Validation of sprinkler actuation data

[sprankleray]

We have been attempting to recreate the CFAST results for the UL/NFPRF tests shown on pg. 344 of the validation guide, specifically those in the image shown for Series 1, Test 4.

Discussion at the start of that page references Figure 4.24 for test thermocouple locations, and describes the “chosen measurement locations” as 56, 68, 86, and 98. The CFAST input file (UL_NFPRF_1_04.in) provides 49 sprinklers in the same locations at those indicated in Fig. 4.24. The thermocouple location numbers presented in Figure 4.24 (red in attached file) are different from the modeled CFAST sprinkler location numbers (blue in attached file). More importantly, the actuation temperatures at these locations do not match. A review of the Surrounding Sensor Gas Temperature output (UL_NFPRF_1_04_devices.csv) for all 49 sprinklers modeled with CFAST shows the temperatures are obtained at different locations (green in attached files). Also included in the attached file is the diagram from page 344 of the V&V Guide overlaid with the output results for green labeled devices in the model. Would you be able to offer any clarification of how the plot for test 4 on pg. 344 of the validation guide was generated? The data seems to indicate that measurement locations 60, 89, 91 and 104 were used to plot the Series 1, Test 4 diagram on page 344 instead of the “chosen measurement locations” stated in the text on page 344. If this is correct, shouldn’t the V&V Guide identify, in Appendix C, the measurement locations actually modeled?
UL-NFPRF Test 4 vs CFAST Model results-1.pdf

[mcgratta]

Let me remind myself of the sprinkler numbering system. I don't think the blue numbering in your figure is correct.

[mcgratta]

I don't know how those CFAST cases were set up, and I'll have to go back and correct them. It's possible that the idea was that in CFAST the only thing that matters is the radial distance from the fire to the sprinkler.

The way it should be is that the dot labeled "98" is sprinkler 1. The sprinkler numbers advance going from left to right along the bottom row. Dot number "99" is sprinkler 8, and so, left to right, row by row, until dot "53" is sprinkler 49. The numbers in the diagram refer to thermocouple channels.

[mcgratta]

Yes, I noticed that the sprinklers are not numbered row by row, which is odd because these CFAST input files were based on the FDS input files for the same experiments. I wonder if the CFAST files assume a different orientation for the coordinate system. In the FDS files, the x direction is from left to right and the y is from bottom to top.

[sprankleray]

The X coordinates in CFAST are on the horizontal (width) axis (left to right) and the Y coordinates on the depth axis. Here is a diagram from the CFAST User's guide.

[mcgratta]

Yes, this much I know. The question is how the cases were originally set up. It appears to me that the orientation of the compartment is consistent with the test report. I just don't know why the sprinklers were arranged in this way.

[sprankleray]

I agree. The only thing we have to go on is the UL_NFPRF_1_04.in file available on the Github repository, which I'm sure you have already considered. If you agree that the data plots for this test, on page 344 of the V&V Guide are presenting results from sprinklers 11, 34, 36, and 49, we can treat the list of "chosen measurement locations" in the text on that page as a typo.

[mcgratta]

Let me fix the files now and we can check the results before I forget.

[rpeacoc]

Let me know when you change the files. I’ll run a full V&V to highlight any differences that result.Sent from my iPhoneOn Mar 24, 2026, at 2:01 PM, Kevin McGrattan ***@***.***> wrote: Let me fix the files now and we can check the results before I forget. —Reply to this email directly, view it on GitHub, or unsubscribe.You are receiving this because you are subscribed to this thread.Message ID: ***@***.***>

[mcgratta]

I'm running cfastbot now. If things look OK, I'll push up my changes and you can check them. I have ordered the sprinklers in the UL input files to conform to the convention used during the experiments and in the FDS files.

[rpeacoc]

As I see it running the Windows version of CFASTbot, only the Ceiling Jet temperature predictions changed, with the bias changing by 2%, going from 1.04 or 1.06 and the sigma changing by 1%, going from 0.43 to 0.44. Seems like what we'd expect for the changes in the input files. Thoughts?

