See discussions, stats, and author profiles for this publication at: https://www.researchgate.
net/publication/341099710
Probabilistic assessment of off-the-shelf fire sprinkler Head activation time
through laboratory experiments
Conference Paper · April 2020
CITATIONS READS
0 57
3 authors, including:
Mohd Zahirasri Mohd Tohir Mohamad Syazarudin Md Said
Universiti Putra Malaysia Universiti Putra Malaysia
74 PUBLICATIONS 701 CITATIONS 45 PUBLICATIONS 474 CITATIONS
SEE PROFILE SEE PROFILE
All content following this page was uploaded by Mohd Zahirasri Mohd Tohir on 02 May 2020.
The user has requested enhancement of the downloaded file.
IOP Conference Series: Materials Science and Engineering
PAPER • OPEN ACCESS
Probabilistic assessment of off-the-shelf fire sprinkler Head activation
time through laboratory experiments
To cite this article: M A S Tajudin et al 2020 IOP Conf. Ser.: Mater. Sci. Eng. 778 012075
View the article online for updates and enhancements.
This content was downloaded from IP address 42.189.39.5 on 02/05/2020 at 10:37
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
Probabilistic assessment of off-the-shelf fire sprinkler Head
activation time through laboratory experiments
M A S Tajudin1, M Z M Tohir1* and M S M Said1
1
Safety Engineering Interest Group (SEIG), Department of Chemical and
Environmental Engineering, Faculty of Engineering, Universiti Putra Malaysia,
Serdang, Selangor, Malaysia.
Abstract. The activation time of a sprinkler has a great impact on the fire growth inside a
compartment. The longer time it takes for a sprinkler to activate, the higher the risk of fire to
start to spread inside the compartment. In this study, the activation times of off-the-shelf half
inch 68°C fire rated pendant sprinklers were assessed through experiments in laboratory settings
and eventually best-fit probabilistic curve for the Response Time Index (RTI) were generated
from the experimental results. The experiments were conducted in two sets; one set without
smoke effect and the other set with smoke effect to determine the activation time and temperature
of the sprinkler heads. The data recorded were then used to calculate the RTI and Palisade
@RISK statistical analysis software was used to analyze the best-fit probabilistic curve for the
RTIs. It was observed that, average time required for sprinkler to activate was 60 seconds. It was
also found that the calculated RTI were within the range of the international sprinkler design
guideline. However, the calculated RTI were found to be inconsistent with specified RTI from
manufacturer. Based on @RISK software, the two best-fit RTI probabilistic curve for the two
experiments were Gamma and BetaGeneral distribution curve. The probabilistic curves obtained
from this work can be used as input for fire in enclosure simulations.
1.0 Introduction
Fire protection system is an integral component which needs to be considered in designing a building.
Without fire protection system installed, the life safety of occupants inside the building are at high risk
in the event of fire. Thus, having proper active and passive fire protection systems inside buildings are
in principle will be able to provide a cost-savvy of risk mitigation to safety of the building occupants
and also to property destruction. One of the most common active fire protection systems used inside
buildings is fire sprinkler system. Fire sprinkler system is a network of pipes installed through ceiling
of a floor in a building holding water under pressure. The fire sprinkler head operates by detecting heat
through head sensitive element, commonly glass bulb or fusible link. Heat responsive liquid commonly
in the form of glycerin solution inside frangible glass bulb of fire sprinkler head is a heat sensitive
element. As the sprinkler head detects heat, the temperature of liquid in the bulb will rise and expand.
As the liquid temperature reach its rated temperature, the glass bulb will be shattered. Hence, the water
(sealed just below of glass bulb) start to discharge towards the deflector. Water will be distributed with
certain spray pattern depending on the design of deflector use. Water spray pattern play an important
role in controlling the fire. During real fire, the activation of sprinkler will provide extra time to Fire
Brigade to arrive while keeping fire under control [1].
Fire sprinkler head activation time plays a major role in determining the fire growth within the
compartment. The sprinkler activation time can be defined as the time at which the temperature of the
Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
sprinkler head reaches the nominal activation temperature [2]. The sprinkler activation correlates
directly to sprinkler sensitivity which depends on its thermal parameters known as Response Time Index
(RTI) and Conduction Factor (C) [3]. There has been a substantial amount of research in fire sprinkler
head activation in the past. Most of the research utilized plunge test and ramp test to determine sprinkler
sensitivity by simulating fire conditions in a wind tunnel. The plunge test was developed in the USA at
the Factory Mutual Research Corporation [4] and the principle of the test is to have air flowing past a
test section at selected air velocity and temperature that higher than sprinkler’s operating temperature.
