Probabilistic neural network based tolerance-circuit diagnosis

An approach to fault diagnosis for analog circuits with tolerance is presented based on probabilistic neural networks. In order to overcome the difficulties in BP network based diagnosis such as slow learning speed for convergence and easily falling into local minimum value, probabilistic neural network is introduced to tolerance-circuit diagnosis. Fault samples including soft faults and hard faults in tolerance circuits are generated by Monte Carlo analysis. Fault features are extracted by using the largest deviation path so as to obtain appropriate training samples. Simulation results show that the proposed diagnosis method has high speed and accurate recognition even for soft faults in circuits with tolerance.