Novel non-destructive machine learning-assisted spectroscopic method for quality control of bleached kraft pulps

Abstract

Lipophilic wood extractives often cause process and quality issues in chemical pulping processes, claiming a significant share of financial losses within the industry. The method currently used for extractive monitoring is not suitable for online process quality control. In this work a quick and non-destructive approach utilizing near infrared (NIR) and Raman spectroscopies combined with machine learning for detection and quantification of extractives in bleached birch and conifer kraft pulps with gas chromatography (GC) as a reference method is proposed. By using either of the proposed spectroscopies, it was possible to classify the dry pulp type with 100 % classification accuracy (CA). NIR combined with logistic regression classified the conifer and birch pulps depending on the batch with 92.0 % and 96.1 % CA, respectively. Spectral data pretreatment followed by machine learning-assisted calibration resulted in successful estimation of extractive group contents including fatty acids or triterpenoids and prenols based on NIR spectra. Models built using Raman spectra could also estimate contents of e.g., total GC extractives. This method is applicable to kraft pulping products quality monitoring and can potentially improve efficiency of the process and fibre industry.