Modeling And Controling Molecular Weight Distribution For Polymerization Processes Via Neural Networks

How to build models and actualize control of molecular weight distribution in polymeric reaction by neural networks are the main targets in this paper. Due to the multiplicity and randomicity of polymeric reactions, the polymers are composed by a large amount of monomers with different chain lengths. The molecular weight distribution has an important influence on the quality of polymers. On-line control of molecular weight distribution is an important method to improve the microscopic quality of polymers. The molecular weight distribution as controlled variable is different from other systems. Unlike single argument or multivariable systems, the controlled parameter here is highly associated parameters' combination. This paper firstly uses the hybrid B-spline networks to build the static models and gets analytic expressions, and then derives control strategies. The effectiveness of models and control strategies have been proved through the simulation studies on an experimental rig in which bulk polymerization of styrene takes place. On basis of static results, this paper's main goal is to get dynamic models and realize dynamic control. Firstly, B-spline network is used to express the nonlinear functions between chain length and molecular weight distribution. Secondly, the linear recurrent neural network with dynamic characteristics is utilized to describe the function between the control variable and network distribution output. Linear recurrent neural network's output can be regarded as the B-spline network's weights. Based on these, the functions between model weight distribution and control variable and chain length can be expressed by analytic expressions. Optimization methods are applied to solve the analytic expressions to get the dynamic control strategies. These principles are used to the simulation studies of styrene polymerization to prove the feasibility of this method.This paper's research supplies a new way and a new thinking to solve the questions of molecular weight distribution's modeling and control, and it is of creativity and generalization in this research field. The results of this paper can be used not only to the microscopic quality control of polymers but also to the control of optimal distribution shapes for analogous distributed parameter systems such as density distribution or temperature distribution and so on.