Data-driven incremental flight dynamics and center of gravity control for tailsitter drones

Abstract

This paper presents a novel design to enhance the performance of a tailsitter flying wing during its transition phase by adjusting the center of gravity (CG) location and employing temporal index Gaussian process-based derivative incremental nonlinear dynamic inversion (TIGPD INDI) control to address uncertainties arising from CG shifts. In the forward and backward transition phases, the tailsitter relies on elevon and thrust control to manage its pitch axis control. During the forward and backward transition, pitch moment must be generated to rotate large pitch angle range under low dynamic pressures. If restoring pitch moment is too high or too low, the static stability hinders the transitions and flight maneuvers. By adjusting the CG location, the restoring pitching moment relative to the angle of attack is effectively manipulated. The tilt-rotor, tilt-wing and hybrid VTOL integrate rotating structural mechanism or motors for transitions, creating mechanical complexity and increasing mass and energy consumption due to redundancy. There exists extended flap, propeller pitch control, double wing configurations for tailsitter to solve lack of pitching moment issue. Taking an innovative approach, the proposed approach eliminates the need for additional multiple actuators or motors typically required for these configurations by employing simple battery moving mechanism with single actuator without any structural changes at the external frame. To address the complex moment dynamics introduced by CG shifts, a robust TIGPD-INDI control is employed, treating the effects of the CG shift in the model as uncertainties. This controller design requires state derivative terms, for which Gaussian process regression—a data-driven model estimation method—is proposed. Simulation results demonstrate that the proposed CG control integrated tailsitter enables smoother transitions and maintains a narrower angle of attack range compared to a tailsitter without static margin adjustments. Also, the proposed TIGPD INDI controller showed robustness and better estimation performance than other methods.