Design of fault detection and isolation via wavelet analysis and neural network

Abstract

A knowledge-based FDI scheme is developed by integrating the time-frequency signal processing technique with neural network design. Wavelet analysis is applied to capture the fault-induced transients in the measured signals and, furthermore, the decomposed signals can be used to extract details about the fault. A Regional Self-Organizing feature Map (R-SOM) neural network is then used to isolate the fault. The R-SOM neural network proposed in this paper has achieved higher clustering and matching-up precision compared with the conventional SOM network, especially when noise, disturbance and other uncertainties occur in the system.