Conference Paper

Interactive airfoil design using artificial intelligence

Authors:
  • University of Applied Scienes Wiener Neustadt
To read the full-text of this research, you can request a copy directly from the authors.

Abstract

Aerodynamic shape design has a long history of extensive and detailed development, including different methods of optimization based on the various technologies that have been made available over the years, ranging from simple manual iteration to numerical inverse design. As artificial intelligence sees increasing popularity, it has also penetrated the field of fluid dynamics. We present a machine learning workflow for developing an artificial neural network that predicts the two-dimensional distribution of the coefficient of pressure along the perimeter of parameterized airfoils in variable steady-state subsonic high-Reynolds flow conditions. The data set is obtained from computational fluid dynamics simulations of pseudo-randomly generated parameter sets comprising of profile shape parameters and flow conditions. Several artificial neural networks are trained and ranked by performance, based on methods and quality metrics commonly used in machine learning. The highest-ranking network is further validated using several methods, including comparison to theory and experimental data. A software implementation of the neural network including a graphical user interface achieves real-time prediction and display of the distribution of the coefficient of pressure in response to variations of the shape and flow parameters. Other software features have been implemented, including interfaces to numeric solvers that can verify predicted results by executing the same automated workflow that has been used for creating the training data. The developed prototype toolbox has the potential to contribute in industry - in both optimization and conceptual design - as well as in education of professionals and students by providing a means to analyze in detail how the flow responds to a shape variation.

No full-text available

Request Full-text Paper PDF

To read the full-text of this research,
you can request a copy directly from the authors.

ResearchGate has not been able to resolve any citations for this publication.
ResearchGate has not been able to resolve any references for this publication.