A Novel Method for Analog Fault Diagnosis Based on Neural Networks and Genetic Algorithms

A systematic method based on a neural network that utilizes a genetic algorithm (GNN) and the deviation space to diagnose faulty behavior in analog circuits under test (CUTs) is presented in the paper. To reduce the computational requirement of network simulations, we derive a unified fault feature, which can be extracted from measurable voltage deviation in the deviation space. The extracted unified feature vectors for single, double, and triple faults are characterized on the basis of measurable voltage deviation in the deviation space. Then, the faults can be classified by applying a neural network (NN) whose inputs are extracted from independent measurements-the transfer impedances at accessible nodes or the corresponding feature of various faults. It is applicable to linear circuits as well as nonlinear ones. The method presented minimizes the online measurements and offline computation. Illustrative examples verify the effectiveness of the proposed method.