Inferential product quality control of a multistage batch plant

Abstract

A modeling framework is developed that addresses the connectivity of a particular, but commonplace, batch plant configuration. Traditional inferential model building techniques such as partial least squares (PLS) and subspace identification are modified to allow for the identification of models that capture the correlation structure across the plant, so that when a disturbance or deviation is detected in one reactor, such information can be shared with other reactors to improve overall control performance. The ability to use information from other running reactors, as well as from previous batch runs, translates into more informed control decisions leading to better overall plant performance. The developed modeling strategy will be applied using data generated from simulating a rigorous model of the Nylon 6,6 process. (C) 2004 American Institute of Chemical Engineers