Analog Circuit Fault Diagnosis Based On Fisher Classifier

With the rapid development of semiconductor technology, electronic circuits has been applied widely. But the scale of circuits become huger, the environment changes of circuits become more complex, and the limited by nonlinear, tolerance or test node of circuits make analog circuits’s fault diagnosis difficulty. According to statistics that failure probability of analog circuits about 80 percent in the entire circuit. So the research on analog circuit fault diagnosis is significant in the theory and practice.Fisher discriminant analysis studies and trains the selected fault samples by using statistic method to separate them first, and then calculate and compare distances between fault samples, At last realize pattern recognition are presented of fault samples. On the basis of fisher discriminant analysis method, the paper mainly from three aspects:the higher order cumulants with Fisher, Fisher combined with wavelet packet and kernel function optimizes Fisher. The researches of the paper are mainly as follows:In this paper, a new method of analog circuit fault diagnosis was proposed based on higher-order cumulants and Fisher discriminant analysis. First, selected circuit model to do simulation and Monte Carlo analysis, sampling response information by these points which can be measured, and then higher-order cumulants analyzes the original sampling information, extracting signal’s feature vector with kurtosis and skness structure of the corresponding mode, finally, Fisher discriminant analysis method for fault pattern classification gets the final pattern recognition result. The simulation result shows that the proposed method has the capability to diagnose faults in tolerance circuits and achieves satisfactory accuracy, the accuracy has exceed 90.46%.At the same time, in order to optimizing the accuracy of the training sample, wavelet packet can be used sampled signals, and extracted the band energy value as the fault feature vectors, then combined with Fisher discriminant analysis method for diagnose and analyze the circuit; In order to confirm the versatility of proposed method, selecting typical filter circuits verifies the validity of wavelet packet and Fisher discriminant analysis method for fault diagnosis of analog circuits. Analog diagnosis examples illustrate that this method is effective for fault location and the accuracy was over 90.77%.In consideration of the complexity of practical problems and nonlinear relationships between datas, Fisher discriminant analysis sometimes identifies nonlinear problems between failure modes difficultly. So the paper uses kernel function to optimizing the Fisher discriminant analysis, mapping datas to kernel space, changing nonlinear problem into a linear problem, and then the Fisher discriminant analysis recognises problems. The result is improved by 5.38%.The simulation results of examples given in this dissertation show that the fault diagnosis methods proposed above have good diagnosis effect and feasibility in analyzing the fault response of analog circuits and can locate the faults in analog circuits correctly.