13 replies to this topic

[tgugcl]

Hello, 

I work as a process engineer in a refinery. We are trying to extend and improve our current atmosferic and vacuum tower simulations with HYSYS. 

 

We have been able to predict and compare the volumetric performance of our model with the the real data of our plant. So far the model's volumetric flows fall within 0.4 or 2% error for almost all the products (list below)

 

Unfortunately we have not been able to reconcile the performance of ours heavy products (Pitch and Topped crude oil). They both have an error of 4% 

 

LPG

Gasoline

Naftha
Kerosene

Diesel
AGO

HGO

LGO

 

Pitch

Topped crude oil 
 

As expected, when we arbitrarily raise the temperature of the inlet (fresh) crude to the atmosferic tower, the heavy product's performance seems to show a better agreement with operational data. 

 

Is this error acceptable? What could be the problem and how can it be fixed? 

Any help is welcomed =) 

Thank you very much

Edited by tgugcl, 22 March 2018 - 04:27 PM.

[Nikolay_]

Hello,

 

You can use back blending to get feed stream. Feed stream is credible problem.

 

Regards,

Nikolay

[Pilesar]

4% does not sound too bad for vacuum distillation. You must change the feed composition for the vacuum tower. Do not rely on a D86 for the vacuum column or the simulated results from the atmospheric column. D1160 is minimum requirement but even then you cannot expect exact simulation matching. If you are relying on any assay composition analyses over a year old, expect them to be wrong since the oil from the reservoir changes with time. It helps to add more assay cuts in the vacuum distillation section -- do not rely on the same cuts you used for atmospheric distillation. Plan on NOT matching the overhead temperature of the vacuum column! The goal is generally to match the product streams and the pumparound flows and duties.

[tgugcl]

Hello,

 

You can use back blending to get feed stream. Feed stream is credible problem.

 

Regards,

Nikolay

 

 

4% does not sound too bad for vacuum distillation. You must change the feed composition for the vacuum tower. Do not rely on a D86 for the vacuum column or the simulated results from the atmospheric column. D1160 is minimum requirement but even then you cannot expect exact simulation matching. If you are relying on any assay composition analyses over a year old, expect them to be wrong since the oil from the reservoir changes with time. It helps to add more assay cuts in the vacuum distillation section -- do not rely on the same cuts you used for atmospheric distillation. Plan on NOT matching the overhead temperature of the vacuum column! The goal is generally to match the product streams and the pumparound flows and duties.

 

For simplicity I used the product stream from the atmosferic unit to feed the vacuum tower's simulation. I haven't tried backblending for neither of the two.

 

My problem is not so much that I have 4% error in one of the product flows. My problem is that ONLY the heavy products seem disagree with the plant's operational data. For that reason I suspect that I could have a topology or specfication error. Although, I do not discard a feed composition error, it just seems more unlikely.

 

Also Pilesar....
The overhead temperature of the vacuum column is exactly one of my specifications. Why should this temperature not be specified?

 

I must add that I used a 3 stages simplified model (as explained in Gerald L. Kaes “A practical guide to Stedy State Modeling of Petroleum Processes”)

That basically means the vacuum column got divided in 3 parts o zones, from top to bottom, one for the LGO separation and pump-around, one for the HGO and the one for the flash zone. 

[Technical Bard]

There are numerous possible issues here.  What thermo package are you using?  Have you tried PR-LK, Grayson Streed and BK10?  Each will give different results and it is almost impossible to say ahead of time which one will match better for a given crude oil.  

 

Second - the characterization of the crude oil is very important.  The best results are obtained by taking high temperature simdis lab data for the products (VR, HVGO, LVGO, Diesel, etc) and rebuilding the crude unit feed from them.  A SimDis or D86 of the feed is rarely got enough detail or accuracy.  D1160 is ok, but SimDis methods are better.

 

The flash zone pressure is critical (along with the whole pressure profile in the vacuum tower) as is the stripping steam rates.  Small changes in stripping steam can change the cut point in the simulator significantly.  Make sure you are matching those accurately.

