Optimization of polymer processing in batch reactor using neural network

The needs for effective control performance in the face of highly process interactions have call for better plantwide process control system synthesis method. As a practical illustration, a vinyl acetate monomer plant was considered. The aim was to develop a suitable control model and then its performance was analyzed. This research underwent several stages. First, data was generated from the simulation of vinyl acetate monomer process. This simulation was performed using Matlab 7.1. This was followed by analyses of dynamic response of the process. An inferential estimator was developed using Artificial Neural Network (ANN) and Partial Least Squares. This estimator will provide a reliable prediction on vinyl acetate concentration and as a platform for controller as a reference trajectory. Both estimators give a satisfactory result however Partial Least Squares show superiority than Artificial Neural Network (ANN). Based on the reference trajectory from estimator, Model Predictive Control was proposed on separator and vaporizer operations. First, transfer functions was developed using First Order Plus Time Delay (FOPTD) equation. These transfer function are then used to represent separator and vaporizer model in development of Model Predictive Control (MPC). Lastly, model testing of vinyl acetate monomer process is done using Simulink environment and followed by tuning process. The optimum value of prediction horizon (P) and control horizon (M) is determined from the tuning process. The result lead to the conclusion that the Model Predictive Control is better than PI controller specifically in optimize the desired production of vinyl acetate