Circuit Fault Location Based on Dynamic Bayesian Network

Abstract

To improve the accuracy of analog circuit fault diagnosis, on account of the problem that is difficult to obtain a high accuracy of the test results for a single model, based on combinatorial optimization theory, an analog circuit fault diagnosis model based on dynamic Bayesian network is proposed. Firstly, circuit fault features are extracted, and then hidden Markov model and least squares support vector machine are used to establish combination diagnosis model of analog circuit fault, and finally the simulation experiment is used to analyze the performance of combination models. The results show that compared to other analog circuit fault diagnosis models, the proposed model not only improves the accuracy of analog circuit fault detection, but also has faster speed of fault diagnosis. Introduction With the development of ultra-large-scale integrated circuits and increasingly complex analog circuits, because of electric parameter drifts and non-linear characteristics, there are numerous causes for the occurrence of analog circuit fault, and the analog circuit fault diagnosis has been the difficult problems in mixed-signal circuit fault diagnosis [1]. Analog circuit fault diagnosis is essentially a problem of pattern classification, including two key steps of feature extraction and classifier design [2]. Feature extraction is the first step in an analog circuit fault diagnosis, because of the complexity of the analog circuit itself, as well as impacts of a variety of external factors, it is featured by nonlinear change. There is currently a lot of methods for analog circuits feature extraction, such as wavelet analysis, Volterra series methods [2], in which Volterra series can better capture the nonlinearity of circuit signal change with certain advantages, thus this paper employs the extracted fault features of analog circuit. The most original analog circuit fault diagnosis classifier is established by multiple linear regression method, but it is difficult to establish accurate fault classification model, and the correctness and timeliness of diagnosis results cannot meet the actual needs of the project [3]. Hidden Markov Model (HMM) has outstanding capabilities of pattern classification and has been widely used [3] in fault diagnosis. In recent years, with the development of nonlinear theory, among it, the neural network is most representative with outstanding nonlinear fitting ability. Analog circuit fault diagnosis has been successfully applied, but because the neural network is a kind of machine learning algorithms based on empirical risk minimization principle, there exist defects such as a large number of training samples, easily occurred over-fitting and others, and its applications in fault diagnosis are affected [3]. Support vector machine is excellent in performance, but the computational complexity of the training process and low efficiency cannot meet the real-time requirements of analog circuit fault diagnosis. In allusion to the problems of complexity in analog circuit fault and unavailable test results with high accuracy for single model, based on combinatorial optimization theory, this paper has proposed a combined analog circuit fault diagnosis model based on dynamic Bayesian model. Firstly, features of circuit fault are extracted, and then the combined analog circuit fault diagnosis model is established using dynamic Bayesian model, and finally the diagnostic performances of the model are analyzed by simulation experiments. 3rd International Conference on Mechanical Engineering and Intelligent Systems (ICMEIS 2015) © 2015. The authors Published by Atlantis Press 646 Extracting features In 1880, Italian mathematician Vito Volterra proposed concept of Volterra series, and as for nonlinear systems, he could recommend the transfer function similar to linear systems. Firstly, the energy values of Volterra series sequence h1 (m), h2 (m1, m2), h3 (m1, m2, m3) and h4 (m1, m2, m3, m4) are calculated and taken as fault features of circuit to be diagnosed [4].