基于物理信息神经网络的参数化热仿真方法用于快速功率模块热设计

This article proposes a parameterized 3D thermal simulation methodology based on physics-informed neural networks (PINNs) to achieve rapid design space exploration for power module thermal design. Leveraging the capability of PINNs to quickly approximate the solutions to the parameterized partial differential equations describing the thermal behavior of power modules, a thermal field simulation framework for a SiC three-phase half-bridge power module is developed for parameterized simulations. After a single unsupervised training session, the PINNs-based model can quickly predict the thermal field distribution results of the power module for different combinations of input parameters. The comparison results show that the PINNs predict results are approximately consistent with both COMSOL numerical simulations and experimental measurements in different combination cases. Moreover, in the task of large design space exploration for parameter optimization, the simulation process can be hundreds of times faster than traditional numerical simulation methods, significantly reducing the time cost required for thermal simulations.