An integrated diagnostic framework for lithium-ion battery internal short circuits: bridging physical models with deep learning

Early diagnosis of internal short circuits (ISC) is crucial for the safe operation of electric vehicles (EV). This paper proposed a novel hybrid ISC diagnostic framework integrating physics-based modeling and deep learning. In this framework, an equivalent circuit model (ECM) specific to the operating condition is first established based on electrochemical principles. Subsequently, leveraging the model mismatch concept, an Iterated Extended Kalman Filter (IEKF) is employed to estimate the battery’s State of Charge (SOC) and extract ISC-related features. A designed LSTM-Transformer hybrid network then acts as a complex mapping function to predict the true internal resistance, a parameter difficult to measure directly. Finally, using the predicted internal resistance, an Extended Kalman Filter is utilized to simultaneously estimate the short-circuit resistance and the true SOC. The proposed framework was validated through ISC experiments under various charging rates and ambient temperatures, achieving a diagnostic accuracy of up to 99%.