Data-driven aerodynamic shape optimization of rotor airfoils based on deep learning

Abstract As the fundamental component of rotor blades, the airfoil dominates the aerodynamic performance of helicopter rotors. The aerodynamic shape optimization of rotor airfoils is thus an effective means of improving their aerodynamic performance. Given that the aerodynamic requirements for the airfoil vary across different states of flight, such as hovering, maneuvering, and forward flight, multi-objective optimization of its aerodynamic design is necessary. In this study, we propose a scheme that simultaneously considers the optimization of the tip and root airfoils to design rotor blades that deliver satisfactory aerodynamic performance. We combine a data-driven surrogate model with a multi-island genetic algorithm to efficiently and globally optimize the rotor airfoils by using the Helishape-7A rotor as the benchmark. We first used Latin hypercube sampling to establish a database of airfoils, and then applied deep neural networks to train surrogate models and obtain the aerodynamic coefficients of the tip and root airfoils. Following this, we optimized the aerodynamic shape of the tip and root airfoils using the multi-island genetic algorithm while considering multiple objectives, including the states of hovering, maneuvering, and forward flight of the helicopter. Finally, the baseline Helishape-7A rotor's aerodynamic performance was compared with that of the optimized rotor. The outcomes demonstrated that, in comparison to the baseline, the optimized rotor produced noticeably better aerodynamic performance.