A Comparative Study of QSPR Generalized Activity Coefficient Model Parameters for Vapor–Liquid Equilibrium Mixtures

Generalized
activity coefficient models are often essential for
predicting the extent of liquid nonideality in a mixture in the absence
of experimental data. This work is focused on generalizing the interaction
parameters of three widely used activity coefficient models, nonrandom
two-liquid (NRTL), universal quasi-chemical (UNIQUAC), and Wilson.
Specifically, we applied a theory-framed quantitative structure–property
relationship (TF-QSPR) modeling approach for the purpose of generalization.
In this modeling approach, theoretical frameworks, such as the NRTL
model, are used to describe the phase behavior properties, and QSPR
methodology is used to generalize the binary interaction parameters
of the models. In this study, a binary VLE database consisting of
916 systems was compiled and employed to develop the QSPR models.
Interaction parameters of the NRTL, UNIQUAC, and Wilson models were
determined by performing data regression analyses. QSPR models were
developed to predict the interaction parameters found in the regression
analyses. The structural descriptors of the molecules were used as
inputs in the QSPR models. The phase equilibria properties estimated
using the generalized QSPR models resulted in about 2 times the error
as compared to the results found in the data regression analyses.
Overall, the quality of property predictions from the QSPR models
is comparable to those of the UNIFAC-2006 group-contribution model
when all of its group-interaction parameters are available; however,
the UNIFAC model produced worse predictions when such parameters are
lacking. Thus, our methodology offers a viable complement when UNIFAC
parameters are missing.