Handbook of probabilistic models pijush samui - Download the ebook and start exploring right away

Page 1


https://ebookmass.com/product/handbook-of-probabilistic-

Instant digital products (PDF, ePub, MOBI) ready for you

Download now and discover formats that fit your needs...

Basics of Computational Geophysics Pijush Samui

https://ebookmass.com/product/basics-of-computational-geophysicspijush-samui/

ebookmass.com

Probabilistic graphical models for computer vision Ji Q

https://ebookmass.com/product/probabilistic-graphical-models-forcomputer-vision-ji-q/

ebookmass.com

Integrating Disaster Science and Management: Global Case Studies in Mitigation and Recovery Pijush Samui

https://ebookmass.com/product/integrating-disaster-science-andmanagement-global-case-studies-in-mitigation-and-recovery-pijushsamui/

ebookmass.com

Electronic and Experimental Music

https://ebookmass.com/product/electronic-and-experimental-music/

ebookmass.com

Organizational Behavior: Human Behavior at Work: Human Behavior at Work 14th Edition – Ebook PDF Version

https://ebookmass.com/product/organizational-behavior-human-behaviorat-work-human-behavior-at-work-14th-edition-ebook-pdf-version/

ebookmass.com

Hero: The Mountain Man (Mountains and Curves Book 3) Betty Harwin

https://ebookmass.com/product/hero-the-mountain-man-mountains-andcurves-book-3-betty-harwin/

ebookmass.com

Trauma Plating Systems. Biomechanical, Material, Biological, and Clinical Aspects 1st Edition Edition

Amirhossein Goharian

https://ebookmass.com/product/trauma-plating-systems-biomechanicalmaterial-biological-and-clinical-aspects-1st-edition-editionamirhossein-goharian/ ebookmass.com

Comparative Criminal Justice Systems: A Topical Approach 7th Edition, (Ebook PDF)

https://ebookmass.com/product/comparative-criminal-justice-systems-atopical-approach-7th-edition-ebook-pdf/

ebookmass.com

Beyond Turnout: How Compulsory Voting Shapes Citizens and Political Parties Shane P. Singh

https://ebookmass.com/product/beyond-turnout-how-compulsory-votingshapes-citizens-and-political-parties-shane-p-singh/

ebookmass.com

What Is Race?: Four Philosophical Views Joshua

https://ebookmass.com/product/what-is-race-four-philosophical-viewsjoshua-glasgow/

ebookmass.com

Handbookof ProbabilisticModels

DepartmentofCivilEngineering,NITPatna,Bihar,India

DieuTienBui

DepartmentofBusinessandIT,SchoolofBusiness UniversityofSouth-EasternNorway(USN),Telemark,Norway

SubrataChakraborty

DepartmentofCivilEngineering,IndianInstituteof EngineeringScienceandTechnology,Howrah,India

RavineshC.Deo

UniversityofSouthernQueensland Springfield,QLD,Australia

Butterworth-HeinemannisanimprintofElsevier

TheBoulevard,LangfordLane,Kidlington,OxfordOX51GB,UnitedKingdom 50HampshireStreet,5thFloor,Cambridge,MA02139,UnitedStates

Copyright©2020ElsevierInc.Allrightsreserved.

Nopartofthispublicationmaybereproducedortransmittedinanyformorbyanymeans, electronicormechanical,includingphotocopying,recording,oranyinformationstorage andretrievalsystem,withoutpermissioninwritingfromthepublisher.Detailsonhowto seekpermission,furtherinformationaboutthePublisher’spermissionspoliciesandour arrangementswithorganizationssuchastheCopyrightClearanceCenterandtheCopyright LicensingAgency,canbefoundatourwebsite: www.elsevier.com/permissions . Thisbookandtheindividualcontributionscontainedinitareprotectedundercopyrightby thePublisher(otherthanasmaybenotedherein).

Notices

Knowledgeandbestpracticeinthis fieldareconstantlychanging.Asnewresearchand experiencebroadenourunderstanding,changesinresearchmethods,professional practices,ormedicaltreatmentmaybecomenecessary.

Practitionersandresearchersmustalwaysrelyontheirownexperienceandknowledgein evaluatingandusinganyinformation,methods,compounds,orexperimentsdescribed herein.Inusingsuchinformationormethodstheyshouldbemindfuloftheirownsafety andthesafetyofothers,includingpartiesforwhomtheyhaveaprofessionalresponsibility. Tothefullestextentofthelaw,neitherthePublishernortheauthors,contributors,or editors,assumeanyliabilityforanyinjuryand/ordamagetopersonsorpropertyasamatter ofproductsliability,negligenceorotherwise,orfromanyuseoroperationofanymethods, products,instructions,orideascontainedinthematerialherein.

LibraryofCongressCataloging-in-PublicationData

AcatalogrecordforthisbookisavailablefromtheLibraryofCongress

BritishLibraryCataloguing-in-PublicationData

AcataloguerecordforthisbookisavailablefromtheBritishLibrary

ISBN:978-0-12-816514-0

ForinformationonallButterworth-Heinemannpublications visitourwebsiteat https://www.elsevier.com/books-and-journals

Publisher: MatthewDeans

AcquisitionEditor: MatthewDeans

EditorialProjectManager: JoshuaMearns

ProductionProjectManager: KameshRamajogi

CoverDesigner: MilesHitchen

TypesetbyTNQTechnologies

Dedicatedtomygrandfatherandgrandmother

Contributors

FaridWajdiAkashah,CentreforBuilding,ConstructionandTropicalArchitecture (BuCTA),FacultyofBuiltEnvironment,UniversityofMalaya,KualaLumpur, Malaysia

HassanZ.AlGarni,DepartmentofElectricalandElectronicEngineeringTechnology, JubailIndustrialCollege,JubailIndustrialCity,SaudiArabia

MumtazAli,SchoolofAgriculturalComputationalandEnvironmentalSciences, UniversityofSouthernQueensland,Springfield,QLD,Australia;Deakin-SWU JointResearchCentreonBigData,SchoolofInformationTechnology,Deakin University,Burwood,VIC,Australia

ArmandoA.Apan,SchoolofCivilEngineeringandSurveying,UniversityofSouthern Queensland,Toowoomba,QLD,Australia

AnjaliAwasthi,ConcordiaUniversity,CIISE,Montreal,QC,Canada

RahimAzadnia,DepartmentofAgriculturalMachineryEngineering,Universityof Tehran,Tehran,Iran

HamoonAzizsoltani,DepartmentofComputerScience,NorthCarolinaStateUniversity,Raleigh,NC,UnitedStates

MouradBelgasmia,DepartmentofCivilengineering,SetifUniversity,Setif,Algeria

OmidBozorg-Haddad,DepartmentofIrrigationandReclamationEngineering,FacultyofAgriculturalEngineeringandTechnology,CollegeofAgricultureand NaturalResources,UniversityofTehran,Tehran,Iran

