Premium
Strategies for implementation and validation of on‐line models for multivariate monitoring and control of wood chip properties
Author(s) -
Jonsson Pär,
Sjöström Michael,
Wallbäcks Lars,
Antti Henrik
Publication year - 2004
Publication title -
journal of chemometrics
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.47
H-Index - 92
eISSN - 1099-128X
pISSN - 0886-9383
DOI - 10.1002/cem.845
Subject(s) - multivariate statistics , process analytical technology , chemometrics , calibration , partial least squares regression , near infrared spectroscopy , raw material , environmental science , spectrometer , process engineering , computer science , biological system , mathematics , statistics , chemistry , engineering , machine learning , bioprocess , physics , organic chemistry , quantum mechanics , chemical engineering , biology
Here we present an approach for on‐line control and monitoring of pulpwood chip properties based on near infrared (NIR) spectroscopy and multivariate data analysis. In addition, this paper suggests how to deal with large multivariate data sets in order to extract information which can be used as a basis for changes in raw material or process conditions in the drive towards more optimal intermediate or end product properties within the pulp and paper industry. The pulpwood chips used as raw material in a pulp and paper making process were characterized at‐ and on‐line using NIR spectroscopic measurements. Collected NIR spectra were used in multivariate calibration models for prediction of the moisture content as well as the between‐ and within‐species variation in the studied raw material. Statistical experimental design was used to form a calibration data set including most of the variation occurring in a ‘real’ on‐line situation. NIR spectra for all designed samples were measured at‐line and the estimated calibration models were used for carrying out predictions on‐line. Predictions of the moisture content (% dry weight) as well as the percentage contents of pine and sawmill chips in the raw material were carried out using partial least squares projections to latent structures (PLS) methodology. NIR spectra were collected subsequently on‐line once every minute, and, to reduce the problem with noise in the time series predictions, the measured signals were filtered using a moving average of 100 predicted values. This provided smoother predictions more suitable for process monitoring and control. To validate the quality of the predictions, wood chips from the studied process were sampled and analysed in the laboratory before being subjected to predictions in the on‐line model. Comparison of the filtered on‐line predictions with the results obtained from the laboratory measurements indicated that moisture and pine chip contents could be well predicted by the on‐line model, while predictions of sawmill chip content showed less promising results. Copyright © 2004 John Wiley & Sons, Ltd.