A robust fault diagnosis scheme based on signal modal estimation

A real-time fault detection and diagnosis technique for linear dynamic control systems is proposed. It provides fault detection and diagnosis using neither observer residuals nor parameter estimation errors, instead, it relies on the estimation of the underlying modal parameters of the dynamic system, and compares the estimates with the pre-calculated characteristic patterns which are represented as a set of root loci of physical parameters. The modal estimation is carried out using a numerically robust least square algorithm based on SVD (Singular Value Decomposition). A pattern recognition technique based on linear multiprototype distance functions is used to classify the faults according to the variation of physical parameters. The method possesses several advantages over the existing techniques: (i) the nature of the fault can be easily identified since the scheme uses physical parameters, rather than model parameters, for classification; (ii) the effect of disturbance on diagnosis is minimized because the modal estimation algorithm treats the disturbance as additional dynamics which are eliminated in the classification stage using truncated SVD; (iii) it is sufficient to use only one measurement signal, since any signal within the control loop contains all necessary modal information for fault diagnosis; and (iv) faults which cause various amount of parameter variation can be easily accommodated by proper selection of parameter ranges in constructing root loci. The method has successfully been applied to a DC servo system