.png)
By Nils Johansson, Christian Pelo, Johan Anderson, Robert McNamee
The Swedish Chapter of SFPE has carried out a “round robin” in the field of CFD modelling. Nine fire engineers performed the same simulation task with FDS. The results shows a relatively large spread, which can mainly be explained by the participants making different engineering choices. This highlights the importance of good education and guidelines for engineers who perform simulations with models like FDS.
In recent years, different comparisons, so-called round robins, have been conducted to study the variation in results from individual users of complex software. Such comparative studies show that, in many cases, the results are far from equal, even when participants receive identical instructions have been given and use the same software.1-2 Based on this, the Swedish Chapter of SFPE (locally called BIV) initiated a new study to investigate how much engineering choices are involved in modelling with the CFD model Fire Dynamics Simulator (FDS). The chapter chose FDS for the study because it is widely used in performance-based fire safety design and has a large number of parameters and inputs that can be adjusted.
A Previous Round Robin Study with FDS
In 2015, a student conducted a round robin for a thesis at Lund University, Sweden.
2 Eight experienced fire engineers each conducted a simulation of the same scenario in FDS. The scenario was based on an actual fire experiment, but the participants performed the task a priori, i.e., without having access to the test results.
The participants received a description of the scenario, the type of fuel and its mass loss level, but were responsible for defining appropriate inputs based on the description. The participants had at least four years of experience of working with FDS and were all graduated fire engineers with a university degree.
Although the participants were given relatively detailed information about the setup and environmental conditions, the difference in estimated temperature increase was more than 50%. There were relatively large differences in how the fuel and burner were described in FDS, which resulted in major differences in the resulting heat release rate (HRR). In addition, several of the participants made minor mistakes when setting up their input files.
The study was followed up with a questionnaire to the participants, and the general impression from them was that the task did not relate directly to what they do in their profession. A natural continuation of the work in the thesis was, therefore, to carry out further round robin studies with scenarios that are more similar to those in performance-based design.
Participants in the Study
Participation in the study was on a pro bono basis, with a total of nine participants from eight companies active in the fire consultancy industry. Participants were recruited mainly through established channels (e.g., mail/newsletters, as well as information on a website) and direct contact with potentially interested parties. All participants (individuals as well as firms) were treated anonymously, because the purpose of the study was to get a picture of how these calculations are performed in the industry, not to compare individual actors’ results against each other.
The participants had worked in the consulting industry for one to five years with a mean of three years and had all been used CFD modelling during their studies. Furthermore, all the participants had degrees in fire protection engineering.
The Simulation Tasks
The nine participants were asked to perform two simulation tasks.
The first simulation task (Case 1) was a simple warehouse fitted with an automatic water sprinkler system, where the time for activation had to be calculated. The fire was specified with a value for the fire growth and a maximum HRR. The participants were asked to take the sprinkler system into account and to present the gas temperature at a couple of locations in the enclosure.
The second simulation task (Case 2) was an enclosure with a more-complex geometry: a theatre on two levels (see Figure 1). In this case, technical systems were not in focus; instead, the study sought the time to critical conditions (ASET) in the room. There are clear criteria for critical conditions in the Swedish building code that are related to thermal radiation, maximum temperature and visibility. The parameter that becomes critical first is usually visibility.
Results
Case 1The calculated time to sprinkler activation is shown in Figure 2. The difference is more than 110 seconds between the longest and shortest calculated time. The large variation is due to having only a few participants; the majority of the participants were within a 20 second interval. Participants F and H had the shortest activation times, which is probably due to the fact that they both used a function in FDS to calculate sprinkler activation.
The other participants used a different computer program (DetactT2) that gives longer, and in this case, more-conservative times to activation. In addition, participant G had a short activation time, because the HRR specified by that participant was slightly faster than the value (0.012 kW/s2) specified in the instructions (see Figure 3).
The HRR used by the participants are displayed in Figure 3. Participants F and G both used the “Spread Rate“ function in FDS and a slightly higher value than that corresponding to the required growth rate of 0.012 kW/s2. This resulted in an approximately 20% higher HRR after 200 seconds for participants F and G.
These participants did this to compensate for the fact that the fire is not circular in FDS. This procedure is recommended in a Swedish best practices guide for CFD simulations [3]. However, the assumption becomes non-conservative when sprinkler activation is part of the analysis. An alternative to the “Spread Rate“ function is to let the fire grow in intensity but maintain its area; this means that in the early stages of the simulations, the fire has an intensity less representative of a real fire.
