Bearing Scratch Fault Detection by Three‐Dimensional Features and a Support Vector Machine

Abstract

Induction motors play a crucial role in various industries owing to their high robustness. The demand for early fault detection is getting attention to avoid serious damage to the machines. Bearing fault is the most common failure in induction motors and the possibility of scratches has a higher probability among the various classes of the bearing faults. Recently, the effective diagnosis method considering the progression and orientation of the scratch fault by using a machine learning algorithm have been reported. However, the diagnosis of such scratch faults with high accuracy is still required and challenging for early fault detection. In this article, the scratch faults are diagnosed with high accuracy rate and the detailed analysis are conducted. A support vector machine algorithm is used to diagnose the faults, where sideband frequency components of the load current are used as the features. In order to obtain high accuracy rate, additional features are considered, where in this study, the rotating speed of the motor and a higher order frequency component are selected as candidates and compared. The rotating speed along with sideband components of load current shows high accuracy rate. Furthermore, the robustness of the method is tested against the multiple faults. The results show that the rotating speed is an effective feature for highly sensitive diagnoses of bearing scratch faults.