Beruflich Dokumente
Kultur Dokumente
art ic l e i nf o a b s t r a c t
Available online 4 February 2014 In part I of the paper, an assembly building was analyzed in order to compute the failure probabilities,
Keywords: and thus the safety level, of current code-compliant buildings. In this second part, various fire protection
Performance-based design systems are modeled within the fire and egress simulations in order to quantify their magnitude of
Life safety impact. Since all fire protection systems can fail to perform as designed on demand, the potential failure
ASET–RSET along with its probability is accounted for in an event tree analysis. Comparing the resulting failure
Probabilistic risk assessment probabilities of the performance-based analyses with and without fire protection systems yields
Event tree information about the magnitude of impact of the fire protection systems on the level of safety and
Sensitivity analysis hence allows a direct, objective, and quantitative comparison to other systems and designs. Accounting
for the cost of the systems, a direct cost–benefit analysis can be conducted.
& 2014 Elsevier Ltd. All rights reserved.
0379-7112/$ - see front matter & 2014 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.firesaf.2014.01.002
82 C. Albrecht / Fire Safety Journal 64 (2014) 81–86
Table 2
Results of the system analysis for a detection system.
Detection system ps=f ; detec: sys: (%) p^ f ; scenario (%) p^ f ;s jp^ f ;f (%) p^ f ; sys: (%)
Table 1
Results of the reliability analyses and sensitivities for the three predominant
variables for the smoke detection. system, this is an improvement of the safety level by a factor of
approximately 2.6 for both criteria.
Scenario α^ tg α^ num: occ: α^ premove: p^ f (%)
An even greater impact can be observed for the hidden fire
Standard with detection, Opt. 0.7 0.4 0.5 21.4 scenario: also considering the 90% reliability, the p^ f ;f becomes 3.5%
Standard w/o detection, Opt. 0.6 0.5 0.6 68.2 for the optical density threshold and 0.9% for the FED criterion.
Standard with detection, FED 0.9 0.2 0.2 1.7
The p^ f ;s 's are 10% and 4%. The total probabilities considering a 10%
Standard w/o detection, FED 0.9 0.2 0.4 5.4 failure of the alarm system yield 13.4% for the optical density and
4.9% for the FID. Comparing these values to the failure probabilities
Hidden with detection, Opt. 0.6 0.4 0.6 3.9
Hidden w/o detection, Opt. 0.6 0.6 0.6 99.2 without an alarm system installed results into an improvement
of the safety level by a factor of around 7-8 times for both criteria.
Hidden with detection, FED 0.9 0.2 0.2 1.0
Hidden w/o detection, FED 0.7 0.4 0.5 39.9
All results are shown in Table 2.
Concluding, an alarm system2 is a very efficient way to shorten
the alarm and pre-movement times which provides significantly
amounts of computational resources. “Recycling” in this case more time for the occupants until untenable conditions are
means that the initial support points from the CFD fire simulations reached. Being alarmed at an early stage of the fire also reduces
can be used again as only the support points within the evacuation the danger of impassable exit routes. The highest impact can be
simulation have changed. For the subsequent iteration steps observed for fires that otherwise would grow to a fully developed
within the solving process, new support point has to be calculated, stage undetected.
of course, for both fire and evacuation simulation. To analyze the
influence of a smoke detection and alarm system, the standard fire
scenario was analyzed for the comparison with the other systems. 3. Smoke and heat exhaustion
Additionally, the hidden fire scenario was also analyzed as an
early detection will have a very large effect here. Even though it Smoke and heat exhaustion (SHE) systems have the effect that
was found (see part I) that the smoke detectors actuate after the harmful impacts of the fire on the occupants are delayed and/
around 40 s on average, only the stochastic model for the pre- or mitigated due to the extraction of smoke and heat. Hence, SHEs
movement time was adjusted to 90/25 s in order to remain aim at the ASET part of the limit state. For the consideration in the
conservative and to retain comparability. The reliability of a analyses, the SHE system has to be included into the fire simula-
commercial smoke detection system is chosen to be 90% according tion. The reliability of an SHE system to work as designed on
to BS 7974 [2]. The results are summarized in Table 1. demand is given with also 90% in the BS 7974 [2].
