Predicting Laminar-Turbulent Transition on Rotorcraft Airfoil Using Multi-Fidelity Artificial Neural Networks

Laminar-turbulent transition is a significant aerodynamic phenomenon that is often neglected in rotorcraft design. Rotorcraft comprehensive code like RCAS and UMARC rely on accurate airfoil polar table to calculate aerodynamic forces acting on the rotor blades. However, Computation Fluid Dynamics (CFD) based airfoil tables are almost always constructed with fully turbulent flow cases, resulting in an underestimation of the predicted performance when compared to one with transition effect factored in. This study developed a machine learning based surrogate model to predict airfoil polars on rotorcraft airfoil using combination of low and high fidelity data. Sobol sequence sampling is used to generate the diverse set of flow condition varying Mach number, Reynolds number, and turbulence intensity. In total, over 1,000,000 XFoil simulations and 100,000 CFD are carried out in the exploration and construction of the multi-fidelity surrogate model.