Analysis of Vapor–Liquid Equilibrium in Ternary Systems for an Adequate Planning of Their Experimental Determination and Correlation

Experimental vapor–liquid equilibrium (VLE) data, and the activity coefficient model parameters obtained by their correlation, are frequently used in separation processes design. The inspection of this information for ternary systems has revealed some problems that are not widely known and could lead to serious mistakes. A small sampling of ten ternary systems, with VLE data sets at constant pressure, arbitrarily selected has been considered to illustrate these problems. The experimental VLE data and the NRTL model parameters for these systems have been taken from different sources, one database and two commercial simulation packages. A MATLAB code has been developed to prepare 3D graphs where the T/x and T/y surfaces calculated by the activity coefficient model (NRTL) are represented along with the experimental VLE data. The results of the analysis of this small sample are very worrying. In fact, half of the selected systems present severe inconsistences between the experimental VLE data and the T/x,y surfaces calculated using the correlation results published or implemented in the simulation software. These inconsistences are of different natures, but the most notable is the existence of parameter sets that do not correlate at all the VLE behavior of the system that they supposedly represent. These irregularities may occur only in the ternary region, only in the binary subsystems, or, in some cases, in both. These problems go frequently unnoticed for different reasons. As discussed in this work, the typical 2D graphs and deviations in T or P and vapor composition are not enough in some cases so as to have an adequate idea of both the quality of the data and the suitability of the model to represent them. This is particularly true when considering, as is strongly advisible, the complete set of data, i.e. the ternary and binary VLE data simultaneously. Some recommendations are presented in this work to overcome the problems detected both when planning the experimental design to determine VLE data and to correlate them with a model.