It should be noted that for the FED reliability analyses, the To show the effect, a total of eight 1 m2 natural in-roof
algorithm has to shift the input for the fire development time to SHE vents were added to the simulations. This is far more than
reach 1 MW t g to approximately 10 s to obtain the design point required by the German assembly building codes [4], but as the
(ASET ¼ RSET).1 Such short times for the fire growth phase may be building serves as a multi-purpose assembly room they might also
mathematically correct but are unrealistic for the actual fire be used for natural lighting during the daytime. The vents have an
development. Hence, the “real” values for the failure probabilities assumed opening time of 15 s after the first smoke detector
are likely to be far less conservative. The formulation of other fire actuates. The fire scenario was chosen to be the standard scenario
scenarios and better (stochastic) modeling may be the subject of while two cases were assumed. In the first case, the detection
further research work. system is only installed to open the vents but not to alarm the
Analyzing the calculated failure probabilities with the event occupants, whereas in the second case, a full alarm is issued upon
tree approach shown in Fig. 1 and assuming a 90% reliability (or a actuation so that the pre-movement time is also reduced (see
10% probability of failure) of the smoke detection and alarm above). The results for the optical density threshold are compiled
system to work as designed on demand leads to p^ f ;s ¼ 19:3% and in Table 3.
p^ f ;f ¼ 6:8% for the standard scenario using the optical density The FED criterion was not reached for any of the calculated
threshold and p^ f ;s ¼ 1:6% and p^ f ;f ¼ 0:5% for the FED criteria. support points until the fire simulations were terminated at 1200 s
Hence, the total probabilities of failure including the alarm system (20 min), which was defined as the absolute maximum time3 in
and its possible malfunction are 26% for the optical density and 2% which a fire compartment has to be fully evacuated, regardless of
for the FED criterion. Compared to the case without an alarm the fire effects. This means that the SHE system herein mitigates
1 2
The shift herein describes the iterative search of the optimal input para- It is considered in this paper that an alarm system is effective in alarming the
meters in the algorithm. In this case, this means ASET can theoretically only be occupants. Unfortunately, many conventional alarm systems fail to accomplish this
equal to RSET if t g is close to 10 s, which is a theoretical value (mathematically goal. The need for further research in proper and effective alarming of the
correct but not practically applicable) since reaching 1 MW in 10 s implies an occupants is underlined by the outcome of the analyses herein.
3
extremely fast, deflagration-like fire growth. This also constitutes a serviceability limit for the normal use of the building.
C. Albrecht / Fire Safety Journal 64 (2014) 81–86 83
Table 3
Results of the reliability analyses and sensitivities for the three predominant
variables for the SHE cases using the optical density threshold.
Table 4
Results of the system analysis for a SHE system.
Fig. 2. Modeling of the SHE system with a coupled alarm system.
SHE system ps=f ; detec: sys: (%) p^ f ; scenario (%) p^ f ;s jp^ f ;f (%) p^ f ; sys: (%)
Table 5 Table 7
Results of the reliability analyses and sensitivities for the three predominant Results of the reliability analyses and sensitivities for the three predominant
variables for the sprinklered standard scenario. variables for the sprinklered ultra-fast scenario.
Scenario α^ t g α^ num: occ: α^ premove: p^ f (%) Scenario α^ tg α^ num: occ: α^ premove: p^ f (%)
Sprinkler, Opt. 0.6 0.5 0.6 67.9 Sprinklered, ultra-fast, Opt. 0.6 0.6 0.6 99.5
Standard w/o sprinkler, Opt. 0.6 0.5 0.6 68.2 Un-sprinklered, ultra-fast, Opt. 0.6 0.6 0.6 99.5
Sprinkler, FED 0.8 0.2 0.5 0.015 Sprinklered, ultra-fast, FED 0.4 0.3 0.6 0.9
Standard w/o sprinkler, FED 0.9 0.2 0.4 5.4 Un-sprinklered, ultra-fast, FED 0.7 0.3 0.6 14.3
Table 6 once again, that the results are very specific for the building and
Results of the system analysis for a sprinkler system. the fire scenarios considered and, therefore, cannot be generalized.
