A Comparative Study of RANS and Machine Learning Techniques for Aerodynamic Analysis of Aerofoils

The design of aerospace applications necessities precise predictions of aerodynamic properties, often obtained through resource-intensive numerical simulations. These simulations, though they are accurate, but are unsuitable for iterative design processes due to their computational complexity and time-consuming nature. To address this challenge, machine learning, with its data-driven approach and advanced algorithms, offers a novel and cost-effective solution for predicting airfoil characteristics with exceptional precision and speed. This study explores the application of the Back-Propagation Neural Network (BPNN), a machine learning model, to forecast critical aerodynamic coefficients such as lift and drag for airfoils. The BPNN model is fed with input parameters including the airfoils name, flow Reynolds number, and angle of attack in relation to incoming flows. Training the BPNN model is accomplished using a dataset derived from CFD simulations employing the Spalart–Allmaras turbulence model on three distinct NACA series airfoils under varying aerodynamic conditions. The data from these simulations are divided into training (70%) and validation/testing (30%) subsets. The BPNN demonstrates a high level of accuracy in predicting these coefficients, evident through low root mean square error (RMSE) and a close alignment between predicted and actual values.