Airfoil Design Parameterization and Optimization Using Bézier Generative Adversarial Networks

What is it about?

Aerodynamic design often requires a large number of expensive CFD simulations. This expense can become impractical when there are a large number of design variables, particularly when adjoint techniques are not available. We propose a deep generative model for compactly and sufficiently representing aerodynamic geometry. It allows fast design space exploration and empirically accelerates airfoil optimization convergence by at least two times compared to state-of-the-art parameterization methods.

Featured Image