Optimization of Wind Turbine Airfoils using Generative Adversarial Networks

Abstract

This work aims to contribute to the optimization of airfoils in the wind energy domain by randomly creating a database of novel airfoil designs with high aerodynamic performance. For this purpose, a Matlab code was established using Generative Adversarial Networks, XFOIL and the BEM theory. First, the two networks of the GAN were defined and trained over the UIUC database. The trained model was used to generate new shapes that were post-processed and then evaluated in XFOIL. Based on this evaluation results, airfoils with the highest lift to drag ratios were selected. Subsequently, the best airfoil among the selected ones was applied to the NREL phase VI wind turbine. Using the BEM method, performance of both the original and modified turbines were calculated and then compared. The results have shown that the turbine with the selected airfoil was slightly more efficient than the original one, and that without any optimization of the blade design being applied. This serves as a validation of the established code’s ability to produce highly efficient airfoils.