Numerical Instability with Burn-Away Features in FDS

[Bastian Stahra]

Hello,

I've modeled a building featuring a balcony structure composed of burnable blocks, see FDS-File WindBrand-WB0062 attached. These blocks ignite at a specified temperature and burn away according to the fuel load of each cell.

Everything works quite well initially. However, I encounter a numerical instability at a certain point in the simulation(usually after 20 – 30 minutes simulation time) , typically when many cells are burning or burning away simultaneously. This happens on a Cluster as well as on my workstation.

This numerical instability arises when a cell disappears from the simulation. Within milliseconds, the temperature and velocity in that cell spike to "infinity," leading to the instability.

Notably, the affected cell varies with each simulation run, indicating that the issue is not due to a simple geometry/model error.

 The instability does not occur when I disable the 'burn away' feature.

In an attempt to resolve this issue, I've reviewed the FDS guides and forum discussions. It is commonly noted that numerical instabilities can arise from sudden changes in the simulation. This is observable when a cell burns away, as illustrated in the attached picture Test-Case-default.JPG. To mitigate this, I attempted to ramp the burning rate at both the start and end of the process (over 30 seconds), as shown in the next picture 
Test-Case-RAMP.JPG  .

However, this approach is only effective for individual cells where the exact number of simultaneously burning surfaces is known, making it impractical for larger scales.

Is there a way to determine the moment when, for example, 90-95% of the fuel in each cell has burned away, so that I can then gradually reduce the Heat Release Rate (HRR) for that cell over 10-20 seconds before it disappears? This might reduce the likelihood of numerical instability by ensuring the change in HRR isn’t as abrupt when the cell vanishes.

I am very grateful for any input!

Bastian

[Jonathan Hodges]

This sounds like the same issue we were seeing a couple weeks ago (https://github.com/firemodels/fds/issues/12450) which Kevin has fixed. Try the nightly build at (https://github.com/firemodels/test_bundles/releases) and see if you still have the issue.

[Kevin McGrattan]

Which of the three FDS input files should I test? Which one has the instability?

[Bastian Stahra]

This one has the instability:

WindBrand-WB0062

Kevin McGrattan <mcgr...@gmail.com> schrieb am Mi., 21. Feb. 2024, 22:28:

Which of the three FDS input files should I test? Which one has the instability?

--
You received this message because you are subscribed to a topic in the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fds-smv/lFkUcJcPozg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fds-smv+u...@googlegroups.com.
To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/CAAJimDEefZNMSo9SWEkDv_ub-nXGOVYUY8AO2PhghEF%3DFRtuLQ%40mail.gmail.com.

[Kevin McGrattan]

The WindBrand case has run 230 s and is still going. I am using the latest source code. What version are you using?

[Bastian Stahra]

At the beginning the simulation operates quit effectively. However, the instability typically arises between 1200 and 1400 seconds. At this juncture, the fire significantly enlarges, resulting in numerous cells concurrently combusting. 

We are using:
 Revision         : FDS-6.8.0-0-g886e009-release

[dr_jfloyd]

Looking at your input file you have this surface type which appears to be used for a window:

&SURF ID='Doppelverglasung Tbruch=600°C',
      RGB=153,96,255,
      TRANSPARENCY=0.34902,
      HRRPUA=250.0,
      IGNITION_TEMPERATURE=550.0,
      BURN_AWAY=.TRUE.,
      MATL_ID(1,1)='Wärmeabzugsmaterial',
      MATL_MASS_FRACTION(1,1)=1.0,
      THICKNESS(1)=1.0E-3/

You are modeling this as something that combusts. If this is just glass and you simply want to remove it cell-by-cell once it reachs 550 C you could try something like:

&MATL ID='Wärmeabzugsmaterial',
      FYI='Bastian Stahra',
      SPECIFIC_HEAT=0.84,
      CONDUCTIVITY=0.96,
      DENSITY=2500.0,
      ABSORPTION_COEFFICIENT=323.0

      REFERENCE_TEMPERATURE=550.0,

      PYROLYSIS_RANGE = 5/

&SURF ID='Doppelverglasung Tbruch=600°C',
      RGB=153,96,255,
      TRANSPARENCY=0.34902,

      BURN_AWAY=.TRUE.,

      MATL_ID(1,1)='Wärmeabzugsmaterial',
      MATL_MASS_FRACTION(1,1)=1.0,
      THICKNESS(1)=1.0E-3/

Now the MATL is something more like glass including some radiation transparency with a solid phase reaction that will start near 550 C and finish just above 550 C consuming all the mass of the window. Howeverm the reaction will not produce anything; the window will just disappear once consumed.

