Towards Robust Operation of VTOL Tailsitters

Abstract

As a result of the recent progress in robotics, Unmanned Aerial Systems find a quickly increasing number of practical applications, especially for in-air data acquisition in fields search as cartography, aerial photography, inspection or such and rescue. Most of these UAVs can be classified into two categories with their own strengths and weaknesses. Fixed-wing systems have higher efficiency and therefore longer flight times, they also require additional infrastructure or wide open spaces for take-off and landing. Rotary-wing systems, on the other hand, are generally more agile and have the ability to hover. For many applications however, this trade-off between rotary- and fixed-wing systems poses several challenges. Therefore a third hybrid system emerged that combines the efficiency with the ability of vertical take-off and landing. This thesis proposes the design and the control and estimation structure of such a hybrid system while the main focus lies on the reliability against wind and internal failures. The thesis is split into two main parts. In the first part of the thesis, we present the design system as well as the necessary low-level control architecture in order to operate such a hybrid system. We start with a brief comparison of different subcategories and proceed with the system design of a tailsitter. With the focus on computational efficiency, we then present the control synthesis with a novel globally stabilizing error function with monotonic properties. For the mapping from desired moment to actuator signals, an empirically derived polynomial model is presented. Starting off with a simple transition strategy with linearly changing setpoints a model-based transition optimization approach is presented in order to minimize the altitude deviation during this maneuver. With practical applications in mind, the reliability of UAVs is of utmost importance. Therefore this will be the main focus of the second part of the thesis. The design of tailsitter implies that during the hover phase the entire surface of the wing is exposed to horizontal winds. Therefore, the knowledge of the influence of the wind can be a crucial element to consider when controlling the system. In our work, we present an approach to model the influence of the wind and how to use it to estimate the direction and magnitude of the wind while hovering without additional sensors. The wind estimate during cruise flight typically relies on the airspeed measurements of a pitot tube. As this sensor is exposed to the environment it can happen that the pitot can get fully or partially clogged, which strongly affects the measurement. We present a way of detecting and mitigating this effect to ensure a robust wind estimate even in the cases of clogging. Finally, we analyzed the failsafe possibilities with respect to actuator failure. We present adaptations to the presented controller that allows us to control the system in case of a flap failure in both cruise and hover flight. In the case of a motor failure, we could show how to maintain controllability of the system during cruise flight. As a consequence of our developments, we were able to present the design and control and estimation algorithms, that enable a VTOL tailsitter to operate robustly in real-world conditions, including wind and a variety of internal failures. All of the above-presented methods are verified with outdoor experiments.