Deep Transfer Learning Based Fault Diagnosis of Electric Vehicle Motor

Abstract

Electric motors are the core component in the Electric Vehicles (EVs) for providing rotational motion to the transmission system of the vehicle. Compared to diesel engines, these electric motors have no carbon emissions and are more reliable in operation. An early fault diagnostic system is necessary to ensure its effectiveness and reliability for EV applications. It can aid in the improvement of operational safety and the avoidance of unplanned breakdowns. Traditional diagnostic techniques are less effective in dealing with real-time and changing operating conditions. This work proposed an acoustic signature-based transfer learning for the identification of EV motor faults at an early stage using Constant Q-Transform (CQT) based on time-frequency scalograms. Transfer learning-based approach is adopted to facilitate the faster prediction in case of limited dataset availability. The proposed technique is validated on two distinct datasets: the bearing dataset is used as the source dataset, and the electric vehicle motor dataset is used as the targeted dataset. The suggested method facilitates and accelerates the training of an acoustic-based deep learning to provide accurate fault identification in a shorter time period. The experimental findings show that the suggested approach for identifying EV motor problems is substantially accurate and reliable.