Abstract:
In order to solve the difficulty of feature extraction in analog circuits fault diagnosis and to accurately classify fault pattern, a new analog circuit diagnosis method based on the combination of optimal wavelet basis, fuzzy theory and self-organizing feature map is proposed. The response signals of the analog circuit are preprocessed by wavelet transform to extract features consisting of energy, mean and variance, and the optimal wavelet coefficients with better classification capacity are obtained by using the minimum cosine separating degree. Then the feature set that is composed of optimal wavelet energy, mean and variance is inputted into a self-organizing feature map (SOM) to identify different fault cases. Simulation and experiment results show that the optimal wavelet transform based on cosine separating degree can extract analog circuit features effectively, and the fuzzy SOM neural network can classify fault cases accurately.
