Affiliations: College of Information Science and Engineering, Northeastern University, Shenyang, China
Correspondence:
[*]
Corresponding author. Nan Wei, College of Information Science and Engineering, Northeastern University, Shenyang, Liaoning Province, 110819, China. E-mail: 167486084@qq.com.
Abstract: Various fluid mechanics software, due to inherent factors such as algorithms and boundary conditions, cannot quickly simulate 3D flow fields in batches, and the calculation of each model still takes a lot of time.In order to realize the rapid prediction of the three-dimensional flow field around the airfoil, this paper uses a new SDF geometric expression to describe the shape of the airfoil, and combines the prediction accuracy of the velocity and pressure channels, and proposes a two-stage Unet3d convolution prediction model based on the SDF expression, which greatly improves the prediction accuracy of the pressure channel.In addition, the introduced two-stage convolutional network is optimized by combining lightweight network and attention mechanism. On the premise of ensuring the accuracy of the network, it can effectively reduce the parameters of the network model and improve the operating efficiency of the network. The two-stage method was tested on the Naca0012 and RAE2822 three-dimensional datasets, and the average accuracy rates were 95.44% and 98.22% respectively, which were 2 to 3 percentage points higher than the original method.
Keywords: deep learning, 3D flow field prediction, lightweight network, two-stage convolution, attention mechanism
