A deep learning based prediction approach for the supercritical airfoil at transonic speeds

In traditional ways, the aerodynamic property of the aircraft is obtained by solving Navier-Stokes equations or performing tunnel experiments. However, these methods are time consuming for the aircraft design and optimization. In comparison, the deep learning technique is capable of handling high dimensional parameters and can describe compressible flow structures clearly and efficiently. For these, an efficient and accurate prediction approach based on the deep neural network is proposed for the compressible flows over the transonic airfoils in this study. By investigating the effects of the input coordinate features of the deep learning method on the prediction accuracy and robustness, the aerodynamic characteristics, such as lift, drag, and pitch coefficients, are obtained from the predicted flow fields. Results indicate that the proposed deep learning prediction method is with a high resolution and efficiency. It is promising to be extended to the optimization and design process of the supercritical airfoil.

Automated summary

The traditional methods for aircraft aerodynamic properties are time-consuming, but the deep learning technique is efficient in handling high dimensional parameters and describing compressible flow structures. This study proposed an accurate prediction approach based on deep neural networks for compressible flows over transonic airfoils, which showed high resolution and efficiency and has potential for optimization processes.