A rapid aerodynamic simulation framework for subsonic and transonic airfoil flow fields

Obtaining accurate flow fields is crucial for aircraft design and aerodynamic analysis. However, a large number of computational fluid dynamics (CFD) simulations and wind tunnel tests significantly increase the computational costs and time expenditures. The emergence of deep learning (DL) technology provides a new perspective for effectively utilizing historical data to accelerate airfoil flow field simulations. However, existing research has struggled to achieve generalization across different resolutions and operating conditions, limiting deep learning further development in the field of aerodynamics. Therefore, this paper proposes a rapid aerodynamic simulation framework for the subsonic and transonic flow field of airfoils with arbitrary resolution. By fusing features of different spatial positions in the flow field solely along the channel dimension, sensitivity of the deep learning model to resolution is eliminated. Additionally, a memory pool module is employed to ensure consistency in resolution between the model’s input and output. The incorporation of the free-stream encoder ensures that the model can generalize well across flow fields with different operating conditions. Extensive qualitative and quantitative analyses are conducted on given airfoil flow field datasets, demonstrating that the deep learning model proposed in this paper can generalize well across flow fields with different resolutions and operating conditions. The prediction speed of the model for single airfoil flow fields is at least two orders of magnitude faster than CFD methods. Particularly, our approach achieves accurate simulation of shock waves and separated flows at high Reynolds numbers simultaneously.