Multi-fault diagnosis for Lithium-Ion batteries under diverse operating conditions based on multi-source domain generalization

Lithium-ion battery fault diagnosis is crucial for ensuring the safety and reliability of electric vehicles and energy storage systems. Machine learning is a promising solution due to its powerful feature extraction capabilities. However, the method relies on the training data with extensive experimental cost. Moreover, domain shifts caused by varying operating conditions may result in false diagnostics. To address these issues, this paper proposes a novel multi-source domain generalization framework based on battery simulation model for multi-fault diagnosis, which is suitable for diverse operating conditions and limited data availability. Specifically, multi-fault (short circuit, low capacity and connection faults) dynamics are holistically coupled into battery electrochemical-aging model. The model offers dense and high-quality training dataset under different cycle conditions, temperatures, fault types and severities. A forgetting factor nonlinear transformation is proposed to process battery raw data, which captures temporal dependencies and amplifies anomalies. Domain adversarial transfer learning is coupled with residual network to extract fault features. This effectively reduces the effect of operating conditions on fault diagnosis. The performance of the multi-fault diagnosis method is validated through multiple fault experiments with different operating conditions, and further verified with different battery types between training and test cells. The results demonstrate high feasibility and diagnostic accuracy of the proposed method, with correct diagnoses for all test samples and robust performance across different battery types.