Lift-to-Drag Ratio Optimization of Four-Digit NACA Airfoil Based on the Artificial Neural Networks Meta-Model Trained on CFD Data

In the present paper, the changes in the geometry of the optimized airfoil were determined with the maximum lift-to-drag ratio criterion for selected angles of attack and Reynolds numbers. For this purpose, a neural surrogate model was used for the prediction of airfoil aerodynamic coefficients from the four-digit NACA family. The input database for training Artificial Neural Networks was formed by cases calculated by the computational fluid dynamics method using the ANSYS FLUENT. Gaps in the literature were analyzed and the method of building parametric airfoil models of the selected family was discussed. The method of creating the computational grid and numerical results was presented. The results were compared with the data from the literature. Artificial Neural Networks were shown to be able to make predictions of lift coefficients (Cl) and drag coefficients (Cd) for airfoils for a specified range with high accuracy (regression R=0.9998). Optimization of airfoils was performed with the criterion of maximizing lift-to-drag ratio for selected Reynolds numbers and angles of attack that change airfoil geometry. The optimization resulted in an improvement of lift-to-drag ratio with respect to the NACA 2412 airfoil by 1.395 times for the angle of attack α=3°. Possible relation between flow conditions, the aerodynamic and geometric properties of the airfoil are shown. The drawbacks of the presented approach and directions for further research are also indicated.