TanmoyChatterjee,CollegeofEngineering,SwanseaUniversity,BayCampus, Swansea,UnitedKingdom;DepartmentofCivilEngineering,IndianInstituteof TechnologyRoorkee,Roorkee,Uttarakhand,India

SouravChoudhary,DepartmentofCivilEngineering,NationalInstituteofTechnologyPatna,Patna,Bihar,India

RajibChowdhury,DepartmentofCivilEngineering,IndianInstituteofTechnology Roorkee,Roorkee,Uttarakhand,India

XuefengChu,DepartmentofCivilandEnvironmentalEngineering,NorthDakota StateUniversity,Fargo,ND,UnitedStates

KavinaS.Dayal,SchoolofAgriculturalComputationalandEnvironmentalSciences, UniversityofSouthernQueensland,Springfield,QLD,Australia

MichaelDelichatsios,FireSafetyEngineeringResearchandTechnologyCentre (FireSERT),UniversityofUlster,Newtownabbey,UnitedKingdom

RavineshC.Deo,SchoolofAgriculturalComputationalandEnvironmentalSciences, UniversityofSouthernQueensland,Springfield,QLD,Australia

S.Dey,DepartmentofMechanicalEngineering,NationalInstituteofTechnology Silchar,Silchar,Assam,India

NathanJ.Downs,SchoolofAgricultural,ComputationalandEnvironmentalSciences, UniversityofSouthernQueensland,Springfield,QLD,Australia

SubhrajitDutta,AssistantProfessor,DepartmentofCivilEngineering,National InstituteofTechnologySilchar,Silchar,Assam,India;NationalInstituteofTechnologySilchar,DepartmentofCivilEngineering,Silchar,Assam,India

AmirH.Gandomi,FacultyofEngineeringandInformationTechnology,Universityof TechnologySydney,Ultimo,NSW,Australia;SchoolofBusiness,StevensInstitute ofTechnology,Hoboken,NJ,UnitedStates

J.RamonGaxiola-Camacho,DepartmentofCivilEngineering,AutonomousUniversityofSinaloa,Culiacan,Sinaloa,Mexico

AboubakerGherbi,DepartmentofCivilengineering,ConstantineUniversity,Constantine,Algeria

AchintyaHaldar,DepartmentofCivilandArchitecturalEngineeringandMechanics, UniversityofArizona,Tucson,AZ,UnitedStates

SoheilSadatHosseini,DepartmentofElectricalEngineering,CapitolTechnology University,Laurel,MD,UnitedStates;DepartmentofElectricalEngineeringand ComputerScience,TheUniversityofToledo,Toledo,OH,UnitedStates

PravinJagtap,DepartmentofCivilEngineering,IndianInstituteofTechnology(IIT) Delhi,NewDelhi,India

MohsinM.Jamali,CollegeofEngineering,TheUniversityofTexasofthePermian Basin,Odessa,TX,UnitedStates

MadanK.Jha,AgriculturalandFoodEngineeringDepartment,IndianInstituteof TechnologyKharagpur,Kharagpur,WestBengal,India

ShahjahanKhan,SchoolofAgricultural,ComputationalandEnvironmentalSciences, CentreforAppliedClimateSciences,UniversityofSouthernQueensland,Toowoomba,QLD,Australia

AnoopKodakkal,ChairofStructuralAnalysis,DepartmentofCivil,GeoandEnvironmentalEngineering,TechnischeUniversitatMu ¨ nchen(TUM),Munich,Germany

R.R.Kumar,DepartmentofMechanicalEngineering,NationalInstituteofTechnology Silchar,Silchar,Assam,India

YuankaiLi,UniversityofElectronicScienceandTechnologyofChina,Chengdu, China

NeharMandal,DepartmentofCivilEngineering,NationalInstituteofTechnology Patna,Patna,Bihar,India

TekMaraseni,SchoolofAgricultural,ComputationalandEnvironmentalSciences, UniversityofSouthernQueensland,Springfield,QLD,Australia

SamanMaroufpoor,DepartmentofIrrigationandReclamationEngineering,Faculty ofAgriculturalEngineeringandTechnology,CollegeofAgricultureandNatural Resources,UniversityofTehran,Tehran,Iran

VasantMatsagar,DepartmentofCivilEngineering,IndianInstituteofTechnology (IIT)Delhi,NewDelhi,India

AnkeMeyer-Baese,DepartmentofScientificComputing,FloridaStateUniversity, Tallahassee,FL,UnitedStates

BehshadMohebali,DepartmentofScientificComputing,FloridaStateUniversity, Tallahassee,FL,UnitedStates

TanmoyMukhopadhya,DepartmentofAerospaceEngineering,IndianInstituteof TechnologyKanpur,Kanpur,India

ShahbazMushtaq,CentreforAppliedClimateSciences,UniversityofSouthern Queensland,Toowoomba,QLD,Australia

S.Naskar,SchoolofEngineering,UniversityofAberdeen,Aberdeen,UnitedKingdom

ThongNguyen-Huy,SchoolofAgricultural,ComputationalandEnvironmentalSciences,CentreforAppliedClimateSciences,UniversityofSouthernQueensland, Toowoomba,QLD,Australia

WeiNing,SchoolofCivilEngineering,ChongqingUniversity,Chongqing,China

RachidOuache,SchoolofEngineering,FacultyofAppliedScience,Universityof BritishColumbia,Okanagancampus,Kelowna,BC,Canada

K.M.Pandey,DepartmentofMechanicalEngineering,NationalInstituteofTechnologySilchar,Silchar,Assam,India

ThendiyathRoshni,DepartmentofCivilEngineering,NationalInstituteofTechnologyPatna,Patna,Bihar,India

ZhangRunhong,SchoolofCivilEngineering,ChongqingUniversity,Chongqing, China

VikasKumarSharma,InstituteofInfrastructuteTechnologyResearchandManagement(IITRAM),DepartmentofMathematics,Ahmedabad,Gujarat,India

S.Sriramula,SchoolofEngineering,UniversityofAberdeen,Aberdeen,United Kingdom

AmirhessamTahmassebi,DepartmentofScientificComputing,FloridaStateUniversity,Tallahassee,FL,UnitedStates

AnthonyTeckCheeGoh,SchoolofCivilandEnvironmentalEngineering,Nanyang TechnologicalUniversity,Singapore

S.MohsenVazirizade,DepartmentofCivilandArchitecturalEngineeringand Mechanics,UniversityofArizona,Tucson,AZ,UnitedStates

