|dc.description.abstract||This thesis focuses on the estimation of wind velocities, angle of attack and aerodynamic coefficients for small unmanned aerial vehicles (UAVs) and in addition, the use of these estimates for two different applications. The main motivation behind this estimation problem is to acquire accurate knowledge of these states to support safer and more efficient UAV operations. Since UAVs face more severe restrictions on size, cost and weight than larger manned aircraft when it comes to equipping additional sensors to measure these variables, knowledge about them is often unavailable. Furthermore, wind tunnel data is often hard and costly to obtain for specific UAV platforms making it hard to use model based estimation approaches.
Therefore, in this thesis methods are investigated that solely use data from sensors that are typically part of a standard UAV autopilot sensor suite, while avoiding the need for prior knowledge. This suite consists of a Global Satellite Navigation System (GNSS) receiver, an inertial measurement unit (IMU) and a pitot static tube measuring the airspeed of the UAV. This sensor data is combined with kinematic, aerodynamic and stochastic wind models in order to estimate wind velocities and aerodynamic coefficients. The aerodynamic model used, is an approximate lift model whose coefficients are estimated on-line. The stochastic wind model separates the wind velocity in two components, a steady wind velocity and a turbulent wind velocity, modeled by a Dryden wind model.
As a first step a simulation study has been carried out, where the sensor data and models have been combined in two different estimators, an Extended Kalman Filter (EKF) and a Moving Horizon Estimator (MHE). In a comparison it showed that the MHE achieves a faster convergence rate and lower estimation biases than the EKF.
Based on the results of the simulation study, the MHE approach was further refined and tested in an experimental study. Flight tests were performed with two different air frames and two different sensor payloads, one consisting of a high-precision IMU and RTK-GPS while the other sensor set uses the autopilot’s low cost sensors. In addition, a high-quality multi-hole probe was used as a reference to assess accuracy. With each of those platforms two test flights were performed covering a variety of different maneuvers. Results show root-mean-square errors around 1for the angle of attack estimation for both airframes and both sensor sets and wind velocity estimation errors below 1m=s.
Two different applications of this estimator are presented in this thesis. In the first application the lift coefficient estimates are used to detect icing on an UAV. For this, first the effects of icing on an UAV are studied and then an icing flight simulator is developed. It shows that for severe icing the lift coefficients are decreased. Simulation results show that this change is detectable using the MHE approach. In the second application the wind velocity estimates are used to increase the path following performance of an UAV by incorporating the wind velocity information into the path planning algorithm. The path planning algorithm considers the aircraft’s kinematics, flight envelope and wind estimate. Simulation results show an improved path following performance and a better exploitation of the flight performance of an unmanned aircraft by the use of the wind adaptive path planning algorithm.||nb_NO