Study On Feature Extraction And Support Vector Machines Ensemble Method In Fault Diagnosis For Analog Circuit

Fault diagnosis for analog circuit is still a challenging subject in the circuit test research field. Due to the inherent characteristics of analog circuit, such as its nonlinearity, continuous response and tolerance on component parameters, etc, inducing the diversity and complexity of fault types of the circuit, it is difficult for the conventional fault diagnosis theories and methods to achieve the expected results in practical engineering. Hence, it is very important to explore some efficient fault diagnosis theories and methods to meet the development of analog circuit. Ensemble learning and multi-fractal analysis, which have been successfully applied in pattern identification, provide a promising solution for fault diagnosis for analog circuit. Combined with the modern test theory, signal processing and pattern identification theory, the fault diagnosis method for analog circuit based on support vector machine (SVM) ensemble learning and multi-fractal analysis is deeply researched in this dissertation. The main research contents and achievements of the dissertation are as follows:(1) The study on feature extraction for analog circuit fault diagnosis. How to quickly and efficiently extract the fault features, which reflect the state of analog circuit, is difficult and critical in fault diagnosis for analog circuit. Considering that the imperfect of feature extraction method commonly used in fault diagnosis for analog circuit, the joint time-frequency domain (JTFD) and dual-tree complex wavelet transform fractal (DTCWTF) feature extraction methods are proposed. The feature vectors, extracted by the JTFD method, represent the high order statistical characteristics of signal in time domain and frequency domain. They can reflect the different circuit state which helps to identify different fault types, and computing complexity of the JTFD method is low. The DTCWTF method, based on dual-tree complex wavelet transform and fractal analysis, makes full use of the consistency process in understanding things from coarse-to-fine and the same nature of self-similarity, and it can extract both the amplitude and phase information. The feature extracted by DTCWTF method has a certain anti-noise capability which makes it easy to design classifier.(2) The research on fault diagnosis for analog circuit based on multi-fractal analysis. Due to the bias shifting of nonlinear device and lacking generally fault model of the nonlinear circuit, fault diagnosis for nonlinear analog circuit is more difficult than that for linear analog circuit. The traditional theories and methods are not capable of solving the complex fault diagnosis for nonlinear analog circuit, but the recent development of multi-fractal theory provides a good idea to this problem. In the dissertation, the multi-fractal detrended fluctuation analysis (MF-DFA) method and the wavelet leader multi-fractal analysis (WL-MFA) method for fault diagnosis for nonlinear analog circuit are proposed. Multi-fractal attributes are estimated from each fault types of the nonlinear circuit and used as classification features within SVM classification procedure. Experiment results show that the multi-fractal features has higher fault diagnosis accuracy than other features do, such as the WP features, the JTFD features and the DTCWTF features, etc.(3) The study on fault diagnosis for analog circuit based on feature selection and SVM parameters optimization. The irrelevant and redundant features variables in actual fault diagnosis system for analog circuit spoil SVM's classification performance seriously. Too many variables increase computing times and result in real-time performance degradation. The appropriateness of selecting SVM parameters also affects the classification result. In this dissertation, two methods based on the hybrid particle swarm optimization (HPSO) and the cross entropy methods (CEM) have been proposed to choose the effective feature and to optimize SVM parameters. The two methods can perform feature selection and optimize SVM parameters concurrently. Analog circuit fault diagnosis experiments verify the effectiveness of the proposed two optimization method.(4) The study on fault diagnosis for analog circuit based on SVM ensemble method. Multi-class support machine classifier by combining of multiple binary SVMs using One-Against-All (OAA) or One-Against-One (OAO) strategy can not meet classification performance requirement in fault diagnosis for analog circuit. In the dissertation, three kinds of fault diagnosis method for analog circuit based on SVM ensemble are proposed. The first method combines hierarchical support vector machine (HSVM) with Dempster-shafer (D-S) theory to improve classification results compared to the simple decision-tree-like method. The second method combines support vector data description (SVDD) with D-S theory, which describe each class only need one class data and can deal with outlier sensitivity problem. The third method is based on improved AdaBoost-SVM, which solves the accuracy/diversity dilemma in AdaBoost algorithm by selecting more diverse weak learners, meanwhile overcomes the difficulty of selection of weak learners'parameters by logistic chaotic mapping. Fault diagnosis experiments on analog circuit show that the proposed three kinds of method have competitively learning ability and acquire better fault diagnosis accuracy than traditional fault diagnosis method based on multi-class SVM do.