Hybrid computational fluid dynamics-machine learning optimization of Darrieus wind turbines: Aerodynamic improvement and noise reduction through wake and vortex interactions

This research uniquely explores the effects of wake interactions between adjacent Darrieus wind turbines on their aerodynamic performance and noise emissions, a critical consideration for optimizing wind farm design and operation in proximity to populated areas. Additionally, it examines the vortex interactions between rotor blades and analyzes the dynamic stall phenomenon, offering valuable insights into the unsteady aerodynamic behavior. By utilizing Large-Eddy Simulation, the study analyzes complex turbulence patterns and rotor interactions, thereby deepening the understanding of their aerodynamic and aeroacoustic effects. A multiobjective evolutionary algorithm integrates machine learning with computational fluid dynamics (CFD) to improve rotor designs for maximum power efficiency and reduced noise, considering wake interactions. The study assesses the impact of physical and geometric parameters on rotor performance, creating a database via Design of Experiments to replace time-intensive CFD model with an Artificial Neural Network for performance predictions. The nondominated Sorting Genetic Algorithm II refines aerodynamic and aeroacoustic attributes, with optimal design parameters identified using the linear programing technique for multidimensional analysis of preference (LINMAP). The LINMAP-optimized rotor outperforms the Point O rotor in aerodynamic and aeroacoustic performance. Its wider blade spacing enhances airflow and torque coefficient (C _T), while the Point O rotor suffers from increased vortex interactions. In the downwind region, the LINMAP rotor maintains positive C _T values, whereas the Point O experiences negative torque. Furthermore, the LINMAP design produces stronger low-frequency noise, while the Point O rotor exhibits higher sound pressure levels above 100 Hz.