A Deep Learning-Based Surrogate Model for Complex Temperature Field Calculation with Various Thermal Parameters

Abstract

Surrogate models of temperature field calculation based on deep learning have gained popularity in recent years because it does not need to establish complex mathematical models. However, the existing models cannot generate the temperature field for different boundary conditions or thermal parameters. In addition, it is also challenging to generate the details of the complex temperature field. In this paper, we propose a Parameters-to-Temperature Generative Adversarial Networks (PTGAN) to generate temperature field images with high-quality details for different thermal parameters. The PTGAN model mainly includes temperature field generation module and thermal parameter encoding module. Additionally, we use a joint loss function to improve the quality of the generated temperature field image. The temperature field of the armored vehicle is calculated by the computational fluid dynamics (CFD) method to obtain data set to verify the proposed PTGAN. The results show that the temperature field generated by PTGAN has high accuracy, and the average relative error is only 0.205%. The attempt to integrate thermal parameters into the temperature field image generation is successful. The temperature field database can be generated quickly and accurately, which is of great significance for the further integration of deep learning and heat transfer.