Data-driven modal parameterization for robust aerodynamic shape optimization of wind turbine blades

This paper proposes a data-driven modal parameterization to address the curse of dimensionality issue in robust aerodynamic shape design optimization of wind turbine blades. The proposed approach reduces the geometric dimensionality to tens by identifying and reformulating the feasible and meaningful geometric space for aerodynamic design optimization. This is achieved by four steps: building two-dimensional airfoil databases, training deep-learning-based airfoil generative models, developing a constrained generative sampling method of blades, and deriving blade modal parameterization from vast feasible blade samples. An effective surrogate-based optimization framework for wind turbine blade shape design is established by leveraging the benefits of this low-dimensional modal parameterization. The effectiveness and robustness of the proposed approach are demonstrated in aerodynamic shape optimization of the NREL 5 MW wind turbine blade under various sets of constraints and targets. Results show that wind turbine blade shape optimization using the proposed approach efficiently converges within hundreds of aerodynamic simulations. The optimized shapes and performances exactly meet the imposed requirements. This work lays the foundation for efficient robust shape design optimization of wind turbine blades using high-fidelity simulations.