…

________________________________ From: Kevin McGrattan ***@***.***> Sent: Tuesday, March 24, 2026 4:01 PM To: firemodels/cfast ***@***.***> Cc: Richard Peacock ***@***.***>; Comment ***@***.***> Subject: Re: [firemodels/cfast] Validation of sprinkler actuation data (Discussion #2234) I'm running cfastbot now. If things look OK, I'll push up my changes and you can check them. I have ordered the sprinklers in the UL input files to conform to the convention used during the experiments and in the FDS files. — Reply to this email directly, view it on GitHub<#2234?email_source=notifications&email_token=ADBYUCHFT27VXX2B4TQLQ6T4SLSTNA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRSHE4DMNZSUZZGKYLTN5XKOY3PNVWWK3TUUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16298672>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/ADBYUCCEWPVDKWGOCOCACXL4SLSTNAVCNFSM6AAAAACW4SPPQCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMMRZHA3DOMQ>. You are receiving this because you commented.Message ID: ***@***.***>

[mcgratta]

I am not surprised. I chose new locations to sample the near-ceiling temperatures. All of these locations are outside of the sprinkler spray. Within the spray zone, neither the model nor the measurements can be trusted. I'd rather see how well the model handles the overall cooling of the jet and upper layer. BTW, Test 11 of the UL/NFPRF series did not involve active sprinklers, so you can see how the temperatures in that case rise above the 200 C. I set the RTI in this case to a fictitiously high value so that it would still collect gas temperature. Take a look at the UL/NFPRF plots in the appendix of the Validation Guide to see if anything looks amiss.

[mcgratta]

I committed new versions of the UL_NFPRF input files to the CFAST GitHub repository. I also changed the measurement locations for the ceiling jet comparisons. Now all four of the chosen locations are outside of the sprinkler spray. I also corrected Test 11 so that it no longer has sprinklers that activate.

@sprankleray, are you able to access the latest files?

@gforney @rpeacoc, is there a publicly-facing "nightly build for CFAST?

[gforney]

yes. https://github.com/firemodels/test_bundles/releases/tag/CFAST_TEST

(fds, smokeview and cfast nightly bundles are below https://github.com/firemodels/test_bundles/releases/ )

there is not a cfast bundle there at the moment. there was a problem with the bundle. after I fix it I'll post a cfast bundle from the latest cfastbot pass

[sprankleray]

I am able to access the new files and I see that you re-arranged the sprinkler head labels. However, we’re trying to recreate the plot on page 344 of the V&V Guide. The text says the locations chosen are 56, 68, 86, and 98 which corresponds to sprinkler heads 7, ?, 31, and 43, respectively in the previous input file. We assumed these sprinklers were the ones plotted but the results didn’t match. We were able to recreate the plot but only by using sprinklers 11, 34, 36, and 49 (previous input file). Re-numbering the sprinklers as you’ve just done, only adjusts which 4 sprinklers we find to match the plot presented on page 344. Using the re-numbered input file, we find the 4 sprinklers that match the plot are 2, 38, 40, and 43. Ultimately, we need to be able to recreate the plot. Also, we’re using the surrounding gas temperature data in the devices.csv output file, which seems the appropriate parameter based on the description in the V&V Guide. If the plot is based on a different parameter (perhaps something related to the ceiling jet), then please identify how those plots were generated. A legend would help greatly. Thanks Ray On Mar 25, 2026, at 11:39 AM, Kevin McGrattan ***@***.***> wrote:  I committed new versions of the UL_NFPRF input files to the CFAST GitHub repository. I also changed the measurement locations for the ceiling jet comparisons. Now all four of the chosen locations are outside of the sprinkler spray. I also corrected Test 11 so that it no longer has sprinklers that activate. @sprankleray<https://github.com/sprankleray>, are you able to access the latest files? @gforney<https://github.com/gforney> @rpeacoc<https://github.com/rpeacoc>, is there a publicly-facing "nightly build for CFAST? — Reply to this email directly, view it on GitHub<#2234?email_source=notifications&email_token=AFZGHI45Q6DOQOB6VSR5YJL4SP4TDA5CNFSNUABIM5UWIORPF5TWS5BNNB2WEL2ENFZWG5LTONUW63SDN5WW2ZLOOQXTCNRTGEYTKOJSUZZGKYLTN5XKO3LFNZ2GS33OUVSXMZLOOSWGM33PORSXEX3DNRUWG2Y#discussioncomment-16311592>, or unsubscribe<https://github.com/notifications/unsubscribe-auth/AFZGHI53HTKVTKVWYNKBKSD4SP4TDAVCNFSM6AAAAACW4SPPQCVHI2DSMVQWIX3LMV43URDJONRXK43TNFXW4Q3PNVWWK3TUHMYTMMZRGE2TSMQ>. You are receiving this because you were mentioned.Message ID: ***@***.***>

