Bearing defect signature analysis based on a SAX-based association rule mining

Abstract

Association rule mining provides the feasibility by taking an inverse approach for bearing defect signature analysis to directly mine associations between labeled defects and defect features instead of traditional forward fault diagnosis steps. Different from the common uniform partitioning approach used in association rule mining, a novel association rule mining approach has been proposed, based on a discretization method Symbolic Aggregate approXimation (SAX). In the presented method, the extracted features from sensing measurements are discretized and transformed into symbolic sequences according to the equalized distribution of the data. Next, the association relation between discretized features and labeled defect modes (or defect severities) is dug to generate the rules. The presented method partitions data equiprobably and avoids centralization or dispersion of the data, thus achieving more effective association rules for analyzing bearing defects. Experimental studies on bearing test data reveal that the proposed method is capable of generating a number of meaningful association rules in bearing defects and outperforms the discretization methods based on equal density and equal width technique.