Surrogate Modeling of the Aerodynamic Performance for Airfoils in Transonic Regime

Advancements in generative AI models have notably enhanced the automation of 3D shape generation, presenting transformative possibilities in the design of wings for aerospace applications. The optimization of such shapes relies on a large number of numerical simulations, which pose a computational challenge in the preliminary design stages. In this paper, we compare different machine learning models for surrogate modeling of the aerodynamic performance of airfoils for the transonic regime. We propose a new representation of the airfoils by combining geometric and aerodynamic features to comprehensively characterize the airfoil and its operating flight conditions. A training dataset that includes eight different transonic airfoils was generated, where we examined each airfoil under various operational flight conditions, encompassing a wide range of Angle of Attack (AoA) and freestream Mach numbers (M_∞). This resulted in a dataset comprising 1, 362 data points. The surrogate models employed in our study are primarily ensemble learning methods, including Random Forest, Gradient Boosting, and Support Vector Machines, complemented by deep learning techniques. We conduct a comparative analysis of these models to evaluate their efficacy in predicting aerodynamic coefficients. Our experiments show that different surrogate models can accurately and efficiently predict aerodynamic coefficients with an R² of 99.6% for unseen flight conditions. The dataset and code used in our study are accessible to the public at: https://github.com/Mohamedelrefaie/TransonicSurrogate.