Physics-Based Modeling for Control and Autonomous Operation of Unmanned Aerial Vehicles

Abstract

UAS are widely employed in commercial and military applications, and their utilization is growing at a rapid pace. Effective predictive models for aeromechanics, body dynamics and control are critical in trajectory planning and optimization, autonomous operations, and decision-making. Aeromechanical and wind models that are currently used in the control and guidance of UAS are typically simplistic and often do not represent the essential physics to an adequate degree. Therefore, the performance and versatility of such vehicles may be limited in extreme flight conditions. At the other end of the spectrum, there exist high fidelity models that are computationally expensive, and thus not applicable in path planning, optimization, and onboard flight controllers.

The major goal of this dissertation is to bridge the gap between physics-based models and onboard decision-making. Multi-disciplinary models of appropriate fidelity are developed and integrated into a comprehensive flight simulation software suite. These models are experimentally validated and utilized in trajectory planning, optimization, onboard control and autonomous flight. Studying the impact of models of different fidelity for the environment and the aerodynamics determines the impact of modeling uncertainty on system-level goals.

A vortex-based model for lifting surfaces is developed, using which control surfaces and couplings therein can be efficiently represented. Using this model, the interaction of the propeller wake with a downstream wing is studied, and it is demonstrated these models are effective tools in predicting the propeller-induced span-wise loading. Such a model is beneficial for trajectory planning and optimization applications to improve flight stability and trajectory tracking.

Next, a novel HBEM model is developed to predict rotor forces over a wide range of flight conditions. The HBEM model is self-contained and combines blade element theory, momentum theory and a linear inflow model to determine the {em unique} inflow that is {em consistent with all theories}. The model utilizes the blade geometry and the flight condition as inputs to determine the relationship between the forces/moments and the rotor RPM. A detailed set of wind tunnel experiments is conducted to validate the model across a very wide range of flight regimes. Further, a semi-empirical model for the RIPF is developed using experimental data. It is noted that these models can be executed in real-time which makes them useful for implementation in flight software.

A custom quadrotor is built and equipped with an ultrasonic wind sensor and RPM sensors. The HBEM and RIPF models are embedded in quadrotor flight software, and it is illustrated these models are fully integrable and efficient enough to run on a typical onboard compute module. To evaluate the ability of these models to function in harsh environmental conditions, motion capture aided autonomous flight is realized in the presence of strong wind gusts generated by a large industrial fan. A feedforward controller is designed to incorporate physical insight into flight mechanics and to provide estimates of the state. Flight tests are conducted in and out of strong crosswind conditions to further show the impact of computationally efficient models that are capable of being executed onboard in real-time. It is shown that the wind sensing and physics-based models along with the feedforward controller improve trajectory tracking in extreme environmental conditions.