Abstract

The solvent dehydration tower in Purified Terephthalic Acid (PTA) production is characterized by strong multivariable interactions, slow vapor–liquid dynamics, and high sensitivity to upstream disturbances, often limiting the effectiveness of conventional proportional–integral (PI) control. Despite increasing interest in model predictive control (MPC) for separation systems, its application to industrial-scale PTA dehydration under realistic disturbance scenarios and operational constraints remains limited. This study develops a 5 × 5 multivariable model predictive control (MMPC) strategy for an industrial PTA dehydration tower based on a validated nonlinear first-principles UniSim® Design R500 model and a complete 25-element first-order plus dead time (FOPDT) prediction model identified from systematic dynamic tests. The proposed MMPC was evaluated against the existing industrial PI controller under four representative industrial disturbance scenarios, including feed temperature, feed flow rate, and feed composition variations in two inlet streams. The results show that the proposed MMPC reduced the Integral Absolute Error (IAE) and Integral Squared Error (ISE) by approximately 87–100%, depending on the disturbance scenario and controlled variable. The greatest improvement was obtained under feed composition disturbances, where the MMPC achieved IAE and ISE values of 117.9 and 21.5 for Stream 1, and 8.1 and 0.1 for Stream 2, respectively. The only exception was the inlet temperature disturbance, for which the existing industrial PI controller remained slightly superior because of the predominantly local thermal dynamics and relatively weak process interactions. These results demonstrate that MMPC is particularly effective for strongly coupled multivariable disturbances and provide a practical framework for implementing advanced control in industrial PTA dehydration systems using validated process models.

IPC Classification

Keywords