Multivariate Methods for Process Data Analysis - A Batch Process Application

Abstract

This article aims to review one particular application of the partial least squares regression (PLS) method using a real-world example from the chemical industry. Moreover, it will highlight some interesting aspects of resampling and data pretreatment which are often encountered with chemical process data. The background of the investigation is a batch process for making dye pigments which BASF operates worldwide at three different sites. The aim of the investigation was to identify critical quality factors using available process data and to generate knowledge from the data to further improve the quality of the product. We will first give a short description of the structure of the process data. In the second step, we will describe appropriate resampling and data pretreatment techniques and subsequently apply a PLS method to analyze the data. In the last step, the results from PLS analysis will be interpreted and used for defining better operating conditions.