Physics Guided cWGAN-gp for accurate airfoil generation

Abstract

There has been research on airfoil design using deep generative models such as generative adversarial network (GAN). However, in prior methods, the generated results do not always satisfy the governing equations. This paper reports the results of an attempt to construct a physics guided deep generative model and use it for fine-tuning. Computational software that calculates the aerodynamic performance of shapes was placed on a network. An objective function was expressed in terms of the relationship between the generated and required performance. As a result, it is confirmed that desirable shapes that accurately satisfies requirements were obtained, but on the other hand, a drawback was found in that the variety of shapes was reduced.