Problems with sprinklers in FDS version 6.7.4

[Maxwell Franzen]

Hello,

I'm running a comparison case between sprinklered and non-sprinklered runs for a hydrocarbon pool fire and getting vastly different results in HRR prior to sprinkler activation. I've attached a snip of the difference in heat release rate prior activation. You can see at the time of activation that HRR is 16x greater in the sprinklered run.

I cannot post the entire FDS files due to confidentiality; however, I have confirmed with notepad++ compare that the only difference between the two files is the the addition of spec id 'water vapor', a description for water particles, the sprinkler linkage,  and devc ID's for several sprinklers. I can post parts of the FDS file, without any geometry, if there is a specific part of the file of interest. My descriptions used are below.

&SPEC ID='WATER VAPOR'/

&PART ID='Water',

      SPEC_ID='WATER VAPOR',

      DIAMETER=499.999,

      MONODISPERSE=.TRUE.,

      COLOR='BLUE',

      AGE=30.0,

      DENSE_VOLUME_FRACTION=0.0/

&PROP ID='Generic Commercial Link01_Water Spray01',

      QUANTITY='SPRINKLER LINK TEMPERATURE',

      ACTIVATION_TEMPERATURE=73.888889,

      RTI=49.687825,

      PART_ID='Water',

      K_FACTOR=116.091903,

      OPERATING_PRESSURE=0.969403,

      ORIFICE_DIAMETER=0.01905,

      SPRAY_ANGLE=60.0,75.0/

&DEVC ID='SPRK01', PROP_ID='Generic Commercial Link01_Water Spray01', XYZ=45.375,31.177232,39.090449/

My initial thought was that this had something to do with the description of water vapor as an extra species. I have not been able to replicate this result in a simpler model and have not seen this difference in other models I have run; models are otherwise virtually identical prior to sprinkler activation.

I read in another forum post response from 2015 by dr_jfloyd that this was an issue in FDS 5 due to the addition of water vapor in the background species of air; however, it sounded like things were a bit different in Version 6. 

"The way we handled species for simple chemistry in FDS 5 is a little different than in FDS 6. In FDS 5, the air background species had no water vapor in it if there were no sprinklers.  If sprinklers were present, then we added a little water vapor to the air background species to account for the ambient humidity. This slightly changes the oxygen volume fraction and slightly changes the radiation (water vapor is absorbing).  Keep the &PART line in the input file without sprinklers and you should see the same results prior to sprinkler operation."

I'm running FDS Version 6.7.4. Is it still the case in this version that water vapor is being added to the background species of air in the sprinklered run due to the added description of sprinklers?

Thank you in advance for you help. I'm a new poster, but very much appreciate the advice provided on this forum.

[Kevin McGrattan]

Look at the two .out files and compare these lines below. I wonder if there is a significant difference in water vapor.

 Tracked (Lumped) Species Information

   AIR
   Molecular Weight (g/mol)            28.75198
   Ambient Density (kg/m^3)            1.195
   Initial Mass Fraction               1.000
   Enthalpy of Formation (J/kg)     -8.50E+04

   Sub Species                    Mass Fraction     Mole Fraction
   NITROGEN                       7.624699E-01      7.825727E-01
   OXYGEN                         2.309974E-01      2.075588E-01
   CARBON DIOXIDE                 5.914650E-04      3.864118E-04
   WATER VAPOR                    5.941209E-03      9.482037E-03

[Maxwell Franzen]

'AIR' is identical to each other in both runs. Both runs are ongoing, but I assume this would not change during the course of the run.

 AIR

   Molecular Weight (g/mol)            28.74476

   Ambient Density (kg/m^3)            1.190

   Initial Mass Fraction               1.000

   Enthalpy of Formation (J/kg)     -9.07E+04

   Sub Species                    Mass Fraction     Mole Fraction

   NITROGEN                       7.621486E-01      7.820466E-01

   OXYGEN                         2.309001E-01      2.074193E-01

   CARBON DIOXIDE                 5.912157E-04      3.861520E-04

   WATER VAPOR                    6.360087E-03      1.014801E-02

The sprinklered run has an additional tracked species 'water vapor', not present in the non-sprinklered run.

 WATER VAPOR

   Molecular Weight (g/mol)            18.01528

   Ambient Density (kg/m^3)            0.746

   Initial Mass Fraction                     0.000

   Enthalpy of Formation (J/kg)     -1.34E+07

   Sub Species                    Mass Fraction     Mole Fraction

   WATER VAPOR                    1.000000E+00      1.000000E+00

 

