In the process of designing robust control systems, the designer often finds that the robust stability and robust performance conditions are not met for the problem as originally posed. Then, it becomes imperative to determine the next step in the control design study so that an effective search is initiated. This can be unequivocally accomplished by a trade-off analysis. Based on the structured singular value paradigm, we present a method that helps the control designer assess the relative importance of information regarding the feedback system. Such information typically consists of process knowledge through the model, control system complexity quantified by the controller order and performance requirements as set forth by the designer. The trade-off analysis is built into the robustness criterion, and simulation examples are utilized to demonstrate the effectiveness of this approach.
IntroductionIn recent years, significant effort has focused on feedback control problems originating from the mismatch between the actual plant and the model used for controller design. This so-called robustness issue has been recognized for quite some time (Horowitz and Sidi 1972) and in the last decade, Doyle and Stein (1981) constructed the robust stability and performance conditions using the spectral norms. Doyle (1982) extended his approach to cover more structured uncertainty descriptions and defined the structured singular value (/1). Parallel to this, process control literature also witnessed similar trends in handling the ever-present issue of robustness. The reader is referred to a special issue of Computers and Chemical Engineering (1989) as well as to a book by Morari and Zafiriou (1989) for an in-depth account of the area of robust process control.The robust control design process typically proceeds by compiling all the available information regarding the process and the designer's expectations about the overall feedback response. The former consists of the process model along with some estimate of the model uncertainty. The latter includes performance specifications such as speed of response, offset-free tracking etc. To this list, one could also add certain specifications regarding the controller structure. Possible scenarios would include single-loop and multivariable controllers as well as low order controllers like the ones with proportional-integral-derivative (PI D) elements. This would limit the order of the controller during implementation and one should weigh the consequences of approximating high-order controllers with low-order ones within the context of the robust control design.