Multi-Field Distributions Generation for Complex Objects Using a Deep Learning-Based Image Processing Method

Chenhao Wu, Shuojin Yang, Zhongwei Huang, Yuntao Zou, Jiang Liu

Abstract: Accurately and rapidly determining the field distribution for complex objects is crucial, especially in the energy sector. Computational Fluid Dynamics (CFD) methods have been employed to simulate thermal field flows by numerically solving the Navier-Stokes (NS) equations. This methodology requires the implementation of complex procedures, such as the establishment of the computational domain, the discretization of the partial differential equation, and so forth. Consequently, it is both time-consuming and labor-intensive. This paper presents a novel approach for predicting thermal field distributions using a deep learning-based image processing method. Generative Adversarial Networks (GANs), which are known for their ability to generate complex images, present a new opportunity for simulating thermal field distributions. In contrast to conventional CFD techniques, GANs are capable of directly generating thermal field distribution images from input object images without the necessity of solving NS equations, thereby significantly simplifying the process. The approach is based on the StarGAN v2 architecture, which has been shown to be efficient for generating complex images. The method is applied to a dataset constructed using traditional CFD methods, and the results demonstrate that the generated thermal field images closely match the original images, with a Structural Similarity Index (SSIM) value exceeding 0.92, this confirms the effectiveness and precision of the proposed method.