SVDD-SVM based fault class diagnosis method for incomplete analog circuits

A method combining support vector data description (SVDD) and support vector machine (SVM) is proposed to deal with the fault diagnosis problem of analog circuits with incomplete fault categories. First the new feature vectors are sorted by the contribution of the data by applying principal component analysis, and a weighted Gaussian kernel function is constructed for the contribution of the feature quantities to improve the difference of features between classes. Then the kernel parameters are optimized using the mucosal algorithm (SMA) to improve the robustness of the diagnostic model and the correct identification rate of the sample data to facilitate efficient diagnosis of normal and known fault samples. Finally, SVDD is used to partition the sample space for diagnosing unknown fault data. Experiments show that this method can efficiently diagnose known faults in analog circuits while judging unknown faults.