PhD Defense - Kurt Geebelen
Start: 23/04/2015, 10:00
Location: Auditorium tweede hoofdwet (Thermotechnical Institute)
Title
Design and Operation of Airborne Wind Energy Systems
- Experimental Validation of Moving Horizon Estimation
for Pose Estimation
Abstract
Global climate change as well as water and air pollution lead to a demand for a sustainable energy supply. Renewable energy sources such as wind and solar power, hydro-electric powerand geothermal energy, together with energy efficiency measures, are a viable alternative to fossil fuels. Focussing the attention on wind energy it is interesting to note that Airborne Wind Energy (AWE) has the potential to capture wind energy at a fraction of the cost achievable by current wind turbines. In the method explored in this thesis, an aeroplane flies a crosswind trajectory while it is tethered to a ground based winch. This winch consists of a drum connected to a motor/generator. The tether is wound up on the drum, and electricity is produced using the ‘pumping cycle’. In the first phase of the pumping cycle, the aeroplane delivers a high traction force on the tether while it is being reeled out, causing the generator to produce electricity. Once the tether is fully unrolled, the aeroplane is controlled such that the force on the tether is reduced and the tether is reeled in using only a fraction of the electricity produced in the firstphase.
Unfortunately the benefits of airborne wind energy come at a cost. While a wind turbine only needs to be aimed towards the wind to operate, an AWE system needs to be constantly controlled to fly a certain crosswind trajectory. Because of this, AWE systems need an automatic control system, which in turn needs a reliable estimate of the system state.
The first part of this dissertation discusses the development of two experimental AWE test set-ups, one set-up for indoors use and one set-up for outdoors use. The goal of these set-ups is to analyse and experimentally validate the rotation start, a start-up method for AWE systems in which the tethered aeroplane is brought up to speed by an arm rotating around a central vertical axis. The dissertation presents the development and design choices for both set-ups.
The indoors set-up is equipped with sensors that allow estimating the position and orientation of the aeroplane, including an Inertial Measurement Unit (IMU) that measures the acceleration and angular velocity of the aeroplane, and a stereo vision system that offers measurements of the aeroplane’s position and orientation. The outdoors set-up is larger than the indoors set-up and mobile to allow experiments to be done at remote locations using larger aeroplanes. The set-ups will be used to implement and experimentally validate advanced, optimisation-based state and parameter estimation techniques and in the future advanced control techniques.
The second part of this dissertation investigates methods to fuse the different sensor measurements to form a reliable state estimate. Moving Horizon Estimation (MHE) is presented as a technique that can reliably fuse the information from the nonlinear system and measurement models and compared to traditional methods such as the extended and unscented Kalman filter using both simulations and experimental data obtained on the indoors set-up. MHE is shown to have both a better start-up behaviour and average estimation performance than Kalman filtering techniques.
Estimators based on both a kinematic and a dynamic model are presented. The developed moving horizon estimator based on a dynamic model of the system is shown to have better estimation performance than estimation based on a kinematic model, at the cost of an increase in computation time.
A main concern of the developed estimators is the limited sampling rate caused by the computational load of the MHE. To relieve this limitation and achieve higher update frequencies, a novel approach using a second inner MHE is presented. This inner MHE uses the IMU measurements to update the state estimates and is shown to perform better than traditional dead reckoning for long prediction times.