The Research On Method Of Fault Diagnosis For Analog Circuits Based On Genetic Algorithm, Wavelet Analysis And Neural Networks

As modern electronic technology develops very rapidly, the research on the fault diagnosis theories and methods for analog circuits are becoming very popular and also challenging. The applications of large-scale integrated circuits allow the scale and the structure of a network to be more and more functional and modularized with the development of modern electronic industry. So it is an urgent subject in practical project to study how to identify the fault sub-circuits and the fault elements correctly from large-scale circuits with tolerance by using modern diagnosis technique, and it is also a key step for practical application of analog networks fault diagnosis theory. As appearance and rapid development of wavelet theory, while increasingly maturation of neural networks theory and its application, it has been becoming a hot studied project to apply wavelet analysis with neural networks to locate faults. The wavelet analysis is used to analyze and process fault signals at first, then neural networks is used to classify and locate faults. Many results of research show that they provides a new way for fault diagnosis of analog circuits. The new methods for fault diagnosis of analog circuits based on the theories of genetic algorithm(GA), wavelet analysis and neural network(NN) is in-depth studied. The author's research results has been introduced, which mixed-signal circuits and electro- mechanical systems testing based on the DSP, NN and expert system. The main works in the dissertation are as follows:(1) The research on the neural network based method of fault diagnosis for analog circuits has been done. Firstly, the structure and principle of analog circuits fault diagnosis system based on the NN is analyzed in detail, the research on the application of the BP neural network(BPNN) in fault diagnosis is also made. Then due to the fact that the BPNN usually converges to local minimum, a new improved way is considered in this dissertation. At Last, a fast approach of module level fault diagnosis for large-scale analog circuit based on the neural network and the crossover tearing technology according to conventional circuit decomposition technology in large-scale circuits is presented in this dissertation.(2) In order to overcome the weakness of the diagnosis technology based on neural network, a new analog circuit fault diagnosis approach based on information fusion D-S evidence theory and NN is developed. It makes full use of the redundancy and complementation of multi-sensor information, the robustness of D-S evidence theory for processing imprecise and uncertain information, integrates efficiently and applies the classification capability of BP network. Thus, the testability of the diagnosed circuit is improved with less strict requirements for the topological structure of it, and the satisfactory accuracy and applicability of the approach is achieved at a low computational cost.(3) The research on wavelet nueural network method for analog circuit fault diagnosis has been done. The method includes two ways. One way uses good time- frequency localization property of wavelets, firstly wavelet transform preprocesses fault signals of analog circuit and distills feature of fault, and uses neural network to process the fault feature, then it is applied in fault diagnosis for analog circiuits. Another, used wavelet functions instead of the activation functions of the traditional BPNN, and then it is applied to diagnosis analog circuits.(4) The research on genetic algorithm based wavelet nueural network for analog circuit fault diagnosis has been done. A systematic approach combining neural network , wavelet analysis and genetic algorithm is proposed for fault diagnosis of analogue circuits. The presented neural network is developed with the improved network weighted reasoning method. The optimal feature sets are extracted to train the network by using wavelet analysis as a preprocessor. This ensures a simple architecture for the neural network and minimizes the size of the training set required for its proper training. And the adjusting of connection weights and optimization of membership functions are performed with genetic algorithms. The reliability and comparison of this method with other methods are shown by active examples, and the results of experimental tests show that this method can satisfactorily detect and identify the faults.(5) The research on mixed-signal circuits and electro-mechanical system testing has been done. In this chapter, firstly, the test theory of the testing, hardware block diagram, the software procedures have been analysed; then the communication design of DSP and PC for the testing has been introduced.