Generalized Nonrandom Two-Liquid (NRTL) Interaction Model Parameters for Predicting Liquid–Liquid Equilibrium Behavior

The
nonrandom two-liquid (NRTL) model is an activity coefficient
model used widely in phase equilibria calculations. The NRTL model
has three adjustable parameters that are determined through regression
of experimental data for a specific system. A generalization for the
model parameters would reduce the time, money and effort expended
on the collection of experimental data. This work focuses on the application
of a theory-framed quantitative structure–property relationship
(QSPR) modeling approach for the estimation of NRTL parameters. A
database of 342 low-temperature binary (10–40 °C) liquid–liquid
equilibria (LLE) systems was employed in this work. Data regression
analyses were performed to determine the NRTL model parameters. Structural
descriptors of the molecules were generated and used in developing
a QSPR model to estimate the regressed NRTL parameters. The newly
developed QSPR model uses 30 significant descriptors as inputs. The
model yielded binary predictions with 8, 38, 51 and 44% absolute average
deviation for the mole fractions (x1 in
1-rich phase and x1 in 2-rich phase) and
partition coefficients (K1 and K2), respectively. These errors are approximately
3 to 4 times the errors found from the regression analyses. Further,
we observed an 11% prediction failure rate, in which cases the model
fails to converge to an equilibrium solution. The application of the
popular and often employed UNIFAC-1981-LLE model resulted in 3 to
7 times the errors obtained through regression analyses and a prediction
failure rate of 36%. These results demonstrate the efficacy of our
QSPR model in providing improved predictions and an increased range
of applicability when compared to the UNIFAC-1981-LLE model for LLE
systems.