Model Re-engineering And On-line Prediction For Polymerization Process Based On Microstructural Quality Indices

As a category of capital,resource and technology intensive industries,process industries have large market and high industrial relevance,and continue to be the bedrock on which China economy is built.Product quality is important to process industries.Typically,how the monomer molecules are connected in the polymer chains determines the quality,variety and premiums of products in polyolefin industries.Therefore,the ability to predict and monitoring polyolefin microstructures is significant for mastering polyolefin manufacturing technology.Microstructural modeling is key to the prediction and monitoring for polyolefin processes.However,challenges lie in the large-scale nature and complexity in the microstructural-based process models.Furthermore,solving such models in real-time is problematic.The thesis proposed a series of methods including modeling and reduction,optimization and reformulation,computation and algorithms.The thesis discusses model re-engineering and on-line prediction for polymerization processes based on microstructural quality indices.From the prospective of modeling and computational methodologies,the thesis focused on the model reduction for optimization,and on-line prediction algorithms.The detailed work can be summarized as follows:1.Model re-engineering for thermodynamics models for polymerization process optimization.PC-SAFT EOS is predictive and popular in modelling thermodynamics of polymer systems.The original model employs a complicated function with iterative nature combined with derivatives of state functions.This feature leads to computational issues in traditional "service-roll" framework for process simulation and optimization.To address this problem,three methods are proposed based on model reduction or reformulation:a thermodynamic-free surrogate model to reduce the PC-SAFT EOS,a cubic equation of state to surrogate and reduce the PC-SAFT EOS,and an equation-oriented approach to handling PC-SAFT EOS in simulation and optimization2.Dynamic reduced order models for molecular weight distribution.Conventional models exhibit the nature of large-scale dynamic system.The numerical solution is computationally expensive and is challenging when for optimization problems.In this chapter,we present a kind of model reformulation based on null-space proj ection method.The large-scale model is reduced to a manageable one without essentially losing accuracy,which provides the prospection of dynamic optimization,state estimation and model predictive control based on molecular weight distribution.3.Fast and reliable computational strategies for on-line application based on rigorous models.On-line monitoring of molecular weight distribution is significant for polymerization processes.A rigorous model-driven soft sensor is developed.Solving large-scale models in real-time is problematic in such on-line application.A moving finite element method is proposed to improve the computational efficiency.The sensitivity information of the nonlinear equation systems contributes to a convergence enhancement strategy.The proposed computational framework significantly improves the computational efficiency and robustness.4.Moving horizon estimation for molecular weight distribution.A common problem encounters in the model-based estimators due to the drifting and distortion of the model parameters.Thus,to improve the prediction performance,parameters need to be estimated simultaneously with the states by introducing measurement.In this chapter,a moving horizon estimator is designed based on molecular weight distribution.Numerical results show that by introducing molecular weight distribution as state variables,the prediction accuracy is significantly improved.It is advantageous to introduce the slow measurement of molecular weight distribution for polymer quality monitoring/control although the different measurement rates need to be coordinated.