A tutorial on inferential control and its applications

In this tutorial, we discuss approaches to control of one or more primary variables in a process using secondary measurements. This approach is useful when the primary variable is not easily measured, or has large time delays or lags associated with it. It is also useful when the secondary measurements contain information about disturbances that affect the primary variable. We start by discussing some classical approaches to this problem. Then we present inferential control strategies that use process models to predict the effect of the disturbance variable on the primary output and use this prediction to regulate the process. Next we present a framework for incorporating the inferential control strategy within the framework of the often-used model predictive control (MPC). This framework, termed as model predictive inferential control (MPIC), is general enough to accommodate multiple secondary measurements as well as nonlinear estimators and controllers. The concept is also extended to end product quality control in batch processes using intermediate measurements available during the middle of the batch. The advantages of inferential control are established using the Shell challenge case study problem, which employs linear transfer function models. Problems of collinearity among the secondary measurements ( which frequently arises) is addressed using principal component analysis (PCA) during the construction of the dynamic estimator. Numerous applications demonstrate the advantages of the inferential control strategy.