FranciscoJavierVillegas-Mercado,DepartmentofCivilandArchitecturalEngineeringandMechanics,UniversityofArizona,Tucson,AZ,UnitedStates Contributors

xxii Contributors

ZhangWengang,KeyLaboratoryofNewTechnologyforConstructionofCitiesin MountainArea,ChongqingUniversity,Chongqing,China;SchoolofCivilEngineering,ChongqingUniversity,Chongqing,China;NationalJointEngineering ResearchCenterofGeohazardsPreventionintheReservoirAreas,Chongqing University,Chongqing,China

LiYongqin,SchoolofCivilEngineering,ChongqingUniversity,Chongqing,China

JianpingZhang,FireSafetyEngineeringResearchandTechnologyCentre(FireSERT),UniversityofUlster,Newtownabbey,UnitedKingdom

Chapter1

Fundamentalsofreliability analysis

AchintyaHaldar

DepartmentofCivilandArchitecturalEngineeringandMechanics,UniversityofArizona,Tucson, AZ,UnitedStates

1.Introduction

Thepresenceofuncertaintyineveryaspectofengineeringanalysisanddesign hasbeenunderconsiderationoveralongperiodoftime.Infact,afamous mathematicianPierre-SimonLaplace(1749 1827)wrote“. theprincipal meansofascertainingtruth induction,andanalogy arebasedonprobabilities;sothattheentiresystemofhumanknowledgeisconnectedwiththe theory(ofprobability). .Itleavesnoarbitrarinessinthechoiceofopinions andsidestobetaken;andbyitsusecanalwaysbedeterminedthemost advantageouschoice.Therebyitsupplementsmosthappilytheignoranceand weaknessofthehumanmind.”(Laplace,1951).

Theaforementionedstatementsbyawell-knownscholarclearlyjustifythe needforthishandbook.Morerecently, Freudenthal(1956),AngandTang (1975),Shinozuka(1983),and HaldarandMahadevan(2000a) madesimilar commentsjustifyingtheneedsforstructuralsafetyandreliabilityanalyses. Therelatedareasgrewexponentiallyinthe1970sand1980s.Itappearsthat mostofthedesignguidelinesandcodeseitherhavebeenorareintheprocess ofincorporatingtherisk-baseddesignconcept,atleastintheUS.This handbookisexpectedtobeextremelyvaluableinmovinginthatdirection.

However,beforemovingforward,itisimportanttofigureoutwhatisthe conceptofuncertainty,probability,reliability,stochasticity,etc.andhowthey implicatetheengineeringanalysisanddesign.StochosisaGreekwordfor stochasticityoruncertainty.Ingeneral,mostobservablephenomenaofinterest toengineersproducemultipleoutcomesandcannotbepredictedwith certainty.Multipleoutcomesmaynothaveanypattern,andsomeoutcomes mayoccurmorefrequentlythanotherscoveringdifferentregionsofinterest. Testingofidenticalspecimensmaynotproduceidenticaloutcomes.The designwindvelocityorrainfallatasiteduringthelifetimeofastructure

cannotbepredictedwithcertainty.Wemayknowtheirupperandlowerlimits orboundsandthemostlikelyvaluebutnotthedesignvalueforwhicha specificstructureneedstobedesigned.Thistypeofunpredictabilityis generallyrepresentedbyuncertaintyorrandomness.Complexityofaproblem maynothaveanythingtodowithuncertainty.Thereisnodoubtthatlanding onthemoonconsistsofnumerouscomplicatedandcomplexprocesses,butwe nevermissedthemoonwhenweattemptedtolandonit.Eventheoutcomeof averysimpletaskoftossingacoincannotbepredictedwithcertainty. Uncertaintyisassociatedwithmostoftheanalysisanddesignofinterestto engineers.Consideringtheunpredictabilityofmostofthedesignvariables,the basicchallengeistoassuresatisfactoryperformanceofengineeringsystemsin thepresenceofuncertainty.Thepresenceofuncertaintycannotbecompletely eliminated,butwithreasonableefforts,itsimpactonthedesigncanbe appropriatelymanaged.Thisobservationclearlyindicatesthatengineering systemscannotbedesigned“full-proof”or“risk-free.”Therewillalwaysbe someriskorprobabilityoffailure.Foracceptabledesign,theamountofunderlyingriskneedstobeminimizedtoanacceptablelevelormitigated appropriately.Riskandreliabilityarecomplementarytermsandneedtobe mathematicallyestimatedusingprobabilitytheory.Probabilityandstatistics arenotsynonymousterms.Statisticsisthemathematicalquantificationof uncertainty.Probabilitytheoryusesstatisticalinformationtoestimatethe likelihoodofspecificevents.Withthisintroduction,fundamentalsof reliabilityanalysisarebrieflydiscussedinthisintroductorychapter.

2.Importantstepsinreliabilityevaluation

Mostengineeringproblemsconsistofmultiplerandomvariables(RVs).The firststepisthenthequantificationofrandomnessinthem,onevariableata timeandjointlywhenpossible.TheuncertaintyinanRVisgenerally describedpictoriallyintermsofhistogramandfrequencydiagramsorprobabilitydensityfunction(PDF)andanalyticallywiththehelpofmean,variance,standarddeviation,coefficientofvariation(COV),skewness,etc.They aresometimescollectivelydenotedasthestatisticsofanRV.FormultipleRVs, itisdescribedasjointPDF,correlationcoefficients,etc.Meanisthecentral tendency,varianceandstandarddeviationindicatethedispersionfromthe mean,COVistheratioofstandarddeviationandmeanandrepresents theamountofuncertaintyinanondimensionalway,andskewnessrepresents thesymmetryinthedata,generallyexpressedintermsofskewnesscoefficient. ForaknownPDF,meanisthecentroidaldistancefromtheorigin;itisalso knownasthefirstmoment.VarianceisthemomentofinertiaofthePDFabout themean;itisalsoknownasthesecondmoment.Skewnessisthethird momentofthePDFaboutthemean.Forsymmetricdata,theskewness coefficientwillbezero;morespreadinthedataabovethemeanwillhave positiveskewnesscoefficient,etc.Inthecontextofquantifyingrandomnessin

oneRV,threeadditionalparametersarecommonlyused.Theyaremode, median,andpercentilevalueofanRV.ThemodeormodalvalueofanRVis thevalueofthehighestPDF.ThemedianvalueofanRVisthevalueforwhich itisequallylikelytobeaboveorbeloworthe50thpercentilevalue.The percentilevalueistheprobability,expressedasapercentage,thevalueofthe RVwillbelessthanthespecifiedvalue.Formorecompletediscussionson thetopics,readersarerequestedtoreferto HaldarandMahadevan(2000a). WhendealingwithmultipleRVs,theinformationoncovarianceandcorrelationisveryimportant.ThecovarianceoftwoRVs X and Y,denotedasCov (X,Y),isthesecondmomentabouttheirrespectivemeans.Normalizingthe covariancewithrespecttothestandarddeviationsofthetwoRVswillresult thecorrelationcoefficient.Thecorrelationcoefficientrepresentsthelinear dependencybetweentwoRVs.Itvariesbetween 1.0andzerowhentwoare uncorrelated.Thevalueofcorrelationcoefficientexactlyzeroisrarely obtainedfromobserveddata. HaldarandMahadevan(2000a) suggestedthatif itislessthan 0.3,thetwovariablescanbeconsidereduncorrelated.

