Fault detection in dynamic systems based on fuzzy diagnosis

The detection and classification of faults in time-invariant dynamic systems involve tasks associated with system identification and pattern recognition. The purpose of the paper is to present the design of a process of fault detection and classification. The faults are characterized by a permanent perturbation on physical parameters of the original system, an event that is detected by monitoring a state-space model of the system, subject to recursive parameter estimation. The main component of the estimation process is a Hopfield-type neural network. The evolution of the parameter values at the output of the parameter estimator is continuously analyzed and if their behavior matches some pattern of permanent perturbation, the process of fault diagnosis indicates the source of the fault. This is a pattern recognition problem, and its implementation is accomplished using fuzzy rules, designed from a signed directed graph.