Activity: Examinations and supervision › Supervisor activities
This project investigates the ability of neural nets to predict turbulent fields from nacelle-mounted lidar measurements, with the goal of increasing the accuracy of the predicted aeroelastic response of a wind turbine. The project examines different techniques for reduced-order modelling (ROM) of turbulent fields and determines each method's accuracy and computational burden. Different neural net models are then trained in conjunction with a selection of ROM techniques on synthetic turbulence data, and their computational cost and ability to produce accurate wind turbine loads are evaluated. Finally, a selection of methods are tested on real turbine data, provided by collaborators from Siemens-Gamesa Renewable Energy.