Multiobjective optimization of UAV winglet performance using surrogate models and CFD

Abstract This paper presents a framework for optimizing UAV winglet design by combining high-fidelity computational fluid dynamics with surrogate modeling and multiobjective optimization. Two distinct optimization problems are addressed. First, aerodynamic multiobjective optimization is used to maximize lift and minimize drag, identifying Pareto-optimal winglet geometries. Second, integrated aerostructural optimization is performed, where lift-to-drag ratio (L/D) and root bending moment (RBM) are used as the aerodynamic and structural performance indicators, respectively. Reynolds-Averaged Navier-Stokes simulations are employed to evaluate performance indicators, with surrogate models used to reduce computational cost during optimization. The Non-dominated Sorting Genetic Algorithm II is utilized to solve the two multiobjective optimization problems and find Pareto fronts. Results from the aerodynamic optimization show that part of the Pareto front improves both lift and drag, compared to a baseline design from the literature, where lift can be increased by 0.93% for the same drag, or drag can be reduced by 0.48% for the same lift. For the integrated aero structural optimization problem, again part of the Pareto front improves both performance indicators, with L/D increasing by 0.56% for the same RBM, or RBM reduced by 1.63% for the same L/D. The performance of several surrogate models is compared showing that Kriging and Gaussian Process Regression are more effective for predicting RBM with limited data, while polynomial regression offers superior accuracy in predicting aerodynamic performance indicators when trained on larger datasets. The study concludes that using 50 samples provides the best trade-off between accuracy and efficiency for both aerodynamic and aerostructural optimization.