Collaborative Research: An Economic Iterative Learning Control Framework with Application to Airborne Wind Energy Harvesting

这个项目的目标是以重复的方式控制系统的新技术的先驱，使用来自先前迭代的信息来改进每个连续迭代的性能。不像现有的技术只寻求改善沿着路径的位置跟踪，这个研究项目将集中在经济指标的迭代到迭代的改进上，比如能量产生/支出，总迭代时间，或者货币成本。在这项研究中创造的控制技术将适用于各种各样的系统，其中重复控制是必不可少的，并且存在一个明确的经济目标，可以从一个迭代到下一个迭代进行改进。示例应用包括装配线和制造操作（一个超过1万亿美元的行业，其中数千个零件被生产出来，最大限度地减少制造时间和能源消耗是至关重要的），主动控制外骨骼（控制重复的人类行走步态），以及机载风能系统。这项研究将特别关注空中风能系统，用绳索和提升体取代传统的塔架，用很少的材料驾驭高空风。这些系统可以通过重复的侧风飞行而不是固定的运行来产生大量增加的能量，并且存在着从一个重复到下一个重复来提高能量产生性能的重要机会。这项研究将通过教育和推广活动得到加强，包括创建以风能为灵感的本科课堂模块，在夏洛特工程大学早期高中开展风筝设计活动，以及在地区航空风能公司Windlift的暑期机会。该项目将为一个独特的经济迭代学习控制框架提供新的控制理论知识，该框架侧重于最大化或最小化盈利指数，而不仅仅是跟踪绩效。具体来说，学习框架将融合两种机制，这两种机制将在点对点迭代学习控制中转变当前的技术状态。首先，与传统的点对点迭代学习控制方法不同，在传统的方法中，路径点是预先指定的，并且只有路径点之间的行为可以从一次迭代调整到下一次迭代，该框架将使路径点本身能够从一次迭代调整到下一次迭代。其次，内循环灵活时间迭代学习控制模块允许航路点到达时间和总迭代时间在每次迭代中变化。这将使研究团队能够通过迭代学习框架来解决时间优化和能量优化问题，这在一般意义上迄今尚未完成。考虑到航点自适应律和灵活时间迭代学习模块由两个相互关联的子系统组成，采用小增益稳定性分析框架推导航点每次迭代允许变化的边界。最后，将ILC框架应用于AWE系统的重复侧风飞行，将为侧风飞行的优化提供一种在线学习机制，由于这些系统的动态模型复杂且不确定（这通常会使离线侧风轨迹优化无效），这一点尤为重要。

The objective of this project is to pioneer new techniques for controlling systems that operate in a repetitive manner, using information from previous iterations to improve the performance at each successive iteration. Unlike existing techniques that seek only to improve position tracking along a path, this research project will focus on the iteration-to-iteration improvement of economic metrics such as energy generation/expenditure, total iteration time, or monetary cost. The control techniques created in this research will be applicable to a wide variety of systems for which repetitive control is essential and there exists a clear economic objective to improve upon from one iteration to the next. Example applications include assembly line and manufacturing operations (an over $1 trillion industry where parts are produced by the thousands and minimizing manufacturing time and energy expenditure is critical), actively-controlled exoskeletons (which control a repetitive human walking gait), and airborne wind energy systems. This research will focus specifically on airborne wind energy systems, which replace the conventional tower with tethers and a lifting body to harness high altitude winds using very little material. These systems can generate substantially increased energy through repetitive crosswind flight, rather than stationary operation, and there exists a significant opportunity to improve the energy generation performance from one repetition to the next. The research will be augmented by education and outreach activities, including the creation of wind energy-inspired undergraduate classroom modules, development of a kite design activity at the Charlotte Engineering Early College High School, and summer opportunities at a regional airborne wind energy company, Windlift, Inc.This project will derive new control theoretic knowledge for a unique economic iterative learning control framework that focuses on maximizing or minimizing a profitability index rather than mere tracking performance. Specifically, the learning framework will blend two mechanisms that will transform the state of the art in point-to-point iterative learning control. First, unlike traditional point-to-point iterative learning control approaches where the waypoints are pre-specified and only the behavior between waypoints can be adapted from one iteration to the next, the framework will enable adaptation of the waypoints themselves from one iteration to the next. Secondly, an inner loop flexible-time iterative learning control module will allow the waypoint arrival times and total iteration time to vary from one iteration to the next. This will enable the research team to tackle time-optimal and energy-optimal problems through an iterative learning framework, which has not been accomplished in a general sense to-date. Considering that the waypoint adaptation law and flexible time iterative learning module comprise two interconnected subsystems, a small gain stability analysis framework will be used to derive bounds on the allowable variation of waypoints from one iteration to the next. Finally, the application of the ILC framework to repetitive crosswind flight of AWE systems will provide an online learning mechanism for the optimization of crosswind flight, which is especially important due to the complex and uncertain dynamic models of these systems (which can often render offline crosswind trajectory optimizations ineffective).