     Viscosity (kg/m/s) Ambient,  294 K:  1.06E-05

                                  500 K:  1.77E-05

                                 1000 K:  3.58E-05

                                 1500 K:  5.21E-05

   Therm. Cond. (W/m/K) Ambient,  294 K:  1.97E-02

                                  500 K:  3.46E-02

                                 1000 K:  8.21E-02

                                 1500 K:  1.37E-01

        Enthalpy (J/kg) Ambient,  294 K: -1.34E+07

                                  500 K: -1.30E+07

                                 1000 K: -1.20E+07

                                 1500 K: -1.07E+07

    Spec. Heat (J/kg/K) Ambient,  294 K:  1.86E+03

                                  500 K:  1.96E+03

                                 1000 K:  2.29E+03

                                 1500 K:  2.63E+03

   Diff. Coeff. (m^2/s) Ambient,  294 K:  2.11E-05

                                  500 K:  5.78E-05

                                 1000 K:  1.98E-04

                                 1500 K:  3.95E-04

 

   Liq. Enthalpy (J/kg)     Melt  273 K: -1.60E+07

                                  323 K: -1.58E+07

                            Boil  373 K: -1.56E+07

 

   Liq. Spec. Heat (J/kg/K) Melt  273 K:  4.23E+03

                                  323 K:  4.18E+03

                            Boil  373 K:  4.22E+03

 

   Heat of Vapor. (J/kg)    Melt  273 K:  2.50E+06

                                  323 K:  2.39E+06

                            Boil  373 K:  2.27E+06

Also the conversion matrix contains an extra column for water vapor in the sprinklered run

Mass Fraction Transformation Matrix to Convert Species Mixtures (Columns) to Primitive Species (Rows)

                                         AIR                    REAC_FUE      PRODUCTS      WATER VA

   REAC_FUEL                    0.000000      1.000000        0.000000         0.000000

   NITROGEN                     0.762149      0.000000        0.708746          0.000000

   OXYGEN                         0.230900      0.000000        0.000000           0.000000

   CARBON DIOXIDE        0.000591       0.000000        0.205230          0.000000

   CARBON MONOXIDE   0.000000      0.000000        0.000841          0.000000

   WATER VAPOR             0.006360       0.000000       0.079577          1.000000

   SOOT                            0.000000         0.000000       0.005605          0.000000

   HYDROGEN                 0.000000         0.000000      0.000000           0.000000

There is an additional line under surface conditions for 'PERIODIC WIND' in the sprinklered run, vs. 'PERIODIC FLOW ONLY' in the non-sprinklered. I don't know if that has any relevance.

[Kevin McGrattan]

OK, can you create a simple test case and start an issue on the Issue Tracker?

[Maxwell Franzen]

Will do.

However, I have not been able to replicate this result in any other case. Do I need to be able to replicate a difference in fire growth in the simple case?

[Kevin McGrattan]

Yes, we can't fix the problem if we don't have a clear example. 

[dr_jfloyd]

Are you predicting the HRR or prescribing the HRR? If you are predicting the HRR, do you see large differences in the INCIDENT HEAT FLUX to the pool surface or in the INTEGRATED INTENSITY or ABSORPTION COEFFICIENT in the gas above the pool at the start of the run?

[Maxwell Franzen]

Dr. Floyd,

I'm predicting the HRR. I can post relevant REAC, MATL, and SURF lines if helpful. 

I noticed INCIDENT HEAT FLUX at the pool surface is much larger for the sprinklered run. This is noticeable immediately in smokeview; however, my output quantity for boundary conditions is only writing at 10 s intervals. I believe heat flux is increasing surface temperature, which is leading to greater evaporation of the liquid which is then driving the increase in HRR. I do not currently have a slice file for INTEGRATED INTENSITY or ABSORPTION COEFFICIENT. I can add these in the next iteration, although it will be a few days before I can free up a computer to do so. I'll also instruct FDS to output at shorter intervals to get a better idea.

As an FYI, I've started a third run where I include sprinklers, but have them spray a negligible amount of water upon activation. In this run I'm still seeing the increase in HRR as opposed to the run without sprinklers.

From your understanding of FDS, would you expect the fact that water vapor is being tracked separately in the sprinklered run to create a difference in how radiation is being calculated prior to activation? 

Kevin,

I'll do what I can to strip down the model as far as I can without eliminating the difference in HRR.

Thank you both for looking at this.

[dr_jfloyd]

Water vapor being tracked seperately should not be having a noticeable effect prior to activation. Water vapor for the sprinkler species should just be getting summed with water vapor in the ambient and in the product species when determining the absorption coefficient in the gas. Since water vapor for the sprinkler is zero at the start, sprinklers or no sprinklers shouldn't change the result. If you get rid of all the geometry, except for what is needed to define the pool surface, do you still a difference?  

[Maxwell Franzen]

I tried removing all geometry other than the pool surface. I was not able to replicate a difference in HRR between sprinklered and non-sprinklered runs. I kept all meshes the same for now.

I should also note that removing geometry causes the interior zone to merge with exterior, I’ll assume for now this is not relevant to the issue.

Now for the strangest part. I had previously been running everything on multiple processors using 14 MPI processes (I have 14 meshes).  I re-ran the original sprinklered result (including geometry) with only a single MPI process (OpenMP is still being used). It now matches the non-sprinklered result up until sprinkler activation. 

I similarly tried comparing the no-geometry runs on 1 vs. multiple MPI processes, but found no differences.

