Fault Detection and Diagnosis for Nonlinear System Based on Neural Network on-line Approximator

A fault detection and diagnosis method based on neural networks on-line approximation structure for nonlinear system with uncertainties was presented. The approximator, which was realized by radial basis function networks, was used for learning the nonlinear fault functions to monitor the abnormal behavior of dynamic system. When faults occured, the on-line approximator could not only detect all possible unknown faults, but also estimate the faults vector. By the definition of dead area function, it was proved that the scheme had good robustness against modeling error and uncertainties. At last, the simulations and experiment results of a three-tank system illustrate the effectiveness of the proposed method