Tidsskrift: Journal of Near Infrared Spectroscopy, vol. 13, p. 265–276, 2005
Open Access: none
Most industries face a growing challenge concerning data handling due to the large data storage capacity available today. In many cases it is difficult to navigate through these amounts of data in search for relevant information, and it is therefore important to find good methods to help us do just that. An important tool in this context is statistical process control (SPC), which enables us to discover possible process drift or other problems as early as possible.
In this work the potential of using near infrared spectroscopy as a multifunction tool for SPC in the context of pulp monitoring has been investigated Both principal component analyses (PCA) and partial least square regression (PLS) as tools for extracting useful information from the NIR spectra have been tested. The two methods have been compared based on interpretation of score plots and explained variance. We have also tested classification tools for prediction classes and various types of validation since our data came from designed experiments. It has been demonstrated that PLS is a useful tool both for forward and backward predictions. Another topic considered is discovery of instrument drift and outlier detection. It has been demonstrated that PLS is a useful tool in both contexts. The robustness of PLS predictions has been investigated. It was found that PLS score plots can reveal useful information early in the process.
This study is a feasibility study, and the models can hence not be used directly in any large scale installations. This work has however demonstrated the usefulness of multivariate techniques in such processes, and found a good basis for further model development.