Application of Wavelet Packet Analysis and Probabilistic Neural Networks in Fault Diagnosis

Abstract

In order to enhance fault diagnosis precision, the wavelet packet analysis and probabilistic neural networks (PNN) are combined effectively. First, by selecting proper parameters, the power spectrum of fault signals are decomposed by wavelet analysis, which predigests choosing method of fault eigenvectors. Second, a method of fault diagnosis based on PNN is presented. The method uses Bayesian classifying and decision making theory to constitute the mathematic model of system, with Gauss function as activating function. The model possesses the characteristics of strong nonlinear processing and arti-interfering ability, by which the rotating machinery fault can be identified and diagnosed effectively. The fault set data are entered noises and both back-propagation neural networks (BPNN) and PNN are used to fault diagnosis of rotating machinery fault. The simulation results show that when the sample sets do not contain any noises or the noises are comparatively small, the diagnosis success rates of both BPNN and PNN are quite high. When noises rise, the diagnosis success rate of PNN is much higher than that of BPNN, which shows the PNN validity in anti-jamming ability and diagnosis success rate.