Multi-classifier fusion approach based on data clustering for analog circuits fault diagnosis

When there are large amount of fault classes in analog circuits, normally single multi-class classifier cannot achieve satisfactory diagnosis accuracy because of its difficult training process. A method of multi-classifier fusion diagnosis approach based on data clustering is presented in this paper to improve fault diagnosis veracity. After extracting fault feature vectors by wavelet transform, fuzzy C-mean clustering algorithm is used to pre-partition the feature space into multiple sub-class groups as binary tree. According to the structure of the fault tree, multi-classifiers are created to form hierarchical diagnosis system. Simulation experiments demonstrate that the proposed approach for analog circuit fault diagnosis is superior to conventional ones. The fault diagnosis accuracy is greater than 98%. It has good performance in tackling large number of fault classes in analog circuits.