Fault detection in linear discrete dynamic system based on detectability

The author proposes three effective fault detection methods for a linear discrete system in the framework of a generalized-likelihood-ratio (GLR) method. These are a reduced-order step-hypothesized GLR (SHGLR) method which models the anomaly vector function by a reduced-order step function, a reduced-order tracking functional subspace method which estimates the anomaly vector function by using reduced-order system information; and a pattern recognition based detection method (PRBM) which detects the fault by monitoring the pattern of the (reduced-order) SHGLR. The PRBM is superior to the other two methods in that it is easy to implement the case of complicated multiple-faults in time-varying systems using it