Prediction of aerodynamic coefficients for multi-swept delta wings via a hybrid neural network

This study investigates the prediction capabilities of a hybrid neural network model for the aerodynamic coefficients of multiple swept delta wings, focusing on rapidly estimating coefficient slopes and aerodynamic characteristics. By leveraging a machine learning approach combining image classification and conventional feed-forward neural networks, the study aims to provide an efficient alternative to resource-intensive computational fluid dynamics simulations or wind tunnel experiments during the early design phases of aircraft. The training dataset is derived from existing wind tunnel measurements of multiple swept delta wing configurations with varying leading-edge sweeps, incorporating both baseline configurations and scenarios with deflected control surfaces and applied sideslip angles. Parameters such as Mach number, Reynolds number, aspect ratio, angle of attack, angle of sideslip, and control surface deflections are considered additional input variables, representing a comprehensive range of flow conditions encountered in practical aerodynamic analyses. The findings demonstrate that the proposed hybrid neural network effectively predicts aerodynamic coefficients with high accuracy in forecasting lift coefficients and their derivatives. While the model exhibits excellent predictive performance for the trends in coefficient slopes, it shows limitations in accurately predicting the absolute values of pitching moment coefficients. Overall, the results underscore the potential of machine learning techniques for rapidly evaluating aerodynamic characteristics and slope trends, offering significant time and cost savings in preliminary aircraft design.