Enhancing Aerodynamic Prediction of Airfoils Using Advanced Machine Learning Models

This paper presents a comparative study on the application of advanced deep learning architectures—Multilayer Perceptron (MLP) and Capsule Networks (CapsNet)—for predicting the aerodynamic performance of airfoils, quantified through the lift-to-drag ratio (LDR). A dataset comprising 6,283 airfoil geometries generated from high-fidelity computational fluid dynamics (CFD) simulations was used to train and validate the models. The MLP achieved a coefficient of determination (R2) of 0.926, outperforming CapsNet, which recorded an R2of 0.700. These results highlight MLP's capability to accurately capture nonlinear aerodynamic behavior while offering a computationally efficient alternative to CFD. The comparative analysis provides insight into the suitability of different neural architectures for aerodynamic modeling. Future work will focus on hybrid and physics-informed neural approaches to improve generalization and flow-field prediction accuracy.