Chemical reaction neural networks to map lithium-ion battery thermal runaway gas generation

The widespread application of lithium-ion batteries in electric vehicles is hindered by safety concerns, notably thermal runaway. Understanding gas generation during thermal runaway is crucial for battery safety design, early fault detection, and fire hazard assessment, but existing kinetic models remain inadequate. We investigate gas generation behavior of 18650-type LiNi 1 / 3 Co 1 / 3 Mn 1 / 3 O 2 cells across varied states of charge, constructing a training dataset to develop a chemical reaction neural network model that autonomously explores unknown reaction pathways and corresponding chemical kinetic parameters. By incorporating an additional non-linear neural network to predict the mean molar weight of generated gases, the model trains a chemical kinetic mechanism comprising 7 species and 16 reactions. The resulting mechanism accurately predicts key parameters, including gas generation rates and amounts, during thermal runaway. This methodology addresses the current limitations in thermal runaway gas prediction, providing a robust framework for enhancing battery safety. • CRNN model is employed to autonomously explore the unknown reaction pathway • A first-ever chemical kinetic mechanism related to LIB TR gas generation is established • Learned pathways and kinetic parameters align well with experimental observations • Gas generation rates and amounts are accurately predicted across cells with varied SOCs Here, Zhang et al. extended a CRNN model that autonomously maps gas generation reaction pathways in lithium-ion battery thermal runaway. The first-ever kinetic model bridges modeling gaps, enabling accurate predictions of safety-critical parameters such as gas generation rates and amounts, advancing the understanding of thermal safety in LIBs.