Modeling, Identification and Validation of Uncertain Systems; Theory and Experiments

提出的工作目标是开发一套集成的工具和方法，用于不确定系统的建模，用于反馈控制设计目的。我们的建模框架是基于鲁棒控制理论，这是由一些复杂的控制设计方法的支持。这些模型实际上是集合描述，其中集合是由模型内未知但有界的动态扰动产生的。这些模型使设计师能够捕捉到物理系统将产生的各种不确定行为。尽管这是一个强大而灵活的建模框架，但可用于物理系统建模的工具是有限的。我们在本提案中解决了这一领域，最终目标是开发一套用于控制系统建模的建模技术和软件工具。我们之前对鲁棒控制模型的模型验证工作构成了本文提出的工作的基础。这种方法导致了一些工具，这些工具允许设计师通过实验量化物理系统与鲁棒控制模型集中最接近的成员之间的差异。在理论模型发展领域，我们建议扩展我们目前的模型验证方法来描述非线性和参数变化系统。在实验应用领域，将研究两种系统：金属有机气相沉积（MOCVD）反应器中半导体生长速率和成分的控制；不稳定旋转磁轴承系统的控制。这两种应用都具有很大的工业相关性，同时也为理论工作提供了具有挑战性的测试平台。在算法开发工作中，我们建议改进结构化算法来处理在将模型验证方法应用于数据驱动问题时出现的非常大的问题。强大的实验成分具有显著的更广泛的好处。MOCVD实验系统是一个非传统的控制工作领域，促进了控制理论界与半导体和材料科学家之间的互动和思想交流。这些实验也被整合到UCSB的工程课程中，目前被用作本科生控制课程和研究生水平设计实验室的示范范例。

The objective of the proposed work is to develop an integrated suite of tools and approaches for the modeling of uncertain systems for feedback control design purposes. Our modeling framework is based upon robust control theory, which is supported by several sophisticated control design methods. These models are in effect set descriptions, where the sets are generated by unknown but bounded dynamic perturbations within the model. These models allow the designer to capture the variety of uncertain behaviors that a physical system will generate. Although this is a powerful and flexible modeling framework, the tools available for modeling physical systems are limited. We address thisarea in this proposal, with the ultimate goal being the development of a suite of modeling techniques and software tools for control system modeling.Our prior work on model validation for robust control models forms the basis for the work proposed here. This approach has led to tools which allow the designer to experimentally quantify the difference between a physical system and the closest member of a robust control model set. In the area of theoretical model development we propose extending our current model validation approaches to describe non-linear and parametrically varying systems. In the experimental application area, two systems will be studied: control of semiconductor growth rate and composition in a metalorganic vapor deposition (MOCVD) reactor; control of unstable rotating magnetic bearing systems. Both applications have a great deal of industrial relevance, as well as providing challenging testbeds for the theoretical work. In the algorithmic development work, we propose refining structured algorithms to handle the very large problems that arise when applying model validation methods to data drivenproblems.The strong experimental component has significant wider benefits. The MOCVD experimental system is a nontraditional area for control work and fosters interaction and the exchange of ideas between the control theory community and semiconductor and materials scientists. The experiments are also integrated into engineering course work at UCSB, and are currently used as demonstration examples in undergraduate control courses and graduate level design laboratories.