The development of a hybrid activity coefficient model utilizing the solution of groups concept.
During the course of this thesis the UNIFAC method (group-based method) was regressed to individual Px(T) binary datasets, and the results are compared to the regression results using the Wilson, NRTL, and UNIQUAC equations (component-based models). It is shown that these component-based methods best represent the experimental data when the comparisons are restricted to those systems defined by only two UNIFAC maingroups. For those systems requiring three or more maingroups, however, the regressions using the UNIFAC method (i.e. the group-based approach) are shown to provide the best reproducible results. Evaluations are also presented on the ability of the UNIFAC and mod. UNIFAC (Do.) methods to reproduce experimental activity coefficients at infinite dilution for single and co-solvent systems. For the case of single solvent-systems the newly developed MRR combinatorial expression (Moller, 2010) is evaluated as a direct combinatorial replacement for both methods, although it was originally developed only for estimating activity coefficients at infinite dilution in alkane-solvents. Overall, it is shown that the best results are obtained using the mod. UNIFAC (Do.) method, and that poor results are obtained when trying to use the MRR combinatorial as a direct combinatorial replacement in either method (for systems other than alkane-solvents). Given the favourable results obtained using the mod. UNIFAC (Do.) method, the model was used to generate pseudo data points at multiple temperatures for regression using the NRTL equation, where parameters quadratic in temperature were fitted. It is shown that one may introduce unnecessary errors when translating these predictions into the model parameters of the NRTL equation. In order to eliminate these potential “losses in translation,” a new liquid activity coefficient model/methodology is being proposed. Instead of using group contribution methods as second-choice data generators, it is proposed that these predictive methods be employed in a more direct fashion in process simulations. Instead of regressing experimental data using component-based methods such as NRTL and Wilson, the error in the predicted results are regressed by layering one of these methods on top of a group contribution method like mod. UNIFAC (Do.). This is the fundamental idea behind the proposed hybrid methodology/models. Results are presented for two hybrid models, where the NRTL and Wilson equations are used to correct for the predictions made using the mod. UNIFAC (Do.) method. These methods are being called NRTL-FAC(Do.) and Wilson-FAC(Do.) respectively. In most cases, it is shown that the overall regression results using these new models are as good as or better than the individual models making them up. All experimental data used in this dissertation was obtained from the Dortmund Data Bank (DDBST Software and Separation Technology GmbH, 2009), and all predictions made using the UNIFAC and mod. UNIFAC (Do.) methods were calculated using the Consortium parameters (The UNIFAC Consortium, 2008).