您好,欢迎访问三七文档
versãoimpressaISSN0101-7438/versãoonlineISSN1678-5142PesquisaOperacional,v.30,n.1,p.195-219,JaneiroaAbrilde2010195RELIABILITYASSESSMENTUSINGDEGRADATIONMODELS:BAYESIANANDCLASSICALAPPROACHESMartaAfonsoFreitas*Dep.deEng.deProdução/EscoladeEngenhariaUniversidadeFederaldeMinasGerais(UFMG)BeloHorizonte–MGmarta@dep.ufmg.br;marta.afreitas@gmail.comEnricoAntonioColosimoThiagoRezendedosSantosMagdaC.PiresDepartamentodeEstatística/ICEXUniversidadeFederaldeMinasGerais(UFMG)BeloHorizonte–MGenricoc@est.ufmg.brthiagords@ufmg.brmagdacpires@gmail.com*Correspondingauthor/autorparaquemascorrespondênciasdevemserencaminhadasRecebidoem07/2008;aceitoem07/2009ReceivedJuly2008;acceptedJuly2009AbstractTraditionally,reliabilityassessmentofdeviceshasbeenbasedon(accelerated)lifetests.However,forhighlyreliableproducts,littleinformationaboutreliabilityisprovidedbylifetestsinwhichfewornofailuresaretypicallyobserved.Sincemostfailuresarisefromadegradationmechanismatworkforwhichtherearecharacteristicsthatdegradeovertime,onealternativeismonitorthedeviceforaperiodoftimeandassessitsreliabilityfromthechangesinperformance(degradation)observedduringthatperiod.Thegoalofthisarticleistoillustratehowdegradationdatacanbemodeledandanalyzedbyusing“classical”andBayesianapproaches.Fourmethodsofdataanalysisbasedonclassicalinferencearepresented.NextweshowhowBayesianmethodscanalsobeusedtoprovideanaturalapproachtoanalyzingdegradationdata.Theapproachesareappliedtoarealdatasetregardingtrainwheelsdegradation.Keywords:Bayesianapproach;classicalapproach;degradationdataanalysis;reliability.ResumoTradicionalmente,oacessoàconfiabilidadededispositivostemsidobaseadoemtestesdevida(acelerados).Entretanto,paraprodutosaltamenteconfiáveis,poucainformaçãoarespeitodesuaconfiabilidadeéfornecidaportestesdevidanoquaispoucasounenhumasfalhassãoobservadas.Umavezqueboapartedasfalhaséinduzidapormecanismosdedegradação,umaalternativaémonitorarodispositivoporumperíododetempoeacessarsuaconfiabilidadeatravésdasmudançasemdesempenho(degradação)observadasduranteaqueleperíodo.Oobjetivodesteartigoéilustrarcomodadosdedegradaçãopodemsermodeladoseanalisadosutilizando-seabordagens“clássicas”eBayesiana.Quatrométodosdeanálisededadosbaseadoseminferênciaclássicasãoapresentados.Aseguir,mostramoscomoosmétodosBayesianospodemtambémseraplicadosparaproporcionarumaabordagemnaturalàanálisededadosdedegradação.Asabordagenssãoaplicadasaumbancodedadosrealrelacionadoàdegradaçãoderodasdetrens.Palavras-chave:abordagemBayesiana;abordagemclássica;análisededadosdedegradação;confiabilidade.Freitas,Colosimo,Santos&Pires–Reliabilityassessmentusingdegradationmodels:Bayesianandclassicalapproaches196PesquisaOperacional,v.30,n.1,p.195-219,JaneiroaAbrilde20101.Introduction1.1BackgroundandliteratureMuchoftheliteraturefocusesontheuseoflifetimetomakereliabilityassessments.Forproductsthatarehighlyreliable,assessingreliabilitywithlifetimedataisproblematic,however.Forapracticaltestduration(orafixedobservationperiodoftheperformanceonthefield),feworperhapsnofailuresmayoccursothatmostoftheobservationsarecensored.Suchdataprovidelittleinformationabouttheproportionofproducts,survivingawarrantyperiodthatisordersofmagnitudelongerthanthetestduration.Recently,degradationdatahaveshowntobeasuperioralternativetolifetimedatainsuchsituationsbecausetheyaremoreinformative(Chiao&Hamada,1996,2001;Lu&Meeker,1993;Lu,Meeker&Escobar,1996;Tseng,Hamada&Chiao,1995).Mostfailuresarisefromadegradationmechanismatwork,suchastheprogressionofachemicalreaction,forwhichtherearecharacteristicsthatdegrade(orgrow)overtime.Weconsiderthesituationinwhichfailureisdefinedintermsofanobservablecharacteristic.Forexample,acrackgrowsovertime,andfailureisdefinedtooccurwhenthecrackreachesaspecifiedlength.Anotherexampleisthebrightnessoffluorescentlightsthatdecreasesovertime.Itsfailureisdefinedtooccurwhenthelight’sluminositydegradesto60%orlessofitsluminosityat100hoursofuse.Suchfailuresarereferredtoas“soft”failuresbecausetheunitsarestillworking,buttheirperformancehasbecomeunacceptable.Toconductadegradationtest,onehastoprespecifyathresholdlevelofdegradation,obtainmeasurementsofdegradationatdifferentfixedtimes,anddefinethatfailureoccurswhentheamountofdegradationofaexperimentalunitexceedsthatlevel.Thus,thesedegradationmeasurementsmayprovidesomeusefulinformationtoassessreliabilityevenwhenfailuresdonotoccurduringthetestperiod.Thereareimportantreferencesthathaveuseddegradationdatatoassessreliability.Nelson(1981)discussedaspecialsituationinwhichthedegradationmeasurementisdestructive(onlyonemeasurementcouldbemadeoneachitem).Nelson(1990,chap.11)reviewedthedegradationliterature,surveyedapplications,describedbasicideasandusingaspecificexample,showedhowtoanalyzeatypeofdegradationdata.Intheliterature,therearetwomajoraspectsofmodelingfordegradationdata.Oneapproachistoassumethatthedegradationisarandomprocessintime.Doksum(1991)usedaWienerprocessmodeltoanalyzedegradationdata.Tang&Chang(1995)modelednondestructiveaccelerateddegradationdatafrompowersupplyunitsasacollectionofstochasticprocesses.Whitmore&Sch
本文标题:RELIABILITY ASSESSMENT USING DEGRADATION MODELS
链接地址:https://www.777doc.com/doc-3377090 .html