Fault Feature Extraction And Intelligent Fusion Approaches For Analog Circuit Diagnosis

Analog fault diagnosis has been an active area of research since the 1960s with significant work and methods carried out. Unfortunately, the progress of analog circuit fault diagnosis from the fundamental theory to sophisticated technology has been hampered by nonlinear effects, component tolerances, poor fault models and other factors. Information fusion is widly applicated to dealing with pattern recognition, identification and diagnosis for fault location. It is hopeful that application of data fusion to the area of analog circuits diagnosis may achieve better results. In this dissertation, the research is focused on seeking analog circuit diagnosis approaches based on intelligent information fusion. The fault feature extraction and decision-fusion based intelligent diagnosis algorithms for analog circuits are dealt with in detail.In order to solve the problem that accessible nodes are too limited to get sufficient diagnosis information in analog circuits, especially in large scale circuits, an algorithm for temperature information extraction is present based on K-nearest-neighbor classitifcation rule.In order to overcome the weaknesses of Dempster-Shafer theory including conflict fusion and one-ticket-veto, a new analog circuit diagnosis method based on heterogeneous information fusion is proposed. The characteristic temperature information is extracted by K-nearest-neighbor classification rule and then corresponding primary fault identification is performed. Meanwhile, another separate primary fault diagnosis is dealt with by a neural network based on accessible node voltages. By calculating prior weight coefficient and correlation weight coefficient under the consideration of information inconsistency, the synthetic diagnosis is completed and reliable fault diagnosis results are obtained.Meanwhile,another analog-circuit fusion diagnosis approach is developed by using transferable belief model so as to overcome the weaknesses of Dempster-Shafer theory. By using transferable belief model, the basic belief assignment, belief function value and combination evidence are obtained from the separate primary fault identification based on accessible node voltages and temperature information, respectively. And then, by using the algorithm of pignistic transfermation in transferable belief model, the synthetic diagnosis is completed and the accuracy of fault diagnosis is increased.The simulation experiment results show that the proposed methods have the capability to diagnose faults in analog circuits and gain satisfactory accuracy.