[mcgratta]

The original CFAST input files for this case were set up by a collaborative partner for a V&V study with the US Nuclear Regulatory Commission. Rather than try to figure out what was done, I decided to choose new sprinkler locations away from the spray. Neither CFAST nor the experimental measurements are valid in the spray.

Why do you need to match those plots exactly?

[gforney]

yes. nightly cfast bundles are here: https://github.com/firemodels/test_bundles/releases/tag/CFAST_TEST

(fds, smokeview and cfast nightly bundles are below this url: https://github.com/firemodels/test_bundles/releases/ )

there is not a cfast bundle there at the moment. there was a problem with the bundle. after I fix it I'll post a cfast bundle from the latest cfastbot pass

[sprankleray]

Kevin, We are going through a V&V process for using the current version of CFAST for regulatory purposes. One aspect of this process ensures that we can re-validate certain often-used parameters, such as sprinkler activation. Our approach is to select a few cases where NIST validated the parameter and ensure we can reproduce the same output data. Section C.18 of the V&V Guide addresses the ability of CFAST to model sprinkler activation. That is the only parameter we’re addressing there and we’re not assessing anything after sprinkler activation. The only output data available is the plots in Appendix C of the Guide which is described in the text preceding the plots on page 344. The text says the “thermocouple temperatures can be compared to ceiling jet temperature predictions,” and the CFAST User’s Guide says the Sensor Surrounding Gas Temperature is the ceiling jet temperature at the device location. So, we’re plotting Sensor Surrounding Gas Temperature data. There are 49 locations, only 4 of which are presented. Our problem is that the sensor locations you say you’ve chosen do not appear to be those presented in the plot. Again, we were able to re-produce your plots, but at different locations than those identified in the text. Since the plots do not include a legend, we are seeking some affirmation that CFAST is performing properly; We’re hopefully assuming that the reason the text and the plots differ is a “problem” in the text (a typo) and not a problem in the model. Can you confirm which 4 locations, from the new input files, are plotted in the diagram for Series 1, Test 4. Message ID: ***@***.***>

[mcgratta]

Pages from CFAST_Validation_Guide.pdf

This is what the new section in the CFAST Validation Guide is going to look like. The numbering and coloring schemes are written at the top. The confusion is caused by the different way that the TCs and sprinklers were numbered. In the CFAST input files, the four measurement locations were specified via the sprinkler numbers, which were all mixed up. Rather than try to sort everything out, I renumbered the sprinklers properly, and then I decided to choose better locations for comparison. During the original V&V study 15 years ago, the points were chosen for all models, zone and CFD. In hindsight, the choice was bad, so I'll take this opportunity to improve things. The measured temperatures in the spray cannot be trusted because the TC is wet to some extent. How much is hard to say.

One thing you should think about---CFAST does not have an algorithm to cool the upper layer or ceiling jet. Rather, CFAST is just decreasing the fire's HRR based on a correlation that predicts exponential decrease based on the sprinkler flow rate. In these experiments, the fire was a heptane spray. It's HRR did not decrease appreciably because of the spray, or let's just say that any decrease in HRR was not measured. Rather, the sprinkler spray(s) cooled the upper layer gases, as you can see in the plots. CFAST and the measured temperatures show the same basic trends, but for different physical reasons.

[sprankleray]

Excellent!! This resolves all our issues with the validation documentation on sprinkler activation. I didn't expect you'd alter all the test input files and you've also revised the empirical test data plots, as well as the text in the V&V Guide. I will use this post to document our results. Thank you very much for the responsiveness and assistance.
Ray

[mcgratta]

It is not unusual to find mistakes like this. Organizing all these validation cases is difficult, especially for someone who was not involved in the experiments. I was, and I wrote the report, which is why I still remember this after 30 years. My one mistake was not to mention in the report (at least I could not find it) how the sprinklers were numbered.