Construction of Statistical Models for Predicting the Presence of Azeotropy at Any Pressure in Separation Processes

Abstract

Distillation is one of the dominating separation processes, but there are some problems. One of those problems is that inseparable mixtures are formed in some cases. This phenomenon is called azeotropy. It is essential to understand azeotropy in any distillation processes since azeotropes, i.e., inseparable mixtures, cannot be separated by ordinary distillation. In this study, to construct a model which predicts the azeotropic formation at any pressure, a novel approach is presented with support vector machine (SVM). The SVM method is used to classify data in the two classes, that is, azeotropes and nonazeotropes. 13 variables including pressure were used as explanatory variables. From the result of comparing the SVM models which were constructed with data measured at 1 atm and data measured at any pressure, the 1 atm model shows a higher prediction performance to the data measured at 1 atm than any pressure model does. Thus, for improving the performance of the any pressure model, we focused on intermolecular forces of solvents. The SVM models were constructed with only data of the solvents having the same subgroups. The accuracy of the model increased and the model predicted change of the presence of azeotrope according to pressure. It is expected that this proposed method will be used to predict azeotropic formation at any pressure with high accuracy.