Adaptive supervisory control of a fixed-bed autothermal reactor

Conventional adaptive control has been shown to be successful when changes in operating condition and process nonlinearities would result in poor performance of standard non-adaptive controllers. Adaptive controllers modify the control law as a function of on-line estimated model parameters and user-specified design parameters. After a particularly severe model change, the design parameters may not be applicable at the new conditions and, if they remain unchanged, may result in performance degradation even if the controller adapts as well as it can.;To maintain the desired control system performance, supervisory control algorithms have been developed that make use of the estimated process models to select 'optimal' design parameters on-line. This study has focused on a class of multivariable pole-placement self-tuning controllers. Algorithms have been developed that deal with selection of the closed-loop pole locations and the controller structure as a function of desired performance and the robustness of the resulting control system. These algorithms have been applied to a high-order nonlinear model of a fixed-bed chemical reactor. Extensive use has been made of known process information to improve the quality of the control system design and to reduce the computational effort required to optimize the design parameters. In the case of on-line closed-loop pole selection, the supervisory algorithm applied to the reactor system has shown to greatly enhance the dynamic closed-loop performance over the entire operating region of interest. The proposed algorithm for selecting the controller structure includes the effects of modeling uncertainty via a structured singular value robustness analysis. Because the modeling uncertainty is considered, this algorithm indicates that no supervisory control is necessary for the reactor system contrary to the results obtained with existing controller selection methods.