A robust fault diagnosis scheme based on signal modal estimation

摘要

A real-time fault detection and diagnosis technique for linearndynamic control systems is proposed. It provides fault detection andndiagnosis using neither observer residuals nor parameter estimationnerrors, instead, it relies on the estimation of the underlying modalnparameters of the dynamic system, and compares the estimates with thenpre-calculated characteristic patterns which are represented as a set ofnroot loci of physical parameters. The modal estimation is carried outnusing a numerically robust least square algorithm based on SVD (SingularnValue Decomposition). A pattern recognition technique based on linearnmultiprototype distance functions is used to classify the faultsnaccording to the variation of physical parameters. The method possessesnseveral advantages over the existing techniques: (i) the nature of thenfault can be easily identified since the scheme uses physicalnparameters, rather than model parameters, for classification; (ii) theneffect of disturbance on diagnosis is minimized because the modalnestimation algorithm treats the disturbance as additional dynamics whichnare eliminated in the classification stage using truncated SVD; (iii) itnis sufficient to use only one measurement signal, since any signalnwithin the control loop contains all necessary modal information fornfault diagnosis; and (iv) faults which cause various amount of parameternvariation can be easily accommodated by proper selection of parameternranges in constructing root loci. The method has successfully beennapplied to a DC servo system