 

If this is a fairly heavy crude, one area that can be an issue is the molecular weight prediction of the simulator.  If you can get an actual MW curve (via Mass Spec) for the crude, that can help a lot.  Check out ASTM Manual 50 by Riazi on methods for predicting these parameters that may not yet be supported by the simulator vendors.

 

How are you modeling the overflash and slop wax?  Is it matching the actual topology of the plant?  Kaes book is good - also check out papers by Golden et al and Grande from the last 15 years or so.

[Pilesar]

Jerry Kaes is an excellent resource. I took a class from him many years ago and I wish I had a copy of his book, but it is very pricey!

  For an atmospheric tower, you do not need much detail for the heavy components since they are not distributed. It is common to have maybe 4 assay cuts in the topped crude range which will give you good results for the crude tower. Errors in the heavies character are not a big deal in the crude column, but they are an order of magnitude greater in the vacuum column. The topped crude may be only a third of the crude column feed, but it is the total feed to the vacuum column. These same 4 cuts become distributed in the vacuum tower and it becomes difficult to match the real world with so few cuts. Where you have products you are separating, you need to include several assay cuts that overlap between products.

  You need to know the source of the characterization of the heavy components. Even a vacuum lab analysis (D1160) will typically leave a large portion of undistilled residue. Was a good sample taken? Was good lab technique used? Are the results reproducible? How was the characterization of the heaviest cuts extrapolated? 

  Are you accounting for cracking in the lab analysis and in the column products? You will have light gases in the products that were not present in the feed to the atmospheric tower! This is one of the reasons that back blending the products will give you better results. The top temperature of the vacuum column is influenced by cracking in the feed furnace and non-condensibles entering the vacuum tower through leaks. It is common to add these as small extra feeds to make your simulation match better.

  You don't say whether your column is producing lube oils or cat cracking feed stock. I don't know what specs you are trying to match. Are you matching tower pressures? heavy gas oil rates? light gas oil rates? If you have a converged simulation and are only off by 4% of a product rate, it sounds like you have done really well. Consider your overall mass balance. Do the measured vacuum products match the measured topped crude rate in the field? The converged simulation will exactly mass balance even if the measured values do not! Tune the feed. Make small changes and observe the results. Stop when you have 'good enough' answer.

[tgugcl]

Technical Bard Pilesar

Thank you very much for taking the time to respond! This discussion has been great. 

I seems that we pretty much agree on one thing, a better characterization of the crude entering both towers would help a lot in predicting the separation taking place. This can be done either with better characterization techniques (Mass spec.) or thermodynamic packages or by bettering lab data in the form of precise SimDis/D86/D160 or with more cuts for the crudes' the heavy fractions.

 

To answer your questions:
1.- For the vacuum tower, the only pressure measurement we have is in the flash zone. For that reason the pressure profile from the top to the flash zone was lineally interpolated assuming a top pressure of 20mmHg-abs. Likewise, the pressure profile from the flash zone to the bottom was lineally interpolated assuming a bottom pressure of: (flash zone pressure)+45mmHg. A better estimate or real plant measurements would certainly help.

 

2.- Back-blending is not really an option right now.

 

3.- I assumed that the vacuum tower's measured super-heated water flow is correct. 

 

4.- The mass balance of both towers are always off. That brings uncertainty to the flow measurements and to the comparison of the simulated and measured performance of the towers.

 

5.- To characterize the crude feed I used assay data from our labs. Many of the assays data were not updated. 5/8 assays were made in the year 2014 and 2015.

 

6.- The vacuum tower is producing cat. cracking feed stock. I am trying to match all of the products rates.

 

7.- The slop wax is re-injected to the flash zone of the vacuum column. There is a wash-oil flow from HGO zone to the top of the flash zone. The topology of the vacuum tower is as shown in the following image

8.- To be honest I don't know if I am modelling the actual overflash of the atmospheric column. 

 

Edited by tgugcl, 26 March 2018 - 11:50 AM.

