Machine learning-based prediction of coolant velocity profiles for thermal management and energy minimization across drive cycles

This study addresses the limitations of existing battery thermal management systems (BTMS) that typically use constant coolant velocities to regulate the temperature excursions. In real-world applications, the power drawn, thereby, heat generation from batteries is not constant due to varying vehicle speeds, influenced by traffic and terrain. Under such conditions, maintaining constant coolant velocities may lead to overcooling, resulting in wasted energy, or undercooling, which could trigger thermal runaway. To overcome these challenges, a Smart Thermal Management System (STMS) is developed in the present study to identify smart velocity profile of the coolant fora user given conditions: (i) C-rate profile with respect to state of charge, (ii) initial temperature of the system (Tint), (iii) inlet temperature of the coolant (Tinlet), and iv) maximum allowable temperature of the batteries (TSPT). As a part of STMS development, an integrated serpentine channel based cold plate geometry, sandwiched between two pouch-type Li-ion batteries (LIBs), is designed numerically. Using Latin Hypercube Sampling (LHS), 600 samples were generated in the respective ranges of input parameters, namely, C-rate (0.5- 5), initial DoD (0- 0.8), initial temperature of the system (15- 35 degrees C), inlet temperature of the coolant (15- 35 degrees C), velocity of the coolant (0.01- 0.5 m/s), and final DoD (0.05- 0.85). Numerical simulations are conducted on the considered geometry using the NTGK model available in ANSYS Fluent for these 600 samples and the corresponding maximum temperatures on LIBs (Tnax) are noted. The GPR model was trained on these parameters to predict maximum temperatures, enabling the identification of smart coolant velocity profiles for popular drive cycles under specified user conditions (Tint, Tinlet, and TSPT). In addition, the percentage of energy being saved with smart velocity profile over constant velocity profile is also calculated. As the developed STMS model seeks information from user (Tint, Tinlet, and TSPT) to calculate smart velocity profile, it can be considered robust, independent of battery chemistry and BTMS designs, and applicable for any geographic or climatic conditions.