Deep Transfer Learning Technology-Based Condition Monitoring and Fault Diagnosis of Electric Vehicle Electric Powertrain Systems: A Review
Haoxiang Xu, Zicheng Liu, Dong Jiang, Ronghai Qu, Jie Tian
Abstract
The safety and reliability of battery, motor, and power electronics systems in Electric Vehicles (EVs) are critical determinants of vehicle operation and passenger safety. Precise condition monitoring and fault diagnosis are essential for ensuring optimal electric powertrain systems performance. EVs operate under complex and variable conditions with diverse onboard system configurations, resulting in dynamic changes in monitoring data's statistical characteristics and distribution discrepancies between newly collected and training data. Traditional intelligent methods are vulnerable to data distribution bias, leading to significant degradation in monitoring and diagnostic performance when models encounter new data. Deep Transfer Learning (DTL), which integrates Deep Learning's (DL) feature extraction capabilities with Transfer Learning's (TL) knowledge transfer mechanisms, effectively enhances model applicability and robustness. This paper addresses the prevalent issue of data distribution differences in cross-domain maintenance of EV electric powertrain systems by first introducing DTL's fundamental concepts and general procedures. It then analyzes specific implementation strategies and recent research developments of DTL across different crossdomain scenarios in the three core systems. The study further conducts an integrated experimental and multidimensional analysis to evaluate DTL paradigms' cross-domain maintenance performance, exploring their necessity, task-dependent, and engineering applicability. Finally, it examines current technical challenges and future research directions.