Study On Fault Characterization, Classification And Diagnosis In Analog Circuit

In recent years, with the rapid development of electronic technology, the reliability of electronic equipment put forward higher requirements. Testing and fault diagnosis of analog circuit are hotspot research in the circuit test field. And the theory and methods for analog circuits fault diagnosis is an extremely challenging research topic. It is difficult for the traditional fault diagnosis theory and methods to achieve the expected results in practical application. The analog circuit fault model has no simple universal representation model, and the inherent characteristics of analog circuits, such as tolerances on component parameter, continuous variation of the element parameter, nonlinearity response, etc, are effecting the development of the analog circuit test and diagnostic methods. Hence, it is important to explore highly efficient fault diagnosis theory and methods to meet the development of analog circuits. The main works and contributions are summarized as follows:(1)Since the integration circuit is continuously improved and the information provided by the limited test nodes is not enough to reflect all components’ status, the test circuit is bound to have some ambiguity groups. How to find ambiguity groups and determine the diagnosable components set in the test circuit becomes an important part of the fault diagnosis. In this study, a method base on circuit topology is firstly described. The test matrix can be calculated by the transfer function between test nodes and inputs of the test circuit, and the ambiguity groups are determined on the basis of the test matrix, and then the separable potential faulty components under the assumption of single fault can be finally determined. In our study, we also propose another two new methods. One method is based on the Pole-zero sensitivity of the transfer function, which is suitable for the single-output test point circuit and has fast calculated rate. In the single-fault assumption, this method is a simple way to find the ambiguity groups and the diagnosable components set. The other method is based on the voltage slope between the different test points, which is suitable for the multi-test point circuit and does not need analysis of the transfer function. In linear analog circuits, we can use the superposition theorem and the substitution theorem to find all components’ slope characteristics. According to the slope characteristics, the surely testable elements and globe ambiguity groups can be found accurately.(2)A new analog circuit fault diagnosis method based on improved Mahalanobis Distance is proposed. The Mahalanobis Distance was first introduced by an Indian statistician P. C. Mahalanobis to represent the covariance distance between data. It takes into the consideration of the association between different characteristics and can effectively calculate the similarity of two unknown sample sets, independent of measurement metrics. In our study, the Mahalanobis Distance is introduced into analog circuit fault detection, and then is improved according to the characteristics of analog circuit. After that, the suspicious components could be classified using the improved Mahalanobis Distance according to the feature values of the test points. The introduction of the improved Mahalanobis Distance can exclude some potential faulty components that have no faults occurred in advance in most cases so as to effectively reduce the possible types of faults and enhance the speed of fault detection.(3)A new analog circuit fault diagnosis method based on Support Vector Machine(SVM) with parameter optimization is proposed. SVM is an efficient machine learning method and it has been developed rapidly in recent years. SVM has perfect theoretical foundation, when to solve the classification problems of small sample, nonlinear or/and high dimensional data, it has advantages of simple structure and strong generalization ability. After a detailed study on fault diagnosis performance by SVM kernel function type, kernel parameters and penalty parameters, we recognized the urgent need to find an effective method of parameter optimization of SVM. In order to get the optimal parameters, a new parameter optimization method with Chaos-PSO is proposed and compared to other methods, such as the multiple parameters grid method, the GA method and the PSO method. Since the introduction of chaos algorithm, the parameter optimization process can be made more easily escape from the local extreme points, and the problem of premature convergence of PSO can be solved. This method is easy to implement, and the parameters obtained by this method let the SVM get higher accuracy of classification than other methods.(4)Power system fault and analog circuit fault have many similarities, e.g. They both have complex fault model, their complete fault samples are difficult to obtain, their responses are continuous and nonlinear, and they are both affected by noise and so on. Nowadays, traditional fault diagnosis method can no longer meet the development requirements of power system which is becoming more and more complicated, thus it is necessary to apply the fault diagnosis method based on support vector machine with parameter optimization in power system. The research indicates that SVM-based protection method has more advantages than the artificial neural network method and other conventional methods. After the training, SVM protection classifier can correctly respond to bus normal operation or fault and can meet the requirement of accuracy protection. Owing to the application of parameter optimization, support vector machine has better fault recognition ability and generalization ability as well as higher fault examination accuracy rate. The bus protection system and transformer system can both get satisfactory results.