Predicting aerodynamic characteristics of airfoils using artificial neural network

Abstract

In this study, an artificial neural network (ANN)-based method is proposed to predict the aerodynamic characteristics of airfoils, such as NACA 0012, NACA 0015, NACA 0018, NACA 0021, and NACA 0025, approximating the flow around airfoils as a function of the Reynolds number (Re), angle of attack (α), airfoil coordinates (X, Y ), and predicting the lift coefficient (CL) and drag coefficient (CD) without using extensive software packages. Wind turbine data were obtained for CL and CD for different α (0◦ ≤ α ≤ 180◦ ) and different values of Re (104 ≤ Re ≤ 107 ). An ANN model was trained to achieve a root mean square error (RMSE) of less than 0.12 and 0.025 for CL and CD, respectively. For CL and CD, the RMSE of the trained model used to evaluate the new data was less than 0.09 and 0.12, respectively. Subsequently, the results were validated in a two dimensional numerical domain using RANS-CFD simulations and experimental data, showing that the proposed ANN approach is in good agreement for predicting the stall shape and aerodynamic characteristics at an angle of attack (α) ranging from (0◦ ≤ α ≤ 30◦ ).