Thesecondstepinthereliabilityanalysiswillbetheperformance requirementsthatneedtobesatisfied.Theperformanceofengineering systemswilldependontheuncertaintyintheRVspresentintheformulation; however,theywillnotaffecttheperformanceequally.Thepropagationof uncertaintyfromthevariableorparametertosystemleveldependsonmany factorsincludingtheindividualnatureofuncertaintyinanRV,functional relationshipbetweenvariables,typesofanalysis;linearversusnonlinear,load pathtofailure,etc.Ingeneral,asystemneedstobedesignedsatisfyingperformancerequirementsattheelementandsystemlevels,andtheyareexpected tobefunctionsofoneormultipleRVs.Thiswillrequiretopropagatethe uncertaintiesinRVstothecorrespondinglevelofinterest.Onlythen,the underlyingriskortheprobabilityoffailurecorrespondingtoaspecificperformancerequirementcanbeestimated.Theuncertaintyinsomeofthevariablesmaybeamplifiedatthesystemlevel,someofthemmaybedeamplified, andsomeotherwillhaveveryminoreffect.Reliability-basedengineering analysisanddesignareexpectedtobemoreinvolvedorcomplicatedin comparisonwiththedeterministicapproachescommonlyusedbytheprofessioninthepastwithoutchangingthebasicunderlyingmathematicalprinciples.Itisamajorchallengetothereliabilitycommunitytomakesurethat whattheyaredoingissimilartopracticesfollowedbythedeterministic community.Themostrecenttrendistoextractthereliabilityinformationby conductingmultipledeterministicanalyses.However,tomaintainthebasic simplicityintheformulation,effortshouldbeexercisedtoreducethesizeof theproblemasmuchaspossiblebyconsideringlesssignificantRVsas deterministicattheirmeanvalues.Itneedstobepointedoutthatinatypical deterministicanalysisanddesign,nominalvalues(severalstandarddeviations abovethemeanfortheload-relatedvariablesandseveralstandarddeviations belowthemeanvaluefortheresistance-relatedvariables)areused,indirectly

introducingthesafetyfactorconcept.Ifformulatedproperly,itmayalso satisfytherequiredunknownunderlyingrisk.Intherisk-baseddesign,the uncertaintyinavariableisexplicitlyexpressedintermsofthestatistical information(mean,varianceorstandarddeviation,PDF,etc.).

Theperformancelevelcanbeattheelementorsystemlevel,asmentioned earlier.Element-levelperformancerequirementsaregenerallystrengthrelated. Thesystemlevelrequirementsgenerallyreflectserviceability-relatedperformancescausingexcessivelateral(interstoryoroverall)deflectionordrift causedbyfailureofstructuralelementsorinadequatestiffnessofelements, etc.Theperformancerequirementsaregenerallyexpressedintheformof performanceorlimitstatefunctions/equations.Atypicalperformancefunction consistsofallload-andresistance-relatedRVspresentintheformulation. Performancerequirementsaregenerallysuggestedinthedesignguidelinesor specifiedbytheowner.Sometimes,performancefunctionscanbeexplicit,but inmostcasesofpracticalinterest,theyareimplicitinnature.

Thethirdstepistheestimationofrisk/reliabilitycorrespondingtoaperformancerequirement.Someoftheestimationprocedureswillbediscussedin moredetaillaterinthischapter.Essentially,theinformationinformsthe designertheunderlyingriskforthespecificdesign.Iftheriskisnotacceptable evenwhenalldesignrequirementsweresatisfied,thedesignneedstobe alteredtosatisfytheownerorthestakeholder.Thisstepmakesthereliabilitybasedanalysisanddesignmoreattractivethantheconventionaldeterministic evaluation.Ithelpstocomparedifferentdesignalternatives.Usinginformation onriskoftwodesignalternatives,themostdesirableorappropriateoptioncan beselected.

3.Elementsofsettheory

Essentialstepsinestimatingriskhavebeendiscussedintheprevioussection. Itisnownecessarytoquantifytheriskassociatedwithadesign.Forthe reliabilityevaluation,theconceptofsettheoryinsteadofconventionalalgebra needstobeused.Theconceptofsettheoryisverybrieflydiscussedinthis section.

Atypicalengineeringproblemmusthaveasamplespace,discreteor continuous,consistingofmutuallyexclusivesamplepoints.Aneventof interestmustcontainatleastonesamplepoint.Eventscanbecombinedusing unionorintersectionoperation,andtheycanbemutuallyexclusive,statisticallyindependent,orcorrelated.Theinformationonriskorreliabilitycanbe extractedwiththehelpofsettheoryandthethreeaxiomsofprobability.Three axiomsofprobabilityare(1)theprobabilityofanevent E,generallywrittenas P(E), willalwaysbenonnegative,(2)theprobabilityofthesamplespace, S,or P(S), willbe1.0;thus,theprobabilityofaneventwillbebetweenzeroand one,and(3)ifthetwoevents E1 and E2,aremutuallyexclusive,theprobability

oftheirunionwillbethesummationoftheirindividualprobability.Itis generallyexpressedas

Thecomplementofanevent,denotedas E ,canbeshowntobe

Eq.(1.2) indicatesthattheprobabilityofsurvivalcanbeestimatedas1.0 risk Ingeneral,eventsarenotmutuallyexclusive.Inthatcase, Eq.(1.1) canbe showntobe

Theprobabilityofintersectionorjointoccurrenceofevents, P(E1E2)asin Eq.(1.3),needstobecalculatedusingthemultiplicationrule.Thegeneral multiplicationrulefortwoeventscanbeshowntobe

P(E1jE2)or P(E2jE1)isgenerallyknownastheprobabilityoftheconditional eventsindicatingtheoccurrenceofoneeventgiventheoccurrenceoftheother event.Itreducesthesizeoftheoriginalsamplespace.Toestimatetheprobabilitywithrespecttotheoriginalsamplespace,theconditionalprobability needstobemultipliedbytheprobabilityoftheeventconditionedonwith respecttotheoriginalsamplespace.

Iftheeventsarestatisticallyindependent,indicatingtheoccurrenceofone doesnotdependontheother, Eq.(1.4) becomes

Iftheeventsaremutuallyexclusive,i.e.,iftheoccurrenceofoneprecludes theoccurrenceoftheother,theprobabilityofjointoccurrencesofthemwillbe zero.Iftheunionsofalltheeventsconstitutethesamplespace,theyarecalled collectivelyexhaustiveevents.Morediscussionsonsettheory,axiomsof probability,multiplicationrules,andotheroperationscanbefoundinthestudy by HaldarandMahadevan(2000a).

