No abstract

This paper presents a combined plant and controller optimization process for airborne wind energy systems (AWEs) that fuses numerical optimization with labscale experimental results. The methodology introduced in this paper, referred to as experimentally-infused optimization, addresses several challenges faced by AWE system designers, including a strong coupling between the controller and plant design, significant modeling uncertainties (which require the use of experiments), and high costs associated with full-scale experimental prototypes. This paper presents an initial case study of the proposed experimentally-infused optimization, where experiments were conducted on a 1/100 th -scale model of Altaeros Buoyant Air Turbine (BAT), which was tethered and flown in the University of North Carolina at Charlotte 1m × 1m water channel. The lab-scale experimental platform reduced the cost of evaluating flight dynamics and control by more than two orders of magnitude, while resulting in substantially improved flight performance, quantified by a 15.2 percent improvement in an objective function value, as compared to a purely numerical optimization.

No abstract

No abstract

This paper presents a methodology for optimally fusing experiments and numerical simulations in the design of a combined plant and control system. The proposed methodology uses G-optimal Design of Experiments to balance the need for experimental data with the expense of collecting a multitude of experimental results. Specifically, G-optimal design is used to first select a batch of candidate experimental configurations, then determine which of those points to test experimentally and which to numerically simulate. The optimization process is carried out iteratively, where the set of candidate design configurations is shrunken at each iteration using a Z-test, and the numerical model is corrected according to the most recent experimental results. The methodology is presented on a model of an airborne wind energy system, wherein both the center of mass location (plant parameter) and trim pitch angle (controller parameter) are critical to system performance.