Resilient fault diagnosis and recovery framework for driving modules of electric vehicles using continuous wavelet transform-based fault-aware deep learning

초록

In complex and interdependent systems such as electric vehicles, maintaining operational resilience requires timely detection and mitigation of component-level faults. Among these, current sensor faults in Permanent Magnet Synchronous Motor (PMSM) widely adopted for their high efficiency and controllability can propagate through control loops and lead to severe system-level disruptions. Such faults, if undetected, may result in degraded motor performance, unstable outputs, and even safety-critical failures. Therefore, robust and real-time fault diagnosis is crucial for ensuring resilient operation of PMSM-based EV systems. In order to address this, this study proposes a new and effective fault diagnosis and signal recovery framework based on Continuous Wavelet Transform (CWT)-based data conversion and a deep encoder-decoder architecture integrated with a confidence-guided fault-aware weighting mechanism. This process enables both accurate fault diagnosis and effective restoration of gain and zero-offset fault signals through a dual encoder-decoder model, in which the contribution of fault-specific encoders is adaptively weighted based on fault classification confidence. The proposed framework demonstrates higher fault diagnosis accuracy and resilient signal recovery capability compared to existing methodologies through experiments conducted using an implemented EV driving simulator that replicates real road driving cycles. These analyses demonstrate that the proposed CWT-based fault-aware deep learning framework contributes to real-time fault diagnosis and resilient control of PMSMs during driving.

키워드