Enzyme-assisted high-consistency fiber refining: enhancing cellulose materials performance in the paper industry through process and physics-informed machine learning modeling

Abstract Despite growing interest in enzymatic fiber modification, the impact of high-consistency enzymatic refining on the mechanical performance of paper remains unexplored. Unlike conventional low-consistency systems, high-consistency enzymatic refining offers a more energy-efficient and industrially scalable pathway for surface modification of cellulose fibers. This study investigates, under industrially relevant conditions, how high-consistency enzymatic refining of bleached kraft eucalyptus pulp with endoglucanases can support the rational design of fiber-based bioproducts. Pulp consistency (3–15 wt% ), enzyme dosage (0–300 mg/kg), and treatment time (15–60 min) were systematically varied. High-consistency enzymatic refining significantly enhanced mechanical performance: breaking length increased up to 89% and internal bonding up to 387%, without substantial freeness reduction. To enable predictive design and process optimization, machine-learning models were developed first based on process variables (consistency, time, dosage). LightGBM model achieved the best results with high predictive accuracy for property prediction (R 2 up to 0.955). To overcome data scarcity, a physics-informed generative augmentation strategy was implemented that incorporates freeness to generate 30 synthetic datapoints. The augmented dataset enhances predictive performance, validating the quality of the synthetic data. A physics-informed Gaussian process regression model was used to extrapolate performance at 400 mg/kg enzyme dosage, and a targeted experiment confirmed its prediction. The results suggest that high-consistency enzymatic refining enhances interfiber bonding through the generation of nanoscale fibrillar elements at the surface, influencing the final bonding of the sheet. Altogether, high-consistency enzymatic refining coupled with predictive modeling shows as a viable pathway towards producing all-cellulose materials with improved mechanical properties, while reducing energy consumption. Graphical Abstract