Deconvolution of Microstructural Distributions of Ethylene/1-Butene Copolymer Blends using Artificial Neural Network

Abstract

Polymer blending is a useful approach to tailor-make microstructural distributions (e.g., molecular weight distribution (MWD), chemical composition distribution (CCD)) and product properties. A technique to help identify polymer components and their weight fractions in the unknown blends is desirable for the product development. In this work, artificial neural network (ANN) models were developed to help identify this information from microstructural distributions and validated with simulated datasets of various binary blends of polyolefin with different characteristics. The proposed models are multilayer perceptron network with 2 hidden layers; the backpropagation algorithm is used for the network training. Three types of input data were compared: (1) MWD, (2) CCD, and (3) MWD+CCD. Optimum topologies for each types of input data were also determined.