Once the temperature inside test section reach equilibrium, the sprinkler head plunge into the hot air
flow until it activates. Due to the fact that full-scale fires are inconvenient to perform, most of the work
from literature were using computational modelling software to predict sprinkler responses.
Automatic fire sprinkler systems are designed to activate if a fire develops in the area of
protection by limit or suppress the further development of the fire. The knowledge of the effectiveness
of sprinkler systems in reducing the risk from fire is important evaluating a building design that
incorporates sprinkler for fire safety [5]. There has been debate as to how far the sprinkler systems is
effective since automatic sprinkler systems were originally invented and developed in the 1800s [6]. An
early reference to estimates of sprinkler effectiveness can be found in the Preliminary Report of the New
York State Factory Investigating Commission, which was released in 1912 following the Triangle
Shirtwaist fire. This report stated that the efficacy of sprinkler systems is varying with sprinkler proper
working limit in between 75% and 95% [7]. The effectiveness of a fire-safety system can be considered
as the product of its efficacy and its reliability [8]. Effectiveness of sprinklers can be determined through
statistical and other historical data. Efficacy is the degree to which a system achieves an objective given.
Different objective means different efficacy of a system.
Globally, there are several renowned manufacturers of fire sprinkler heads, namely Dyne, Tyco,
Globe, Viking, and Reliable; and then there are also smaller market companies which manufacture fire
sprinkler heads. The products from these companies are known to be of high quality and reliability based
on previous research by Frank et al. [5]. However, nowadays there are more and more unknown
manufacturers which are also producing fire sprinkler heads at a fraction of the price as compared to
what has been sold by the renowned companies. Even though, the cheap sprinkler heads are claimed to
be the same quality as those sold by the renowned companies, there is no guarantee that the sprinkler
heads are going to activate according to the specifications. The market distribution of these cheap
sprinkler heads are becoming much more extensive in Malaysia and are available off-the-shelf at many
DIY stores. This has led to a growing concern in Malaysia that construction companies are opting for
cheaper fire sprinkler heads while compromising the life safety of the building occupants.
Therefore, it is the aim of the study to conduct probabilistic assessment on the activation time
of off-the-shelf fire sprinkler heads from repetitive experiments. The main objective is divided into sub-
objectives as follows: (1) to measure activation time required for fire sprinkler head to activate with and
without smoke effect; (2) to assess on variability of sprinkler head Response Time Index (RTI); (3) to
identify appropriate distributions for experimental RTI values.
This research study could provide information on the reliability of the fire sprinkler heads
available in the market is reliable. Besides, the research outcome will be valuable for probabilistic study
of fire growth dynamics using computational modelling. In the end, the findings would be useful as a
baseline study to improve current local codes and standards regarding designing, installation and
inspection of fire sprinkler system, specifically in Malaysia.
2.0 Methodology
The research was performed through laboratory experiments in Combustion Laboratory at the Faculty
of Engineering, Universiti Putra Malaysia. The main principle of the experiments were to have hot air
flowing through a test section at a selected velocity and air temperature [9]. The specifications of the
fire sprinkler head selected was a standard type sprinkler head (pendant type) with K-factor of 80.6
LPM/bar1/2, temperature rating of 68⁰C with standard coverage of 20.9 m2 (specifically for ordinary
2
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
hazard). There were a total of 68 fire sprinkler heads have been used for the two sets of experiments and
the sprinkler heads were bought at a random hardware store. Both sets of experiments use a fixed value
of hot air velocity (1 m/s) with thermal heat flux of 300 kW/m² except set number two had additional of
smoke effect. Smoke had been generated by a fog machine (Antari) while the hot air was generated by
a heat gun (AEG Powertherm 560). The sprinkler head plunged with perpendicular orientation to the
hot air flow inside metal chamber with size of 0.5 m (length) x 0.5 m (wide) x 0.5 m (height). The metal
chamber is a part of customized Fire Smoke Detector and Water Sprinkler System apparatus as can be
seen in Figure 1. The apparatus consists of two metal chambers where one chamber installed with Fire
Smoke Detector and other one metal chamber installed with fire sprinkler head. Only the chamber
installed with sprinkler head used in this experiment. The air temperature inside the chamber was
measured using 0.5mm diameter Type K (chrome/alumel) thermocouple.
