The Research On The Wavelet Neural Network Based Fault Diagnosis Approach For Analog Circuits

With the rapid development of electronical industry and consistent expansion of integrated scale of components, the analog circuit diagnosis has attracted more and more attentions from all aspects. The study on the analog circuit diagnosis has been performed for more than 40 years, but it is rare to be put into practice. The main reason is that the theory has restriction so that it is hard to diagnose the tolerance and the nonlinear circuits especially the ultra large-scale analog and the digital-analog circuit .Currently, the research and application for the wavelet transformation theory and neural network has become a hot spot in the field of fault diagnosis. The main purpose of this article is to explore a new method that can solve problems of analog circuit diagnosis by means of combining newest progress of analog circuit diagnosis with that of wavelet neural network.Firstly the analog circuit fault diagnosis research's situation is summarized, and the neural network and the wavelet analysis elementary theory are investigated. We discuss two ways of the neural network and the wavelet analysis in the electric circuit fault diagnosis on the basis of the Back Propagation neural network:1.the loose: The wavelet has good localization nature in the time and frequency domain and neural network's character, so the paper discusses that the fault signal is pretreated with the wavelet transformation, which can reduce the input nodes of the neural network. In addition, according to the character of the wave packet and the energy of the signal, the paper gives the examination method and the simulant circuit; 2.the Compact: According to wavelet function's fast astringency, Mexican Hat wavelet substitutes network's S function, which constructs one kind of new wavelet neural network. The article makes use of the noise to be testing signal, which can reduce many unnecessary steps (example: the noise were filtered in the traditional methods) and will reduce costs in practical. Finally the article makes use of the phase to act as fault feature vector approach. Two examples were simulated with Orcad10.0 and Matlab6.5. The test results indicate the methods are feasible and effective.