A three-dimensional flow field reconstruction method of wing based on SE-3DUnet

Convolutional neural networks are used instead of computational fluid dynamics (CFD) solvers to simulate flow fields in design, analysis, and optimization problems related to aerodynamics. However, in the flow field distribution, the flow field near the aircraft wing changes It is more severe at the nearer end. When reconstructing the flow field information, we should focus on the near-end How field characteristics of the aircraft and increase the weight of its features. However, feature redundancy will appear when using the existing method for sample feature extraction. This paper proposes a 3D airfoil flow field prediction method based on improved SE-3DUnet, which improves SENet and introduces global maximum pooling. The method in this paper can fuse shallow feature maps with more detailed texture information into advanced Among the characteristics. The experimental results on the actual data set show that compared with Unet, the average precision of the method in this paper is increased by 5.76%