Soft fault detection and isolation in analog circuits: some results and a comparison between a fuzzy approach and radial basis function networks

This paper provides a comparison between two techniques for soft fault diagnosis in analog electronic circuits. Both techniques are based on the simulation before test approach: a “fault dictionary” is a priori generated by collecting signatures of different fault conditions. Classifiers, trained by the examples contained in the fault dictionary, are then configured to classify the measured circuit responses. The suggested classifiers have similar structures. The first is based on a fuzzy system, obtained by processing fault dictionary data for automatic generation of IF–THEN rules, and the second classifier is based on a radial basis function neural network. The two classifiers are used to detect and isolate faults both at the subsystem and component levels. The experimental results point out that both classifiers provide low classification errors in the presence of noise and nonfaulty components tolerance effects. The fuzzy approach provides better results due to an efficient generation method for the IF–THEN rules that allows adding IF parts in the input space regions where ambiguity occurs.