Induction Motor Bearing Fault Diagnosis Using Statistical Time Domain Features and Hypertuning of Classifiers

Abstract

Condition monitoring of induction motors plays a significant role in avoiding unexpected breakdowns and reducing excessive maintenance costs. In the majority of cases, bearing faults are found to be an issue in the failure of induction motors. The detection and valuation of irregularities at an early stage can help prevent disastrous failures. In this paper, the detection and classification of bearing faults in an induction motor are performed using machine learning techniques. The current signal from two different phases is recorded for three motor conditions: healthy, inner race fault and outer race fault. The statistical features are then applied for dimensionality reduction. Finally, the statistical features are used as the input of classifiers, including support vector machines (SVMs), random forests (RFs), and k-nearest neighbor (KNN). The grid search method is used to estimate the best-suited metaparameters for each classifier to achieve the best performance in fault classification. With the regularization parameters, all the classifiers achieve over 98% classification accuracy.