Process Data Analysis Using Multivariate Statistical Methods

Nearly all industrial chemical processes are heavily computerized to collect large volumes of data By studying and analyzing data, improved understanding produces higher quality products and increases profitability. The fallacy in this reasoning is the assumption that the right tools are in place to analyze and make sense out of the data. Only when data contains information is data valuable. This short paper examines two multivariate statistical methods, Principal Component Analysis (PCA) and Partial Least Squares (PLS), to analyze and interpret data from a large chemical process. In an example, PCA and PLS were used to identify the correct correlations between the measurements and the output to reduce the dimensionality of the process data and to build a model to predict the output from the known measurements.