[Bastian Stahra]

Hello dr_jfloyd,

Thank you for your suggestion. I created a test case based on it, but now the glass wall disappears after 0.2 seconds, even though the heater isn't hot yet.

Could you please provide some guidance on how to implement your idea using my input file?

[dr_jfloyd]

I hadn't run with those values was just giving an example but it turns out not a great example. With REFERENCE_TEMPERATURE and PYROLYSIS_RANGE, FDS uses Eq. 9.4 and 9.5 in the User's Guide to get an A and E.  5 K turns out to be too small of a value for that REFERENCE_TEMPERATURE and the resulting value for A overruns the real number type which leads to nonsense. If you look at the .out file and and find the MATL reaction you will see A=infinity.  Need to increase PYROLYSIS_RANGE to get a better A and E. 20 K seemed to work OK.

Since in your actual case you have a window that is exposed on both sides to the FDS domain I would not set TMP_BACK or BACKING='VOID'.

[Kevin McGrattan]

The case that was originally posted to this thread is still running on my cluster. It has reached about 900 s. I'll just let it continue.

[Bastian Stahra]

Ok, perfect, thanks a lot!

[Bastian Stahra]

Thank you dr_jfloyd. I will review the User Guide and conduct some experiments with these values.

[Kevin]

The case is still running, 1313 s.

[Bastian Stahra]

This scenario stopped in my simulations between 1032s and 1395s. Therefore, your run should be within the "critical time".

I have a model with the same geometry but with about 20 different surfaces to represent different items in the model (like chairs, tables, facade, wardrobe, wpc panels etc.) With so many surfaces it is even more critical for numerical instability.

These are the breakup times in 2 of my attempts with the model you have right now.

First attempt:
 Time Step: 115357, Simulation Time:   1032.37 s

ERROR: Numerical Instability - FDS stopped (CHID: WindBrand-WB0046)

Second attempt (after slightly cleaning the modell in some areas (this is the scenario you have)):
 Time Step: 148488, Simulation Time:   1395.90 s

ERROR: Numerical Instability - FDS stopped (CHID: WindBrand-WB0062)

Did anything "strange" occure, yet?

[Kevin McGrattan]

The job failed with a Numerical Instability at 1441 s.

[Kevin McGrattan]

I looked again at the input and output files. One thing I noticed is that you have 364 sealed volumes within your domain. Some have as few as 1 cell. These little cavities can cause trouble when the surrounding solid burns away. 

I suggest that you not use the BURN_AWAY feature and that you try to avoid creating empty pockets within the domain.

[Bastian Stahra]

Thank you for your advice. As a first step, I am now eliminating all pockets. I hope this adresses some, if not all, of the issues.

[Nikos Sarikoudis]

@Kevin How did you determine the amount of sealed volumes? I could use this to troubleshoot imported geometries.

[dr_jfloyd]

there is a summary of pressure zones in the .out file. 

[Bastian Stahra]

Next to the information of the number in the *Out File, you can also make a 3-D Plot of the Pressure Zones. This ist helpful to find the undesired zones in the model.

--

You received this message because you are subscribed to a topic in the Google Groups "FDS and Smokeview Discussions" group.
To unsubscribe from this topic, visit https://groups.google.com/d/topic/fds-smv/lFkUcJcPozg/unsubscribe.
To unsubscribe from this group and all its topics, send an email to fds-smv+u...@googlegroups.com.

To view this discussion on the web visit https://groups.google.com/d/msgid/fds-smv/e3298076-af37-405f-adcc-fe08cb488076n%40googlegroups.com.

[Bastian Stahra]

Hello Kevin, thank you so much for running and checking my simulation.

I closed all "empty pockets" and checked it in the out-file and with the 3-D slice "pressure zones". There are no empty pockets anymore in my model.

The numerical instability still occurs with burnaway feature on.

With burnaway feature off, the instability does not occur.  

In my analysis, the problem with the numerical instability comes with a high number of pressure iterations in situations where lots of blocks disappear due to the burnaway feature at the same time. 

Back to my initial question. Is it possible to ramp the HRR of a cell down before the cell disappears/burns away? This could lead to a less abrupt drop of HRR and hence to a smother run?!

Kevin McGrattan schrieb am Sonntag, 3. März 2024 um 23:56:15 UTC+1:

[Kevin McGrattan]

There is a parameter called MINIMUM_LAYER_THICKNESS on the SURF line that defines the thickness of the obstruction when it is removed from the simulation. It is by default 1e-6 m, which is very small. Try setting this to, say, 1e-4 and see if that helps. I am looking into this issue in general, where very thin obstructions can have wildly oscillating temperatures just before removal.