Combined method based on neural network for control system fault detection and diagnosis

In this paper, a combined method based on neural network for control system fault detection and diagnosis is proposed to overcome the drawbacks of the existing methods. In order to make the fault feature in residual clearer and more recognizable, the residual is prolonged by using multi-scale wavelet transform. Then the prolonged residual is sent to a neural network, which is made as an intelligent classifier. After being trained properly, the neural network can exactly detect and diagnosis smaller faults than those detected by using the existing methods. The most important advantage of the method is that the probability of false alarm and miss alarm can be suppressed at the same time by training the neural network on-line and off-line alternately. Although the Kalman Filter-Based method is taken as an example in this paper, it can be made as a general combined method for control system fault detection and diagnosis.