The Utilization of Generative Adversarial Networks for the Production of Airfoil Geometries

As the world increases the use of data mining and artificial intelligence to improve everyday life, machine learning algorithms and practices have become more widely studied and utilized. One such machine learning algorithm is a generative adversarial network (GAN) that uses a series of convolutions and neural layers to create new instances of data that resemble real instances of data very closely. This study applied a GAN to generate unique airfoil geometries based on a set of airfoil performance data. Typically, airfoil geometry is designed using Computational Fluid Dynamics (CFD) and optimization algorithms. By applying a GAN, new geometries can be created in a fraction of the time reducing the resources spent during the design and rendering process. The results of the study show promise for GANs as an alternative to traditional design methods, however the results are far from perfect. Additional methods exist that could further improve the model but they require additional data and higher computing power.