Author/Authors :
WilliamL.Oberkampf، نويسنده , , TimothyG.Trucano، نويسنده ,
Abstract :
Verificationandvalidation(V&V)aretheprimarymeanstoassesstheaccuracyandreliabilityofcomputationalsimulations.V&Vmethodsandprocedureshavefundamentallyimprovedthecredibilityofsimulationsinseveralhigh-consequencefields,suchasnuclearreactorsafety,undergroundnuclearwastestorage,andnuclearweaponsafety.Althoughtheterminologyisnotuniformacrossengineeringdisciplines,codeverificationdealswithassessingthereliabilityofthesoftwarecoding,andsolutionverificationdealswithassessingthenumericalaccuracyofthesolutiontoacomputationalmodel.Validationaddressesthephysicsmodelingaccuracyofacomputationalsimulationbycomparingthecomputationalresultswithexperimentaldata.Codeverificationbenchmarksandvalidationbenchmarkshavebeenconstructedforanumberofyearsineveryfieldofcomputationalsimulation.However,nocomprehensiveguidelineshavebeenproposedfortheconstructionanduseofV&Vbenchmarks.Forexample,thefieldofnuclearreactorsafetyhasnotfocusedoncodeverificationbenchmarks,butithasplacedgreatemphasisondevelopingvalidationbenchmarks.Manyofthesevalidationbenchmarksarecloselyrelatedtotheoperationsofactualreactorsatnear-safety-criticalconditions,asopposedtobeingmorefundamental-physicsbenchmarks.Thispaperpresentsrecommendationsfortheeffectivedesignanduseofcodeverificationbenchmarksbasedonmanufacturedsolutions,classicalanalyticalsolutions,andhighlyaccuratenumericalsolutions.Inaddition,thispaperpresentsrecommendationsforthedesignanduseofvalidationbenchmarks,highlightingthecarefuldesignofbuilding-blockexperiments,theestimationofexperimentalmeasurementuncertaintyforbothinputsandoutputstothecode,validationmetrics,andtheroleofmodelcalibrationinvalidation.ItisarguedthattheunderstandingofpredictivecapabilityofacomputationalmodelisbuiltonthelevelofachievementinV&Vactivities,howcloselyrelatedtheV&Vbenchmarksaretotheactualapplicationofinterest,andthequantificationofuncertaintiesrelatedtotheapplicationofinterest.