Figure 1. Fire Smoke Detector and Water Sprinkler System Apparatus
The experiments are divided into two sets where; Experiment 1 - without smoke effect in the
experimental chamber and Experiment 2 - with smoke effect in the experimental chamber. Fire sprinkler
head activated time and temperature data were recorded for both of the experiments. RTI value was
calculated using RTI equation before analysis using Palisade @RISK Software were conducted. The
RTI equation used in this study given by:
−𝑡𝑟 √𝑢
𝑅𝑇𝐼 = 1−∆𝑇𝑒𝑎 (1)
𝑙𝑛( )
∆𝑇𝑔
Where tr is a sprinkler head activation time, u is the velocity of hot gas flow, Tea is a temperature
surrounding of fire sprinkler head at time of activated and Tg is a hot gas flow temperature.
Using the @RISK statistical analysis software, the best-fit probabilistic curves were generated
for both experiments. The software used Monte Carlo simulation and includes tools to automatically fit
distributions to data using different types of fitting statistics and fitted distributions are ranked according
to the goodness of the fit. The fitted statistics such as the mean, the minimum value, the maximum value
and the standard deviation are provided. @RISK can also apply the Akaike Information Criterion (AIC),
Bayesian Information Criterion (BIC), Chi-Squared Statistics, Kolmogorov-Smirnov Statistic and
Anderson-Darling Statistics for estimating distribution. The Kolmogorov-Smirnov estimation was
selected for the fit statistics because it is appropriate for continuous data and it focuses on the central
region of the input values. The RTI experimental data were compared with the theory and manufacturer
data as to compare the extent that the data deviates and eventually assess the time effect on property loss
due to failure of sprinkler to operate within time.
3
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
3.0 Results and discussion
A total of 34 experiment repetitions were performed for Experiment 1 and Experiment 2. The results
and analysis of the experiments are shown in the following subsections.
3.1 Sprinkler Activation Temperature and Activation Time
Experiment 1 was conducted under constant hot air flow without smoke effect. Sprinkler activation
times were established visually by seeing the activation through the window and identified by the noise
of glass bulb breaking. In other words, the activation time of the sprinkler head is the time at which the
temperature of the sprinkler link reaches the nominal activation temperature. At this stage, the data
logging and video recording is halted. On average, it takes around 60 seconds time for the sprinkler to
activate. It was observed that sprinkler head with a unique number 1-27 gave fastest activation time at
31.82 seconds at 78.8⁰C. While, sprinkler head with a unique number 1-25 gave the longest activation
time at 89.13 seconds at 94.1⁰C. The difference in activation time was 57.31 seconds. Further details on
the distribution of frequency for sprinkler head activation time is shown in Figure 2(a). Also from the
figure, out of the 34 sprinkler heads, seven sprinkler heads required 55 seconds time average to activate,
while five sprinkler heads activated within time range of 37.5 seconds until 65 seconds. From here, it
can be concluded that average of sprinkler will achieve activation time less than 60 seconds.
Experiment 2 was conducted under constant hot air flow with additional smoke effect. It was
observed that the average time required for sprinkler head to activate is around 60 seconds. Sprinkler
head with a unique number 2-1 gave fastest activation time (35.3 seconds at 87.3⁰C). While, sprinkler
head with a unique number 2-11 gave the longest activation time (85.14 seconds at 96.2⁰C). The
difference in activation time was 49.84 seconds. Further details on the distribution of frequency for
sprinkler head activation time is shown in Figure 2(b).
8 4.5
7 4
6 3.5
3
Frequency
Frequency
5
2.5
4
2
3 1.5
2 1
1 0.5
0 0
75
30
35
40
45
50
55
60
65
70
80
85
85
30
35
40
45
50
55
60
65
70
75
80
90
Time Range (s) Time Range (s)
(a) (b)
Figure 2. (a) Frequency distribution of the activation time of sprinkler head for Experiment 1 (b)
Frequency distribution of the activation time of sprinkler head for Experiment 2
Based on the results from Experiment 1 and Experiment 2, additional smoke effect, has a slight
impact on the activation time of fire sprinkler. From the results, it was observed that without the
additional smoke effect, the maximum temperature achieved at glass bulb break was 102.8⁰C while with
additional smoke effect, the maximum temperature achieved was 109.6⁰C. From the average, it is found
that about 6.6% temperature increase due to additional smoke effect while the activation time between
both experiments decrease by 8%. From these experiments, it can be concluded that additional smoke
effect contributes to quicker fire sprinkler activation time. This is in agreement with the work by Beyler
[10] where it was found that the presence of smoke increased the gas temperature at ceiling level which
contributes directly to quicker activation time for fire sprinkler heads.
