Modeling of Drying Kinetics and Color Change of Bacterial Cellulose: a Comparative Study of Theoretical, Semi-Empirical, and Machine Learning (ANN, SVR, GPR) Approaches

Despite its superior biological and mechanical properties, bacterial cellulose (BC) faces processing challenges in industrial applications due to its high-water content. This study represents the first investigation in literature applying the Refractance Window (RW) technique (at 45, 55, and 65 °C) for BC dehydration, analyzing drying kinetics through theoretical, semi-empirical, and machine learning (ANN, SVR, GPR) models. Experimental results indicated that increasing the temperature significantly reduced the drying time from 144 to 72 min. Regarding the diffusion mechanism, the Dincer and Dost approach provided the best statistical fit, revealing that the process is controlled by both internal and external resistances. While the Midilli model excelled among thin-layer models, Gaussian Process Regression (GPR) outperformed ANN and SVR algorithms in machine learning comparisons, demonstrating superior stability and lower RMSE and χ2 particularly suitable for limited datasets. Furthermore, the thermal properties of BC (specific heat, thermal conductivity, and diffusivity) were evaluated as a function of moisture content; the sharp decline in thermal conductivity during drying was identified as the primary mechanism driving the falling rate period behavior. Color analysis showed that higher temperatures (65 °C) triggered non-enzymatic browning reactions. These findings demonstrate that RW drying is an efficient technique for BC, and GPR serves as a robust tool for predicting drying kinetics, paving the way for the development of energy-efficient, large-scale industrial processing of high-performance biopolymer.