Intelligent Diagnosis Method for Open-Circuit Faults in On-Board Chargers Based on Adaptive Attention Convolutional Residual Network

The rapid deployment of electric vehicles (EV) is accelerating the electrification of transportation. During peak demand periods, EV can feed power back to the grid through onboard chargers (OBC), enabling vehicle-to-grid (V2G) interactions and enhancing grid flexibility. Consequently, accurate diagnosis of open-circuit (OC) faults in OBC is essential to ensure the reliable operation of EV and the safe functioning of the power grid. To address the complex and nonstationary characteristics of OC fault signals in OBC, this paper proposes an intelligent diagnostic model based on an adaptive attention convolutional residual network (AACRN). The proposed model integrates depthwise separable convolution (DSC) for efficient local feature extraction and residual network (ResNet) structures for deep hierarchical representation learning. Furthermore, a convolutional block attention module (CBAM) is incorporated to adaptively emphasize fault-relevant features while suppressing redundant information, thereby improving feature discrimination and enabling accurate fault localization. In addition, the model's interpretability is enhanced through a unified framework that integrates t-distributed Stochastic Neighbor Embedding (t-SNE) visualization, attention heatmaps, and shapley additive explanations (SHAP) analysis. This framework highlights fault-relevant regions and quantifies key feature contributions, thereby enhancing the transparency and reliability of fault diagnosis. Experimental evaluations demonstrate that the proposed method achieves a diagnostic accuracy of 96.54% under noise-free conditions and maintains an accuracy of 93.21% under real electromagnetic interference (EMI) in physical tests, confirming its strong anti-interference robustness in practical engineering scenarios.