Mathematical modelling of gas-phase ethylene/hexene copolymerization with metallocene catalyst

Gas-phase ethylene/hexene copolymerization using silica-supported (n-BuCp) 2ZrCl2 metallocene catalyst has been investigated in a 2-litre laboratory reactor. Replicate experimental runs were conducted to confirm the reproducibility of measured responses, such as polymerization rate, reactant concentrations and copolymer properties. Comparisons of polymerization rate profiles, catalyst activity and product properties were made using a number of designed experimental runs. The experiments revealed that Al(Bu)3 scavenger was the most important cause of low catalyst activity, and a low initial polymerization rate, followed by a rate increase. The effects of other influencing factors, such as residence time, temperature, pressure, concentration of reactants, catalyst and cocatalyst, were also investigated. It has been found that hydrogen concentration and hexene concentration have significant effects on molecular weight and short-chain branching, respectively. In addition, hexene enhances the polymerization rate and catalyst activity while cocatalyst and hydrogen both lead to lower polymerization rate.; Several models were developed to describe gas-phase ethylene homopolymerization and ethylene/hexene copolymerization using a supported metallocene catalyst. Estimability analysis was conducted to determine subsets of estimable parameters and reduce the number of unknown parameters. An iterative estimability analysis procedure was used to estimate unknown parameters to fit experimental data. Although the single-site models provided good fits for the polymerization rate, gas composition and most product properties, they failed to accurately predict the molecular weight data and its distribution. Sequentially, two-site models were built for both homopolymerization and copolymerization, which showed significant improvements over the single-site models. The two-site models were finally validated using the data that were not employed in the parameter estimation process. Most of the model predictions fall within the 95% confidence intervals of the experimental data.