Abstract:

Airfoil parameterization method plays a very important role in airfoil manufacturing， aerodynamic and stealth optimization design. To further enhance the representation capability of the airfoil parameterization method， avoid abnormal geometric shapes during the optimization process， and improve the efficiency of airfoil optimization design， in this paper， first， based on the existing airfoil databases， we propose a new airfoil parameterization method： CST-GAN， which combines more flexible Class and Shape Transformation （CST） method and Generative Adversarial Network （GAN） model where latent data distribution can be learnt. Then， the effect of design dimension on CST-GAN airfoil parameterization is studied by examining the geometric quality and representation error of the generated airfoils. Moreover， the representation accuracy of CST-GAN is compared with that of Bezier， B-spline， and Principal Component Analysis （PCA） method. Finally， airfoil optimization design based on the proposed parametrization method is conducted. The results show that the proposed method can generate smooth and effective geometric shapes and describe the airfoil shape more precisely. Compared with other commonly used parameterization methods， the CST-GAN method exhibits faster optimization convergence speed and better optimization results， which contributes to optimization efficiency improvement and computational cost saving. In addition， the proposed method is robust and easy to implement， with potential applications to parametric modeling and aerodynamic optimization design of three-dimensional wings and the entire aircraft.

Key words: airfoil, parameterization, generative adversarial network, aerodynamic optimization design, class and shape transformation