Improved diagnostics for the incipient faults in analog circuits using LSSVM based on PSO algorithm with Mahalanobis distance

Diagnostics of incipient faults for analog circuits is very important, yet very difficult. Traditionally, the soft faults can only be diagnosed under the occurrence of large variation of component parameters. In this paper, a novel approach for diagnosing incipient faults in analog circuits is proposed. First, a statistical property feature vector composed of range, mean, standard deviation, skewness, kurtosis, entropy and centroid is proposed. Then, the least squares support vector machine (LSSVM) is used for diagnostics of the incipient faults. Conventionally, multi-fault diagnosis for analog circuits based on SVM usually used a single feature vector to train all binary SVM classifier. However, in fact, each binary SVM classifier has different classification accuracy for different feature vectors. Thus, the particle swarm optimization (PSO) based on Mahalanobis distance (MD) is proposed to select a near-optimal feature vector for each binary classifier. The experimental results for three analog circuits show: (1) the accuracy using the near-optimal feature vectors is better than the accuracy using a single feature vector, and is also better than the accuracy using the optimal single feature vector; (2) the accuracy using the near-optimal feature vectors is close to the accuracy using the optimal feature vectors selected by the exhaustive method; (3) the accuracy using the near-optimal feature vectors based on LSSVM is better than the accuracy obtained by hidden Markov model; (4) the consuming time of the near-optimal feature vectors selected by MD is reduced by about 98% in comparison to the time of the optimal feature vectors. & 2014 Elsevier B.V. All rights reserved.