From the observation of both experiments, the temperature recorded at time when glass bulb is
broken were on average higher than temperature rating given by manufacturer (68⁰C). The temperature
4
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
recorded is also known as glass bulb surface temperature. However, this phenomenon is considered
normal as Sze [11] mentions that the glass bulb surface temperature is technically a bit higher than 68⁰C
because it still needs more time for the hot gas to enter through the glass bulb surface into the liquid
inside. Therefore, there is a thermal lag of bulb mass [12].
3.2 Response Time Index (RTI) analysis
Figure 3 shows the plotted RTI for each of the sample experiment repetition from Experiment 1. The
RTI values are scattered across the test condition for a given bulb diameter and orientation. Sprinkler
head with the unique number of 1-23 gave highest RTI value with 72.13 m1/2·s1/2, while sprinkler head
with the unique number of 1-27 gave lowest RTI value with 16.01 m1/2·s1/2. Figure 4 shows the plotted
RTI for each of the sample experiment repetition from Experiment 2. Sprinkler head with the unique
number of 2-16 gave highest RTI value with 82.20 m1/2·s1/2, while sprinkler head with the unique number
of 2-4 gave lowest RTI value with 15.37 m1/2·s1/2
80.00
70.00
60.00
50.00
RTI (m.s)½
40.00
30.00
20.00
10.00
0.00
0 5 10 15 20 25 30 35
Sample unique number
Figure 3. The plotted RTI for each of the sample experiment from Experiment 1
90.00
80.00
70.00
60.00
RTI (m.s)½
50.00
40.00
30.00
20.00
10.00
0.00
0 5 10 15 20 25 30 35
Sample unique number
Figure 4. The plotted RTI for each of the sample experiment from Experiment 2
RTI values observed were almost similar across different set of experiments for a given bulb
diameter and orientation. Based on Figure 3 and Figure 4, both experiments have shown almost the
5
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
same pattern of distribution except for Experiment 1 which had the widest scatter in the data. Also from
both experiments, it was observed that as the activation times were reduced, RTI value also reduced.
The similar analysis was also carried by Sze [11] to investigate the relationship between RTI and
activation time. It was found from the work that, if the activation time is getting smaller, the RTI also
will be getting smaller.
The RTI distribution for both experiments were fitted using @RISK to get the best fit probabilistic
curve. For Experiment 1, using the Kolmogorov-Smirnov (K-S) technique, the goodness-of-fit
comparison ranks Gamma, Log normal, and Weibull distribution as the top 3 fittest distribution curve.
Figure 5 illustrates the RTI data distribution comparison between 3 types of distribution for Experiment
1. It was observed that Gamma distribution curve is the fittest when compared to the experimental data
while Lognorm and Weibull distribution little bit skewed to the left and skewed to the right. Hence,
Gamma distribution was ranked first as the fittest distribution curve for Experiment 1. Gamma
distribution is a two-parameter family of continuous probability distributions with a shape factor (also
known as α value) and rate factor (also known as β value). As can be seen in Figure 6, the shape factor
and rate factor were 11.207 and 4.1762. Therefore, these values can be applied into Gamma distribution
equation to predict the time of activation for no smoke effect.
Figure 5. RTI Comparison between 3 types of distribution for Experiment 1
Figure 6. RTI Gamma Distribution for Experiment 1
While for Experiment 2, using the Kolmogorov-Smirnov (K-S) technique, the goodness-of-fit
comparison ranks Weibull, Gamma, and BetaGnereal distribution as the top 3 fittest distribution curve.
Figure 7 illustrates the RTI data distribution comparison between 3 types of distribution for Experiment
2. It was found that BetaGeneral distribution is the fittest distribution for Experiment 2 data while
6
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
Weibull and Gamma distribution little bit skewed to the left and skewed to the right respectively. Hence,
BetaGeneral distribution was ranked first as the fittest statistics for Experiment 2. BetaGeneral
distribution are characterized by two continuous shape parameter, α1 and α2, and two continuous
boundary parameter i.e. min and max. As can be seen in Figure 8, the α1 value, α2 value and max value
were 5.02, 7.13 and 112.5. Therefore, these values can be applied into BetaGeneral distribution equation
to predict the time of activation with smoke effect.
