Leveraging genetic algorithms and neural networks for stiffness and weight optimisation of an unmanned aerial vehicle (UAV) support arm

• Advanced surrogate models with significantly reduced computational costs. • Integration of GA with ANN into a multi-objective optimisation framework. • Design, modelling, optimisation, manufacturing, and testing of a support arm. • Ultra-lightweight design of a UAV. • Stiffness maximisation and mass reduction of UAV components. The increasing demand for lightweight and high-performance structures in aerospace applications has spurred the development of advanced design frameworks for 3D-printed composite components. This study proposes a multi-objective optimisation approach that integrates the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) with an Artificial Neural Network (ANN) to optimise the design of a UAV support arm. The dual objectives are to minimise weight and maximise stiffness. A finite element (FE) database is first constructed using four geometric parameters: height, width, lattice element count, and thickness; under combined thrust and torsional loading. This is expanded with three additional parameters − edge thickness, lower reinforcement thickness, and upper reinforcement width − for ANN training. The ANN predicts mass and stiffness with over 99% accuracy while reducing computational time by 99% compared to full FE simulations. NSGA-II then generates a Pareto front, from which an optimal design is selected and fabricated using fused filament fabrication (FFF) with polylactide (PLA). The final prototype achieves a 42% weight reduction and a 51% increase in specific stiffness relative to the baseline, demonstrating the potential of the proposed framework for solving complex composite design problems.