Multi-task learning framework for fault detection in energy storage system lithium-ion batteries: From degradation to slight overcharge

Fault detection and state of health (SOH) estimation are both critical for ensuring the safety and reliability of lithium-ion battery energy storage systems (BESS), yet conventional methods often treat them separately, leading to inefficient data utilization and limited fault sensitivity under subtle conditions such as slight overcharge. To enhance diagnostic efficiency and address the challenges of data scarcity, this study proposes a multi-task learning framework that leverages both aging data and limited fault data to simultaneously achieve SOH estimation and overcharge detection in BESS. To overcome inefficiencies of separate models, the framework introduces a shared feature factor within a comprehensive loss evaluation mechanism, allowing a single model to effectively handle both SOH estimation and slight overcharge detection. To further tackle the issue of data imbalance in slight overcharge scenarios, a Mixup data augmentation method was developed, ensuring consistency with the original fault dataset. Testing results demonstrated that the framework achieved an SOH estimation accuracy of 97.8 % and a fault detection accuracy of 97.1 %. Additionally, to meet practical deployment needs, the framework was optimized for efficiency, with memory usage and floating-point operations per second as key criteria, reducing inference time by 55.5 %.