Multirotor Design and Modeling for Indoor Applications

In recent years, multirotor Unmanned Aerial Vehicles (UAV) came to the attention of the scientific world. Ingenuity experiment, the first drone to fly on another planet is their last success. UAVs are widely used in different sectors thanks to their small dimensions, the possibility to hover, the great precision of maneuvers and the limited costs. In fact, UAVs are employed for military and scientific applications. They are also used in rescue and healthcare operations, precision agriculture, shipping and delivery services, as well as for hobby, for filming and for FPV races. The purpose of this thesis is designing, modeling and assembling a multirotor for indoor applications and designing and testing on-board guidance algorithms. The first part consists of designing a small drone. The motors, the propellers, the battery and the Electronic Speed Controllers (ESC) are dimensioned to have the best compromise between endurance and weight. Then, 3D printing is used to produce the UAV frame. This technology allows rapid prototyping, low costs, reduced weight. The plates are printed in PolyLactic Acid (PLA) whereas the arms are made from aluminum. Finally, the whole chassis is tested through a static analysis to verify that it can endure the loads. Motors and propellers are the principal part of a drone because they produce the thrust that allows the drone to fly. It is therefore important to properly characterize them with thrust tests. Experimental thrust tests are carried out on the propellers to obtain a model of the actuators to be used in simulation. In the second part of the thesis project, guidance and on-board algorithms are analyzed. A trajectory planner is implemented as guidance algorithm. Two improved versions of classic algorithms are considered and compared: an improved Artificial Potential Field (APF) and a Rapidly exploring Random Tree Star Fixed Nodes with APF (RRT*FNDAPF) that is a revisited Rapidly exploring Random Tree (RRT) algorithm for unstructured environments. Both algorithms are implemented in the MATLAB Simulink model of the drone. Khatib’s APF has a low computational effort and can avoid static obstacles and new obstacles that were not predicted before the flight. However, it cannot map all the environment before the flight, it cannot avoid moving obstacles and it could present the local minima problem and the Goal Non-Reachable with Obstacles Nearby (GNRON) problem. These disadvantages can be solved by modifying the repulsive potential field of APF. The latter path planning algorithm is derived from a classic RRT. The classic RRT can map the environment before the flight, and it is simple to implement. However, it has a computational effort that depends on the dimensions of the environment, it cannot avoid dynamic obstacles, and it cannot find the optimal path. In addition, the convergence could be slow because the search of nodes is completely random. RRT*FNDAPF can solve these problems. For both algorithms, simulations are performed in MATLAB Simulink. The simulations are also made using Unity as plant to have more realistic results.