Machine learning-assisted sedimentation analysis of cellulose nanofibers to predict the specific surface area

Abstract

This study introduced a novel machine learning (ML) approach for predicting the specific surface area (SSA) of cellulose nanofibers (CNFs) at various fibrillation stages by leveraging sedimentation profiles from their aqueous slurries. Both sedimentation speed and sedimentation heatmap images, derived from the sedimentation profile data, formed the basis of the ML-assisted prediction model, achieving a coefficient of determination (R²) of up to 0.94 for SSA prediction. The high R² values can be obtained through the appropriate ML algorithms used for the prediction model, including extreme gradient-boosting (XGBoost) regression and convolutional neural networks (CNN) for sedimentation speed and sedimentation heatmap images, respectively, which are effective to deal with these sedimentation data, enabling accurate predictions. Furthermore, the predicted SSA values were used for the construction of the prediction model for impact strength of polypropylene/ wood-derived CNF composite materials by integrating with the infrared spectrum data of the CNFs, achieving the improved R² of 0.88, as compared to the conventional models based on experimentally obtained SSA with R² = 0.79. This sedimentation analysis method therefore enables the acquisition of information related to the morphology of CNFs, which can be widely applied in the quality control of CNFs as well as in the material applications.