4.Quantificationofuncertaintiesinrandomvariables

Thenextessentialtaskinthereliabilityestimationisthequantificationof uncertaintiesinalltheRVspresentintheformulation.Thequantificationof uncertaintyinanRVrequiresthecollectionofdataonitfromasmanysources aspossible.Thesamplesizeisveryimportantintheuncertaintyquantification. Largersamplesizeisalwayspreferable.Itmaybeimpracticaltocollectdata fromavailablesourcesfromallovertheworld.Thecollecteddatacanbe graphicallyrepresentedwiththehelpofthehistogramandPDF.PDFisa

histogramwiththeareaof1.0.Inmostcases,aPDFcanbegeneratedbyfitting apolynomialwhoseparameterscanbeestimatedfromthesamplesusedto generateit.Inmostengineeringapplications,two-parameterPDFsare routinelyused.Itcanbeshownthattheseparameterscanbeestimatedfrom thefirsttwomoments,themeanandvariance,ofthesamples.Themost encouragingpartofreliability-basedanalysisanddesignisthatmostofthe designvariablesusedinroutineengineeringapplicationsarealready researchedandwidelyavailableintheliterature.TheinformationonthePDF anditsparametersofanRVofinterestcanbeeasilyobtainedwithacasual literaturesearch.

Forthecompletenessofthisdiscussion,commonlyusedcontinuousRVs arerepresentedbynormal,lognormal,Beta,Rayleigh,andexponentialdistributions.AnormaldistributionwillbevalidwhenanRVisvalidfromminus infinitytoplusinfinity.AlognormaldistributionisusedwhenanRVisvalid fromzerotoplusinfinity.ABetadistributioncanbeusedwhenanRVisvalid betweenalowerandupperlimits.SomeofthecommonlyuseddiscreteRVs arebinomial,geometric,Poisson,etc.ParametersrequiredtodefinetheseRVs aretabulatedinmanybooks(AngandTang,1975;HaldarandMahadevan, 2000a).

5.Transformationofuncertaintyfromparametertothe

systemlevel

Aswillbediscussedlater,ifaperformanceorlimitstateequation(LSE)or functionisalinearfunctionofoneRV,itwillberelativelysimpletoestimate theriskusingtheavailablestatisticalinformationontheRV.However,inmost casesofpracticalinterest,theperformanceofastructuralelementmaydepend onnonlinearfunctionofoneRVormultipleRVs.Forastructuralsystem consistingofnumerousstructuralelements,theLSEwillbeafunctionof numerousRVs.Theunderlyingriskwilldependontheformulationsandthe statisticalcharacteristicsofalltheRVsinvolvedintheformulation.Forthe sakeofdiscussion,theaxialloadcarryingcapacityofasteelcolumnwillbe denotedastheresponsevariableandwilldependontheloadactingonit,grade ofsteelused,cross-sectionalarea,radiusofgyration,lengthofthecolumn, supportconditions,etc.Collectively,theywillbedenotedhereafterasthe basicRVs.Uncertaintiesinthesebasicvariableswilldictatetheoverall uncertaintyintheaxialloadcarryingcapacity.Functionalrelationship betweenthesedesignvariablestotheloadcarryingcapacityisgenerally suggestedindesigncodesorguidelinesorcanbemathematicallyderived. WhenthefunctionalrelationshipofallRVsandtheappropriateperformance requirementisavailable,itwillbeconsideredasanexplicitLSE.Asteelframe mayconsistofnumeroussuchcolumns.Thelateraldeflectionatthetopofthe frame(responsevariable)willdependonthepropertiesofthesecolumnsand otherstructuralelementspresentintheframe,essentiallyallthevariables

involvedintheestimationofthelateraldeflectionandtheloadsactingonit. However,anexplicitexpressionforthelateraldeflectionintermsofallthe RVsmaynotbepracticaltodevelop;itmaybeavailableinanalgorithmic form,foranexample,intheformoffiniteelementsrepresentation.Inthis situation,theLSEcanbeconsideredasimplicitinnature.

TransformationofuncertaintyfromthebasicRVstotheresponsevariable willdependonmanyfactorsincludingthetypesofthefunctionalrelationship, linearornonlinear,thetotalnumberofRVspresentintheformulation,their statisticalandcorrelationcharacteristics,etc.Obviously,theunderlyingrisk estimationprocedurewillbedifferentforeachsituation.Eachscenariowill producedifferenttypesofchallengesandwillrequiredifferentlevelsof sophisticationorexpertisetoextractthereliabilityinformation.Onlyinfew cases,thepropagationofuncertaintiesfromthevariabletosystemlevelcanbe accomplishedexactly.Insomecases,thepropagationcanbeachieved partially;insomeothercases,thepropagationcanbeachievedapproximately; insomeothercases,theinformationcanbegeneratednumerically;andin somecases,theunknownrelationshipneedstobegeneratedfromtheavailable data.Theyarediscussedbrieflyinthefollowingsections.

Fortheeaseofpresentationandconsideringthespacelimitation,the discussionsaresubdividedintotwoparts:whenfunctionalrelationshipis availableandwhenitisnot. HaldarandMahadevan(2000a) discussedmany relatedissuesinmoredetailandshouldbereferencedifdesired.

5.1Explicitfunctionalrelationship

Evenwhenexplicitfunctionalrelationshipsareavailable,theymaybeof differentforms.Someofthemarediscussedinthefollowingsections.

5.1.1Exactsolution

Supposethefunctionalrelationshipislinear,theresponsevariable Y isa functionofonebasicrandomvariable X.Therelationshipcanbeexpressedas

where a and b areknownconstants.Itcanbeshownthatthedistributionof Y willnotchange,butitsmean E(Y)andvarianceVar(Y) needtoberecalculatedas

Supposetheknownfunctionalrelationshipisnonlinear,responsevariable Y isafunctionofonebasicrandomvariable X.Therelationshipcanbe expressedas

ThePDFof Y canbeshowntobe

where x ¼ g 1(y)and n isthenumberoftermsaftertheinversion; n willbe twoforaquadraticrelationship.

Suppose Y isaknownfunctionofmultipleRVs,theexactsolutioncanbe obtainedonlyforfewspecialcases.Someofthemarediscussedinthe following.

Case1 MultipleRVswithknownjointPDF

Thiscaseisexpectedtoberare,buttheprocedurediscussedforthesingle basicRVusing Eq.(1.10) canbefollowed,asdiscussedin Haldarand Mahadevan(2000a).

