Data driven Mathematical Modelling and Analysis of a VTOL UAV system

Thanks to new technologies in terms of design of software and hardware, Unmanned Aerial Vehicles (UAVs) or drones are spreading all over the world. Drones are effective and compatible to many different fields such as agriculture, military operations, surveillance, scientific studies, cinema and the delivery sector. Due to Multicopters flight efficiency, companies and purchasers are still sceptical of their endurance. Vertical Take Off and Landing (VTOL) drones have been introduced as a solution for both endurance, flight distance and restricted landing and take-off area. These drones are capable of extreme flight conditions but only if correctly controlled. Open-source software often provide basic flight controls and sluggish drones’ dynamic responses. These control laws need to cover a whole variety of different configurations and are not specific for each drone dynamics. ArduPilot is an open-source extremely versatile software. It offers compatibility with several different drones’ configurations. Despite being robust, it provides simple and limited flight control which makes many flight phases energy-expensive and unoptimized. New control laws are proposed to optimize some of these manoeuvres to minimize the time spent, energy loss and increase safety. The developed control laws have the objective of cooperating with the main flight control system inside the autopilot to improve its overall performances. In a brief description of the current drones’ State of Art, the differences that arise between the various designs are highlighted, concluding that the configuration in exam can be inserted into the market delivery sector. In this thesis, two drones of two different configurations, Tailsitter and Bellysitter, are introduced and analysed. As no data of these drones was available beforehand, the most fundamental physical, aerodynamic and thrust properties such as mass, inertia, centre of gravity and neutral point were calculated. A flight campaign has been performed and many different flight conditions have been studied. The data was generated from on-board sensors and then postprocessed in MATLAB. The analysis of the logs, Bench Tests results and other data sources allowed the estimation of both propellers’ and wing-aerodynamic coefficients. In order to study the many possible control algorithms, a six Degrees of Freedom Simulator is developed in MATLAB/SIMULINK environment. After many simulations and tests, some algorithms have been either written in Lua programming language or in C++ to be inserted directly into the firmware to cooperate with the rest of the code during these difficult manoeuvres. Tests in lab have been performed to check the consistency of the results obtained and the robustness of the algorithms. A lookup-table-based control was developed as a starting point for the landing phase. Later a PID control law was designed as well as Sliding Mode control to increase both performances and robustness. Later, given the extreme complexity of some flight phases, the option of a Model Predictive Control has been analysed.