Assessing lignin content in Nordic hardwood and softwood species using models based on near-infrared (NIR) spectral data and partial least squares regression (PLSR)

Continuous kraft cooking digesters face challenges affecting product quality, making it valuable to improve control through advanced techniques like near-infrared (NIR) spectroscopy, model predictive control, and machine learning models. The primary goal of this study was to use NIR spectra to predict the amount of lignin in hardwood and softwood samples. This study investigated the correlation of NIR derivative spectra with the amounts of lignin relative to other constituents, namely cellulose, hemicellulose, and water, in wood chip samples of varying chip sizes and shapes from six Nordic wood species. It employed partial least squares regression (PLSR) on the NIR data to construct a model that predicted the lignin fraction and the relative fraction of acid-soluble lignin. When trained on a group of five wood species, the model achieved a satisfactory predictive ability, striking a balance between a wide range of lignin content and a consistent chemical environment. The accuracy increased further when the model was restricted only to spruce and pine, reflecting the benefits of a more homogenous dataset. Additionally, the optimal number of latent variables was identified as two, indicating that three distinct chemical components — cellulose, lignin and water — can be effectively differentiated using NIR.