Abstract
It is important to know whether or not mixtures form an azeotrope for the design of distillation processes since azeotropes cannot be separated by ordinary distillation. Thus, there have been many methods to predict the presence of azeotropy, e.g. equation of state, quantitative structure-property relationships, non-random two-liquid model, UNIFAC, UNIQUAC, thermodynamic equations and so on. In this study, to construct a model predicting the azeotropic formation and improve the prediction performance, we propose a novel method based on σ-profiles. The σ-profiles are major information of a COSMO-RS theory and the distribution of molecular surface areas which have charge density σ. Since the σ-profiles are characteristic of not only an objective molecule but also a relationship between the molecule and a solvent, they will be able to be used as descriptors to express interaction between solvents. The model predicting the presence of azeotropic formation is constructed with support vector machine (SVM).Through the analysis of data from the Dortmund Data Bank, it was confirmed that the proposed model has higher prediction performance compared to previous ones where structure descriptors are used as explanatory variables. Moreover, the performance of the SVM model was improved by adding structure descriptors to σ-profile descriptors.
