Wavelet Transform based fault diagnosis in analog circuits with SVM classifier

In this work, the diagnosis of hard and soft faults in analog circuits has been addressed using Wavelet Transform as a preprocessor and Support Vector Machine (SVM) as a classifier. Test circuits have been excited with random analog signal and the output responses have been analyzed with Daubechies Wavelet Transform. Principal component analysis (PCA) has been implemented to reduce the dimension of extracted features and faults are classified in principal component spaces with the help of supervised machine learning. The proposed algorithm is validated for two benchmark circuits (simulated with UMC-180nm PDK in CADENCE Virtuoso and processed using MATLAB 2019): Two-Stage OPAMP and second-order Sallen-Key band-pass filter. The use of a random signal in the proposed method minimizes the cost of the generation of the test signal. The potentiality of Wavelet Transform for time-frequency analysis of output responses has been utilized for characterization and subsequent fault diagnosis of the circuits. The accuracy and other performance parameters have been measured to show the effectiveness of the proposed method.