The Research Of Analog Circuit Fault Diagnosis Method Based On Wavelet Analysis And Neural Network

With the rapid development of modern electronic technology, electronic products gradually develop toward integrated, modular and refinement direction. Fast and accurate analog circuit fault diagnosis technology has become an objective need no matter from the perspective of cost structure or safety performance.This paper fully summarizes the traditional and new modern diagnostic techniques. The research is based on the BP neural network, and wavelet analysis and information fusion technology are used as analytical tools. Fault feature extraction, pre-processing of the input vector, selection of diagnostic methods and the improvement of diagnosis rate are four aspects we focus on. The main results are as follows:(1) Analyze and introduce the current development status of analog circuit fault diagnosis technology, summarize the existing diagnosis methods. Expound the advanced neural network theory, BP neural network algorithm and diagnostic procedure, and illustrate the superiority of BP neural network in fault diagnosis with examples.(2) In the aspect of fault feature extraction method, wavelet analysis for fault signal, which combining with the concept of entropy theory kurtosis, can effectively extract wavelet entropy and kurtosis of reconstructed signal. According to the the signal energy distribution characteristics and the degree of signal deviating from the normal state, the circuit fault feature is extracted. From the point of the example analysis, the wavelet entropy and kurtosis can be used to distinguish the different fault condition, effectively reduce the processing of information, which laid a foundation for the effective diagnosis.(3) In the aspect of pre-processing of the input vector, with the theoretical basis of wavelet analysis, multi-resolution analysis and information fusion, the sum of the absolute values of coefficient sequences for each frequency band signal and energy feature are extracted as characteristic signals. In this paper, an analog circuit fault diagnosis method based on wavelet analysis and information fusion is put forward using feature fusion technology of information fusion. Finally the effectiveness of this method in the aspect of single and multiple soft failures diagnosis is illustrated with a practical example. The experimental results show that this method has high diagnostic accuracy and fast diagnosis characteristics, and it can improve the ability to identify failure category.(4) In the aspect of improving the diagnosis rate, this article uses wavelet function as the transfer function between the input layer and the hidden layer of the neural network to form embedded WNN(Wavelet Neural Network). In this paper, on the basis of embedded WNN, the wavelet entropy and kurtosis characteristics are extracted as eigenvectors to carry out fault diagnosis, combining with the information fusion method. The results illustrate the effectiveness and feasibility of the method.