Data-driven design of doped SnS2 for thermal runaway gas detection

Abstract

Lithium-ion batteries release harmful gases during thermal runaway, making high-performance gas-sensitive materials for early warning systems crucial. This study constructed a dataset of adsorption energies for 28 noble metal-doped SnS₂ systems with 6 gases (C₂H₄, C₃H₆, CH₄, CO, CO₂, H₂). We developed machine learning models to predict adsorption energies, incorporating both doping atoms and the system's electronic structure (Feature set II), which improved model generalization compared to traditional methods relying only on doping atoms or gas properties (Feature set I). The Gaussian Process Regression model achieved the best performance under Feature set II (R² = 0.82). The Au-doped SnS₂ system was further analyzed using density of states (DOS), electron localization function (ELF), charge density difference (CDD), and I–V characteristics. Results indicate that gas adsorption significantly alters the material's electronic and transport properties, especially for gases like CO andC₃H₆, demonstrating selective sensing and conductivity modulation. This study proposes a multi-scale design strategy combining first-principles calculations and machine learning, providing insights for developing gas sensors and thermal runaway warning systems in batteries.