An Efficient Electrothermal Coupling Simulation Method Based on Neural Network-Aided Power Loss Model for Power Module Thermal Optimization

This article introduces a highly efficient electrothermal coupling simulation method that leverages a data-driven neural network power loss model to enhance the simulation efficiency in the thermal optimization of power modules. First, methods to build the neural network power loss model and to extract data automatically for neural network training are introduced. Then, an indirect coupling strategy is applied for the bidirectional coupling between the power loss model and COMSOL thermal model. To address the challenge of balancing simulation accuracy and efficiency in the proposed coupling strategy, an adaptive time step adjustment algorithm based on Lagrange interpolation functions is designed. Finally, quantitative comparisons of different simulation methods and experiments validate the high precision and efficiency of the proposed method. The results indicate that the proposed method achieves the same level of simulation accuracy as existing methods, while improving simulation efficiency by approximately five to seven times compared to the thermal network method and by about two to five times compared to the COMSOL-PSpice method across 100 cases.