Machine learning techniques to estimate the dynamics of a slung load multirotor UAV system

Vargas Moreno, Aldo Enrique (2017) Machine learning techniques to estimate the dynamics of a slung load multirotor UAV system. PhD thesis, University of Glasgow.

Full text available as:
[img]
Preview
PDF
Download (26MB) | Preview
Printed Thesis Information: https://eleanor.lib.gla.ac.uk/record=b3289887

Abstract

This thesis addresses the question of designing robust and flexible controllers to enable autonomous operation of a multirotor UAV with an attached slung load for general cargo transport. This is achieved by following an experimental approach; real flight data from a slung load multirotor coupled system is used as experience, allowing for a computer software to estimate the pose of the slung in order to propose a swing-free controller that will dampen the oscillations of the slung load when the multirotor is following a desired flight trajectory. The thesis presents the reader with a methodology describing the development path from vehicle design and modelling over slung load state estimators to controller synthesis.

Attaching a load via a cable to the underside of the aircraft alters the mass distribution of the combined "airborne entity" in a highly dynamic fashion. The load will be subject to inertial, gravitational and unsteady aerodynamic forces which are transmitted to the aircraft via the cable, providing another source of external force to the multirotor platform and thus altering the flight dynamic response characteristics of the vehicle. Similarly the load relies on the forces transmitted by the multirotor to alter its state, which is much more difficult to control. The principle research hypothesis of this thesis is that the dynamics of the coupled system can be identified by applying Machine Learning techniques.

One of the major contributions of this thesis is the estimator that uses real flight data to train an unstructured black-box algorithm that can output the position vector of the load using the vehicle pose and pilot pseudo-controls as input. Experimental results show very accurate position estimation of the load using the machine learning estimator when comparing it with a motion tracking system (~2% offset). Another contribution lies in the avionics solution created for data collection, algorithm execution and control of multirotor UAVs, experimental results show successful autonomous flight with a range of algorithms and applications. Finally, to enable flight capabilities of a multirotor with slung load, a control system is developed that dampens the oscillations of the load; the controller uses a feedback approach to simultaneously prevent exciting swing and to actively dampen swing in the slung load. The methods and algorithms developed in this thesis are validated by flight testing.

Item Type: Thesis (PhD)
Qualification Level: Doctoral
Keywords: uav, neural networks, recurrent neural networks, slung load, controller, multirotor, drone.
Subjects: T Technology > TL Motor vehicles. Aeronautics. Astronautics
Colleges/Schools: College of Science and Engineering > School of Engineering > Aerospace Sciences
Supervisor's Name: Anderson, Dr. David and Thomson, Dr. Douglas
Date of Award: 2017
Depositing User: Aldo Enrique Vargas Moreno
Unique ID: glathesis:2017-8513
Copyright: Copyright of this thesis is held by the author.
Date Deposited: 17 Oct 2017 12:17
Last Modified: 23 Nov 2017 08:29
URI: http://theses.gla.ac.uk/id/eprint/8513
Related URLs:

Actions (login required)

View Item View Item

Downloads

Downloads per month over past year