How auxiliary variables and plant data collection affect closed-loop performance of inferential control

Abstract

The design of property estimators for inferential control is addressed in this paper, and the effects of the auxiliary variables (estimator’s inputs) and of the approach to collect plant data, used to compute the model coefficients, are investigated. The concept of steady-state closed-loop consistency, which is the ability of an estimator to guarantee low offset in the unmeasured controlled variables, is adopted and theoretical results about this property are derived. It is shown how the selection of auxiliary variables represents the most crucial design step that determines the final closed-loop performance of an inferential control system. When this selection is done on a steady-state closed-loop consistency basis, the closed-loop performance is satisfactory, and it is secondary how the dataset is built. On the other hand, when “inconsistent” inputs are used, the performance is, in general, poor and may be significantly affected (in positive or in negative) by the dataset characteristics.