Case2 SumofIndependentnormalvariables

Forthiscase,theresponsevariable Y canbewrittenas

where c0 i s areknownconstantsand X 0 i s arestatisticallyindependentnormal RVswithmeanandstandarddeviationof mXi and sXi ,respectively.

Themeanandvarianceof Y canbeshowntobe

Case3 Productofindependentlognormalvariables

Inthiscase,

Where X 0 i s arestatisticallyindependentlognormalRVswithparameters lXi and xXi .

Itcanbeshownthat

Case4 SumofindependentPoissonRVs

Suppose X 0 i s arestatisticallyindependentPoissonRVswithparameters yXi , theyarerelatedto Y as

Itcanbeshownthat Y isalsoaPoissonRVwithparameter

Case5 Centrallimittheorem

Theaforementioneddiscussionswillnotbecompletewithoutbriefly discussingthecentrallimittheorem.Itstatesthatthesumofalargenumberof RVs,ifnoneofthemdominatethesum,regardlessoftheirinitialdistributions, tendstothenormaldistributionasthenumberincreases.Similarly,theproduct ofalargenumberofRVs,ifnoneofthemdominatetheproduct,regardlessof theirinitialdistributions,tendstothelognormaldistributionasthenumber increases.

5.1.2PartialSolutions

Theexactsolutionsdiscussedintheprevioussectioncanbeverylimitedfrom theapplicationpointofview.Forsomeofthecasesnotdiscussedpreviously, itwillbedesirableifpartialsolutionscanbeobtainedtopropagate uncertaintiesfromthevariabletothesystemlevel.

PartialSolutions linearfunctionofmultipleRVs

Consider Eq.(1.11) again.Supposetheinformationondistributionof X 0 i s isunknownortheyhavedifferentdistributions,itwillnotbepossibleto determinetheexactdistributionon Y. However,itsmeanandvariancecanbe obtainedas

5.1.3Approximatesolutions generalfunctionofmultipleRVs

Ifapartialsolutionisnotpossible,itwillbereasonabletoobtainan approximatesolution.Someofthemarediscussedinthefollowingsections.

Supposetherelationshipbetween Y and X 0 i sisknown,themeansand varianceof X 0 i sareknownwithouttheirdistributioninformation.Therelationshipcanberepresentedas

Itwillnotbepossibletoobtainthedistributioninformationon Y;however, theapproximatemeanandvariancecanbeobtainedbyexpandingthefunction inTaylorseriesaboutthemeanvaluesof X 0 i s.

Theapproximatefirst-orderandsecond-ordermeans,denotedas E(Y 0 )and E(Y 00 ),canbeshowntobe

Inmostcases,onlythefirst-ordervariance,denotedas Var(Y 0 ),canbe estimatedas

where Ei and Ej arepartialderivatives vg=vXi and vg=vXj ,respectively,evaluatedatthemeanvaluesof X 0 i s.Theyareconstants.Thecoefficients E 0 i scanbe consideredasamplificationordeamplificationfactorsbasedontheirnumerical values.TheywillshowtheroleofeachRVinpropagatinguncertaintyfromthe variabletothesystemlevel.Iftheyareverymuchsmallerthan1.0,theycanbe consideredasdeterministicconstantsevaluatedattheirmeanvalues.Ifthey aremuchlargerthan1.0,resourcesshouldbeinvestedtoreducetheiruncertaintieswiththehelpofqualitycontrolandmoreinspections.

5.2Multiplerandomvariableswithunknownrelationship

Inmanypracticalcases,theexactfunctionalrelationshipof Y and X 0 i smaynot beknown;itmayonlybeknowninanalgorithmicform,forexample,withthe helpofthefiniteelementalgorithm.Inthissituation,thefirst-ordermeanand varianceof Y,withoutcalculatingthepartialderivativesasin Eqs.(1.24)and (1.25),canbeobtainedusingtheTaylorseriesfinitedifferenceprocedure. Inthiscase, Y valuesneedtobecalculatedtwomoretimesforeachRVas

and

Eq.(1.26) suggeststhat Y þ i tobeestimatedatthemeanvaluesofallthe RVsexceptthe ithone;ithastobeatthemeanplusonestandarddeviation value.Similarly, Eq.(1.27) suggeststhat Yi tobeestimatedatthemeanvalues ofalltheRVsexceptthe ithone;ithastobemeanminusonestandarddeviationvalue.Usingthecentraldifferenceapproximation,itcanbeshownthat Ei in Eq.(1.25) canbeestimatedas

Thecorresponding Var(Y 0 )canbecalculatedas

5.3Regressionanalysis

RegressionanalysisisgenerallyconductedtodevelopprobabilisticrelationshipbetweentheresponsevariableandoneormultiplebasicRVs.Developing alinearrelationshipbetweentheresponsevariableandonebasicRViscalled thesimplelinearregressionanalysis.Whenlinearrelationshipisrequiredfor multipleRVs,itiscalledthemultipleregressionanalysis.Whennonlinear relationshipisrequired,itiscallednonlinearregressionanalysis.Theconcept behindregressionanalysisisrelativelysimple;however,thereisapotentialfor misuse.Thebasicconceptisbrieflydiscussedinthefollowingsectionfor simplelinearregression.Theinterestedreadersaresuggestedtoreferto HaldarandMahadevan(2000a) and Montgomeryetal.(2012) formore advanceddiscussionsonregressionanalysis.

5.3.1Simplelinearregression

Denoting Y astheresponsevariableand X asthebasicRV, n pairsofdata (x1, y1),(x2, y2), .,(xn, yn)areavailabletodevelopalinearregression-based probabilisticrelationship.Atypicalscatterdiagramofthedataisshownin Fig.1.1.Itcanbeexpressedas

where b0 and b1 aretheinterceptandtheslopeoftheline,respectively,and theyareknownastheregressioncoefficients. Eq.(1.30) representsthemean valueoftherelationship.Someofthebasicassumptionsoftheregression analysisrepresentedby Eq.(1.30) arethatthescatterdiagramindicatesthe relationshipislinear,thespreadofthedataabouttheequationisuniform representedbyanerrorterm ε withazeromeanandunknownbutconstant variance,andtheerrorsareuncorrelatedandnormallydistributed.Thebasic assumptionsinregressionanalysisareshownin Fig.1.1

Generally,theregressioncoefficientsareestimatedusingthemethodof leastsquares.Theyareestimatedbyminimizing ε2,alsoknownastheerror sumofsquares(SSE)representingthedifferencesbetweentheobservedand predictedvaluesusingtheregressionequation. Eq.(1.30) representsthemean valueoftherelationship.Ithasalsoavariance,denotedas Var ðY jX ¼ xÞ¼ S2 Y =x ,knownastheerrormeansquareortheresidualmean square(MSE).Itcanbeestimatedas

whereSSE canbeestimatedas

FIGURE1.1 Scatterdiagramofthedata.

