Analog circuit diagnosis using RBF network and D-S evidential reasoning

In order to solve the possible problems in neural-network based analog fault diagnosis including lack of fault information, slow training speed and difficult converge, a novel data-fusion based fault diagnosis approach for analog circuits is presented by using radial basis function (RBF) networks and D-S evidential reasoning. The manifold transducer information and symptoms were utilized in diagnosis. The map from symptom space to fault pattern space was constructed by the separate RBF network for each kind of symptom information. The output results of every RBF network were then aggregated using the D-S evidential reasoning algorithm. Fault location was accomplished based on the synthesis decision regulation. The experimental results show that the proposed approach can effectively combine the evidences to produce a more accurate diagnosis and has the capability to diagnose catastrophic and parametric faults of analog circuits with tolerance.