Professionally I work at BPT and focus on creating value adding App for Process Simulation. You can find more info on those on the BPT website. Below are posts that should help HYSYS, PetroSIM and UNISIM users alike in their day to day challenges to produce accurate yet fast models efficiently.

Release the true potential of your staff Apps for Process Simulation

Saturday 1 March 2008

Don't blame the simulator, at least not right away

You know the feeling: you just built a nice case of something that any simulator worthy of that name should be able to solve and HYSYS just doesn't converge. You type in different numbers flip the specs around, try all you can think of and curse HYSYS.

Well ... take a deep breath, go have a walk or get a coffee.
Now that you have gotten your cool back, give some thought to what real system you are trying to model and if the way you set up your model really reflects this. Also give some thought to the fact that any steady state simulator enforces a mass balance and you have to give it at least one degree of freedom to compensate for any inaccuracies in your data. I saw two examples with distillation columns in the last two days.

Column one had the condenser modelled outside of the column to later move the model to dynamics. The column had the bottom flow rate specified and the distillate rate was specified on the TEE that splits distillate and reflux. There also was a vapour product off the condenser. So, when you set up the problem you would think you have left one degree of freedom open, the vapour flow off the condenser, hence the column should nicely solve. In this particular case the column indeed solved the first time round. But when you started putting an adjust on top of the model, nothing worked anymore. What was going wrong?
First, the recycle operation that was used to send back the reflux only had the default tolerances which are way to big to converge a column properly. You need to decrease these from 10 to 0.1 for sure. Once you do that you will find out that the column doesn't solve anymore at all. This may be that moment of despair, you got the model running, you fixed a problem and it all got worse.

Time to blame HYSYS?
The vapour product off the condenser is really a purge of inert gas and given the fact that the temperature of the condenser is fixed, whatever you throw at this condenser it is going to produce the same vapour flow rate: the inert gas flow rate + the amount of other vapour that corresponds to the vapour pressure in the reflux drum at the specified reflux drum temperature. So you really don't have a degree of freedom left on the mass balance and that is why things don't converge. Set the flow spec on the reflux side instead of on the distillate side and the column nicely converges in a minimum of iterations

The second column was a crude tower.
The column was set up in what seemed like a reasonable way as far as specifications were concerned. I first got distracted from the nastier problem by immediately seeing that the vapour flow up the column was not big enough to ensure all the draw above the feed could meet their flow spec. The usual trick here is to add a trim duty stream on hte stage of the main liquid feed to allow the column to drive more vapour up the column. So that didn't work ...
Close inspection of the actual setup of the side strippers and pumparounds revealed that the draw of the side strippers was happening 1 stage above the return of the corresponding pump around. This meant that the lowest pump around produced liquid that had nowhere to go but right to the bottom of the column. But there was a specification in place to limit the amount of overflash and that one was constantly fighting with the liquid produced by the pump around. Lowering the draw of the side stripper solved this problem and I am pretty sure that in real life this side stripper is drawing it's liquid from the stage where the pumparound returns, or even lower.

So if the model build went smooth and quick and the model doesn't converge. Take a step back and think beyond the immediate problem think about mass and energy balance and verify if your model accurately reflects the real life situation.

2 comments:

Kiran said...
This comment has been removed by the author.
Kiran said...

Very true...and you are very correct ....we need to do homework ...with pen and paper first before giving any spec ...particularly keeping in view the material balence aspect.

I am simulating one column , having five draws. I have put one leakage air at the top and the corressponding flow from the top to vaccumm pump. Which allow me to specify all the five flows. My top and bottom pressure is abt -3.5 psig to +3.5 psig.
I have removed a condenser from top and instead put one pump around and a side stream as light drag from top.

Would lie to know from yr experience , who to esimate the temperature and liquid or vapour flowrate in estimate pade. I select programme generated estimate and my heat erorr oscillate in the range .01 to .8 and equibrium error goes to zero. I decraese he damping facter or also tried adapative damping. But nor successful. In pumparound i have given delta T as spec.

Any suggestion what i shd do to achive convergence. Material balence is perfect ...sum of all 5 draws matches to feed.
Flow to vacuum pump from top is available for colmn to adjust any anomony in meeting flow imbalence.

Contents By Category