Freeze calculation without error message and incorrect HRR

[Lilian-ISI]

Hello,

We are contacting you regarding calculations performed on a remote calculator.
The software version used for these calculations is 6.9.1.

We have recently run several calculations and different behaviours have appeared.

Here are our observations, 2 cases can be distinguished:

1 - The calculation reaches its end, it runs at a ‘normal’ speed, there is nothing to report during the calculation. However, when we concentrate on the HRR, we don't reach the desired values.

2 - The calculations run normally at first, then ‘freeze’ without any software error messages in the terminal. Once again, the measured HRR is not what was expected.

What's more, we ran identical 3D models (position and flow rates of the fans, geometry of the building, etc.), with only the position and power of the fire differing. One of the calculations reached its end (1800 seconds), the second stopped quickly (less than 200 seconds simulated). In both cases, the HRR curve did not follow the desired curve set in RAMP.

The basic model was built using Pyrosim. We used the ‘Model Check’ function, which allows us to check the HRR max, which is correct.

We ran the model locally to compare with what we obtained on the external calculator. We're still waiting for the results; the calculation is underway.

Does anyone observe the same issue ?

[drjfloyd]

If the HRR in the hrr.csv file doesn't match your expectations it is generally one of three things:

1. You haven't defined the fire like you think you have. Check the MLR in the hrr.csv and multiply it by the fuel heat of combustion. Do you get the HRR you expect? If not then the issue is how you defined the fire.
2. You have a sealed domain and have used up the oxygen.
3. Fuel cannot mix with sufficient oxygen to burn before it leaves the domain. Look at the fire in smokeview. Do you see flames at OPEN VENTs that look like they extend outisde the domain?

[Lilian-ISI]

This is because we don't get the desired HRR when we multiply the MLR by the heat of combustion. However, we made a simplified case with the same fire properties and the HRR is good.
You will find attached the simplified file (‘simple_fire.txt’) and the complete model (‘Escalator_fire.fds’).

We also have the same problem with Sabalcore as reported in the following link:
#14289

One calculation goes to the end of the simulation (1800 s) and another stops at 180 s. The only changes are the position of the fire and the RAMP of the fire.

Discussion_GITHUB.zip

[rmcdermo]

Set SUPPRESSION=F and monitor the OXYGEN in your domain, slice or volume integrals. You need to test suggestions 2 and 3 that Jason made.

[Lilian-ISI]

Thank you both for your suggestions.

Regarding Jason's suggestions :
2- The building is large (7 floors) and we have air pulses on every level. We did a scenario with a small fire at 100 kW and didn't achieve the desired output. We think that the fire has enough oxygen to develop.
Nevertheless, we launched a new scenario by opening the doors on one of the levels of the building to see what happens.

3- There are no flames at the edge of the domain or through an opening. We don't have any openings close to the fire.

Regarding Randy's suggestions, we've followed up by adding your suggestions.
We'll get back to you as soon as possible.

[Lilian-ISI]

We have reviewed the various cases that have been launched to give you more information.

The calculations that ran on Sabalcore before the post:
1 - A scenario where the firepower is good (~1300 kW), but the calculation runs in a vacuum after ~300 s with no error message. This model had been run before the post. The only difference with the same model (same fire, fan, geometry) is the DT_HRR.
DT_HRR = 1.0 in the model with the correct power and DT_HRR = 30.0 in the other.

2- A fire at another level with a DT_HRR = 1 that goes all the way, quickly and without error, but with the wrong firepower.

These calculations were run by assigning 2 threads per MPI on Sabalcore, as discussed in another post https://github.com/firemodels/fds/discussions/14289.

The calculation which runs locally taking into account the suggestions :
3- The same scenario as point 1. Between the two input files the changes are :
Opening a door on the upper level in relation to the fireplace
Addition of SUPPRESSION = F
Addition of PLOT_3D of VOLUME_FRACTION for OXYGEN
DT_HRR = 10.0
The calculation is running at ~200 s, the firepower curve is not as expected.

So we have two ‘problems’. The first may be related to Sabalcore, where calculations stop without an error message. Regarding Sabalcore, and as someone has already mentioned in his post https://github.com/firemodels/fds/discussions/14289 an FDS file was run and no output file was generated. We've contacted them and said that we're also in discussions here.
And the second relates to HRR, but we don't understand how the DT_HRR change affects the FDS HRR for the files in point 1.
We're going to run point 3 with DT_HRR = 1 instead of 10.

Thanks again!

[drjfloyd]

This is what I get for you escalator case for the first 100 s. The HRR is following the RAMP exactly. I suspect you are making an error when comparing the FDS results to what you think your RAMP is doing.

[Lilian-ISI]

Thank you for your feedback.

Yes, it's OK in this case.
For the second case sent, the restaurant fire, here's what we get:

We checked the surface of the fire, the HRRPUA and the RAMP.
Our fire surface is 6.24 m2 (1.2 m x 1.0 m x 4 sides + 1.2 m x 1.2 m top) and the HRRPUA ~ 120 kW/m2, i.e. HRRmax ~750 kW. Theoretically, we should obtain the orange curve.
However, the FDS HRRmax is ~500 kW.

Thank you!

[drjfloyd]

You need to pay close attention to what your inputs are, and how FDS has snapped your inputs to the grid. Your issue is item 1 from list of the three typical causes (your fire is not what you think it is).

Here is your fire:
&OBST ID='Feu_restaurant', XB=50.2,51.4,36.8,38.0,26.4,27.4, SURF_IDS='Feu_restaurant_4eme','Feu_restaurant_4eme','INERT'/

The image below is from Smokeview. Note the floor top is at 26.8 m. So your height is not 1 m it is 0.6 m

(4x1.2x0.6+1.2x1.2)x120=518 kW which is what you have in your plot.

[Lilian-ISI]

Thank you very much!

We missed this and apologise for the inconvenience.
In conclusion, the FDS input files appear to be correct.

The fact that the calculation seems to work but nothing happens is due to Sabalcore. They will be carrying out tests with some of the files affected by the problem.
Do you think that running the calculation in MPI on 1 thread per mesh could solve the problem? Do you have another suggestion?

Thanks

[mcgratta]

I rarely use more than one OpenMP thread per MPI process. Two OpenMP threads is not going to cut the run time in half. At best, it might reduce the total run time by, maybe, 25%, depending on the computer. It is far better to dedicate compute cores to an increased number of meshes. The efficiency is far better that way.