Designing fire-retardant polymer-based electrolytes and separators for high-energy-density lithium-ion batteries via combustion calorimetry and machine learning

Lithium-ion batteries are pivotal to electric vehicles and modern energy systems, playing a key role in global efforts to achieve net-zero CO₂ emissions by enabling the storage of renewable energy. However, their safety remains a major concern due to the flammable nature of polymer-based electrolytes and separator membranes. This study leverages a large dataset of published experimental measurements of Total Heat Release (THR) from various polymer materials obtained through combustion calorimetry, aiming to assess the fire hazard potential of such materials. The performance of Support Vector Regression (SVR), Gaussian Process Regression (GPR), and Hist Gradient Boosting Regression (HGBR) models was evaluated in estimating THR values. Key input features included density (g/mL), solubility (MPa^1/2), molar cohesive energy (J/mol), entanglement molecular weight (g/mol), and glass transition temperature (K), with THR (KJ/g) as the target variable. Three feature configurations (C1, C2, and C3) were assessed to investigate the influence of solubility and molar cohesive energy on THR prediction. Model performance was validated using multiple statistical metrics and visualized through a 2D Taylor diagram and a 3D bubble plot. The HGBR-C3 model (NSE = 0.9858, CC = 0.9932) demonstrated superior predictive accuracy, with C3 configurations consistently yielding low bias and high reliability, except in the GPR model, where C2 showed improved performance. The integration of artificial intelligence techniques with combustion calorimetry data provides a valuable approach for mitigating fire risks in lithium-ion batteries.