| As the rapid development of modern science and technology, the network size of analog integrated circuit is growing, leading to the circuit fault type diversity and testing nodes accessibility, thus increasing the difficulty of component level fault location. At the same time, because of the continuous improvement of the level of circuit modularity, the necessity of circuit component level fault diagnosis is reducing. Large-scale circuit fault diagnosis research focus is turning into a module level fault diagnosis theory and methods of research.After years of research and development, wavelet analysis, neural network method, network tearing method and cross-tear search method has been effectively applied to the study of large-scale circuit fault diagnosis, achieving certain results. A fault diagnosis method combining network tear and neural network avoid large amount of fault information which lead to can’t be accurate fault located and slow diagnosis rate. This method has opened up a new and effective way for large-scale circuit fault diagnosis researching, it is of great research significance.This paper first summarizes the analog circuits fault classification and diagnosis methods, focuses on the application of neural network in circuit fault diagnosis, discuss the design method of wavelet neural network used in the analog circuit fault diagnosis method. On the basic of analysis advantages and disadvantages of network tear method, step by step method, slip tear method and cross search method that used in large-scare circuit, a new large-scale analog circuit fault diagnosis method which combines group tearing method and improved wavelet neural network is proposed. This method tears the large-scale analog circuit according to the topological characteristics and group tear standards to get a low dimension fault feature vector. By using the highly parallel processing ability of improved wavelet neural network and selecting wavelet function with fast convergence property to be the underlying layer excitation function, achieve rapid classification of fault characteristic vectors. Combined with the logical diagnosis of group tear, the diagnosis results are obtained. Diagnosis examples show that this fault diagnosis method for large-scale circuit diagnosis has small amount of work before test, much less amount of diagnosis and calculation, the high ability for more fault detection and wide practicability. |
