Analysis And Optimization Of A Polyurethane Reaction Injection Molding (Rim) Process Using Multivariate Projection Methods 2003.pdf

Chemometrics and Intelligent Laboratory Systems 65 (2003) 17–33
ate/chemometrics
Analysis and optimization of a polyurethane reaction injection
molding (RIM) process using multivariate projection methods
F. Yacoub, . MacGregor*
Chemical Engineering Department, McMaster University, 1280 Main Street West, JHE-374, Hamilton, ON, Canada L8S 4L7
Received 13 April 2002; received in revised form 20 August 2002; accepted 25 August 2002
Abstract
ponent analysis (PCA) and projection to latent structure (PLS) methods are used with industrial data to
essfully diagnose several different problems arising in the manufacturing of rigid polyurethane foam insulation panels. The
PCA and PLS models are used to reveal the spatial variation of quality variables throughout the foamed product, and their
relations with the process variables. Designed experiments are performed in the key process variables identified from the PCA
studies and the results are used to optimize the process.
D 2002 Elsevier Science . All rights reserved.
Keywords: Polyurethane; Reaction injection molding; Projection method
1. Introduction (PLS) provide a way to handle the highly correlated
data collected by these systems. In addition, they deal
In the last two decades, chemical processes, like effectively with multiple response variables and with
many other industries, have been going through a missing data, and they provide a good tool to extract
revolution in their data collection systems. Machine and highlight the systematic variation in these multi-
intelligence, immense data storage capacity, and high variate data sets.
throughput data acquisition systems have driven the The most important property of projection methods
cost per data point down to a very low level. Masses of is the capability to reduce the multivariate dimension of
data are now available by measuring process variables a problem into a low-dimensional space, usually con-
as well as quality variables eith