Diagnostics Method for Analog Circuits Based on Improved KECA and Minimum Variance ELM

Abstract

Kernel entropy component analysis(KECA) is a new method for data transformation and dimensionality reduction. However it is sensitive to a single kernel radius. By analysis of the relation of statistics in the kernel feature space, improved KECA introduces two kernel radii and an adjusting factor to make KECA less sensitive to kernel radius. A method for fault diagnosis of analog circuits based on the combination of improved KECA and minimum variance extreme learning machine(ELM)is presented. Through wavelet decomposition of sampled signals, features are extracted. Improved KECA for feature dimension reduction is used. Then the fault patterns are classified by minimum variance ELM. Case studies on two analog circuits demonstrating our diagnostics method are presented.