Multi-fidelity multi-objective Bayesian optimization for fixed-wing design of training UAV
Anonphong Risanthia, Tharathep Phiboon, Auraluck Pichitkul, Suradet Tantrairatn, Sujin Bureerat, Masahiro Kanazaki, Atthaphon Ariyarit · Smart Science · 2025
Fixed-wing Unmanned Aerial Vehicles are increasingly essential for applications such as surveillance, environmental monitoring, and precision agriculture. To support extended missions, optimizing aerodynamic design to reduce energy consumption is crucial, as efficient energy use extends flight duration and enhances mission effectiveness. Additionally, ensuring ease of landing is vital, especially in challenging environments, to minimize damage risks and extend UAV operational life. Traditional aerodynamic optimization methods, while effective, are often time-consuming and computationally expensive, especially when addressing multiple objectives like reducing drag during cruising and maximizing lift during landing. To overcome these challenges, this study employs a multi-objective Bayesian Optimization (BO) framework, which is well suited for efficiently balancing trade-offs in complex design spaces. By integrating low-fidelity and high-fidelity models, the study reduces computational costs without compromising accuracy. The low-fidelity model, based on the Vortex Lattice Method (VLM), enables rapid approximations, while the high-fidelity model, using Computational Fluid Dynamics (CFD) simulations, provides detailed results. The study focuses on optimizing key design variables, including thickness-to-chord ratio, taper ratio, and twist angle, achieving significant aerodynamic performance improvements during both cruising and final approach phases. Some optimal designs show over a 5% performance enhancement while maintaining the same wing area, demonstrating the effectiveness of the proposed approach.