Modelling, assessement, optimization, virtual monitoring and model based predictive control of the polyvinyl acetate production in a real life industrial reactor

Abstract

This thesis is an applied research thesis. It combines the purely academical research with its application to a real life industrial process. This has been possible thanks to the support of the company Mitol d.d. Mitol is an adhesives production factory located in Se�ana, Slovenia. As a leading company in its field, its initial objective was to increase its production capacity. The capacity of production of certain products in Mitol was periodically saturated, due to seasonal peaks in the demand. Mitol took the challenge of investing in a research and development project to achieve it. As a pilot experience, the polyvinyl acetate production process was chosen by the factory as the target of this research. The main objective of the thesis has been the development of new mathematical models and innovative control algorithms which allow the optimization of polymerization plant. A multi disciplinary top�down approach to the process has been followed. The research started by assessing the existing resources, in order to identify and remove potential bottlenecks not only in the production process under study, but in the plant as a complex and interrelated system. In a second step an optimization of the policies of addition of the raw materials based on a detailed mathematical model of the plant was targeted. It required the development of a first principles model of the reaction on which the innovative control strategies could be tested without any risk for the plant or its workers. The model was used to simulate various policies of addition of the raw materials, which proved a sensible way to reduce the batch time and thus increase the productivity. A second objective was the use of mathematical models to develop intelligent control algorithms and procedures. This second objective was approached in two steps. In a first step a calorimetric monitor was developed. It allowed the estimation of internal variables of the reactor which could not be measured. In a second step a neurofuzzy network was used for the prediction of the temperature in the reactor after a few minutes based on the variables estimated by the calorimetric monitor. A control strategy combining both the internal status of the reaction and the future temperature was proved to increase of the quality, repeatibility and productivity of the plant.