Near-infrared spectroscopy for rapid compositional analysis of cellulose pulps after fractionation with ionic liquids

The composition of cellulose-enriched solids is typically monitored using a laborious and expensive wet-lab analytical method. Here, the development and application of an alternative tool that uses NIR spectroscopy and a software sensor is reported, drawing on a large data set (149 training samples) consisting of untreated grass, hardwood, and softwood biomass and cellulose pulps obtained after fractionation with the low-cost ionic liquids triethylammonium hydrogen sulfate ([TEA][HSO4]) or N,N-dimethylbutylammonium hydrogen sulfate ([DMBA][HSO4]) mixed with water. A partial least squares (PLS) model was trained on compositions determined with the traditional wet-lab procedure, followed by the application of an uncertainty quantification framework to estimate confidence in the predictions. Good agreement with the wet-lab experimental data (mean absolute errors on unseen samples below 5%) was found for ionic liquid fractionated cellulose and purified cellulose samples generated with non-ionoSolv approaches. Cellulose with low crystallinity and isolated lignins generated poor fits, suggesting that more specialised models are needed. The sugar-derived pseudo-lignin (humin) content in the cellulose pulp was estimated by comparing the model with a second PLS model that excluded charred (over-treated) pulps. The study shows that NIR soft-sensors can cost- and time-effectively estimate the composition of ionoSolv-based pulps, speeding up process and product development and facilitating process operation.