Using supervised kernel entropy component analysis for fault diagnosis of rolling bearings

Abstract

Rolling bearings as vital components are present ubiquitously in manufacturing machines and systems, timely fault diagnosis is of great importance in avoiding serious catastrophe, and a variety of methods have been developed to detect faults of rolling bearings. In this study, we introduce labels to adjust the distance of classes and propose a method called supervised kernel entropy component analysis for fault diagnosis of rolling bearings. The method is developed based on kernel entropy component analysis which attempts to preserve the Renyi entropy of the data set after dimension reduction and presents a good performance for nonlinear feature extraction. Simulation and experiments are carried out to verify the method. The intrinsic geometric features derived from original vibration signals are extracted, and then classified by support vector machine. The results prove the feasibility and effectivity of the supervised kernel entropy component analysis in comparison with other similar methods, demonstrating its potentiality for fault diagnosis of rolling bearings.