Sag-flownet: self-attention generative network for airfoil flow field prediction

Flow field prediction is essential for airfoil design. It is a time-consuming task to obtain the flow fields around an airfoil. Convolution neural networks (CNN) have been applied for flow field prediction in recent years. However, CNN-based methods rely heavily on convolutional kernels to process i...

Full description

Saved in:
Bibliographic Details
Published inSoft computing (Berlin, Germany) Vol. 28; no. 11-12; pp. 7417 - 7437
Main Authors Wang, Xiao, Jiang, Yi, Li, Guanxiong, Zhang, Laiping, Deng, Xiaogang
Format Journal Article
LanguageEnglish
Published Berlin/Heidelberg Springer Berlin Heidelberg 01.06.2024
Springer Nature B.V
Subjects
Accuracy
Airfoils
Application of Soft Computing
Artificial Intelligence
Artificial neural networks
Computational Intelligence
Control
Deep learning
Engineering
Flow nets
Fluid dynamics
Mathematical Logic and Foundations
Mechatronics
Modules
Multilayers
Neural networks
Optimization
Performance prediction
Pressure distribution
Reynolds number
Robotics
Velocity
Generative networks
Flow field
Self-attention
Prediction
Online AccessGet full text

Cover

More Information
Summary:Flow field prediction is essential for airfoil design. It is a time-consuming task to obtain the flow fields around an airfoil. Convolution neural networks (CNN) have been applied for flow field prediction in recent years. However, CNN-based methods rely heavily on convolutional kernels to process information within local neighborhoods, making it difficult to capture global information. In this paper, we propose a novel self-attention generative network referred to as SAG-FlowNet, both for original and optimization airfoil flow field prediction. We investigate the self-attention mechanism with a multi-layer convolutional generative network. We use the self-attention module to capture various information within and between flow fields, and with the help of the attention module, the CNN can utilize the information with stronger relationships regardless of their distances to achieve better flow field prediction results. Through extensive experiments, we explore the proposed SAG-FlowNet performance. The experimental results show that the method has accurate and universal performance for the reconstruction and prediction of the flow field both for original and optimized airfoils. SAG-FlowNet is promising for fast flow field prediction and has potential applications in accelerating airfoil design.
ISSN:1432-7643
1433-7479
DOI:10.1007/s00500-023-09602-x