuseoftheMarkovmodeltocalculateinspectionfactorswasfirstdocumentedintheEPRIRI ISITopicalReportinReference[4].TheNRCsafetyevaluationoftheEPRITopicalReportincludesfindingsthatapprovetheuseofthemodelandthesupportingdatafortheEPRIRIISIevaluations.ThisreviewwassupportedbyanNRCsponsoredreviewoftheMarkovmodelandtheBayes'failureratemethodverysimilartowhatweplantouseinthisprojectdonebyLANLinReference[5].ApeerreviewedjournalarticlewithmanyofthemathematicaldetailsofthismethodarefoundinReference[6].TosummarizetheMarkovmodelisusedtoformulateordinarydifferentialequationswhicharesolvedanalyticallyforthetimedependentstateprobabilities.TheinputparametersformodelwhicharethecoefficientsoftheODEaredefinedintermsofpipefailuremechanismsthatproduceflaws,leaksandruptures,aswellasparametersforthefrequencyandreliabilityofprogramstodetectleaksandinspectforflaws.Thenfromthesesolutions,analyticalexpressionsareobtainedforthehazardrate,whichiskindofatimedependentrateofrupture.Duetotheboundaryconditionsoftheequations,thehazardrateincreaseswithplantage(asseeninAppendixDofNUREG 1829).Theinspectionfactoristhehazardrateat40years(or60yearsdependingontheapplication)duetosomespecificintegritymanagementprogram(combinationofleakinspectionandNDE)dividedbythehazardrateatthesametimefortheaveragecomponentwithaverageintegritymanagement.2. Therelationshipbetweentheflowchart(slide12)andequations(1)-(3)(slides8and9)isunclear.Pleaseidentifywhichspecifictermsintheequationsarecalculatedbyspecificstepsintheflowchart.WeplantowalkthroughanexampleinourJuly7presentationwhichwillclarifyeachstepinquantifyingtheLOCAfrequencies.

3. Therearemanyquestionsrelatedtotheflowchart(slide12).a. Whydoesthenumberofleaksprovideinputtoboththefailurefrequencyandconditionalruptureprobability?Thenumberofleakscontributestothenumeratorofthefailurerateestimate.Thenumberofleaksalsocontributestothedenominatoroftheconditionalprobabilityofruptureestimate.b. Whereisdegradationmechanism(DM)susceptibilityinEquations(1)-(3)?Howdoesitfactorintothoseequations?Response:Weknowfromtheserviceexperiencethatsomefailuresoccurduetosomespecificdamagemechanisms.Wecancalculatetheunconditionalfailureratefromanydamagemechanismsimplydividingthenumberoffailuresbythetotalcomponentyearsintheservicedata.ButafterwehaveacompletedRIISIprogramsuchasthecasewithSTPweknowonaweldbyweldbasiswhichweldsaresusceptibletoeachdamagemechanism.Nowweneedtocalculatetheconditionalfailuregivenweknowtheapplicabledamagemechanism.ForthisweneedanestimateofthefractionofweldsinthedatabasethatproducedsomayfailuresduetoeachDMhowmanyaresusceptibletoeachDM.HencethefractionfinthedenominatorofEquationmustbeappliedtoestimatetheconditionalfailurerate.Leavingitoutwouldyieldtheunconditionalfailurerate.WeshallalsoaddressthisquestionintheJuly7meeting.c. Whatexpertsarebeingusedtoprovidevariousestimates?Aresameexpertsusedtoprovideeachdistributionindicatedinfigure?Response:InthecaseofinformationwebringinfromNUREG1829,itistheexpertpanelfromthatproject.Wemayalsoneedtoincorporateexpertjudgmentsfromourteamthatwillbeclearlydocumentedinthesubmittal.d. WhydotheDMsusceptibilityestimatesprovideinputtoboththegenericpriordistributionandintheBayesupdatedistributions?Response:TheestimatesofthefractionofweldsinthegenericpopulationthataresusceptibletothedamagemechanismareusedtodeterminetheparametersofthelikelihoodfunctionsfortheBayes'updateofpriordistributionswhichareintendedtorepresentthefailureratesforcomponentssusceptibletothosemechanisms.Theyneedtobeconsistent.OurJuly7presentationwillaimtoclarifythis.e. How,specifically,istheBayesianupdateofthepriordistributionperformedusingthethreedistributionsgeneratedtoinformtheprior?Response:Thepriorareassumedtobelognormal.WeuseaPoissonlikelihoodfunctiontoupdatethesewithonesetofdataforeachhypothesisofweldpopulationandweld susceptibilityfraction.ThisyieldsseveraldifferentposteriorsthatarecombinedusingwhatisreferredtoasBayes'posteriorweighting.WewillexplainthismoreclearlyduringJuly7presentation.

