Enhanced cellulose nanofiber mechanical stability through ionic crosslinking and interpretation of adsorption data using machine learning

Herein we report the fabrication of cationic functionalized cellulose nanofibers (c-CNF) having 0.13 mmol.g(-1) ammonium content and its ionic crosslinking via the pad-batch process. The overall chemical modifications were justified through infrared spectroscopy. It is revealed that the tensile strength of ionic crosslinked c-CNF (z(c)-CNF) improved from 3.8 MPa to 5.4 MPa over c-CNF. The adsorption capacity of z(c)-CNF was found to be 158 mg.g(-1) followed by the Thomas model. Further, the experimental data were used to train and test a series of machine learning (ML) models. A total of 23 various classical ML models (as a benchmark) were compared simultaneously using Pycaret which helped reduce the programming complexity. However, shallow, and deep neural networks are used that outperformed the classic machine learning models. The best classical-tuned ML model using Random Forests regression had an accuracy of 92.6 %. The deep neural network made effective by early stopping and dropout regularization techniques, with 20 x 6 (Neurons x Layers) configuration, showed an appreciable prediction accuracy of 96 %.

Automated summary

In this study, cationic functionalized cellulose nanofibers (c-CNF) with 0.13 mmol.g(-1) ammonium content were successfully fabricated and crosslinked using the pad-batch process. The chemical modifications were confirmed through infrared spectroscopy. The tensile strength of the resulting crosslinked c-CNF (z(c)-CNF) was improved compared to c-CNF, and the adsorption capacity of z(c)-CNF was evaluated using the Thomas model.