Follow up on inconsistent pyrolysis modeling between FDS 5 and 6 (2242 now closed)

[gforney]

Please complete the following lines...

FDS Version:
SVN Revision Number:
Compile Date:
Smokeview Version/Revision:
Operating System:

Describe details of the issue below:
One of the major problems has been the increase in computation time to see how the
simulation is going,  I have also revised some of the fuel properties and done a bit
more digging.
I realize the files c1(ver 5) and d1(ver 6) are big, but they are identical except
for commands added to d1 to make it work in fds6.  These were
&OMP NUM_THREADS=1
and for the fuels WOOD and PLASTIC SPEC_ID='POLY-A'
File c1 started took a month to run 3000 seconds and file d1 took 2 weeks to run 260
seconds.

I have been testing fuel properties and a recent file started 13 April I note in the
output file, maximum velocity error of about 0.5E-01 and pressure iterations start
at 2 and increase to 4. Could this be causing the increased run time?

Going back through the output files, c1 used the first 2 layers of WOOD and 0.5 layers
of PLASTIC

Some observations.
File c1 (ver 5) used first 2 layers of WOOD and 0.5 layers of PLASTIC.  File d1 (ver
6) used 0.5 layers of both.

The WORD file is an extract from a log of the various simulations showing the parameter
modifications and early results.  Of particular concern is early on when I removed
the ignition source, the drop in heat release rate, even though there should have been
enough heat to sustain combustion.  Changing the fuel properties later on solved this
problem, but the growth rate is faster than desirable.  It also appears to peak sooner
than desirable, although it may not have run long enough to confiirm that.

If the pyrolysis model does require a product specification for each fuel, a complexity
I question given the ability to use a much simpler HRRPUA operation, then the implication
of that is not described in the manual and probably should be.

FDS 5 did a very good job of modeling the water/fire interaction without resorting
to artificially imposed curves, although they are available.  FDS 6 does not seem capable
of getting to the base case without extensive effort in determinng stoichiometric fuel
characteristics.  

1.  I think the computation time is significant.  A month was OK for this.  Six months
for base case is not.
2.  I think the sensitivity of the combustion process to the ignition source availability
is significant and bears investigation.
3.  Are the pressure iterations driving the computation time?  If so, are these significant
to the process?
4.  Is the maximum velocity error of 0.05 significant to the process?  If not can it
be varied?

The attached files include the original ones plus my log of trials and a spreadsheet
of heat release rate vs. time for comparison.

The c1 file derived from a simulation compared to the LTA Tunnel fire testing and was
a very good comparison with the fire heat release rate and temperatures.  As with any
large fire and modeling constraints the results were not exact, but resonable for the
purpose intended.  Fuel properties were researched and set consistent with published
data.  With the d1 files, the same fuel properties yielded significantly different
results, primarily that it would not start nor sustain combustion without significantly
more ignition energy for a longer duration.  The sudden drops when the ignition energy
decreased are also odd.  Changing  the fuel characteristics seems to solve the ignition
problem, but the fire now grows faster and seems to peak earlier than otherwise.  More
experimentation with the fuel properties might fine-tune this, but it seems a very
cumbersome approach, especially as these properties are not regualrly cataloged.
I apologize for the long-windedness, but wanted to show that I took your comments seriously
and have been looking for a reaonable way to make this work.  I have also tried to
summarize what I thought were key aspects, but as you can see it takes a lot of time
to test each scenario.  

Please feel free to contact me if you have any questions.
Thank you.
Ken

Original issue reported on code.google.com by harriskj on 2015-05-06 00:04:19

---

• Attachment: [fds issues.zip](https://storage.googleapis.com/google-code-attachments/fds-smv/issue-2435/comment-0/fds issues.zip)

[gforney]

These cases are too complex for us to analyze. There are also too many issues piled
one on top of the other. We need to work on very simple cases that clearly demonstrate
a problem. We know that FDS 6 takes longer than FDS 5. We know that the pyrolysis model
is sensitive to many input parameters. Let's take things one at a time. Can you submit
a case that has only a few dozen lines that demonstrates a clear problem in the most
current release of FDS?

Original issue reported on code.google.com by mcgratta on 2015-05-06 12:08:21

[gforney]

Kevin,
I'll see what I can do. The difficulty is figuring out what is unimportant
that can be eliminated.  While these input files look complex, the overall
structure is relatively simple, just a lot of stuff.  I will have to
eliminate and test, which as you know is time consuming.
FWIW, I think the error message and pressure iterations in the output file
are significant, but I don't know their relevance from Version 5.  Was the
same situation occurring there, but not being reported or acted on?
Is there a conflict being created with the fuel product being required now,
rather than just working off the vaporization and combustion heats?  The
documentation cautions against duplication of properties.
In the meantime, I will work on producing a smaller file.

Original issue reported on code.google.com by harriskj on 2015-05-06 23:17:47

[gforney]

Big picture -- up until FDS 6, we designed the model to require as few inputs as possible,
and we made some pretty big assumptions for the user. Obviously, the users loved it.
So much less to think about. It's just like the "AutoContent Wizard" in PowerPoint.
But then we worried -- are we turning our users into morons, just like Microsoft does
with these dumb features? So in FDS 6, we decided to stop making such big assumptions
for the user and put the onus back on him/her to make sensible decisions about how
to model things. Obviously, things changed between 5 and 6. But rather than to just
compare 5 and 6 and expect similar results, we are now advising users to create a series
of small verification cases before embarking on that one huge simulation. Yes, it takes
some time, but in the end it's worth it because you have a better sense of what the
model is actually doing. In your case, you may not see the forest for the trees with
such a complex input file. Even more importantly, as model reviewers and AHJs gain
more knowledge about how the model works, you are in a much better position to defend
your modeling approach. 

Original issue reported on code.google.com by mcgratta on 2015-05-07 12:28:23

[gforney]

Fundamentally, I agree as evidenced by some of the postings I've seen.  I
very much admire your staff's patience.  I started a simpler case and in
fact made it too simple and got an instability, so had to make it a little
less simple and am running it now.  I also revised the ignition source and
noticed that the number of pressure iterations decreased significantly and
run times were comparable with FDS5.
On the output file, the sign of the radiation loss is reversed from ver 5,
I don't know if this is significant or nomenclature.
I am revising the input file down slowly to keep the integrity and hope to
have somethng to you next week.  Thanks for your patience.
 And yes, I am learning a lot and will be better able to defend the
modeling approach.

Original issue reported on code.google.com by harriskj on 2015-05-09 20:25:20

[mcgratta]

I am closing this issue. If you want to reopen it, do so using the GitHub issue tracker.