The Study Of Expert System Based On Neural Network For Analog Circuit Fault Diagnosis

Along with the rapid development of electronics industry, the analog circuit fault diagnosis question already aroused the widespread interest, moreover, it is a big difficult problem to the domestic and foreign experts when they design and use the electron system. Some analog circuit diagnosis methods which already existed are only suitable under the special condition, for example, opening, short circuit and so on. It is very difficult to discover the soft fault which caused by electronic device's tolerance change in electric circuit. It is a question that has seldom been studied in, this thesis will research on this problem.In view of the traditional limitation of diagnosis technology, we discussed the plan which use neural network (NN) to diagnose soft fault of analog circuit, and wavelet transformation (WT), which include multi-resolution analysis and wavelet packet analysis to extract the circuit fault character and NN which has misalignment mapping characteristic to approaches the fault diagnosis model, also the thesis proposed a improved fault extraction arithmetic based on wavelet packet analysis. A concrete simulate diagnosis circuit is used to train NN. The diagnosis result indicated that the method which the thesis proposed is effective and this research will provide the new theory basis and the diagnosis method for the complex analogous circuit failure diagnosis even integrated circuit.In addition, the tradition fault diagnosis ES exists insufficient which cannot carry on self-study, auto-adapted, difficult to gain the knowledge, and match conflict when it inference, and so on. This article uses the wavelet analysis and NN technology as part of the construction of ES. Use wavelet analysis extract analog circuit fault characteristic, the ES knowledge gaining part is replaced by neural network(NN)'s training, and use the connection power and the threshold value of back propagation (BP) NN which had been well trained replaces the ES knowledge library. The ES inference part is completed through the operation of the weight data and the NN input data. In view of a low pass filter electric circuit, we research and develop an ES based on BPNN for analog circuit. The system uses MATALB GUI programming to realize the following functions: Firstly, four WT algorithms realization of feature extraction, user can choose the different diagnosis algorithm according to the user's needs to realize to the fault feature extraction; Secondly, set parameters of BPNN. User can set hid level integer, study rate size and so on; Thirdly, NN training and fault diagnosis.This system makes up the flaws which traditional ES could not self-study, auto-adapted, and overcomes the insufficiency of the traditional electric circuit fault diagnosis method and enhances the diagnosis automation and the intellectualized level.