Deep Learning Based Fault Detection Method in DC Motor Start

Abstract

DC motors and solenoid starters used in railway industry are susceptible to erosion due to the high current flow in the DC motor circuit, which can cause a melting effect in copper contacts and disk inside the solenoid. This underscores the importance of having an accurate and reliable fault detection mechanism to prevent disruptions in locomotive operations. Conventional fault detection methods typically rely on either visual inspection and/or signal analysis of current, speed, and voltage. However, signal analysis technique may be less accurate if the solenoid structure causes the disk to rotate after each start-up process. This paper proposes a deep-learning based method to reliably enable fault detection on the surface of solenoid starters of DC motors. The proposed method uses image segmentation as a powerful technique in computer vision which can be effectively used for fault detection by identifying and localizing faults or defects. The results indicate that this method can reliably automate fault detection with balanced accuracy of 98% across a broad range of faults, each with varying degrees of area damage including those invisible to the naked eye.