Thermal degradation of lithium-ion battery cathodes: a machine learning prediction of stability and safety

Abstract

Lithium-ion batteries are extensively utilized due to their diverse applications, but their potential risk of thermal runaway leading to fire or even explosion remains a significant challenge to their sustainable development. The simulation of battery thermal runaway is complex, as it involves multiple reaction mechanisms. This study focuses on the interfacial interactions between reducing gases and cathode materials and explores the factors that influence these interactions during gas crosstalk within the battery. Thermogravimetric analysis coupled with differential scanning calorimetry was used to simulate the thermal attack of argon and hydrogen ($ \mathrm{H}_2 $/Ar) mixtures on battery cathode materials to evaluate the chemical impact on the thermal runaway process. Four key material and environmental parameters, (1) cathode atomic composition; (2) hydrogen gas concentration; (3) gas flow rate; and (4) heating rate, were controlled and paired with thermal analysis curves to compile a database of 55 possible cases. Using seven input variables, this database was trained by an artificial neural network model to predict 11 critical degradation temperatures and rates for assessing material stability and safety. With an overall prediction accuracy above 0.73 (test set), we adopted an analytic hierarchy process to establish a novel scoring mechanism for cathode thermal stability. This work provides valuable insights into battery thermal runaway mechanisms and practical guidance for optimizing battery cathode chemistry.

Keywords

Li-ion battery / reductive attack / thermal stability / artificial neural network / risk assessment