A Data-Driven Design Framework for Wind Turbines via Design-by-Morphing and Bayesian Optimization

Abstract

This study presents a data-driven design optimization framework the integrates Design-by-Morphing (DbM) with Bayesian optimization (BO) to facilitate the discovery of novel, high-performance aerodynamic geometries. Applied to vertical-axis wind turbines (VAWTs), the DbM approach enables the generation of novel shapes that go beyond the conventional constraints of lift-based (Darrieus) and drag-based (Savonius) geometries. Candidate designs were evaluated via expensive high-fidelity Unsteady Reynolds-Averaged Navier-Stokes (URANS) simulations under representative urban wind conditions, with the design space systematically explored to maximize the power coefficient, C_p. From a design space of 5.80 million unique physical configurations, the best-performing configuration was first identified after 13 BO-guided evaluations following a 50 design initialization. The best design achieved a 21.26% higher maximum C_p than the best existing baseline design, while maintaining a stable periodic torque response across the investigated Reynolds number range. Ultimately, this work presents a scalable and computationally efficient DbM-BO framework for aerodynamic shape optimization. While
demonstrated here for a two-bladed VAWT using primarily two-dimensional URANS simulations, the modular framework is not specific to this configuration and can be extended to other aerodynamic shape-design applications.