Fault detection using an adaptive fuzzy system

A fault detection approach is presented for Lipschitz non-linear dynamic systems with bounded measurement noises. This approach does not require any nominal models of actual systems in the presence or absence of faults. It uses an adaptive fuzzy system (AFS) to remember the past behaviour, and it continually updates the mathematical representation of the actual system. The basic framework of the updating process is the same as Wang and Mendel's, but the description of fuzzy rules, the creation of the combined fuzzy rule base, the defuzzification strategy and some other details are greatly different from Wang and Mendel's. This updating process also consists of five steps. The first four steps are devoted to the AFS updating, the last step with the determination of the AFS output mapping. Based on the error analysis on the AFS, this paper proposes a fault-detection strategy that depends on the deviation of the actual system output from the AFS output. Simulations show that this approach can be used effectively in the detection of abrupt or incipient sensor faults. Although this approach is designed for systems with bounded measurement noises, it can also be used for systems with unbounded measurement noises -- if a limited bound that covers most of the measurement noises exists.