AUTHOR=Ahmed Shakeel , Kamal Khurram , Abdul Hussain Ratlamwala Tahir , Louhichi Borhen , Alrasheedi Nashmi H. 

TITLE=Predictive modeling of airfoil aerodynamics via support vector machines

JOURNAL=Frontiers in Physics

VOLUME=Volume 13 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/physics/articles/10.3389/fphy.2025.1621236

DOI=10.3389/fphy.2025.1621236

ISSN=2296-424X

ABSTRACT=The aerodynamic properties of fluids flowing around a wing or an airfoil are typically predicted through wind tunnel testing (experimental) or through computational fluid dynamics (CFD) by solving the Reynolds-averaged Navier-Stokes equations numerically. Although the numerical solutions are considered a low-cost alternative to the experimental efforts with a slight compromise on forecast accuracy, they consume a significant amount of time and computational resources, especially during the initial iterative design phases. The current boom of machine learning in engineering applications, data-driven surrogates such as support vector machines, offers promising potential in aerodynamic modeling. This work investigates the efficacy of support vector machines in forecasting the lift coefficient and the drag coefficient of four different NACA airfoils under varying flow conditions. Six different variants of SVM, including linear, quadratic, cubic, fine Gaussian, medium Gaussian, and coarse Gaussian SVMs, were used to forecast the aerodynamic coefficients of drag and lift. Almost all the models evaluated performed well in predicting the aerodynamic coefficients; however, Cubic SVM outperformed other models, achieving the lowest RMSE of 5.364 × 10-3 for drag coefficient and 40.702 × 10-3 for lift coefficient, and correlation coefficient values exceeding 0.995, indicating excellent correlation between the tested and predicted data. Contrarily, the linear and quadratic SVMs were the least effective for drag coefficient and lift coefficient predictions, with the highest RMSE of 14.156 × 10-3 and 93.703 × 10-3, respectively, with correlation coefficient values above 0.9650. These findings indicate the efficacy of machine learning in aerodynamic prediction and pave the way for faster airfoil design, particularly in applications requiring rapid iteration and low computational cost.