Improving the Accuracy of Activity Coefficient Estimation in Specialty Chemistry Using Local Estimators

The pharmaceutical and specialty industries require the
rapid development
of molecules and their production processes to meet market demands.
Computational tools can accelerate process development by screening
potential alternatives. However, accurate parameter inputs are necessary
for these tools to calculate processes correctly. In this study, we
aim to improve the prediction of activity coefficients in mixtures
involving complex molecules using the UNIQUAC functional-group activity
coefficients (UNIFAC) and UNIQUAC segment activity coefficients (UNISAC)
models. We propose adding a new group contribution model to represent
the complex core structure of the target molecule, enabling the models
to mimic the steric effects induced by the intricate core structure.
By regressing the parameters for the new group contribution model
using data from a similar molecule, we can accurately depict the properties
of the complex structure. We also plan to explore a similar approach
for COnductor-like Screening MOdel Segment Activity Coefficients (COSMO-SAC),
adjusting the global parameters to account for the structural deviations.
Our results show that the accuracy of the models with local estimators
is comparable to vanilla models for vapor–liquid equilibrium
(VLE) data sets involving simple molecules. The adjusted group contribution-based
models outperform the vanilla models for solid–liquid equilibrium
(SLE) data sets based on steroid solutions. These findings suggest
that the local estimator approach enhances group contribution-based
models for complex molecules without sacrificing accuracy. Reliable
extrapolation was not observed for the local estimator based on COSMO-SAC.
The local estimator approach improves group contribution-based physical
property estimation in mixtures of complex molecules during early
process development, enhancing downstream process modeling capabilities
in specialty chemistry.