A hybrid fault diagnostic approach using operational transfer path analysis and denoising deep learning with remote sensors: Application to electric vehicles

Abstract

This study proposes a new approach to diagnose drivetrain faults in electric vehicles through the use of remote sensors placed under the driver's seat. The research aims to address the reduction in diagnostic accuracy caused by the receiver's lower sensitivity when diagnosing faults in the drivetrain (source) located far from the receiver. The proposed approach builds on transfer path analysis (TPA), identifying critical paths between the source, transition locations, and receiver. A denoising deep learning model is proposed for operational TPA (OTPA) to overcome the limitation of conventional TPA methods, such as singular value decomposition. The denoising deep learning model takes the vibration signal captured at the receiver location and converts it through the transition locations into the vibration signal that is intended to closely represent the original vibration signal from the source location. Finally, a diagnostic deep learning model classifies drivetrain faults using the synthetic vibration signals from OTPA. The experimental results of real-scale electric vehicles show that the proposed approach effectively diagnoses the electric vehicle drivetrain faults using remote sensors. © 2025 Elsevier Ltd