The ratio between the characteristic fire diameter and the control volume size (D*/dx) can be used to describe the resolution of the fire in a FDS simulation. Four of the participants were below the recommended value of D*/dx and thus did not use a resolution in line with recommendations. However, no systematic difference in the results can be attributed to the poorer resolution.
Case 2
All participants found that the time to critical conditions (ASET) was less than the time to evacuate the room (RSET). All participants also came to the conclusion that critical conditions would first arise in the theatre stands. Several participants indicated that the visibility was the criterion deemed to be most relevant to study in this case. However, there is a fairly wide spread in the results for visibility, and the difference between the shortest and longest time to 10 m visibility was 60 seconds (see Figure 4).
In Case 2, the participants used a fire scenario (Scenario 1) that is specified in the Swedish Building Code,4 which means a fire with a growth rate of 0.047 kW/s2 and HRR of 10 MW. All except participant F modelled only a growing fire, meaning they did not limit it when 10 MW was reached, probably because they have previous knowledge that critical conditions in similar cases occur before the maximum HRR is reached.
The participants used different ways of representing the fire, but a majority used the Spread Rate function as recommended.3 The HRR used by participants in FDS are shown in Figure 5.
Participant G has used a slightly higher value of Spread Rate that corresponds to a growth rate of 0.047 kW/s2; this means that the HRR was overestimated by 30–40% after 150 seconds. This overestimation is conservative; participant G appears to be aware of this and made it to compensate for the fact that it is not possible to model circular fires in FDS. However, since critical conditions arise before the maximum HRR is reached, this compensation is not considered necessary.
There was a big difference between the participants in the number of cells used, which depends in part on the mesh size, but also the volume (domain) simulated. The resolution was generally better than in Case 1 and all participants are above the lower limit of D*/dx that is as recommended in the Swedish guidelines developed by the local SPFE chapter.3
Conclusion
FDS is a powerful tool that can be used to make advanced and detailed calculations. However, it requires the user to have good knowledge of the program and its features. It is, of course, important to have good knowledge of fire dynamics to be able to choose the most-appropriate way to model the fire process and to interpret the results.
It is important to emphasize that small variations are, in principle, unavoidable in fire engineering design — and not necessarily a problem. Different fire safety modellers can model things differently without necessarily being wrong. However, differences in how the fire is modelled (e.g., growth rate, maximum HRR and area) can have a substantial effect on the results. The question is how big a difference can be accepted.
It is, of course, reasonable to carry out sensitivity analyses, but given the various possibilities available for choices in FDS, the amount of such analyses can quickly become overwhelming. An option is to make conservative assessments consistently and use the most-conservative choices when modelling in FDS. The downside of that option is that it might lead to an unnecessarily high safety margin.
The building code and the guidelines for CFD modelling provide good support for fire safety calculations in Sweden. Without this kind of advice and guidance, it is likely that the variation in this round robin would have been significantly greater.
After the participants completed the simulation task, they filled out a questionnaire. One question was: “Do you think that the task differs greatly [for] typical fire simulation in FDS at your consultancy firm?” A majority responded that it did not. Those who answered yes said that Case 1 differed from what they normally simulate. This indicates that Case 2 resembles an actual task that fire safety consultants might face and that the external validity of the variation found in this case is good.
The participants were also asked about their knowledge about FDS. A majority considered themselves as possessing a high level of knowledge about FDS and a higher level relative to other fire engineers who work with FDS. However, in regard to the HRR curves used by some of the participants, especially in Case 1, the consultants should be self-critical about their own level of knowledge about the software to ensure that the HRR curves used in the model is consistent with what is desired.
BIV intends to continue to hold more round robins to increase knowledge about engineering choices that often have to be made in connection with fire simulations and future round robins are planned.
Nils Johansson is at Lund University. Christian Pelo is with Ramboll Sweden. Johan Anderson is with RISE. Robert McNamee is president of the Swedish SFPE Chapter and is with Brandskyddslaget.
References
1Lange, D., and Boström, L. 2017. A round robin study on modeling the fire resistance of a loaded steel beam. Fire Safety Journal, 92:64–76. DOI: 10.1016/j.firesaf.2017.05.013.
2Johansson, N., and Ekholm, M. 2018. Variation in Results Due to User Effects in a Simulation with FDS. Fire Technology 54(1), 97–116. DOI: 10.1007/s10694-017-0674-y.
3BIV's Application Document 2/2013. CFD calculations with FDS [in Swedish].
4Swedish National Board of Housing, Building and Planning. 2013. General recommendations on the analytical design of a building’s fire protection. Karlskrona: BBRAD. https://www.boverket.se/globalassets/publikationer/dokument/2013/bbrad-bfs-2011-27-tom-2013-12-english.pdf.