Identification of Ill-conditioned plants — A benchmark problem

Abstract

This note provides a simple process example from chemical engineering which is proposed as a challenge problem for multivariable identification. The process considered is a simple heat-exchanger with two inputs and two outputs. It is strongly interactive and also ill-conditioned. A single slow pole, resulting from the interactions, is dominating all the individual open-loop responses. Attempting to identify a model based on fitting the individual transfer-matrix elements will usually result in a multivariable model which incorrectly has this dominant pole repeated. Such a model, although a reasonable model for the open-loop dynamics, yields a poor prediction of the process behavior under feedback control, in particular when considering partial control.

The note includes a description of the process, a file for generating open-loop “experimental” data and an example demonstrating that classical identification employing an ARMAX-type of model yields a model which is poor for feedback control studies of the process.

Financial support from the Royal Norwegian Council for Scientific and Industrial Research (NTNF) is greatfully acknowledged.