Toward predictive digital twins for self-aware unmanned aerial vehicles : non-intrusive reduced order models and experimental data analysis

The concept of the Digital Twin describes the use of comprehensive and authoritative digital models tailored to a unique physical asset that dynamically adapt as the asset evolves over time and are able to inform valuable decisions. A key challenge is to make a digital twin truly predictive so that it can be used to drive high-consequence decisions with quantified confidence. Currently, this can only be achieved through high-fidelity physics-based models. These models are computationally expensive to solve and prohibitive to use in a real-time context. Reduced-order modeling methods have emerged as a powerful tool for enabling high-fidelity simulations with computationally efficient models. This thesis aims to advance the framework for a predictive digital twin for unmanned aerial vehicles using physics-based models and scientific machine learning, as well as hardware experimentation. In particular, this work develops and demonstrates non-intrusive projection-based reduced-order modeling strategies for aerodynamic loading that can be applied to an unmanned aerial vehicle (UAV). In addition, this research presents an experimental data collection and analysis methodology to further the evolution of a digital-twin-enabled self-aware UAV. Self-awareness in this context refers to the ability of the vehicle to collect information about itself and its surroundings and to use this information to alter the way it completes missions via on-board dynamic decision-making. Emulation of wing damage states on the hardware testbed produces data sets that can be used in conjunction with previously developed computational methods in order to enable classification of the UAV structural state in flight. The two-way coupling between estimation of the UAV structural state and dynamic mission replanning is a capability that is critical for realizing the self-aware UAV concept.