A Neural Network Predictor for Reengineering of Resin Manufacturing

We describe the design and testing of a neural network for chemical process prediction and its use for process optimization. Our objective is to design an improved, robust, accurate, and adaptive model for Resin Melt Index prediction, leading to a better prediction of addition of chain stopper during polymerization. We synthesize a neural network to predict the melt index and optimize the chain stopper, while providing learning capabilities to account for process mean drifts as well as accommodating future changes in the lab analysis. We show that this approach results in a predictor that is superior to the existing nonlinear regression approach, as measured by a number of performance indices.