Surrogate-Based Multi-Objective Optimisation of UAV Winglet Designs: Integrating Aerodynamic Performance and Structural Integrity through High-Fidelity Simulation

Abstract

Winglets reduce induced drag and extend UAV endurance but introduce structural penalties through increased loading and mass. Existing optimisation strategies inadequately capture these aero-structural trade-offs, often neglecting deformation, buckling and composite failure constraints. This research addresses these gaps by developing a surrogate-assisted framework that jointly optimises aerodynamic efficiency and structural integrity. The proposed methodology integrates automated parametric geometry generation with steady RANS CFD and composite shell FEA. Three sequential phases are addressed. First, multi-objective aerodynamic optimisation is performed to solve two problems: maximising lift while minimising drag and maximising lift-to-drag ratio (L/D) while minimising root bending moment (RBM). Surrogate models polynomial regression, Kriging and Gaussian Process Regression (GPR) trained on Latin Hypercube Sampling datasets enable efficient design space exploration. Second, selected Pareto-optimal designs undergo structural optimisation to minimise wing mass subject to deformation, Tsai–Wu failure and buckling constraints under pull-up and push-down manoeuvres, with adaptive infill sampling refining surrogate accuracy. Finally, two-way fluid–structure interaction (FSI) simulations validate aeroelastic performance. Results show that for aerodynamic optimisation, 50 CFD samples provide optimal cost-accuracy balance, with Kriging and GPR achieving superior prediction accuracy. For structural optimisation, GPR consistently converges within three infill iterations, starting from 25 FEA samples. Aerodynamic optimisation improves L/D by up to 1% while reducing RBM by 1.6% compared with a baseline design, demonstrating the trade-off between aerodynamic efficiency and structural mass. Two-way FSI validation of optimised designs shows only minor differences in aerodynamic performance (L/D ≤ 0.7%, RBM < 1%) but some larger differences for structural responses: deformation increases ≤ 1.6%, buckling factor reductions ≤ 2.1%, and Tsai–Wu index increases of 16–21%. The developed framework demonstrates that surrogate-assisted optimisation delivers aerodynamically efficient, structurally feasible winglet designs while significantly reducing computational effort.