Integrating CFD and data-driven techniques for the optimization of dielectric coolants in EV battery immersion cooling systems

This study combines Computational Fluid Dynamics (CFD) and data-driven techniques to optimize dielectric coolants for Lithium-ion battery (LIB) thermal management systems (BTMS) using immersion cooling. A CFD-simulated database of 88 coolants was analyzed using machine learning, focusing on enhancing the XGBoost algorithm. Spearman correlation and Sobol sensitivity analyses assessed coolant property interrelationships and sensitivities. The findings highlight the superiority of the XGB-MLP model with near-zero MAE and RMSE metrics and an R 2 metric of up to 0.988. In immersion BTMS , viscosity and thermal conductivity are identified as the most influential physical properties, with viscosity exhibiting the highest sensitivity indices for maximum temperature ( T max ), temperature uniformity ( U Taver ), and energy consumption ( E ) at 0.494, 0.382, and 0.933, respectively, followed by density and specific heat capacity . Coconut Oil, exhibiting properties near theoretical optima, excels in BTMS performance with a T max of 299.04 K, U Taver of 0.24, and E of 6.60 W, making it a prime candidate for efficient BTMS, though its viscosity presents an area for future E optimization. This study optimizes dielectric coolants, enhancing LIB stability and safety, and guiding the design of immersion BTMS for electric vehicles.