Motor Bearing Fault Detection Using Spectral Kurtosis-Based Feature Extraction Coupled With K-Nearest Neighbor Distance Analysis

Abstract

Bearing faults are the main contributors to the failure of electric motors. Although a number of vibration analysis methods have been developed for the detection of bearing faults, false alarms still result in losses. This paper presents a method that detects bearing faults and monitors the degradation of bearings in electric motors. Based on spectral kurtosis (SK) and cross correlation, the method extracts fault features that represent different faults, and the features are then combined to form a health index using principal component analysis (PCA) and a semisupervised k-nearest neighbor (KNN) distance measure. The method was validated by experiments using a machinery fault simulator and a computer cooling fan motor bearing. The method is able to detect incipient faults and diagnose the locations of faults under masking noise. It also provides a health index that tracks the degradation of faults without missing intermittent faults. Moreover, faulty reference data are not required.