Using deep generative adversarial network to explore novel airfoil designs for vertical-axis wind turbines

• Main Author: Santos, Gabriel B. [UNESP]
• Publication Date: 2023
• Other Authors: Pantaleão, Aluisio V. [UNESP], Salviano, Leandro O. [UNESP]
• Format: Article
• Language: eng
• Source: Repositório Institucional da UNESP
• Download full: http://dx.doi.org/10.1016/j.enconman.2023.116849; http://hdl.handle.net/11449/246958
• topic: Airfoil optimization; Airfoil parameterization; Computer fluid dynamics; Generative adversarial network; Sensitivity analysis; Vertical-axis wind turbine

Wind energy has emerged as an attractive alternative to the current fossil fuel-based energy mix. In this context, small-scale H-Darrieus vertical-axis wind turbines (VAWTs) combine interesting characteristics for harvesting wind energy in urban-like conditions. Still, H-Darrieus turbines are reported to experience relatively low aerodynamic efficiency. Even though several devices have been proposed to increase the aerodynamic performance of H-Darrieus turbines, the literature seems to overlook the potential of specifically designed airfoil shapes. In part, this is a by-product of different shortcomings related to the most common airfoil parameterization methods, such as restricted shape variability, high dimensionality, discontinuous spaces, and/or non-orthogonal parameters. Seeking to overcome these drawbacks altogether, we investigate here the benefits of the Bézier-GAN as an airfoil parameterization method for H-Darrieus turbines. For that, we use computational fluid dynamics (CFD) simulations along with sensitivity analysis, metamodeling, and optimization strategies. The results show that the Bézier-GAN integrates nicely with the proposed framework, substantially reducing the total computational cost of the experiment. By expanding the bounds of the latent design space, we can easily explore novel airfoil designs. The sensitivity analysis clearly indicates a lack of two-way interactions between the latent variables, which further simplifies both the metamodeling and the optimization processes. The optimal geometry increased the turbine performance by 20.5% relative to a NACA 0015 and by 9.1% relative to a NACA 0021—two common airfoil shapes used in H-Darrieus turbines. Interestingly, the optimal geometry was found outside the original bounds of the design space, further confirming that the search for novel airfoil designs may open the way for better aerodynamic performance of small-scale H-Darrieus turbines.