Researchers from the University of Oxford, including Oliver Summerell, Gerardo Aragon-Camarasa, and Stephanie Ordonez Sanchez, have explored the potential of neural networks to predict the performance of aerofoils used in wind and tidal energy turbines. Their work, published in the journal Nature Scientific Reports, aims to provide a benchmark for a dataset created by the National Renewable Energy Laboratory (NREL) in the USA.
Currently, the energy industry relies on methods like Computational Fluid Dynamics (CFD), thin aerofoil, and panel methods to assess the lift and drag coefficients of aerofoils. These methods, however, often involve a trade-off between computational speed and accuracy. The researchers investigated whether neural networks could offer a faster and equally accurate alternative.
The study evaluated four types of neural networks: Multi-Layer Perceptron (MLP), PointNet, GraphSAGE, and GUNet. These models were trained on a range of aerofoils at various angles of attack to predict fluid flow and calculate lift coefficients. During the training phase, GraphSAGE and GUNet showed promising results, but their performance dropped during testing. The researchers identified PointNet and MLP as the two most effective models. While MLP provided more consistently correct predictions for fluid behavior, PointNet offered more accurate calculations of lift coefficients.
This research highlights the potential of machine learning to enhance the design and analysis of aerofoils in the wind and tidal energy sectors. By improving the accuracy and speed of performance predictions, neural networks could help optimize turbine design, reduce costs, and increase energy output. The findings provide a valuable benchmark for future research in this area, contributing to the ongoing development of renewable energy technologies.
This article is based on research available at arXiv.