Open AccessIdentification and Classification of Technical Lignins by means of Principle Component Analysis and k‐Nearest Neighbor Algorithm
Author(s) -
Fink Friedrich,
Emmerling Franziska,
Falkenhagen Jana
Publication year - 2021
Publication title -
chemistry ‐ methods
Language(s) - English
Resource type - Journals
ISSN - 2628-9725
DOI - 10.1002/cmtd.202100028
Subject(s) - organosolv , principal component analysis , lignin , kraft paper , attenuated total reflection , chemometrics , kraft process , chemistry , softwood , fourier transform infrared spectroscopy , analytical chemistry (journal) , biological system , infrared spectroscopy , mathematics , chromatography , pulp and paper industry , organic chemistry , statistics , chemical engineering , engineering , biology
The characterization of technical lignins is a key step for the efficient use and processing of this material into valuable chemicals and for quality control. In this study 31 lignin samples were prepared from different biomass sources (hardwood, softwood, straw, grass) and different pulping processes (sulfite, Kraft, organosolv). Each lignin was analyzed by attenuated total reflectance Fourier transform infrared (ATR‐FT‐IR) spectroscopy. Statistical analysis of the ATR‐FT‐IR spectra by means of principal component analysis (PCA) showed significant differences between the lignins. Hence, the samples can be separated by PCA according to the original biomass. The differences observed in the ATR‐FT‐IR spectra result primarily from the relative ratios of the p‐hydroxyphenyl, guaiacyl and syringyl units. Only limited influence of the pulping process is reflected by the spectral data. The spectra do not differ between samples processed by Kraft or organosolv processes. Lignosulfonates are clearly distinguishable by ATR‐FT‐IR from the other samples. For the classification a model was created using the k‐nearest neighbor (k‐NN) algorithm. Different data pretreatment steps were compared for k=1 … 20. For validation purposes, a 5‐fold cross‐validation was chosen and the different quality criteria Accuracy (Acc), Error Rate (Err), Sensitivity (TPR) and specificity (TNR) were introduced. The optimized model for k=4 gives values for Acc=98.9 %, Err=1.1 %, TPR=99.2 % and TNR=99.6 %.