Use of Feature Ranking Techniques for Defect Severity Estimation of Rolling Element Bearings

Bearings are the most common components used in rotating machines. Their malfunction may result in costly shutdowns and human causalities which can be avoided by effective condition monitoring practices. In present study, attempt has been made to estimate the severity of defect in bearing components by a two-step process. Initially, defects of various severities in all bearing components are classified. In the next step, if defect exist in any of the bearing components, i.e. inner race, outer race, and rolling elements, level of severity of defect is estimated. Various statistical features are extracted from the raw vibration signals. Two machine learning techniques; support vector machine and artificial neural network, along with four feature ranking techniques; Chi-square, gain ratio, ReliefF and principal component analysis are used and employed for the analysis. Results show the potential of the proposed methodology in defect severity estimation and classification of rolling element bearings.