Surrogate modeling of aerodynamic coefficients for Unmanned Aerial Vehicle design

Abstract

In the field of aircraft aerodynamic design and optimization, the use of surrogate models has emerged as a powerful tool for reducing computational costs. Computational Fluid Dynamics (CFD) simulations are one of the highest computational expenses of developing a digital twin for an Unmanned Aerial Vehicle (UAV). In order to mitigate this expense, this study aims to evaluate the properties of Gaussian Process Regression algorithms in comparison to N-dimensional linear interpolators and Convolutional Neural Networks (CNNs) for use in the creation of a digital twin for an Unmanned Aerial Vehicle (UAV). An experimental analysis was conducted utilizing actual aerodynamic data from CFD simulations of a vertical takeoff and landing (VTOL) UAV. The results of this study indicate that Gaussian Process Regressors (GPRs) are the most suitable choice for estimating aerodynamic coefficients as a function of roll, pitch, and yaw angles.