Analog Circuit Fault Diagnosis Based On Data Fusion

With the development of industrialization and modernization and the blossom ofelectronic technology, the structure and size of circuit network tend to be modular andfunctional. Studying how to use modern diagnostic technology to accurately diagnosethe fault of the circuit branch and components from the analog circuit has become anurgent practical engineering problem. Analog circuit fault diagnosis, data fusiontechniques and neural network achieved rapid development in the respectiveapplications or areas that paired together, but the combination of data fusiontechniques and neural network for diagnosis of analog circuit fault is still relativelysmall. This paper studied combining data fusion techniques and neural network forfault diagnosis of analog circuits, and achieved some results. Paper work and resultsare as follows:(1)Development of analog circuit fault diagnosis technology is introduced. Andsimple expositions of the existing diagnostic methods, advanced application of neuralnetwork theory and technology and data fusion technology in the field of analogcircuit fault diagnosis are given.(2) The basic principles of the neural network method and its advantages in faultdiagnosis of analog circuits are particularized, including the classification of theneural network and neural network learning rule. Using the BP neural network, themost widely used tool in the circuit fault diagnosis and a dual-stage RC amplifiercircuit is tested.(3)The concept of data fusion technology, the advantages and disadvantages ofthe basic methods and their applications in various fields of military, civilian isintroduced. And specific methods for data fusion, especially the Bayesian statisticalfusion method, fusion method of Dempster-Shafer evidence theory and fuzzy settheory fusion method are listed. A circuit fault diagnosis based on method ofundetermined coefficients and the fuzzy integration method is demonstrated.(4)A new data fusion diagnostic method using the static working point voltageswhich contain components’ DC characteristics and output peak voltages underdifferent frequency motivates which contain components’ AC characteristics as twotypes of information is proposed. Through two neural network diagnosticdecision-making level information fusion and applying the fuzzy theory, fuzzy transform results of the two kinds of fault information are obtained. Diagnosticaccuracy is significantly improved compared with the methods using single neuralnetwork diagnostic information, for lowering the uncertainty caused by insufficientfault information and the interaction of circuit components.The data fusion and neural network-based approach proposed by this paper canachieve accurate real-time rapid diagnosis of analog circuits fault and have a certainpractical value.