Inferential Control of a Distillation Column through the Successive Update of the Soft-sensor and Control Algorithm

Abstract

Inferential control facilitates the direct feedback control of a variable that is di cult to measure in real time and achieves more e cient plant operation. However, when an inferential control system is introduced into the existing plant, the di cult-to-measure variable cannot be estimated accurately because operating conditions during data acquisition differ from those during inferential control operation. Thus, the control performance of the di cult-to-measure variable is poor. The contribution of this research is to propose a method to obtain an inferential control system that has high control performance and robustness against estimation error. In the proposed method, the degree of change in the operating conditions is limited by setting constraints on the inferential control system. Restrictions are relaxed in a step-bystep manner when the model is updated with the newly acquired data under the inferential control operation. The usefulness of the proposed method was evaluated through simulation case studies. In the case studies, control simulations of a vinyl acetate monomer (VAM) plant were performed. The bottom water concentration of the distillation column of the VAM plant was controlled. Four control methods including the proposed method were compared. The results of the case study showed that using the proposed method enhances both control performance and robustness against estimation error.