On constructing efficient UAV aerodynamic surrogate models for digital twins

Aerodynamic modeling and optimization for unmanned aerial vehicles (UAVs) are complex and computationally intensive tasks. Surrogate models have emerged as a powerful tool for increasing efficiency in the aircraft design and optimization process. We review and evaluate some modeling techniques, such as artificial neural networks and support vector regression, showing that Gaussian process regression generally provides a well-performing solution to this type of problem. We propose an active learning algorithm based on the relevance factor, that combines bias estimated from nearest-neighbor Euclidean distance and variance, to achieve higher accuracy with fewer compuational fluid dynamics (CFD) simulations. The obtained performance is evaluated using four 2-D test functions and an experimental CFD case, indicating that the proposed active learning approach outperforms classical random sampling techniques. Thanks to this architecture, the development process of a new commercial UAV can be significantly streamlined by expediting the testing phase through the use of DTs modeled more efficiently.