Abstract

Fast charging of lithium–ion batteries is essential for accelerating a widespread use of electric vehicles; however, its adoption significantly increases battery thermal stress and the risk of thermal runaway, particularly in aged cells. This study proposes a simulation-trained digital twin (DT) framework for probabilistic assessment of thermal runaway and critical charging current estimation under fast charging conditions. A dataset is generated using an electrochemical–thermal Single Particle model, varying current rate, capacity, and internal resistance. Then, an encoder–decoder neural network architecture is developed to map and convert static operating conditions into dynamic temperature evolution, enabling efficient surrogate modeling of thermal behavior. The proposed digital twin achieved an MAE of 3.05 °C and a recall of 97.22% for thermal runaway prediction while estimating critical charging currents of approximately 1.35–1.52C. The proposed methodology provides a computationally efficient tool for risk-aware fast-charging strategies, which can be integrated into battery management systems for enhanced safety. While the current study is applied to specific single-cell chemistry and simulation-based training, the framework can be easily extended to online battery systems and operating conditions.

IPC Classification

Keywords