The Research Of Analog Circuit Fault Diagnosis Method And Its Application

With the rapid development of electronic technology, the integration of analog circuits is increasing and its application is more and more widely. So reliability and stability of the circuit is self-evident. When the circuit breaks down, the ability of accurate and timely locate faulty components as a reflection of the reliability of the analog circuits. Therefore, doing research on analog circuit fault diagnosis is of great theoretical and practical significance. The nonlinear problem and large tolerance issues of the electronic components in analog circuits make the fault diagnosis of analog circuits more difficulty. Recent years, Artificial Intelligence (AI) and signal analysis have developed quickly and have been widely applied in fault diagnosis of analog circuits which have been achieved some results. This paper also does some search on this aspect.Analog circuit fault diagnosis is essentially a pattern recognition problem, it mainly includes two aspects:feature extraction and the design of classification. In this paper, model features use a combination of state variables which collect from several pivotal test points to describe the fault model. By this way, we can avoid the problem of inadequate information when we just sample one test point. Then we use wavelet analysis technology which has better characteristics in time-frequency domain to acquire wavelet features of sample signals. Wavelet features can express the mainly information furthest. In order to improve the classifier's diagnostic efficiency, using PCA to reduce the dimension of wavelet data. About the classifier, Two-pass Classification method with Biomimetic Pattern Recognition (BPR) and Error Correcting SVMs is used. In order to take advantage of BPR's low error rate so that we make it as a pre-classifier. About the second classifier, we introduce in error correction coding technique which views the classification of the various models as the transmission of information. The second classifier we used is Error Correcting SVMs and it can correct a certain number of classification errors. At last, the paper uses the method described earlier to detect the actual circuit and achieves good result. The result verifies the effectiveness of the method we proposed.