Tidal turbine hydrofoil design and optimization based on deep learning

Abstract

The optimal design of hydrofoils is critical to improve the hydrodynamic performance of the tidal turbine. However, the global optimization of hydrofoils is limited by the high dimensionality of the design space, which requires extensive computational fluid dynamics simulations. This paper proposes an interactive framework for hydrofoil design and optimization based on deep learning. Generative adversarial networks are used to parameterize the hydrofoil design, which automatically learns representations from existing hydrofoils and controls new hydrofoil generation using fewer variables to reduce optimization dimensions. Moreover, the surrogate model based on convolutional neural networks is constructed, which realizes the mapping of hydrofoil design and operating parameters to hydrodynamic performance parameters. The framework can generate a large number of smooth and realistic hydrofoils with three design variables and quickly predict the performance, enabling effective optimization design of hydrofoils. The results show that the optimized hydrofoil shapes have larger lift-to-drag ratios than those of the common hydrofoils. Furthermore, the optimized hydrofoil is applied to the design of 3D horizontal axis tidal turbine blades. The simulation results show that the framework is effective and stable, which can facilitate the design of tidal turbine rotors and provide hydrofoils with higher power coefficients.