A New Method for Feature Selection and Gear Defect Detection

This paper presents a new method for gear defect detection. Firstly, the feature subset that describes a gear health state is generated, followed by two sub-steps: (I) features that are robust to data noise are extracted from the collected data utilizing the information-theoretical concept of entropy; (II) the optimal feature subset for gear defect detection is selected using a wrapper approach. Secondly, patterns of the feature subsets that describe the most current gear health states are grouped into a health-map, using the self-organizing map (SOM) method. This health-map shows a clear clustering of healthy and faulty gears, and can be used as a reference for future gear health evaluation. An industrial case study is presented that shows the effectiveness of the presented method.