A quantitative prediction of the viscosity of amine based DESs using Sσ-profile molecular descriptors

In recent years, the preparation of deep eutectic solvents (DESs) using amines as hydrogen bond donors (HBD) has been reported by several research groups. One of the potential use of this type of DESs is in the field of CO ₂ capture, where the viscosity of the solvent before and after the absorption is of paramount importance. Since the number of possible combinations of DESs is huge, a mathematical model for the predicting of the viscosity of DESs at different temperatures is very important. In this work, a new mathematical model for the prediction of amine-based DESs viscosities using the quantitative structure property relationships (QSPR) approach is presented. A combination of multilinear regression (MLR) and artificial neural networks (ANN) methods is used for the development of the model. A data set of 108 experimental measurements of viscosity of five amines-based DESs, taken from the literature, is used for the development and subsequent verification of the model. The more appropriate model is determined by a dedicated statistical analysis, in which the most significant descriptors are preliminary determined. The results show that the proposed models are able to predict the DESs viscosities with very high accuracy, i.e. with a R ² value of 0.9975 in training and 0.9863 for validation using the ANN model and R ² value of 0.9305 for the MLR model. The retrieved model can be considered as a very reliable tool for the prediction of DESs viscosity when experimental data are absent. In turn, this can provide useful guidelines for the synthesis of low-viscosity DESs able to minimize energy requirements associated to their processing (e.g. power required for pumps), thus fostering their industrial-scale implementation.