Theappropriatenessoftheregressionequationisevaluatedbyestimating thecoefficientofdetermination, R2.Denotingthetotalvariabilityin Y as Syy andtheamountofuncertaintyaccountedforbytheregressionequation, SSR, thefollowingrelationshipcanbeobtained:

Then, R2 canbeshowntobe

R2 willhaveavaluebetween0and1.Whenitiscloseto1.0,itrepresentsthat theregressionequationisappropriateorreasonable.Itisimportanttomention thattheextrapolationoftheequationbeyondtherangeoftheobserveddatais notrecommended.

Formultipleornonlinearregressionanalyses,theinterestedreadersare referredtothereferencescitedearlier.

6.Fundamentalsofreliabilityanalysis

Itisexpectedthatthereadersnowdevelopedthebasicunderstandingofhowto quantifyuncertaintyinanRVandhowtopropagateitfromthevariablelevel tothesystemlevel.Asmentionedearlier,thebasicintentionistoestimatethe underlyingriskorreliabilityofengineeringdesigns.Theconcepthasbeen developedovertime,anditisconsideredtobematuredatpresent.Severalrisk estimationprocedureswithvariouslevelsofsophisticationandcomplexityare available.However,beforediscussingthem,thefundamentalsofreliability analysisarebrieflydiscussedinthefollowingparagraph.

Thebasicintentofanyengineeringdesignisthattheresistance R isgreater thanthemostcriticalloadeffect S duringthelifetimeofastructure.Because both R and S cannotbepredictedwithcertainty,theyarerandominnature. RepresentinguncertaintyinthemwiththehelpoftheircorrespondingPDFsin termsofmean m andstandarddeviation s,thebasicconceptisgraphically shownin Fig.1.2.Theshadedareaisrelatedtothepossibilityof R lessthan S, indicatingtheunderlyingriskofthedesign.Obviously,itshouldbeassmallas possiblesatisfyingtheintentofthedesignguidelines.Thefigurealsoindicates thattheoverlappedtailendsof R and S cannottotallybeeliminated,indicating itwillbedifficultifnotimpossibletodesign“fail-proof”systems.

Thefigurealsoindicatesseveralwaystoreducethesizeoftheoverlapped area.Thereductioncanbeachievedbyincreasingthedistancebetweenthe twoPDFsorseparatingthemeanvaluesofthetwoRVsbyreducingthe uncertaintyinthemintermsof s value(maybewiththehelpofadditional inspectionsorqualitycontrol),orbychangingtheshapeofthetwoPDFs representingthenatureofuncertaintyinthem.Thenominalvaluesof R and S areindicatedinthefigureas RN and SN,respectively.Thenominalvalueswere

usedinthepastbythedeterministiccommunity.Thenominalvalueofthe resistance RN isseveralstandarddeviationsbelowthemean,indicatingthe amountofunderestimationofit.Ontheotherhand,thenominalloadeffect SN isseveralstandarddeviationsabovethemean,indicatingtheamountof overestimationofthem.Accordingtodeterministicpractices,iftheratio of RN/SN isgreaterthan1.0,itisanacceptabledesignandifitislessthan1.0, sayeven0.999,itisnotacceptable.Byunderestimatingtheresistanceand overestimatingtheloadeffect,thedeterministiccommunityintroducedthe safetyfactorconceptwithoutestimatingtheunderlyingrisk.Thefigureclearly indicatesthebasicweaknessintheoldpractice.

Thesamedesigncodesorguidelinesforsteel,concrete,orothermaterials areusedtodesignnuclearpowerplants,hospitals,firestations,buildings,etc. intheUS.However,theengineeringprofessionclaimsthatanuclearpower plantismuchsaferthananordinarybuilding.Itistruebecausefornuclear powerplants, RN and SN areselectedmuchmoreconservativelybyconsidering themseveralstandarddeviationsbelowandabovethemeancorrespondingto differentreturnperiods,respectively.Denoting p astheannualprobabilityof occurrenceofanevent,thereturnperiod, T,canbedefinedas T ¼ 1/p (HaldarandMahadevan,2000a).For50yearsofreturnperiod,onanaverage, theannualprobabilityofoccurrencewillbe p ¼ 1/50 ¼ 0.02.Inroutine applications,ordinarystructuresaredesignedfor50-yearreturnperiodevents, butnuclearpowerplantsaredesignedforthousandsofyearsreturnperiod events.

Incorporatingthepresenceofuncertaintyinthedesign,andreferringto Fig.1.2,theprobabilityoffailure, pf, canbedefinedas P(R < S)andcanbe estimatedas

FIGURE1.2 Fundamentalconceptofriskevaluation.PDF,probabilitydensityfunction. Adopted fromHaldar,A.,Mahadevan,S.,2000.Probability,Reliability,andStatisticalMethodsinEngineeringDesignNewYork,NY.WithpermissionfromJohnWileyandSons,Inc.

Numericalevaluationof Eq.(1.35) canbeverychallenginginmostcases. Ingeneral,both R and S areexpectedtobefunctionsofmultipleRVs,andtheir jointPDFsmaynotbeavailable.When R and S areassumedtohavespecific knownPDFsandtheinformationonthecorrespondingparameterstodefine themisavailable,closed-formnumericalsolutionof Eq.(1.35) mayrequire considerableeffort.However,forfewcases, pf canbediscussedinthe following.

R and S areassumedtobenormalRVswithknownparameters,i.e., N (m R , s R )and N ( m S , s S ),respectively.Itisreasonabletoassumethattheyare statisticallyindependent.Then, p f canbeestimatedas

Intheaforementionedequation, V isthecumulativedistributionfunction (CDF)ofthestandardnormaldistributionand b isgenerallyknownasthe reliabilityindex.Theaforementionedequationindicatesthatwhen b islarge, theunderlyingriskwillbesmaller.Inmostdesigns,thevalueof b between 3and5isexpected.

If R and S arestatisticallyindependentlognormalRVswiththesame meansandvariances, pf canbeapproximatelyestimatedas

where d istheCOVofRVs.Intheaforementionedequation,COVofanRVis consideredtobenotlarge,sayitislessthan0.3(HaldarandMahadevan, 2000a).Forthelognormaldistribution,thereliabilityindex b istheterminthe squarebracket.

Inmostcases, R and S arenotexpectedtobenormalorlognormalvariables.Evenoneofthemcouldbenormal,andtheothercouldbelognormal. Ingeneral, R and S areexpectedtobefunctionsofmultipleRVs,andthe functionalrelationshipcouldbeofanyform.Then,theevaluationofriskusing Eq.(1.35) canbeverychallenging.Forthesecases,severalriskevaluation techniqueswithvariousdegreesofsophisticationwereproposed.However, beforediscussingthem,theconceptoflimitstateneedsadditionaldiscussions.

