A Critical Comparison of Linear and Nonlinear Property Estimators in Inferential Control

Abstract In this paper, a comparison of linear and nonlinear estimators with particular emphasis to the closed-loop properties of the resulting inferential control scheme is presented. The concept of closed-loop “consistency” is introduced as an effective criterion for choosing the auxiliary variables. An estimator is consistent if it guarantees low closedloop steady-state offset in the true unmeasured controlled variables. By means of a case study of a high purity distillation column, a number of issues that can arise in inferential control are emphasized, and their implications on the closed-loop stability are discussed. It is shown that the use of some nonlinear estimators, which in general guarantee a superior precision, may be inappropriate because of the presence of zero gains and gain inversions that can lead the closed-loop system to instability. Moreover, in multi-input multi-output (MIMO) systems it is possible that the estimator requires the auxiliary variables to reach values that are not reachable by the actual plant.