f. HowistheBeliczy SchultzcorrelationusedtocreatetheP(RlF)priordistribution?Response:Asyouknow,thiscorrelationwasusedbyBengtLydellasinputtothisdistributionforthebasecaseanalysesinAppendixDofNUREG1829.IntheSTPweplantobasethepriorsforthisdistributionusinginformationfromNUREG 1829whichwillbeexplainedduringtheJuly7meeting.g. HowistheP(RlF)priorupdatedusingBayes?Response:WeperformaBayes'updateforeachdiscreteLOCAcategory,whichisassociatedtoabreaksize.WeuseatruncatedlognormaldistributiontorepresenttheuncertaintyintheconditionalprobabilityofLOCAateachCategoryseparately.Weupdateitwithevidenceofnormally0LOCAsandNfailureswhereNisthenumberoffailuresusedtocalculatethecorrespondingfailurerates.WewillshowthisintheJuly7meeting.1. ThecalculationprocedureandtheapplicationoftheMarkovmodelinboththeflowchartandequationsisunclear.Also,information/documentationontheusedofthismodelforotherNRC approvedapplicationsandothernuclearapplicationsshouldbeprovided.Aretheredifferencesbetweenthemodelbeingusedfortheseestimatesandwhathasbeenapprovedandusedinotherapplications?Response:Pleaseseetheabovequestionresponseontheinspectionfactorandprovidedreferences.Anydeviationsfrompreviouslyreviewedapplicationswillbefullydocumentedinthesubmittalandsupportingreports.2. HowistheMarkovmodeldifferentthanP(RlF)?IsthismodelusedtodetermineP(RlF)?Response:P(R F)isusedtocalculaterupturefrequencies.Failureratesforflaws,leaks,andrupturesareinputtotheMarkovmodeltodeveloptheintegritymanagementfactors.3. IntheMarkovmodelthereisnoprobabilityofrupturegivennodetectibledamage.Whyisthistermneglected?Doesn'tthispresumethatISIisperfect?Response:Thismodelisonlyusedandwillonlybeappliedforrupturesduetodegradationmechanisms.ThereisanotherversionofthemodeldevelopedinReference[1]thatincludesadditionaltransitionsforleaksandrupturesabsentadetectableflaw.ThereasonfornotincludingthosetransitionsisthatpipefailuresduetosevereloadingconditionsaremodeledexplicitlybynonLOCAtypeinitiatorsinthePRAmodel.

4. Thereareanumberofquestionsrelatedtoslide33a. Whatisthebasisofthehazardrate{h(t)}equation?Response:Thisisstandardreliabilityengineeringtheory.Thehazardrateisdefinedasthenegativeoftherateofchangeofthereliabilityfunction(probabilityofnorupture)dividedbythereliabilityfunction.InthisMarkovmodel,thereliabilityisthesumofthestateprobabilitiesforsuccess,flaw,andleak.SeeReference[7]forthemathematicaldetails.b. Whatisr(t)?Response:Thisisthereliabilityfunctionreferredtoinitema.c. Howish(t)normalized?Response:Seetheabovequestion/responseontheIntegrityManagementfactord. Whatsensitivityanalysesonh(t)areperformed??Response:Thisreferstochangingtheassumptionsabout:whetherornotthereisaleakdetectionprogramandifthereis,howoftenitisdoneandwhatistheeffectiveprobabilityofdetection;andwhetherornotthereisanNDEprogram,andifthereishowoftenitisperformedandwhatistheeffectiveprobabilityofdetection.ForeachcombinationofleakdetectionandNDEinspectionparameters,adifferentresultisobtained.Questionsonspecificslides:1. Slide17:WhatisWH(waterhammer)?Yes2. Slide19:a. Howareunconditionalfailureratesdetermined?ByleavingfoutofthedenominatorinEquation(3)b. Arethe"conditional"estimateddeterminedthroughexpertelicitationtodeterminebump upfactorsforthe"unconditional"estimates(i.e.,bump upofapp.5forthermalfatigue)?Ifnot,howweretheconditionalestimatesdeterminedandwhatdotheymean?NotheresultsoftheRI ISIDMevaluationareusedtoresolvedeterministicallywhichweldsaresubjecttoeachDM3. Slide22:Whatare"excessive"LOCAs?VesselfailuresandmultiplepipebreaksandanyLOCAthatisbeyondthecapabilitiesoftheECCSaccordingtothePRAsuccesscriteria.4. Slide38:Whydoesinspectionstillyyieldapositive CDFforreactorcoolantpump(RCP)?Whileit'salowernumberthanifnoinspectionisperformedwhydoesitstilladdrisktotheplant?RCPstandsforreactorcoolantsystempiping.ThereisanincreasebecauseofthefactthatmanyweldswereremovedfromtheRI ISIprogram.ThisistypicalinallRI ISIprograms.References

[1] PipingSystemReliabilityandFailureRateEstimationModelsforUseinRiskInformedInServiceInspectionApplications.EPRI,PaloAlto,CA:1998.TR 110161.[2] PipingSystemFailureRatesandRuptureFrequenciesforUseinRiskInformedIn ServiceInspectionApplications.EPRI,PaloAlto,CA:1999.TR 111880.[3] Mosleh,A.andF.Groen,"TechnicalReviewoftheMethodologyofEPRITR110161",UniversityofMarylandreportforEPRI,publishedasanAppendixtoEPRITR 110161(Reference[6])[4] RevisedRiskInformedIn ServiceInspectionProcedure.EPRI,PaloAlto,CA:1999.TR 112657,Rev.B A.[5] U.S.NuclearRegulatoryCommission,SafetyEvaluationReportRelatedtoRevisedRiskInformedIn ServiceInspectionEvaluationProcedure:EPRITR 112657,Rev.B,July1999,Washington,D.C.,1999.(publishedasaforwardtoTR 112657(Reference[9])[6] Martz,H.,TSA 1/99 164:Final(Revised)ReviewoftheEPRI ProposedMarkovModeling/BayesianUpdatingMethodologyforUseinRisk InformedIn ServiceInspectionofPipinginCommercialNuclearPowerPlants,"LosAlamosNationalLaboratory,June1999.[7] Fleming,K.N.,"MarkovModelsforEvaluatingRiskInformedInServiceInspectionStrategiesforNuclearPowerPlantPipingSystems,"ReliabilityEngineeringandSystemSafety,Vol.83,No.1,pp.27-45,2004.