Multivariable Inferential Feed-Forward Control

Two multivariable inferential feed-forward control strategies are proposed in this paper. In the first strategy, the effects of disturbances on the primary process variables are inferred from uncontrolled secondary process variables that are measured on-line. In the second approach, the effects of disturbances on the primary process variables are inferred from the manipulated variables for those controlled secondary process variables that have fast dynamics. The proposed strategies are particularly useful in situations where some disturbances cannot be easily and quickly measured. Robustness analysis of the inferential feed-forward controllers and the selection of appropriate secondary measurements are discussed. Structured singular value analysis is used in assessing the robustness of the inferential feed-forward control systems. The performance characteristics of the two inferential feed-forward control systems are demonstrated by application to a simulated methanol-water separation column. In the first system, the effects of disturbances in feed composition (and feed rate) are inferred from tray temperatures, whereas in the second system, the disturbance effects are inferred from inventory manipulations. Nonlinear dynamic simulation results demonstrate the superior performance of these strategies. Robustness analysis shows that using multiple tray temperatures can improve the robustness of the inferential feed-forward controller, and this conclusion is confirmed by simulation.