Enhanced bearing fault detection using multichannel, multilevel 1D CNN classifier

Abstract

Electric motors are widely used in many industrial applications on account of stability, solidity and ease of use. Mechanical bearing faults have the highest statistical occurrence percentage among all of the motor fault types. Accurate and advance detection of the bearing faults is critical to avoid unpredicted breakdowns of electric motors. Through early detection of bearing faults, it would be possible to solve the problem at a lower cost by repairing and/or replacing relevant parts. Most of the fault detection works in the literature attempted to detect binary {healthy, faulty} motor fault case based on a single input. In this study, we propose an enhanced performance bearing fault diagnosis system based on multichannel, multilevel 1D-CNN classifier processing vibration data collected from multiple accelerometers mounted on bearings in a test bed. Effectiveness and feasibility of the proposed method are validated by applying it to the benchmark IMS bearing vibration dataset for inner race and rolling element faults and comparing the results with the conventional single-axis data-based fault detection.