Sprinkler system ps=f ; detec: sys: (%) p^ f ; scenario (%) p^ f ;s jp^ f ;f (%) p^ f ; sys: (%)
5. Organizational fire protection and trained personnel
Standard scenario 98 0.015 0.015
FED criterion 2 5.4 0.11 0.12
Organizational fire protection and trained personnel may have
the biggest impact on the system reliability as it can affect
of failure is significantly lower than for the un-sprinklered sce- multiple scenarios and parameters within the fire protection and
nario. As the threshold is reached later ðt 4 1:7 t g Þ and is cumu- life safety system. Yet the effect is usually implicit and thus very
lative and time-integrated, the lower release of heat and hard to model or quantify as no data is currently available.
asphyxiant gases directly affects the ASET. Looking closer at the Additionally, the organizational fire protection includes various
results, one might even argue again whether the scenario is prone aspects of which some may be complied with (e.g. presence of
to failure at all, as the design point of the convolution (see above) evacuation maps) while others are neglected completely (e.g.
pushes t g near zero, which is mathematically correct but not propping smoke doors open). Some of the imaginable effects of
necessarily physically realistic. As for the sensitivities, t g , the pre- highly trained staff or even dedicated fireguard personnel within a
movement time and the number of occupants have a high holistic organizational fire protection concept include, but are not
influence. Additionally, the HCN-yield has a significant impact on limited to:
the results ðα^ yHCN ¼ 0:3Þ based on the stochastic model, which, as
mentioned before, is based on literature values and engineering Effect on the scenarios:
judgment with conservative assumptions. ○ Reduction of the fire initiation frequency due to proper
A system analysis was performed to demonstrate the impact of maintenance of, for example, electric installations.
a sprinkler system, taking into account the possibility of a failure ○ Lower failure probabilities due to well-maintained fire
of 2% thereof. There is no evident impact on the optical density for protection systems.
the reasons stated above. For the FED criterion, the total prob- ○ Higher probability of success of the manual intervention, as
ability of failure is 0.12% and thus about 44 times lower than the the staff is familiar with the use of fire extinguishers and
un-sprinklered scenario with a 5.4% probability of failure as shown responds more quickly to a fire.
in Table 6. Considering the reason stated above, the relative level ○ Lower probability of severe scenarios like the ultra-fast fire
of safety is even higher implicitly. due to the use of inflammable (temporary) decoration and
the prohibition of firework usage.
4.1. Influence of a sprinkler system on the ultra-fast fire ○ Faster egress due to evacuation management (e.g. live voice
communication instead of alarm sound or taped messages).
As mentioned above, the sprinklers did not have a large impact Effect on the parameters used:
when assessing the standard fire for optical density, as they ○ Reduction of fire load.
usually actuate later compared to the other systems. Madrzy- ○ Use of inflammable materials or retardants for slower fire
kowski et al. [6] conclude from a comparison of sprinklered and development, changing t g .
un-sprinklered experiments and simulations that a sprinkler could ○ Limiting the number of occupants by strictly enforcing
have prevented the Station Fire disaster where an ultra fast fire maximum allowable capacities.
occurred. Hence, the ultra-fast scenario from above was re- ○ Reducing the warning and/or pre-movement times.
evaluated together with the sprinkler model introduced. The
results are compiled in Table 7. It can be seen again that the As absolutely no data to quantify the effects of well-trained
sprinklers do not have any significant influence on the visibility personnel could be found, no direct influences are considered
(optical density threshold) as this criterion is reached shortly after herein. The detailed modeling of these effects constitutes an
the fire starts. Yet a significant impact can be seen for the FED interesting topic for further research. Omitting the influence
criterion, which is nearly completely mitigated due to the reduc- entirely penalizes those building operators who invest into orga-
tion of the maximum heat release rate and the early termination nizational fire protection. Hence, a very simplistic approach shall
of the fire. It should be noted that herein, the fire was assumed to be assumed: if the organizational fire protection is considered
start at floor level which conservatively implies a comparatively appropriate within a building, the reliability of manual interven-
late actuation. Also, as stated above, the sprinkler model (plateau tion is assumed to increase to, for example, 75% instead of 25%
and linear decay) can be considered conservative. with low organizational consideration. This has a significant effect,
In conclusion, the sprinkler system does have a very large as the manual intervention is a critical early decision gate within
effect, as the possibility of occupants being severely threatened of the event tree. This means that the success of the manual
incapacitation is mitigated. Yet a risk of delaying the safe egress intervention ultimately leads to a safe state where no hazard is
and thus inflicting injuries due to smoke inhalation could not be imposed on the occupants. Hence, pf ;s always leads to a zero
prevented according to the models utilized. It should be noted, probability of failure so that pf ;f fully controls the sub-tree but is
C. Albrecht / Fire Safety Journal 64 (2014) 81–86 85
Fig. 4. The complete event tree considering all fire protection systems.
