KPCA- KELM Ensemble based Incipient Fault Diagnosis Approach of Analog Circuits

Abstract

An ensemble approach based on kernel principal component analysis and kernelized extreme learning machine, called KPCA - KELM, is proposed in this paper to deal with the incipient analog circuits diagnosis problem of insufficient generalization performance, too much training time and low fault recognition rate. Firstly, we set the single and multiple fault value of the circuit under test (CUT) and collect output response signal by frequency sweeping. Then, kernel principal component analysis (KPCA) is used for preprocessing the output characteristic data, obtaining the data dimension reduction of the corresponding CUT response signals, reducing system operating costs and leading the minimum loss of information. Finally, we use KELM for training and forecast, and add contrast experiment by the original extreme learning machine (ELM). The results verify the positioning ability of component fault diagnosis for CUT is effective.