Figure 7. RTI Comparison between 3 types of distribution for Experiment 2
Figure 8. RTI Gamma Distribution (with smoke effect)
Table 1 shows the summary of best fit distribution curve for Experiment 1 (without smoke
effect) and 2 (with smoke effect). The information provided is sufficient and ready for use into any fire
simulation software as input for RTI or even can be converted into sprinkler activation time. Technically,
in the fire simulation, when the ceiling jet reaches the advertised activation time of the sprinkler, the
distribution comes into place where it will randomly select a time of activation of the sprinkler using the
information from the best fit distribution.
7
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
Table 1. The summary of best fit distribution curve for Experiment 1 and 2
Attributes Experiment 1 (Without Experiment 2 (With smoke
smoke effect) effect)
Best Fit (Ranked by K-S) Gamma distribution BetaGeneral distribution
(Streteched beta
distribution)
K-S value 0.0897 0.0856
α value (Gamma) 11.207 -
β value (Gamma) 4.1762 -
α1 value (BetaGeneral) - 5.0193
α2 value (BetaGeneral) - 7.1307
Max value (BetaGeneral) - 112.50
4.0 Conclusion
The data collected from both experiments were recorded and analyzed through Microsoft Excel and
@RISK statistical analysis software. The study assesses sprinkler activation time through two
experiments; 1) Without smoke effect and 2) With smoke effect. Based on both experiments, on average,
the sprinkler heads activated at around 60 seconds after the ceiling jet inside the experimental chamber
reached the advertised operating temperature of the sprinkler heads. Overall, it was found that the
sprinkler activation time recorded for Experiment 2 was slightly faster compared to Experiment 1. It is
clearly shown that, there is a possibility of sprinkler performance being impaired by the existence of
smoke. Experimental Response Time Index (RTI) of sprinkler was calculated based on activation time
recorded and compared with manufacturer RTI sprinkler. It shows inconsistency of RTI value between
experimental RTI and manufacturer RTI. Experimental RTI value were used to generate probabilistic
curve by using @RISK software. The summary of the best fit probabilistic curves for both conditions is
shown in Table 1.
5.0 References
[1] Stowell F M and Murnane, L 2012 Essentials of Fire Fighting and Fire Department
Operations (Pearson Education, Limited.)
[2] Teanu B, Erban D M and Solomon D D 2014 Analisys of Predicted Sprinkler Activation Time
in Automated Car Parkings International Scientific Conference CIBV 2014
[3] Thomas I R 2002 Effectiveness of Fire Safety Components and Systems Journal of Fire
Protection Engineering
[4] Heskestad G and Smith H F 1976 Investigation of a new sprinkler sensitivity approval test :
the plunge test. Norwood, Mass.: Factory Mutual Research Corporation.
[5] Frank K, Gravestock N, Spearpoint M and Fleichmann C A review of sprinkler system
effectiveness study Fire Sci Rev 2: 6
[6] Grant C 2019 History of NFPA today as the National Fire Protection Association, 1–10.
[7] Wagner R F, Price G M, Porter H F J, Pike V L, Goldmark P D and Pratt E E 1912
Preliminary Report of the Factory Investigating Commission, 1912.
[8] Moinuddin K A M and Thomas I R 2014 Reliability of sprinkler system in Australian high
rise office buildings Fire Saf. J 63
[9] Arvidson M 2018 The response time of different sprinkler glass bulbs in a residential room
fire scenario Fire Tech. 54(5)
8
26th Regional Symposium on Chemical Engineering (RSCE 2019) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 778 (2020) 012075 doi:10.1088/1757-899X/778/1/012075
[10] Beyler C L and Cooper L Y 2001 Interaction of sprinklers with smoke and heat vents Fire
Technology 37(1)
[11] Sze C K 2009 Response time index of sprinklers International Journal on Engineering
Performance-Based Fire Codes (l)
[12] Madrzykowski D 2008 Impact of a Residential Sprinkler on the Heat Release Rate of a
Christmas Tree Fire Nistir 7506 Fire Research
Acknowledgments
The authors would like to thank Master of Process Safety and Loss Prevention program for funding this
project.
View publication stats
