Inverse Design Method of Transonic Airfoil Based on Deep Neural Network

This research introduces a deep neural network-based transonic airfoil inverse design approach. A deep neural network is established using the predicted pressure coefficient distribution, angle of attack, and Mach number around the airfoil as inputs and parameterized airfoil parameters as outputs. The network solves the transonic flow field of an airfoil using the Euler equation. The neural network is trained to grasp the impacts of geometric changes on the pressure coefficient and shock wave at various places on the airfoil surface. Using the finite volume approach and multigrid acceleration method, the Euler equation is employed to batch compute the airfoil database and build the training set. As the training set's output vector, the CST parameterization technique is utilized to parameterize each synchronously generated airfoil into 10 parameters. The results demonstrate that the deep learning approach developed in this research is capable of achieving the reverse design of a transonic airfoil, and that the trained neural network can generate the airfoil with the requisite aerodynamic properties rapidly and directly. The generalization error of the anticipated airfoil geometry information is less than 1.5 percent when compared to the accurate results. The results reveal a high level of precision. The impact of the deep neural network's shape and the activation function of neurons on the subjects under investigation is also examined. The results reveal that the activation function used is critical to the outcome, and the design of the neural network also has an impact on the outcome.