Table 8 The result of the interaction case is shown in Table 8 for the
Results of the reliability analyses and sensitivities for the three predominant optical density, along with some comparative values. Interestingly,
variables for the sprinkler–SHE-interaction standard scenario with alarm compared the effect of the sprinkler is now significant despite having no
to the other cases. effect without the SHE vents. This is most certainly due to the fact
Scenario α^ tg α^ num: α^ premove: p^ f
that the SHE system delays the smoke filling of the room by
occ:
pushing the design point forward in time to a degree that the
Sprinkler–SHE-interaction 0.9 0.2 0.2 3.1 sprinkler system can now affect the optical density by cutting the
SHE w/ alarm, Opt. 0.9 0.2 0.2 5.4 HRR. This shows a significant interdependency of fire protection
Sprinkler, Opt. 0.6 0.5 0.6 68.0 systems and the necessity to appropriately model those effects.
Standard w/o sprinkler or SHE, Opt. 0.6 0.5 0.6 68.2
The FED threshold was not reached in any of the simulations.
Hence, the toxic and heat effects are completely mitigated beyond
directly dependent on the protection system reliability. Thus, the simulation capability for the standard scenario.
failure probability of the protection system acts as a plain multi- If all possible interactions are considered within the event tree,
plier for pf ;f . simple multiplication and addition can be used to evaluate the
A simple example shows the influence of the organizational system: all branches are independent minimal cut-sets of the
fire protection: assuming a 75% failure probability of the manual system and interaction (correlation) effects are modeled within
intervention for the standard scenario using the FED threshold the scenario. The total probability of failure of the system for the
ðp^ f ¼ 5:4%Þ leads to a total system probability of failure of 4%, standard fire scenario using the optical density threshold is
while a 25% failure probability obviously leads to only 1.35% in a pf ; sys: ¼ 0:00157.
system analysis (implying that a good organizational fire protec-
tion increases the safety level by a factor of three for the particular
case).
7. Conclusions and outlook
even allows for a cost–benefit analysis. This ultimately leads to Appendix A. Calculations
more economic fire protection design without sacrificing safety.
Yet it was also found that the results are highly dependent on In the following, the simplified equations for the analyses are
the parameters, scenarios and models used, foremost as the input given for the four scenarios with their respective conditional
data statistics and thus the stochastic models are far from being probability of failure.
sufficiently accurate. Especially for the sprinkler system a different
model will lead to very different results, i.e. even a shorter A.1. Failure probability considering a system and its failure
actuation time would significantly increase the safety level. Hence,
a lot of research is needed in this area in order to sufficiently In order to compute the effect of a protection system, its
account for the systems used. The examples shown herein can be effectiveness needs to be calculated considering that the system
regarded as a first step towards the explicit quantification of the can fail with a given failure probability. It is assumed herein that in
impact of fire protection systems. case of a failure of a protection system, the scenario defaults to the
scenario without the protection system installed. Partial failure
7.1. A semi-probabilistic life safety concept such as, for example, working but under-designed systems are
disregarded herein and constitute interesting further research.
Currently, the application of performance-based life safety In case of the detection system for the optical density, for
concepts is rather arbitrary when it comes to the input data and example, considering a 10% failure probability of the system and a
the safety concept chosen. The classic deterministic approach 21% failure probability if the system is working and the 68% failure
ignores the high level of uncertainties, especially in the input probability of the system is not working, respectively, the overall
parameters and thus only allows for a calculation where the level failure probability of the system considering its possible failure can
of conservativeness remains unquantified. Adding a safety factor be calculated to
κ into the life safety equation ðASET 4 κ RSETÞ does add some Probability of failure in case the system is working:
safety, yet the actual improvement also remains unquantified, so p^ f ¼ 0:9 0:21 ¼ 0:19 ðA:1Þ
that the concept was even considered flawed [11]. In addition, a
Probability of failure in case the system is not working:
direct and objective comparison to the deemed-to-satisfy pre-
scriptive codes is not possible. p^ f ¼ 0:1 0:68 ¼ 0:068 ðA:2Þ
Using the methodology above allows for the quantification but
And the total probability then yields
would even allow the derivation of a semi-probabilistic safety
concept that implicitly ensures the acceptable level of safety by p^ f ; total ¼ 0:19 þ 0:068 ¼ 0:26: ðA:3Þ
using safety factors γ i for the most influential parameters. These
The factor at which the building is safer compared to the case
safety factors can be based on, for example on a 90-quantile of
without the system is a simple comparison of the probability
their underlying distribution, so that for a normal distributed
without the system compared to the probability with the system
parameter
considering the system's potential failure, so that
1
1 þ αi Φ ðpf ;req: Þ V i 0:68
γi ¼ ; ð1Þ κ system ¼ ¼ 2:6: ðA:4Þ
1 þ 1:28 V i 0:26
1
where αi is the sensitivity, Φ ðÞ the inverse normal distribution,
pf ;req: the required maximal allowable failure probability, and V i References
the coefficient of variation of the underlying normal distribution.
