Analog Circuit Fault Diagnosis Using Multi-source Information Fusion

Since 1970s,analog fault diagnosis has been an active field in the circuit research and outstanding achievements have been made.However,because of the diversity of analog circuit faults,dispersedness and widely nonlinear effects of the component,it is not practical either in the theory or in the methods.With the development of large scale integrated circuit,it's difficult to extract sufficient accessible node voltage information as the fault feature in the practical diagnosis.Consequently,it limits the improvement of fault diagnosis accuracy.As a powerful tool of information processing,information fusion makes it possible to overcome the difficulties in analog circuit diagnosis.In the dissertation,the research focuses on analog circuit diagnosis approaches based on multi-source information fusion,and makes a deep research on the fault information processing.Under the consideration of the characteristics of the analog circuit and its diagnosis,the function model of analog circuit fault diagnosis based on multi-source information fusion is constructed according to the latest achievements from domestic and overseas research.Combining the advantages of neural network with D-S evidence theory,a fault diagnosis method is proposed based on heterogeneous information fusion.First,the fault feature with normalization is extracted from accessible node voltages and element temperature.Then preliminary diagnosis is performed separately by two independent neural networks.According to the preliminary diagnosis results,the mass function of each fault state is calculated.The results of the integration diagnosis and ascertains the corresponding belief intervals are obtained.The fault diagnosis results indicate that the proposed approach not only improves the accuracy of fault diagnosis,but also gets the trust value of the diagnosis results.Since the circuit temperature field will change in the fault condition,the heat exchange is introduced in the paper.The non-destructive diagnosis is performed by comparing the circuit board's thermal images in fault states with the normal state.Furthmore,this paper tries to search an automatic diagnosis method for analog circuits by analyzing the temperature field of circuit board.