(171c) A Model-Centric Framework for the Smart Manufacturing of Polymers

Plastic and polymeric materials have an enormous impact on a variety of industries and our daily lives. The synthesis of polymers occurs by reacting small molecules referred as monomers together to form, under certain conditions, long molecules referred as polymers. During polymerizations, it is mandatory to assure the uniformity between batches, high-quality of products, energetic efficiency, minimum impact to the environment, reliability of the process, and labor safety. It remains as a major challenge the establishment of operating, monitoring, and control conditions capable to achieve all mentioned goals together.

In this work, various model-centric strategies towards the smart manufacturing of polymers were designed, implemented, and tested using experimental data from two polymerization systems. The main aims included the combination of different model-centric strategies which permit to obtain operational conditions such that production goals are achieved. After validating the correctness and reliability of the fundamental model, different optimal policies are formulated, advance control elements are developed, and nonlinear online estimators are designed for reaching the joint goal of signal processing and full polymer characterization.

The first system studied is the synthesis of polyacrylamide in aqueous solution using potassium persulfate as initiator in semi-batch fashion. In this system, the proposed framework integrates nonlinear modelling, dynamic optimization, advanced control, and nonlinear state estimation towards producing polymers with target properties. The main objectives include the attainment of desired polymer characteristics through feedback control while a complete polymer motoring is possible during the reaction. Results, in terms of the dynamic set point achievement, are satisfactory for the closed-loop system. An improvement of 42% in average is foreseen when using a hybrid discrete-time extended Kalman filter (h-DEKF) as state observer. In addition, other strategies for state estimation are proposed including different structures of the geometric observer (GO) from which the 4-state GO with passivation shows the best performance overall. The 4-state GO with passivation achieves an improvement of 52% in the estimation of the molecular weight distribution when compared with the h-DEKF.

The second system is the copolymerization of ethylene with 1,9-decadiene using dimethylsilyl (N-tert-butylamido) (tetramethylcyclopentadienyl) titanium dichloride (CGC)/MAO as catalyst in semi-batch fashion. For this system, a data-driven strategy for online estimation of important kinetic parameters was assessed. A global sensitivity analysis was performed for selecting the most representative kinetic parameters of the system. The retrospective cost model refinement (RCMR) algorithm is adapted to the problem and implemented for online parameter estimation. After verifying the stability and robustness of the method, real experimental data permitted to validate its performance. Results show that the estimation of the kinetic parameters of interest is feasible, and converges close to theoretical values without requiring previous knowledge of their magnitudes.