Currently, pf ;req: is unknown and it was found that the resulting [1] G. Ramachandran, The Economics of Fire Protection, E & FN Spon; Routledge,
failure probabilities are highly dependent on the chosen models, London, UK; New York, NY, 1998.
[2] BS 7974, Application of Fire Safety Engineering Principles to the Design of
scenarios and input parameters with their according distributions. Buildings. Code of Practice, Technical Report, British Standards Institution
Thus, pf ;req: has to be derived from repetitive calculations of various (BSI), 2001.
building occupancies, with defined models, scenarios, thresholds, [3] C. Albrecht, D. Hosser, A response surface methodology for probabilistic life
safety analysis using advanced fire engineering tools, Fire Saf. Sci. 10 (2011)
and parameters in order to calibrate mean level of safety. A safety 1059–1072, http://dx.doi.org/10.3801/IAFSS.FSS.10-1059.
concept then has to prescribe these exact features and provide [4] MVStättV, Musterverordnung über den Bau und Betrieb von Versammlungs-
the safety factors in a way similar to the approach outlined by stätten (Muster-Versammlungsstättenverordnung - MVStättV), Technical
Report, IS-Argebau, 2005.
Fleischmann [12]. [5] E. Rosenbaum, et al., SFPE Engineering Guide to Performance-Based Fire
Providing a basis for risk-informed design using performance- Protection, 2nd ed., National Fire Protection Association (NFPA) and Society
based methods by prescribing various input parameters would of Fire Protection Engineers (SFPE), 2007.
[6] D. Madrzykowski, N.P. Bryner, S. Kerber, The NIST Station Nightclub Fire
lead to performance-based designs with the same objectively investigation: physical simulation of the fire, Fire Prot. Eng. 31 (2006) 34–46.
proven safety level as prescriptive codes. This improves the [7] N.P. Bryner, D. Madrzykowski, W.L. Grosshandler, Reconstructing the Station
acceptance and constitutes the performance-based methods as Nightclub Fire: computer modeling of the fire growth and spread, in:
Proceedings of the 11th International Interflam Conference (INTERFLAM
real and equal alternative to prescriptive design. In addition, it
2007), vol. 2, InterScience Communications, 2007, pp. 1181–1192.
would make the process comprehensible and transparent among [8] DIN EN 1991-1-2/NA, National Annex – Nationally Determined Parameters –
all parties. Eurocode 1 – Actions on Structures – Part 1–2/NA: General Actions – Actions
on Structures Exposed to Fire, Technical Report, Normenausschuss Bauwesen
(NABau) im DIN Deutsches Institut für Normung e.V., 2010.
[9] VdS CEA 4001, Spinkleranlagen, Planung und Einbau, Technical Report, VdS
Acknowledgments Schadensverhütung GmbH, November 2008.
[10] K. Frank, M. Spearpoint, N. Challands, Uncertainty in Estimating the Fire
Control Effectiveness of Sprinklers from New Zealand Fire Incident Reports,
This work was supported by a fellowship within the Postdoc- Fire Technology OnlineFirst, 2012, OnlineFirst. http://dx.doi.org/10.1007/
Program of the German Academic Exchange Service (DAAD). The s10694-012-0297-2.
author would like to thank Prof. Brian Meacham and the Depart- [11] V. Babrauskas, J.M. Fleming, B.D. Russell, RSET/ASET a flawed concept for fire
safety assessment, Fire Mater. 34 (2010) 341–355.
ment of Fire Protection Engineering at Worcester Polytechnic [12] C. Fleischmann, Is prescription the future of performance-based design? Fire
Institute for the support. Saf. Sci. 10 (2011) 77–94, http://dx.doi.org/10.3801/IAFSS.FSS.10-77.