Artificial neural network predictions for temperature: Utilizing numerical analysis in immersion cooling systems using mineral oil and an engineered fluid for 32700 LiFePO4

Abstract

Immersion cooling offers high cooling efficiency, due to direct contact with the heat source. This investigation includes the performance of 32700 LiFePO₄ battery cells using immersion cooling with two dielectric fluids: mineral oil (MO), and an engineered fluid (EF). The investigation includes a numerical analysis of 16S1P arranged battery cells under different mass flow rates (0.001, 0.008, and 0.01 kg/s) and discharge rates (1C, 2C, 3C, and 4C). Results show that immersion cooling effectively maintains temperature homogeneity within and between cells. At a mass flow rate of 0.01 kg/s, the average temperature rise stays below 5 °C at a 3C discharge rate and below 10 °C at a 4C-rate across for both fluids. Additionally, an artificial neural network (ANN) model is developed to predict the average temperature of the battery cells with high accuracy. Using coolant type, C-rate, flow rate, and time as input parameters, the ANN achieves good predictive performance with consistently high R-values and low mean squared error across training, validation, and testing datasets. ANN predictions are in good agreement with numerical results, and the maximum prediction error is less than 1 K. This research has shown that flow rate and coolant selection are the most critical parameters in optimizing thermal management, demonstrating the accuracy of ANN in temperature predictions. The present results therefore provide a basis for further investigation into the development of more effective cooling methods, different dielectric fluids, and advanced ANN architectures for performance and safety improvements in LiFePO₄ battery modules.