Analog circuits fault diagnosis based on μSVMs

Analog circuit fault diagnosis problem can be modeled as a pattern recognition problem and solved by machine learning algorithm. SVM is often chosen as the learning machine because of its good generalization ability in small sample decision problem. However, in practical applications, because the fault samples are hard to acquire, the number of fault sample is far less than that for normal samples, which makes fault diagnosis a typical imbalanced problem. And it is found that traditional SVM can not ensure good performance in this situation. So in this paper, we propose an improved SVM-μSVM. In the new method, a parameter μ was introduced into the decision function, so that weight for fault class can be adjusted, and consequently the influence of fault class in decision function can be enlarged. Simulation experiments show that this method is effective in solving the problem of analog circuit fault diagnosis.