Faulty Class Diagnosis of Three Phase Induction Motor Bearing Using Stator Current Spectral Features and Machine Learning Algorithms

Induction motors are widely used in various industries because of their robustness, which makes them attractive for applications in harsh environments. Fault detection is a topic of increasing interest, particularly for bearing faults. Various methods of bearing fault diagnosis have been proposed, including vibration, acoustic, and current signature analysis. To predict the repercussions of bearing faults, the detection of the fault class and number of faults is of particular interest. However, the above diagnostic methods only consider a single bearing fault. In this study, inclusive diagnoses were performed for detecting the class (i.e., holes and scratches) and number of faults by using the frequency-domain features of the load current. In experiments, faults of different classes and numbers were introduced to the outer raceway of bearings and tested at various load levels. The sideband frequency components of the load current were affected by the fault class and number. A support vector machine was applied to fault diagnosis using the sideband frequency components as features. Electromagnetic simulations suggested that the dependence of the feature distribution on the fault class and number could be attributed to the effect of the eddy current on the outer raceway of the bearing. The results demonstrated the robustness of the proposed diagnostic method against the class and number of bearing faults.