Physics-Informed Machine Learning Enhanced Battery Pack Optimization

The popularity of transportation electrification has increased the demand for battery thermal management systems. Among different battery thermal management systems, indirect liquid cooling is a promising candidate to ensure systems' performance, safety, and reliability. Temperature uniformity is important for systems' safety and needs to be optimized. However, the optimization process for indirect liquid cooling is prolonged due to the complexity of the system. Though finite element simulation can greatly save efforts on experimental testing, the computational cost required by the finite element simulation is high, especially for generating a large training dataset for constructing the surrogate model. Thus, adapting physics-informed machine learning to accelerate the optimization process is crucial. By integrating physics-informed loss functions into the surrogate model, together with the optimization framework, the battery thermal management systems can be improved. Based on the developed experimental testing and the high-fidelity multiphysics finite element model, the geometry can be simplified, and the complexity of the problem can be reduced. Based on the simplified heat conduction function, a physics-informed convolutional neural network was developed as a surrogate model to estimate the temperature distribution of the battery pack. This network contains a loss function considering the heat conduction equation based on the finite difference method, helping the convergence of the training process with less data. Finally, the optimal battery layout designs were obtained based on the two different optimization algorithms, showing a 7.2% improvement in temperature uniformity for the battery cells, improving the battery pack's performance and safety.