A generalized LQG approach to self-tuning control Part I. Aspects of design

Abstract

Implicit self-tuners based on single-stage cost-function minimization schemes require exact knowledge of the dead-time of the controlled plant. Explicit pole-placement self-tuners on the other hand, though robust against variations of the time-delay, are very sensitive to the assumptions made about the model order of the plant; wrong assumptions may result in very poor performance. An explicit method based on minimization of a multistage cost-function and a CARIMA plant model formulated in state-space is presented. The method extends the design features of the generalized minimum variance scheme to the multistage case; these can be used to tailor the closed-loop response of the system. The approach is shown to be robust against wrong a priori assumptions made about the plant dead-time or order, thus maintaining good servo and disturbance-rejection properties (see Part II).