Fusing Equivalent Circuit Models and Explainable Machine Learning for Internal Short Circuit Diagnosis in Lithium‐Ion Batteries

Internal short circuit (ISC) in lithium‐ion batteries (LIBs) is one of the key triggers for thermal runaway, making its early diagnosis crucial for ensuring battery safety. To address the accuracy dependency of traditional model‐driven methods and the “black box” dilemma of data‐driven approaches, this paper proposes a novel diagnostic framework that deeply integrates equivalent circuit model (ECM) with interpretable machine learning (ML). First, this study establishes a dedicated ECM for the constant‐current (CC) charging phase and achieves high‐precision state of charge (SOC) estimation through the extended Kalman filter. Subsequently, features are extracted from incremental capacity (IC) curves and SOC charging curves and organically combined into a hybrid feature set to comprehensively capture fault information. Based on this feature set, we construct a LightGBM diagnostic model that achieves 98.61% accuracy in ISC multilevel diagnosis. Most importantly, we employ SHapley Additive exPlanations (SHAP) to quantitatively interpret the model's diagnostic decisions, elucidating the specific contributions of each feature to the diagnostic results and enhancing the reliability of diagnostic outcomes.