Multiobjective Robust Control of Linear Parameter Varying Systems and Applications

点击翻译按钮获取中文摘要

9907051SznaierRobust control of linear time invariant systems has undergone extensive developments. In the past two decades, leading to powerful formalisms such as H-infinity, mu-synthesis/analysis and, more recently, L-1 optimal control, that have been successfully applied to challenging practical problems. In contrast, practical tools for handling linear parameter varying plants have just emerged and are still far from complete. Among others, issues yet to be resolved include non-conservative handling of multiple performance specifications, complexity of both the design method and the resulting controllers, and the development of identification methods compatible with these design techniques. The proposed research is aimed at removing these limitations by incorporating into the LPV framework the capability to deal exactly with a broad class of performance specifications and model uncertainty. Specifically, the objectives of the proposed research are:Development of tools for robust identification and model validation of LPV systems.Development of an analytical framework for synthesizing robust multiobjective LPV systems. This framework will:(a) Handle multiple performance specifications in a non-conservative fashion;(b) Identify the intrinsic limits of performance of the system as well as the limiting factors; and(c) Address the issue of controller complexity.Application of the resulting theory to the synthesis of robust controllers for active vision systems and experimental validation using a Bisight robotic head and dedicated image processing hardware.The proposed research will combine elements from functional analysis and viability theory, successfully used by the PI to solve multiobjective robust identification and control problems for LTI plants. Preliminary results show that this combination has the potential to address, in a unified framework, both the identification and synthesis aspects of the problem.This research effort will result in an expanded robust control framework for LPV systems, capable of addressing realistic problems necessitating neither potentially conservative approxi-mations nor multiple trial and error type iterations. Moreover, in addition to advancing the state of the art in control theory, we expect that by removing some of the limitations of currently available LPV tools, it will foster progress in related areas. An example is computer vision, an area where recent technological advances have rendered a number of practical applications feasible, provided that certain related control issues can be resolved. These applications range from intelligent highway systems to remote surgery and have the potential to broadly impact society. ***