Correlation feature selection analysis for fault diagnosis of induction motors

Abstract

This paper presents a feature selection method for stator winding fault analysis of induction motors by using a Correlation-based Feature Selection (CFS) method. The 14 original motor parameters are selected from the feature selection method with various searching approaches. The classification efficiency of optimal features obtained from the feature selection method is compared with results from the feature extraction method and the original features. In our experiment, we employ a 2.2 kW delta-connected motor which drives a dc generator as a load. The experimental results demonstrate that 4 common selected features for stator winding fault analysis of induction motors are a percent of load (%Load), a power factor (pf), a negative sequence voltage (Vn), and a negative sequence impedance (Zn). The accuracy of the classification using this feature subset is higher than using all original features for three classification methods.