Deep Learning-Based Aerodynamic Analysis for Diverse Aircraft Configurations

In this study, a neural network was developed to predict the aerodynamic characteristics of fixed-wing aircraft with two lifting surfaces of various aerodynamic configurations. The proposed neural network model can incorporate 23 parameters to describe the aerodynamic configuration of an aircraft. A methodology for discrete geometric parameterization of aerodynamic configurations is introduced, enabling coverage of various combinations of relative positions of aircraft components. This study presents an approach to database construction and automated sample generation for neural network training. Furthermore, a procedure is provided for data preprocessing and correlation analysis of the input variables. The optimization process of the hyperparameters of the multilayer perceptron (MLP) architecture is described. The neural network models are validated through comparison with numerical simulations. Finally, several aerodynamic design problems are addressed, and the key advantages of the developed MLP-based surrogate aerodynamic models are demonstrated.