Machine-Learning-Assisted Density Functional Theory Calculations: A New Approach to Screening Thermal Runaway Gas Sensors for Lithium-Ion Batteries.

Lithium-ion batteries face safety risks, such as spontaneous combustion and explosion, during long-term use and will rupture and release combustible gases, such as CH, CO, H, and CO, during thermal runaway. Therefore, the detection of these specific gases becomes the key to prevent thermal runaway accidents of lithium batteries. In response to this demand, this study proposed a method of selecting gas-sensitive materials with the aid of machine learning (ML) to efficiently obtain ideal materials from a huge candidate space. The doped substrates formed by 28 transition metals with MoS were selected as candidates for the study, and the initial stabilized configurations of the four gases were determined. By construction of 8 ML models, the optimal model is selected for adsorption energy prediction. In this process, the importance of each feature in model prediction is then deeply analyzed. The sensing performance of the selected materials is further validated to screen the best performing materials. This study combines density functional theory calculations with ML, which significantly saves the time and cost of screening thermal runaway gas materials for lithium-ion batteries and provides a new research idea for rapid screening of nanomaterials.