Features of Linear Models that May Compromise Model-Based, Plant-Wide Control Techniques. The Case of the Tennessee Eastman Plant

TL;DR

This paper investigates how certain features of linear models affect the reliability of plant-wide control systems, using the Tennessee-Eastman benchmark to analyze eigenvalues, errors, and condition numbers.

Contribution

It identifies specific model features that may compromise control system reliability and provides guidance for designing more effective linear model-based plant-wide controls.

Findings

Eigenvalue analysis reveals stability issues.

Average simulation errors indicate model inaccuracies.

Condition numbers highlight numerical sensitivities.

Abstract

This work examines a set of features that impact the reliability of linear models within the context of plant-wide control design (PWC). The study case is the Tennessee-Eastman (TE) plant. This benchmark problem is well-known for challenging many control design approaches. Analyses involve eigenvalues, average errors between simulations, condition numbers, and loss of rank across frequencies. These studies offer guidance for designing an effective plant-wide control system based on linear models.