Inverse Design of Transonic Airfoils Using Denoising Diffusion Probabilistic Models

Inverse design offers significant advantages in aerodynamic design, such as improved performance and efficiency. In this paper, a denoising diffusion probabilistic model (DDPM) is adopted as a generative model to produce pressure coefficient distribution data for the RAE2822 airfoil. Through the forward noise addition process and the reverse denoising process, the trained DDPM model can sample a large amount of pressure coefficient distribution data from a standard normal distribution. Two neural networks are then employed: one maps the pressure coefficient distribution to geometric parameters, linking the pressure field with geometric parameters, and the other maps the pressure coefficients to lift and drag coefficients. Computational fluid dynamics (CFD) validation of the sampled data shows that the CFD results are close to the generated pressure distributions, demonstrating the effectiveness and reliability of the proposed approach.