6.1Limitstateequationsorfunctions

Riskisalwaysestimatedforaspecific limitstatefunctionorequation,itwill bedenotedhereafterasLSE.Asmentionedearlier,theyarefunctionsof

load-andresistance-relatedparametersandtherequiredperformancelimitor acceptablebehavior.Theconceptisgraphicallyshownin Fig.1.3.

AtypicalLSEcanbeanexplicitorimplicitfunctionofthedesignvariables andthedesignrequirementsexplicitlysuggestedindesignguidelinesorcodes. Theyaregenerallyrelatedtothestrengthand/orserviceabilityrequirements. Inmostcases,multipleLSEsneedtobesatisfied,indicatingmultiple requirementsneedtobesatisfiedunderthesamedesiredrisk.Thisisvery importanttoeliminatetheweakest-linkfailuremode.Therequirementsmay beatthestructuralelementlevelorbasedontheoverallwholestructural behavior.Forframedstructures,someofthecommonlyusedLSEsarepure bendingofmembers,combinedaxialandcompression,andbendingofframe members,sideswayatthetopoftheframe,interstorydrift,maximumdefectionalongthelengthofamember,etc.Itisextremelyimportanttoformulate theappropriateLSEproperly.

6.1.1Serviceabilitylimitstateequationorfunction

TheserviceabilityLSEfortheinterstory,overall,orthemaximumdeflection alongthelengthofamembercanbeexpressedas

where dallow istheallowablevalueofthespecificdeflectionofinterestmostof thetime;suggestedindesignguidelinesorcommonlypracticedand dmax(X)is themaximumvalueobtainedbytheanalysisusingtheunfactoredloads.Itis interestingtonotethatsomedesignguidelinespermitusinghigher dallow valuesdependingontheproceduresusedtoestimatethem.Forexample, ASCE/SEI7 10permitsincreasingitby125%ifitisestimatedforthe seismicloadingapplyingtheexcitationintimedomain.

FIGURE1.3 Limitstateconcept.

6.1.2Strengthlimitstateequationorfunction

SupposeastrengthLSEofasteelmemberofaframeneedstobedefined,the followinginteractionequationsareusedtodesignthemaccordingtotheload andresistancefactordesign(LRFD)guidelinesfortwo-dimensionalelements oftheAmericanInstituteofSteelConstruction(AISC)’s(AISC,2011):

where f istheresistancefactor, Pu istherequiredtensile/compressive strength, Pn isthenominaltensile/compressivestrength, Mu istherequired flexuralstrength,and Mn isthenominalflexuralstrength. Pu and Mu arethe criticalfactoredloadeffects.ThecorrespondingLSEscanbeexpressedas

where b P u and M _ ux areunfactoredloadeffects.Itisimportanttonotethatin LSEs,theloadandresistancefactorsarenotused.

6.2Reliabilityevaluationmethods

Thestageisnowsettoestimateriskorreliability.Toevaluate Eq.(1.35), severalreliabilityevaluationmethodswithdifferentlevelsofsophistication wereproposed.Initially,thedevelopmentswerelimitedtolinearortangentto thenonlinearlimitstateorperformancefunction,leadingtothedevelopment ofthefirst-orderreliabilitymethods(FORMs).In Eqs.(1.36)and(1.37),the limitstateequations(LSEs)aredefinedas Z ¼ R S and Z ¼ R/S,respectively. Ingeneral, R and S arefunctionsofmultiplerandomvariables.Togeneralize thediscussions,atypicalLSEcanberepresentedas

¼ gðXÞ¼ gðX1 ; X2 ; ; Xn Þ (1.43)

Because Xi ʹ shavedifferentstatisticalcharacteristics,itwillbedifficultto estimateexactlythestatisticalinformationon Z.However,asdiscussedearlier, theapproximatemeanandvarianceof Z canbeestimatedbyexpanding Z in TaylorseriesaboutthemeanofRVs.Thesafetyindex b canbecalculatedby takingtheratioofthemeanandthestandarddeviation.However,becausethe TaylorseriesexpansionismadeatthemeanvaluesofalltheRVs,itisknown

asthemeanvaluefirst-ordersecondmoment(FOSM)orsimplyMVFOSM. Mostofthefirst-generationreliability-baseddesignguidelinesarebasedon thisconcept.

Theaforementionedconcepthasseveralmajorweaknessesasdiscussedby HaldarandMahadevan(2000a).Theprocedurefailstoincorporatethe informationondistributionofRVsevenwhenitisavailable.BecausetheLSE representedby Eq.(1.43) islinearizedatthemeanvaluesoftheRVs,itcan introducesignificanterrorbynotconsideringthehigherorderterms.Itwas documentedinthestudyby HaldarandMahadevan(2000a) thatforthesame mechanicallyequivalentLSEsdefinedas(R S < 0)or R=S < 1 ,theproceduredoesnotgivethesamereliabilityindex.Itneedsimprovements.

6.2.1AdvancedFOSMforstatisticallyindependentnormal variables(Hasofer Lindmethod)

Duringthisphaseofdevelopment, HasoferandLind(1974) gaveageometric definitionofthereliabilityindex.Theyobservedthatinthereducedvariable space,asdefinedinthefollowing,thereliabilityindex,denotedas bHL,isthe shortestdistancefromtheorigintotheLSE.Thisobservationisvalidwhenall RVsarenormal.Thereducedvariableof Xi,denotedas X 0 i ,canbedefinedas

TheclosetpointonanLSEinthereducedvariablespacefromtheoriginis generallydenotedasthedesignorcheckingpoint.Ifitscoordinatesare expressedinavectorialformas x 0 *,itcanbeshownthat

Theconceptisshownin Fig.1.4.

Thesignificanceof HasoferandLind(1974) observationcanbeeasily demonstratedwiththehelpalinearLSEconsistingoftwoindependentRVs,as shownin Fig.1.5.Consider Fig.1.5A first.TheLSEisplottedintheoriginal coordinatesystem.Itwillbeastraightlineata45degreesangle.Thesafeand unsaferegionsarealsoshowninthefigure.InMVFOSM,thefunctionis evaluatedatthemeanvaluesofRVs,indicatingthecoordinatesofthemost probablefailurepoint(MPFP)orcheckingpointordesignpoint.Itisnoton LSEandisamajorsourceoferrorintheriskestimation.Thecoordinatesof theMPFP,denotedas(r*, s*),arealsoshowninthefigure.ItisonLSE.The authorobservedthaterrorintheriskestimationincreaseswiththeseparation distancebetweentheMPFPandthemeanvalues.

Consider Fig.1.5B now.ThesameLSEisnowplottedinthereduced coordinatesinthefigurebylineAB.ThecoordinatesofPointAare

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.