A cloud-edge collaborative dual-modal fault detection block for battery packs: Enabling intelligent safety management in electric vehicle battery systems

Intelligent diagnosis of battery pack faults using real-world data has become a key research focus in the era of vehicle-cloud integration. While cloud-based deep learning diagnostics have shown strong capability, edge-based fault diagnosis is increasingly preferred for its real-time responsiveness, lower communication bandwidth demand, and enhanced data privacy. This study aims to address the challenge of simultaneously achieving model light-weighting and high diagnostic accuracy for onboard deployment. To this end, we propose a novel battery-pack fault detection block with dual modeling modes for cloud and edge environments. In cloud mode, the core model is designed to capture spatio-temporal features, and its structural parameters are optimized to improve global fitting and structural modeling capability. The cloud model is then compressed into a lightweight edge-deployable model using a three-stage knowledge distillation strategy. This strategy largely preserves diagnostic accuracy while satisfying the resource constraints of onboard battery management systems (BMS). Embedded-system tests further verify the deployability and efficiency of the lightweight model on resource-constrained microcontroller units (MCUs), achieving an inference time of 27.20 ms and energy consumption of 1.63 mJ. In addition, a multi-scale fault diagnosis method based on residual thresholds is implemented, yielding high performance with an Accuracy of 0.938, Precision of 0.935, Recall of 0.993, and F1-score of 0.963. The proposed framework provides a practical solution for real-time and intelligent battery safety management in electric vehicles, especially in scenarios with limited cloud connectivity and strict onboard resource constraints.