Modeling and Optimization of a Semi-Interpenetrating Polymer Network Process

The integration of modeling, simulation and optimization provides powerful tools for sup-porting advanced decision making in the competitive market. However, when applying the tools to polymerization processing, the challenging task is to accommodate the predictability of the mathematical model and the capability of model-based optimization due to its inherent complexities. In this thesis, novel strategies of modeling and optimization are developed and applied to a complex polymerization system, Semi-Interpenetrating Polymer Network (SIPN) process. By providing a comprehensive study on SIPNs, we show the great potential of advanced modeling and optimization in the polymer industry.

New mathematical modeling strategies are first presented in this work. In the SIPN pro-cess, improving the productivity while preserving the quality of final products are usually contradictory with each other because of a slow seeded polymerization mechanism. The presence and interaction between two polymers imposes complexity in the model. To simu-late the complex interpenetration and networking process, the model needs to consider several key features, such as a particle growth mechanism, intra-particle heterogeneity and polymerization kinetics. Hence, the first highlight of the work is a multi-stage modeling framework which decouples the modeling complexity: a generalized reaction-diffusion model to describe the particle growth and the intra-particle dynamics; and a new multi-population-balance representation to integrate simultaneous crosslinking, grafting and degradation reactions into the kinetic model. The last component enables the prediction of gel content and molecular weight development simultaneously up to full conversion.

Advanced computational tools are developed to tackle the problem of computation. Acquisition of reliable parameters is the key to predict the process performance over a broad range of operation. However, we face the problem of large variation of polymerization pa-rameters in the literature and limitations in obtaining analytical measurements for polymer composites. A hybrid approach is presented to reduce model distortion by selecting the most prominent parameters from a large parameter set. Parameters are ranked through successive orthogonalization of the sensitivity matrix, and then the selection is iteratively refined through statistical inference from simultaneous parameter estimation. Validated models ob-tained are therefore usable for process optimization. New operation policies are explored Abstract Abstract IV through a new profile representation. In addition, Kriging surrogate modeling is introduced and combined with dynamic optimization. An efficient optimization algorithm is developed for the integrated multi-stage model based on surrogate sub-models. Furthermore, we devel-op a robust two-stage algorithm for large-scale multi-scenario dynamic optimization problems by taking advantage of NLP sensitivity analysis. This enables continuous devel-opments in parameter estimation and process optimization applications.

The model derived shows consistent agreement with SIPN pilot plant experiments. New-ly proposed operation policies significantly improve the productivity while maintaining the quality of the final product. The tools developed in this thesis are useful for other complex system analysis and optimization.