Nonlinear Observers for Aircraft Maneuvering in Wind

Abstract

Knowledge of wind velocity is fundamental across fields from atmospheric science to aeronautics, yet direct wind sensing is often constrained by operational limits. This motivates indirect wind estimation methods that infer wind from aircraft motion. However, typical model-based estimators lack rigorous stability guarantees across the full flight envelope --- a major limitation for safety-critical aerospace applications. This dissertation addresses these gaps by advancing nonlinear observer design and flight dynamic modeling to estimate wind from aircraft motion with assured performance. First, a symmetry-preserving, reduced-order state observer is introduced for the unmeasured part of a system's state, leveraging the fact that the system dynamics are invariant under the action of a Lie group. By using a moving frame to construct invariant observer mappings, both the design process and stability analysis are simplified. In cases where the system's nonlinearities comprise the Lie group's action, the nonlinear observer may even yield linear state estimation error dynamics to enable a multitude of design and optimization techniques that improve performance. Next, a quasi-steady nonlinear flight dynamic model for multirotor aircraft is derived from blade-element and momentum theory, ensuring validity over a large operating range while remaining identifiable from flight data. The utility of this model is assessed through a high-fidelity simulation study based on wind tunnel data. Recognizing the challenges of parameter estimation in large-domain models for unstable aircraft, a two-phase data collection methodology is proposed. In the first phase, a set of linear time-invariant models is identified at multiple operating conditions to define an uncertain linear parameter-varying (LPV) model. In the second phase, a robust LPV control law with an H-infinity norm bound guarantee is synthesized, enabling automated flights with sufficiently large excitation signals for nonlinear system identification. Finally, the nonlinear observer theory is combined with the large-domain flight dynamic models to achieve provably effective wind estimation for maneuvering aircraft. The framework is extended to uncertain aerodynamics and random turbulence by formulating the system as a stochastic differential equation. A nonlinear passivity-based wind observer is also introduced, serving as a full-order alternative to reduced-order methods. Together, these observers offer stability guarantees applicable to general maneuvering flight, demonstrated on both fixed-wing and multirotor UAVs. Overall, this dissertation contributes to safer, more autonomous aerospace systems.