Dragonfly Rotor Optimization Using Machine Learning Applied to an OVERFLOW-Generated Airfoil Database

NASA’s fourth New Frontiers Mission is the Titan Dragonfly relocatable lander. This coaxial quadrotor vehicle is scheduledto launch on a rocket in 2028 with the goal of exploring Titan’s prebiotic chemistry and habitability. The multirotor designfor this unique application has evolved to meet constraints such as Titan’s cryogenic atmosphere at 95 K (−288◦F), gravity14% that of Earth’s, atmospheric density 4.4 times of Earth’s standard sea level, and the inability to test the entire systemunder these conditions until the first flight on Titan. This paper focuses on rotor design aspects of the Dragonfly landerand introduces a framework for multirotor design optimization considering multiple flight conditions. A new OVERFLOWMachine Learning Airfoil Performance (PALMO) database is first presented. PALMO is then incorporated into a Bayesianoptimization framework and applied to a four-rotor system (one side of the Dragonfly lander). Training data are generatedon each iteration of the optimization using the CAMRAD-II comprehensive analysis to evaluate rotor designs overthe planned mission profile. An optimal design for the four-rotor system was found with approximately 900 rotor designsanalyzed in CAMRAD-II, which required 9 million queries of the PALMO surrogate models. This demonstration case evaluated10,000,000 candidate rotor designs in 5.5 h on 114 CPU cores using uniform inflow, and in 27.8 h using the prescribedwake model.