Cooling channel designs of a prismatic battery pack for electric vehicle using the deep Q-network algorithm

In this study, using the deep Q-network (DQN) algorithm, which is suitable for cooling channel design, a design method that satisfies the specified target inputs, namely, maximum temperature, average temperature, temperature standard deviation, and pressure drop, was proposed. The cooling channel aims to design this shape. The agent designs this shape through grid environment experience and obtains a reward through the analysis results of the generated shape. Finally, one obtains the maximum reward through the learned policy. With the proposed design method, it was possible to obtain the optimal cooling channel and the maximum reward for three examples. The final DQN results were verified for validity by comparing them with the Ansys results. Computational fluid dynamics (CFD) analysis requires high-quality mesh generation and selection of an analysis technique suitable for the problem and high proficiency. Therefore, it is expected that the proposed method will not only shorten the calculation time but also design the cooling channel according to various conditions.