Thermal runaway and flame propagation in battery packs: numerical simulation and deep learning prediction

The widespread application of lithium-ion battery technology faces a significant challenge from the inherent risk of thermal runaway and consequent fire spread. This paper proposes an intelligent framework for predicting the temperature distribution and thermal runaway propagation in a battery pack across diverse conditions, including various battery types, ambient temperatures, and fire heat release rates. First, we generate an extensive numerical database, comprising 36 simulations of battery jet flame and thermal runaway processes that are validated by experimental data. Subsequently, a dual-agent artificial intelligence (AI) model is employed to forecast the cell-to-cell thermal runaway propagation and evolution of temperature field in the battery pack. The results demonstrate the accuracy and reliability of the deep-learning approach in capturing battery thermal runaway dynamics. Quantitatively, the AI-based methodology achieves a relative error below 10% for thermal runaway time predictions in database-contained scenarios and below 30% for extrapolated cases. The model also shows excellent performance in predicting temperature field distributions, with an R² value exceeding 0.99 and a maximal MSE of 1.52 s². This study underscores the potential of AI method in improving the battery safety management, thereby facilitating timely interventions, preventive maintenance and fire safety of battery energy storage system.