多尺度PCA在传感器故障诊断中的应用研究

？Amultiscaleprincipalcomponentanalysismethodisproposedforsensorfaultdetectionandidentification.Afterdecompositionofsensorsignalbywavelettransform,thecoarse-scalecoef-ficientsfromthesensorswithstrongcorrelationareemployedtoestablishtheprincipalcomponentanalysismodel.Amovingwindowisdesignedtomonitordatafromeachsensorusingthemodel.Forthepurposeofsensorfaultdetectionandidentification,thedatainthewindowisdecomposedwithwavelettransformtoacquirethecoarse-scalecoefficientsfirstly,andthesquarepredictionerrorisusedtodetectthefailure.Thenthesensorvalidityindexisintroducedtoidentifyfaultysensor,whichprovidesaquantitativeidentifyingindexratherthanqualitativecontrastgivenbytheapproachwithcontribution.Finally,theapplicabilityandeffectivenessoftheproposedmethodisillustratedbysensorsofindustrialboiler.