Airfoil performance analysis using shallow neural networks

Oztiryaki F. G., Piskin T.

AIAA Science and Technology Forum and Exposition, AIAA SciTech Forum 2021, Virtual, Online, 11 - 15 January 2021, pp.1-9 identifier

  • Publication Type: Conference Paper / Full Text
  • Volume:
  • City: Virtual, Online
  • Page Numbers: pp.1-9
  • Middle East Technical University Affiliated: Yes


© 2021, American Institute of Aeronautics and Astronautics Inc, AIAA. All rights reserved.A shallow neural network was trained to do regression and predict the lift coefficient of a symmetric airfoil for a given thickness, angle of attack, and Reynolds number value. Overall, the predicted values are within 3 percent accuracy, and the neural network depicts the expected behavior of an airfoil correctly. Furthermore, the speed at which the predictions are made is remarkable, up to a few million predictions per second when the data is requested in bulk. Although the study is highly limited in various aspects, due to the simplicity of the flow solver used and the nature of the problem, it stands as a good proof that already existing solutions can be used to predict similar solutions. This is expected to be especially useful in areas where obtaining a solution to a problem is highly expensive, such as turbulence modeling, and where computed similar solutions already exist. It can be further extended to capture the entirety of the flow field.