Esteban Ferrer (UPM) — Wind turbine aeroacoustic simulations using high order solvers

💡 MareNostrum5 makes his debut in this RES Success Story about air flows in wind turbines 💡 

📋 "Wind turbine aeroacoustic simulations using high-order solvers" led by Esteban Ferrer Vaccarezza from Universidad Politécnica de Madrid

Simulating the air flows around wind turbines and wind farms are computationally demanding due to two main factors: the rotating blades, which create unsteady flows and dynamic effects, and a high Reynolds number of the atmospheric flow in which these devices work. 

🔊 There exist many simplified techniques simulating individual blades and the entire rotor that allow to efficiently replicate the effect of the energy extracted from the flow in certain conditions. However, for simulating detailed flow features for fine design or aeroacoustic predictions, they become insufficient as they neglect some important effects such as trailing edges, dynamic stall or 3D effects on lift and drag. 

When comparing accurate, costly simulations with advanced models such as sliding meshes or immersed boundary methods) to simplified models, it is possible to extract corrections for the simple models under certain operating conditions making them more precise but still efficient. That was the aim of the research team. 

🖥 Thanks to RES supercomputer #MareNostrum5 they could accurately simulate horizontal wind turbines using their own open source high-order discontinuous Galerkin solver "HORSES3D". It had very low errors that decayed exponentially instead of algebraically thanks to reinforcement learning to adapt (on the fly) the optimal high order polynomial in each mesh element. More about HORSES3D at the the scientific paper: https://linkinghub.elsevier.com/retrieve/pii/S0010465523000450

The left image shows the divergence of the lamb vector when simulating a 10MW wind turbine in offshore environments, which is related to sources of noise and can help to predict the regions where it is generated in complex situations, while the right image depicts the average polynomial order used in each mesh element.