Machine Learning-based Surrogate for Variable Airfoil Rotor Blade Design

Rotor blade design is a fundamental problem in rotorcraft aerodynamics. Conventional design methods rely on high-fidelity simulations such as Computational Fluid Dynamics (CFD) that are computationally prohibitive at the early design stage, while low-fidelity methods lack the accuracy required to capture complex aerodynamic interactions. This paper presents a surrogate-based framework for a rotor blade design with multiple airfoil geometries along the span. An efficient design of experiments strategy based on Sobol sampling with unique airfoil identifiers reduces the effective sampling dimension during data generation. Training data is generated using an in-house linear inflow model, which provides a computationally inexpensive yet representative dataset. The primary objective of this work is to demonstrate a complete end-to-end methodology for surrogate model development for high-dimensional rotor blade design with varying spanwise airfoil geometries. The linear inflow model is intentionally chosen to keep data generation tractable and focus the contribution on the surrogate pipeline itself, with the extension to higher-fidelity aerodynamic methods reserved for future work. Artificial Neural Network (ANN)-based surrogate achieves higher accuracy for aerodynamic predictions in hover. The surrogate is coupled with a Genetic Algorithm (GA) to minimize the rotor power subject to a constant thrust constraint.