Using generative adversarial networks to represent geometries of vertical-axis wind turbines with winglets

• Grant number: 25/01697-2
• Support Opportunities: Scholarships in Brazil - Scientific Initiation
• Start date: April 01, 2025
• End date: March 31, 2026
• Field of knowledge: Engineering - Mechanical Engineering
• Principal Investigator: Gabriel Bertacco dos Santos
• Grantee: João Pedro Vendrame Spagnuolo
• Host Institution: Faculdade de Engenharia (FEIS). Universidade Estadual Paulista (UNESP). Campus de Ilha Solteira. Ilha Solteira , SP, Brazil

Abstract

Given the urgency of the global energy transition, vertical axis wind turbines, especially the Darrieus turbine, emerge as an interesting alternative for wind energy generation on floating platforms at deep offshore waters. However, the feasibility of this application still faces some challenges, mainly due to the low aerodynamic performance of Darrieus turbines when compared to traditional horizontal axis wind turbines. To improve the performance of Darrieus turbines, aerodynamic optimization studies can be used to identify blade geometries specifically designed for this application. In particular, using winglets can reduce vortex shedding at the blade tips and, consequently, the blade--wake interaction, improving the overall aerodynamic performance of Darrieus turbines. However, to incorporate winglets into the optimization process, the parameterization method must be capable of robustly representing the space of possible geometries, which does not naturally occur when choosing a parameterization method that is based on common geometric parameters, such as the radius and curvature angle of the winglet. In this case, different values of the geometric input parameters produce similar or even identical geometries, which can confuse the optimization algorithm, delaying or even rendering the optimization process unfeasible. As an alternative, generative adversarial networks can be used to learn a latent space that provides a compact and orthogonal representation of the sample space. Thus, the latent variables can be used as proxies in the optimization process, facilitating the search for optimized solutions in a broad optimization process. In this context, this project proposes developing a methodology for using Generative Adversarial Networks in the parameterization of Darrieus turbine blade geometries with winglets.