1 reply to this topic

[KeJoSa2010]

Hello,

I want to fit the NRTL activity model to match some plant data that I have been given (i.e. I want to do a regression). The data that I have is from 3 different plants whose operating conditions are at different temperatures and pressures. Oh, and there are 31 components in the system (which means there are lots of different binary interactions that could occur ;-) ). This is for a 3-phase system (ie VLL) at the cold separator of a refrigeration plant using Ethylene Glycol as a hydrate inhibitor. So hydrocarbon vapor off the top, HC Liquid phase and a EG-Water phase. I will be using the Peng-Robinson EOS to model the vapor phase. I have done a regression for a 2-component system during my undergrad studies however, this seems quite different as I have no VLLE data and the 3 plants are at different conditons.


Thanks very much in advance!

P.S. I am sure there will be questions, and if I am not making, sense I apologize. I am new at this....

Edited by KeJoSa2010, 24 September 2012 - 09:49 AM.

[MrShorty]

What you are trying to do is not a trivial exercise. Probably not one that can be easily replied to in a short post on an internet forum.

The biggest problem I see is that, with 31 components, you have ~500 interaction parameters. If you need to include temperature dependence of those parameters and use 2 parameters, then you've doubled the number of parameters needed. IMO, the first step in a problem like this is to reduce the number of adjustable parameters to a manageable number. Look for binaries that already have data and/or interaction parameters in the literature. For example, ethylene glycol/water probably has enough data in the literature for you to find NRTL parameters for that system. Other binary pairs in your stream may be present in such small quantities that interaction parameters cannot be meaningfully regressed and make little to no difference in the result. Perhaps other binary pairs can be assumed to be ideal (activity coefficient=1). Hopefully you can reduce the problem down to a few binary pairs that you need to regress parameters for.