Battery System Fault Detection: A Data-Driven Aggregation and Augmentation Strategy

In applying machine learning to battery system fault detection, current methods encounter some challenges. Inadequate extraction of discriminative features and data imbalance notably hinder the accuracy and robustness of detection. This study proposes a novel data-driven approach that synergistically combines data aggregation and feature augmentation strategies. The methodology first implements data aggregation to address data scarcity by effectively expanding fault sample representation. Subsequently, an advanced feature augmentation process is employed to enhance feature separability through multi-dimensional transformation techniques. Leveraging these enhanced datasets, this study develop an optimized Light Gradient Boosting Machine model specifically tailored for fault detection tasks. Comprehensive experimental evaluations demonstrate that our approach achieves marked improvements in both detection accuracy and robustness compared to conventional methods. These advancements not only enable more precise battery system diagnostics but also present a generalizable paradigm for other industrial fault detection applications requiring robust performance under data constraints.