Machine Learning Approach to Aerodynamic Analysis of NACA0005 Airfoil: ANN and CFD Integration

This study presents a machine learning approach to predict the unsteady aerodynamic performance of a NACA0005 airfoil. Data generated by computational fluid dynamics (CFD) is used to train the model for Reynolds numbers <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Re</i> ∈ [1000 − 5000] and angles of attack ranging from 9° to 11°. A Scaled Conjugate Gradient (SCG) algorithm is employed for efficient training. The ANN has a two-layer architecture, with 9 fixed neurons in the first hidden layer and a varying number of neurons in the second layer to find the best configuration. The model yielded coefficients of determination (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">R</i><sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup>) of 0.994 (Coefficient of lift (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C_l</i>)) and 0.9615 (Coefficient of drag (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C_d</i>)) for training, and 0.9563 (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C_l</i>) and 0.9085 (<italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">C_d</i>) for testing. Overall mean errors are found to be less than 1%. It offers a powerful surrogate modeling approach for aerodynamic studies at ultra-low Reynolds numbers. Moreover, it provides rapid and reliable alternatives to traditional CFD simulations in aerodynamic analysis for unseen cases.