Aqueous viscosity of carbohydrates: Experimental data, activity coefficient modeling, and prediction with artificial neural network-molecular descriptors

Abstract Food waste is a rich source of carbohydrates. This material can be reprocessed in order to produce purified compounds or other materials with high added value. The implementation of processes that enable this transformation depends on kinematic viscosity data and robust mathematical modeling. In order to meet part of this demand, the objective of this work was to study the kinematic viscosity of binary and ternary solutions involving sucrose, sorbitol, xylose and xylitol in the temperature range between 303.15 K and 363.15 K at concentrations of 0.5 mol‧kg−1 to 3.0 mol‧kg−1. In the correlation of experimental data, the Eyring equation was associated to the Margules, van Laar, Wilson and NRTL models. In the simulation, an artificial neural network associated with molecular descriptors was developed. The experimental results showed to be dependent on the number of OH groups present in the sugar. The mathematical modeling proved to be efficient in the treatment of the experimental data, with the NRTL model being the one with the best performance. Artificial neural networks were satisfactory in the simulation of the data, with the 7–3–3-1 architecture being the one with the best data prediction capacity.