The Research On Fault Diagnosis For Analog Circuits Based On(Multi) Wavelet(Packet) Analysis、Neural Networks And Optimization

Study on the fault diagnosis of analog circuits is one of the most challenging hot points in the field of circuit test. Because of the inherent characteristics such as the continuity and non-linearity of responsiveness and tolerance of component parameters in analog circuits as well as the diversity and complexity of analog circuit faults, expected effects are difficult to achieve in practical application of traditional fault diagnosis theory and methodology. With the rapid development of modern electronic technology, the size and structure of network will be functionized and modularized gradually. It becomes a pressing issue in practical engineering and the key step for practical application of the fault diagnosis theory and methodology in analog circuits to study the way to find a fast, accurate and efficient diagnostic method for analog circuit faults by using modern diagnostic technique. With the emergence and development of wavelet theory, the theory and methodology of neural network become more sophisticated. The wavelet is used to analyze and process the fault signal, and the neural network is used to carry out fault diagnosis. Both of them have become the hot subject for research. A large number of studies have indicated that these methods appear as new approaches for the fault diagnosis in analog circuits. Based on the theories of neural network, wavelet (packet) analysis, multi-wavelet transform, genetic algorithm, particle swarm algorithm and signal processing, the paper intensively studies on the fault feature extraction and fault diagnosis methodology. The researches of the Paper are mainly as follows:(1) Explaining the neural network and wavelet transform and the theory of wavelet packet and multi-wavelet transform; describing the common BP neural networks in detail; discussing several improved BP algorithms in respect of the imperfection of BP algorithm; exploring the emerging research focus in the field of wavelet theory-multi-wavelet transform.(2) On the basis of analysis and elaboration of theoretical study on neural network and (multi) wavelet (packet) transform, exploring their combination-the application of (multi) wavelet (packet) neural network in the fault diagnosis of analog circuits; study on the compact type of wavelet neural network and multi-wavelet neural network’s structure, learning algorithm and approximation properties more profoundly.(3) Discussing the extraction of fault feature vectors, key step for fault diagnosis of analog circuits, in detail. In the Paper, four methods, i.e. the wavelet extraction, the optimal wavelet packet extraction, the analysis extraction on principal components, the multi-wavelet transform extraction, are explored to extract the feature vectors for faults. Based on examples of diagnosis, the advantages and disadvantages of four methods are analyzed and studied.(4) Exploring the method for parameter optimization of neural network. Such deficiency as slow convergence for network and limiting in local optimization commonly exists in the traditional diagnosis for analog circuits based on neural network. This paper studies the optimized structures and parameters of neural network by using genetic algorithm and particle swarm algorithm respectively. Compared with research of traditional neural network, study of neural network provided by these methods include not only the correction of network weights but also the adjustment of other parameters in neural network. Respective diagnosis examples show that the optimized neural network has been further improved in respect of the accuracy and speed of fault diagnosis.(5) Exploring the fault diagnosis methods for analog circuits based on the extraction of signal’s kurtosis and skewness and combined with information fusion from the perspective of high-order cumulants and data fusion. In addition, examples are listed to demonstrate the efficiency and feasibility of this method.(6) To enhance the practicality of the paper, given the MATLAB source code program of analog circuits fault diagnosis by using compact-type neural network, genetic wavelet neural network and particle swarm wavelet neural network, it has been run through on the author’s simulation platform.