(716h) Predicting Mosced Parameters from COSMO-SAC Sigma Profiles | AIChE

(716h) Predicting Mosced Parameters from COSMO-SAC Sigma Profiles


Elliott, J. R. - Presenter, The University of Akron
Commonly used activity models such as UNIFAC and MOSCED (Modified Separation of Cohesive Energy Density) rely on empirically characterized parameters to predict phase behavior for petrochemical mixtures. MOSCED is particularly attractive in many situations because it offers intuitive insights into how to tune solvent-solute interactions to achieve optimized formulations. Unfortunately, only 133 compounds have been characterized with the original MOSCED implementation, compared to UNIFAC which has been developed using several hundred compounds. Furthermore, extending UNIFAC to new compounds is straightforward through the group contribution concept, whereas MOSCED simply requires more experimental data specific to the new compounds.

This presentation details a simple method to estimate MOSCED parameters in order to determine infinite dilution activity coefficients based on density functional theory (DFT) calculations provided by the Virginia Tech database of Ï?-profiles for the COSMO-SAC method. The hypothesis of the work is that the surface charge density of a molecule, provided by a Ï?-profile from the DFT computation, can be used to estimate the α, β, and Ï? parameters used in the MOSCED model. By defining a charge density threshold for regions of hydrogen bonding, the probability of the charge exceeding that threshold and size of the surface can be correlated to determine the parameters for the molecules. By assuming the remaining surface charge potential that does not contribute to hydrogen bonding represents the polarity of the molecule, the polarizability parameter Ï? can also be determined.

The current number of characterized compounds for the MOSCED model is 133, of which only 89 are considered to have non zero acidity and basicity parameters. This number limits the possible use of the method due to the low number of molecules exhibiting interactions other than dispersion forces. The proposed method allows for the expansion of the model to 1432 different compounds along with any molecules that are characterized in the future by a Ï?-profile. The correlated functions were regressed based on 4375 binary solution infinite dilution coefficients provided by Lazzaroni et al. and the deviation from the experimental data was calculated based on the logarithmic ratio of the calculated versus experimental (4). The resulting average deviation for the MOSCED model with correlated parameters was found to be 0.28 while using the original parameters tuned to the experimental data had a deviation of 0.106. The UNIFAC model was also compared for binary solutions in which the functional groups were defined and the resulting deviation was found to be 0.183. The calculated infinite dilution coefficients were then used to interpolate the entire phase behavior of a binary system across the composition range. The "simplified MOSCED" (SSCED), Wilson, and NRTL models were chosen to test the accuracy of the method based on the low number of parameters needed to define the interaction energies for the system. The resulting phase equilibrium predictions were compared to experimentally determined results from the Danner and Gess database (10). The average percent deviations of the pressure for the 39 binary systems tested were 17.39% for Wilson, 18.90% for NRTL, and 13.83% for SSCED. In conducting this work, it was noted that characterizations for amines were omitted from the previous MOSCED publications. A database of amine VLE was compiled to characterize both the compound specific parameters and the Ï?-profile predictions. Deviations were similar to other mixtures as noted above.