Outdoors rotational test at University of Freiburg (27 February 2017)
Rotational test setup of the University of Freiburg (11 July 2016)
University of Freiburg "Half-Betty" wing, used to test rotational flight optimal control algorithms. (11 July 2016)
Jonas Schlagenhauf PhD Researcher University of Freiburg Department of Microsystems Engineering
Systems Control and Optimization Laboratory Georges-Köhler-Allee 102 79110 Freiburg Germany jonas.schlagenhauf@imtek.uni-freiburg.de www.syscop.de
Non-Linear Modeling with Learned Parameter Refinements for NMPC on a
Real-World Aerodynamic System
Jonas Schlagenhauf, Tobias Schöls, Moritz Diehl
Department of Microsystems Engineering (IMTEK), University of Freiburg
In this work we present a combined modeling and con-trol approach to nonlinear model predictive concon-trol of a rotational launch setup. While the continued interest in airborne wind energy (AWE) has produced many promis-ing solutions regardpromis-ing power generation trajectories and kite configurations, the task of autonomously launching and landing one or multiple kites is still a challenge. A major differentiating factor between current AWE launching approaches is the availability of on-board pro-pellers allowing the kite to fly to its crosswind operation region, drastically simplifying take-off and landing. How-ever for lift-mode systems where those motors cannot be reused for power generation, this comes with the cost of increased weight and additional complexity. If imple-mented robustly, a passive take-off and landing proce-dure could potentially increase the efficiency and dura-bility of the system significantly.
We present a modeling and control approach for a ro-tational launch maneuver based on a reduced-DOF real-world setup. We describe the system dynamics using a quadratic state space model, using linear regression in combination with hyper-parameter optimization to iden-tify the model parameters. A nonlinear model predic-tive controller is then designed based on the identified model.
The model performance was empirically evaluated com-paring the simulated behavior with the real-world system behavior. The controller was evaluated using a variety of reference trajectories, analyzing the dynamic behavior in a wide range of scenarios. Based on these experiments on
the real-world system, we could show that aerodynamic systems in a forced rotational motion can be adequately controlled using our approach. A future free-flight setup on the basis of this approach and empirically obtained data is currently in the planning stages [1].
1 2 3 1 4 3 4 2 5 6 6 1 Frame 2 Motor 3 Winch 5 Arm 6 Joints 7 Plane
Illustration of the carousel with attached plane.
References:
[1] Schlagenhauf, J. F.: Nonlinear Model Predictive Control of a Con-strained Model Airplane. MSc Thesis, University of Freiburg, 2017. https://freidok.uni-freiburg.de/data/12264