[Pilesar]

Comments on your answers:

1) The pressure profile in a vacuum tower is typically NOT linear. Your simulation gives vapor rates inside the tower and you can use those to better estimate internal pressure drops. Assuming pressure drop is proportional to the square of vapor velocity is better than assuming a linear profile.

2) The information about the products from the plant is available. If you do not back blend, you are ignoring information that could make your simulation better. If you know what you are trying to match, why not use that information in your input?

4) Since you do not know the real mass balance, then how have you judged your simulation is 4% off on some products? Why is matching an uncertain mass balance within 4% unacceptable? If your real world material balance does not close and your simulation material balance does close, then the two will never match.

5) Old assays are unreliable. You have better information by combining the characteristics of the real plant products.

8) You can try to tune the overflash to get a better product match, but there are large knobs available of feed characterization that will make a bigger impact.

 

Try this exercise: If you consider your feed characterization golden and unchangeable, you can plot temperature vs quantity by hand and make the product cuts on the graph to see how the products distribute. For example, you will never get more HVGO out of the simulation than is in the feed. You can find an 'ideal' distribution this way that you can only approximate with your model since there will be overlap between products. If the cut point temperatures needed to match the flow rates are very different from the real world, it may give you a clue where your material balance is lacking. It is not just the overall material balance that matters, it is the material balance for each assay cut.

Good luck.

Edited by Pilesar, 27 March 2018 - 08:29 AM.

[tgugcl]

Thank you for taking the time to respond.  I'm interested in Nº8

 

1.- I will ajust it based in your recomendation 

 

2.- Thank you again for this. 

 

4.- I know that comparing the performance out of an un-closed mass balance does not make much sense. Unfortunately this is the way the plant is operating right now. I have turned a blind eye on that one. 

 

8.- Could you explain what do you actually mean by overflash? I have seen conflicting definitions.  Is there a way to physicaly measured it in a real distillation column?

[Pilesar]

Overflash is the liquid falling back into the flash zone of the atmospheric column. This will be less than 5% of the crude feed. It is not directly measured, but you can calculate it from the heat and material balance. For your simulation model you control it to a small quantity (perhaps 3%) since it will be controlled to a similar point in the real column. So this will be one of your specifications. The associated variable will be the condenser duty. For the bottom product flowrate spec, you will vary the furnace duty. To meet the 95% point specs for your side products, you will vary their draw rates.

[tgugcl]

Thanks for the clarification. 
That is exactly what I thought and used in my simulation, 3% of liquid overflash.
 

[Nikolay_]

Hello Pilesar,

 

Could you please clarify what is overflash for? Is it implemented in any real plant or it calculation trick? Sketch is desirable.

 

Thank you.

 

Nikolai

[Pilesar]

Overflash is just the reflux that runs back to the flash zone. Every tray in a distillation column needs both liquid and vapor for separation of components to happen. If there were no liquid on the tray then all the vapor components entering the tray from the bottom also leave the tray at the top and the tray might as well not be there at all. The liquid traffic in an atmospheric column is greatest at the top and is reduced on each lower tray. The tray above the flash zone has the least reflux. You don't want that tray to be dry and you don't want too much liquid or it is an energy penalty. This liquid is controlled to 3% to 5% of the crude charge. Somewhere in that range is 'just right' in a refinery. Your steady state simulation does not handle dry trays very well since the calculations are configured for both a vapor and liquid phase. This is why even if you have a complete liquid draw tray, you spec your model to have a very small non-zero liquid flow from that stage. The small liquid flow from total draw trays is a 'calculation trick', but the tray above the flash zone does have liquid run back so I would not call that a calc trick.

  Are you working in a refinery? Are you a student?

[Nikolay_]

Pilesar,

 

Thank you for explanations. I'm a process engineer. I attached screenshot of 2nd atmospheric column profile. If I right understood the overflash is 9020 kg/h.  It was calculated without setting such specification. Why do I should control it when it can’t be controlled in real life? Is it for matching product specification, and finding operation parameters?

Attached Files

•  Screenshot.pdf   182.76KB   13 downloads