Research And Realization Of Fault Diagnosis System Of Analog Circuit Based On Support Vector Machine And Adaptive Resonance Theory

Currently, the theory and method of analog circuit fault diagnosis are in the challenging area of the front and hot research topic in the international. As the semiconductor technology and the rapid development of technics, analog circuit becomes large-scale and integrated, rises circuit complexity, the difficulty of diagnosis is also increased. Therefore, the traditional analog circuit fault diagnosis theory and method in practical engineering is difficult to achieve the desired results. Looking for high efficiency and meeting the development needs of analog circuit fault diagnosis theory and methods is very important. In recent years, support vector machine of statistical learning theory becomes a new method in analog circuit fault diagnosis, and the domestic and foreign hot research topic.Study on circuit fault diagnosis for feature extraction, classification and so on, proposed a multi-class classification based on support vector machine and the adaptive resonance theory. The main research contents are as follows:K Research on the circuit fault signal acquisition and feature extraction. Due to the analog circuits with tolerance and non-linear, using PSpice for the special function of fault analysis of monte carlo state of different signal, wavelet packet transform to fault signals feature extraction. Of the analog circuit output signal changes in frequency components of energy and different fault signal corresponding relationship to the energy vector as fault feature vectors. Different faults for analog circuits, we proposed the optimal wavelet packet transform (Optimal Wavelet Packet Transform, OWPT) and incomplete wavelet packet transform (Incomplete Wavelet Packet Transform, IWPT) signal feature extraction method. Experimental results show that the optimal wavelet packet transform suitable for soft fault feature extraction, incomplete wavelet packet transform suitable for hard fault feature extraction.2^Research on support vector machine combined with the adaptive resonance theory of multi-class classification algorithm. Focuses on the principle and implementation algorithm of support vector machine (Support Vector Machine, SVM) and Adaptive Resonance Theory (Adaptive Resonance Theory, ART). Selection of kernel function and optimization of different parameters were studied and simulated about support vector machine. Analyzes and compares the advantages and disadvantages of three algorithms about the support vector machine one-to-many(1-versus-rest,1-Vr), one to one (1-versus-1,1-V-1) and decision directed acyclic graph (Decision Directed Acyclic Graph, DDAG) on multi-class classification algorithm. For one to one support vector machine classification algorithm when the classifier output close to0or the same number of votes, the voting method will appear in wrong decisions or refuse to issue, a real-time online fault diagnosis algorithm was proposed. Based on multi-class classification algorithm of support vector machines combined with the adaptive resonance theory solve these problems. Meanwhile, discuss the structure of the neural network design, training algorithm. The BP threshold vector is introduced during data preprocessing, and classification accuracy is improved. The simulation results indicate the effectiveness of this algorithm. As the same time with a variety of fault diagnosis methods are compared, results showed that fault diagnosis method about the combination of ART and SVM has higher accuracy.3、Based on Matlab7.1platform, developed analog circuit fault diagnosis system, the design of fault diagnosis system modules, realize the analog fault diagnosis based on improved SVM-ART-b algorithm.