Abstract:  Multiplefaultdiagnosisunderunreliabletestsusingthemultisignalmodelwasstudied.TheoptimalmultiplefaultdiagnosisproblemisNPcomplete,sosuboptimalalgorithmwasusedforlargescalesystemdiagnosis.Agoodfaultestimationunderlowfalsepositiveratewasobtainedbythesubgradientoptimizationmethod.Butifthesizeofthegraphisbigandthefalsealarmrateishigh,thisalgorithmwill notbeabletoeliminatetheinfluenceoffalsepositive,andtheresultingestimationwouldcovereveryfailedtestinsteadoffindingthetruefaultpositionofthesystem.Anewkindofdiscriminatefunctionconsideringbothfalseandtruepositivetestswasused,andanimprovedgeneticalgorithmwasproposedtosearchthefailurecomponents.Simulationresultsshowthatthenewmethodisfasterthansubgradientoptimizationandhasahighercorrectisolationratewhenintercurrentfaultsarefew.

Key words: Genetic algorithm, Multi signal model, Fault diagnosis, False alarm, Spacecraft