Relative Assessment of Selected Machine Learning Techniques for Predicting Aerodynamic Coefficients of Airfoil

In computational fluid dynamics, RANS expressions are solved numerically, as a cheap replacement for experimental work with an acceptable forecast accuracy compromise. Recently, use of machine learning techniques has increased significantly and has been useful in many sectors including aerodynamics. This paper examines the application of three distinct machine learning approaches to compute and predict aerodynamic coefficients of airfoil. We employ back-propagation neural networks, regression trees, and support vector machines to model the complex relationship between airfoil geometry, flow conditions, and the resulting aerodynamic coefficients. Our study investigates the applicability of these machine learning models and compares their performance to identify the most effective model for predicting airfoil coefficients. Overall, among all the different machine learning models examined, back-propagation neural networks demonstrated the best performance in terms of mean squared error and correlation coefficient values. Notably, for predicting coefficient of drag, the fine tree model achieved the lowest mean squared error of 3.1704 \(\times\) 10^–7, while for the prediction of coefficient of lift, the lowest mean squared error of 4.9766 \(\times\) 10^–7 was obtained by the back-propagation neural networks. This research not only offers deeper understanding of how machine learning techniques could play a pivotal role in enhancing airfoil coefficients predictions but also provides a practical application for improving aerodynamic designs in various engineering fields.