An Accurate 3D Thermal Simulation Method Based on Neural Network-Aided Power Loss Model

Abstract:
This paper presents a novel approach to enhance the simulation accuracy by serving a more accurate power loss model. Based on a data-driven neural network power loss model, the proposed method aims to address the issue of low simulation accuracy resulting from inaccurate input power losses. The paper first introduces the method used to construct the neural network power loss model and automate data extraction for training the neural network. Subsequently, an indirect coupling strategy is proposed to facilitate the bidirectional coupling between the power loss model and the COMSOL thermal model. In addition, an interface software incorporating the proposed method has been developed for better compatibility. Finally, the effectiveness of the proposed method is validated through a comparison between simulation and experimental results, which shows a mismatch below 5%.