It would seem both the MPI process and geometry are relevant. Does that make any sense to you? I'm curious if there is any inaccuracy in radiation or species transport during an MPI exchange (similar to pressure).

I'm still investigating which elements of the geometry are relevant.

[Kevin McGrattan]

When you  say you get different results depending on the number of processes, do you also change the number of meshes? Or are you running the exact same input file but using different number of OpenMP threads?

[Maxwell Franzen]

Exact same input file. I'm initiating with  "mpiexec -localonly -n 1" specified  rather than "-n 14". This shows in the .out file that only 1 MPI process was used.

MPI Enabled;    Number of MPI Processes:       1

OpenMP Enabled; Number of OpenMP Threads:      8

The other run was on a different machine with more processors, but the .out file indicates

 MPI Enabled;    Number of MPI Processes:      14

 OpenMP Enabled; Number of OpenMP Threads:     16

I realized this morning that I had inadvertently used a machine with FDS version 6.7.5 on the single MPI run which seemed to fix itself. I believe now that the version number is the real cause of the fix. I've also confirmed by testing on a machine with FDS 6.7.4 with the number of MPI Processes set to 1; the issue still persists in the older version. Sorry for the confusion!

For now I'm going to move forward with just re-running the sprinklered run on FDS version 6.7.5. The results are more in line with other test series and match experimental results better.

[Maxwell Franzen]

I'm now seeing that there is a know issue with water droplets in 6.7.5, which gives me a bit of pause with doing the sprinklered run on this version. 

• A bug was introduced in FDS 6.7.5 that impacts the evaporation of droplets. In cases where the operation time of multiple sprinklers is being determined or where the cooling of gas from water droplets is being determine, it is recommended to use the latest test bundles or FDS 6.7.4

Was this issue impacting all sprinklered runs or just specific instances? 

I cannot seem to find FDS 6.7.6. Has this version been released yet? Github notes April 2021 release.

[Kevin McGrattan]

It has not been officially released, but you can download the latest build.

[Kevin McGrattan]

https://drive.google.com/drive/folders/1X-gRYGPGtcewgnNiNBuho3U8zDFVqFsC?usp=sharing

[Maxwell Franzen]

In case anyone comes across this thread while troubleshooting, I discovered that my different HRR's in the original post were done using two different version of FDS (6.7.5 vs 6.7.4). As a result, I no longer believe the sprinklers were the cause. After some testing, I'm finding version 6.7.5 is consistently predicting lower fire growth and subsequent HRR than version 6.7.4, at least for my set up, YMMV.

I have not seen the validation guide for version 6.7.4 so I'm not sure how they compare, but the early growth phase of high flash point hydrocarbon pool fires does not appear to be well studied in general in the version 6.7.5  validation guide (if someone has a study showing otherwise I'd greatly appreciate!)

I'm also a bit curious about the bug that was noted in the release of version 6.7.5 related to sprinklers. I'm not noticing a huge difference in HRR after sprinkler activation for version 6.7.5 vs the test bundle; although this is only comparing one case so maybe it's not occurring here. If anyone has more information on the bug, I'd be interested.

On Thursday, April 22, 2021 at 11:04:12 AM UTC-5 Kevin wrote:

https://drive.google.com/drive/folders/1X-gRYGPGtcewgnNiNBuho3U8zDFVqFsC?usp=sharing

[dr_jfloyd]

Heat and mass transfer from droplets use a semi-implicit approach that results in a system of equations to be solved by matrix inversion. Between 6.7.4 and 6.7.5 a bug was fixed that dealt with heat transfer to surfaces by droplets. The fix wound up introducing a bug into one of the matrix terms used for the gas temperature. It didn't have much impact on steady-state or equilibrium conditions so it wasn't caught in the verification suite but it did have an impact on transient gas temperatures. 

[Kevin McGrattan]

There is a bug in 6.7.5 related to droplet evaporation. I suggest you use the latest test bundle if your case involves sprinklers.

[Maxwell Franzen]

Thanks for the description, I'll take a look at gas temperature on existing runs and move forward with the test bundle once possible.

I'm now seeing in the validation guide for Revision: FDS6.7.5-0-g71f025606  lists in "Table 16.2: Validation Git statistics for all data sets" that the latest FDS revision string was done with FDS6.7.4-1112-gf459ae1-master. Does this mean the data in the validation guide is not actually from FDS6.7.5-0-g71f025606, but from previous revisions as listed in this table?

[dr_jfloyd]

The verification suite is run every night. Before we do any release, all verification cases must pass.

The validation suite takes a while to run. We don't run the entire suite before a maintenance release (e.g. 6.7.4 to 6.7.5). We do continuously cycle through the cases in the validation suite, but only for major or minor releases do we make sure every case has been updated. If you were to go to the current nightly build of the Validation Guide you would see that almost all of the validation cases have been updated since 6.7.4 was released.

[Maxwell Franzen]

Thank you! I did not realize there was a nightly bundle for the documentation, but now I see. Located here: https://drive.google.com/drive/folders/0B_wB1pJL2bFQUlJwMmNfaHlqME0