(125e) Experience with Using Group Contribution Methods for Phase Equilibria in Industrial Distillation Column Design

It is common to find mixtures with many (10-20) compounds in many, if not most complex chemical separation processes. Activity coefficient models are imperative when modeling the phase equilibria for such systems (vapor-liquid, vapor-liquid-liquid, or liquid-liquid). This effort requires providing binary interaction parameters for each constituent binary in the mixture. For example, with 15 components, one would need to characterize the behavior of 105 binaries.

Ideally, one should find experimentally measured phase equilibrium data for each such binary and regress the activity coefficient parameters for the chosen activity coefficient model. In attempting to do so, there are, unfortunately, several immediate issues to be confronted: (a) such data are scattered throughout the world’s literature (often in foreign language journals), (b) it is generally impractical to attempt to find them in a reasonable amount of time, (c) there is no guarantee that published data are in fact at all reliable – in fact, there are many published datasets that are “manufactured’ or completely fake, and, finally, (d) a random or careless search could well fail to find data that do exist somewhere.

Alternatively, one could access computerized databanks, developed over the years, for storing such data that have been screened by domain experts and made available in an organized way; however, these databanks are not inexpensive. Besides, it may well happen that no such data exist for several binaries in question. In such situations, group contribution methods (GCMs) would be used to estimate phase equilibria for such binaries, provided they are not "key components" in the separation system. If the binaries in question do, in fact, represent such key components, there would be no alternative but to insist on commissioning experimental phase equilibria measurements from a qualified laboratory.

These GCM methods originated many decades ago and have been updated over time. Regrettably, the group interaction parameters for the most reliable GCMs are not in the public domain and require licensing arrangements to access the latest developments.

The major process simulators used in the chemical industry do allow users to provide their own GCM interaction parameters; however, they are not allowed to provide them by default. In addition, while they provide built-in activity coefficient parameters for selected binary mixtures, it is often the case that these are not properly documented as to their sources:

• Were these parameters regressed using measured phase equilibria?
• Were all available data for each binary mixture used, or was some subset used naively without regard to limitations on extrapolation way beyond the range of the data (e.g., for pressure or temperature)?
• Were these parameters estimated using some GCM based on outdated or inaccurate GCM interaction parameters?

This paper will present examples of a few successful industrial projects where the issues outlined above were addressed reliably. The discussion will show how practitioners can avoid serious (but subtle) flaws in the thermodynamic models that lie behind every industrial process simulation. Such errors can lead to disastrous designs that fail to meet throughput or product quality guarantees. There are no easy shortcuts, and it behooves all practitioners to be mindful of, and plan for, the non-trivial effort required to ensure that their designs are successful.