See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/301310022

Use of Monte Carlo Simulation to assess uncertainties in fire consequence

calculation

Article  in  Ocean Engineering · May 2016

DOI: 10.1016/j.oceaneng.2016.03.050

CITATIONS READS

7 862

1 author:

Ahmed.M Salem
King Abdulaziz University
13 PUBLICATIONS   64 CITATIONS   

SEE PROFILE

Some of the authors of this publication are also working on these related projects:

Fire Safety of Nile-Floating Hotels Working in Egypt View project

Vehicle deck fires aboard RoPax ships View project

All content following this page was uploaded by Ahmed.M Salem on 13 February 2018.

The user has requested enhancement of the downloaded file.

 Ocean Engineering 117 (2016) 411–430

Contents lists available at ScienceDirect

Ocean Engineering
journal homepage: www.elsevier.com/locate/oceaneng

Use of Monte Carlo Simulation to assess uncertainties in fire

consequence calculation
Ahmed.M Salem a,b,n
a
Department of Marine Engineering, Faculty of Maritime Studies, King Abdulaziz University, Jeddah, Saudi Arabia
b
Naval Architecture and Marine Engineering Department, Faculty of Engineering, Alexandria University, Alexandria, Egypt

art ic l e i nf o a b s t r a c t

Article history: Over the past years, ro-ro/passenger ships and cruise liners have been involved in several accidents/
Received 1 June 2015 incidents, including fire. In response to this, the IMO has adopted a series of amendments to SOLAS
Received in revised form Convention to ensure the non-reoccurrence of such accidents. In contemporary fire safety design of
28 December 2015
ships, computer fire models can be utilised to estimate the consequences of fire scenarios on the life
Accepted 19 March 2016
safety of crew/passenger on board. The available safe egress time (ASET) from a fire scene is of para-
mount importance when assessing life safety. The utilised fire model asks for input data that some of
Keywords: them have stochastic nature and may be subjected to uncertainty. The most common technique used to
Monte Carlo Simulation explore the effect of propagation of uncertainty from the random inputs into the predicted ASET is the
Consequence analysis
Monte Carlo Simulation. This work presents the results obtained from combining this technique with a
Ro-ro/passenger ships
fire model while prediction the ASET for four different fire scenarios that involve typical ship layouts
Ship layouts
ASET commonly found in accommodation spaces aboard ro-ro/passenger ships and cruise liners. The results
Zone models indicated that ASET is always affected by uncertainties propagated from the random inputs, with the
time to reach untenable condition due to fire toxicity being the most severely affected output in almost
all examined cases.
& 2016 Elsevier Ltd. All rights reserved.

1. Introduction deviate from the other prescriptive regulations provided that a

fire risk assessment technique is followed. Consequence ana-
Ro-ro/passenger ships and cruise liners in operation today lysis is an essential part of any fire risk assessment technique,
are subject to a wide range of standards and regulations cov- where fire models have to be used to predict the consequences
ering every aspect of ship construction and operation. In re- of a given fire scenario (Salem, 2010).
sponse to several accidents/incidents that occurred over the Uncertainty is considered as one of the challenges that need
past years, the International Maritime Organization (IMO) has to be treated carefully when carrying out fire consequence
adopted a series of amendments to the International Conven- analysis using fire simulation models (Kong et al., 2014). This is
tion for the Safety of Life at Sea (SOLAS), 1974 to ensure the due to the fact that some of the input parameters that are
non-reoccurrence of such accidents/incidents by enhancing the considered when conducting fire modelling are of stochastic
nature. Random input parameters are not easily to determine
safety requirements, including those relating to fire safety
and may be subjected to uncertainty. Example of such para-
measures (such as escape routes and fire protection systems for
meters are fire location, fire size and fire growth parameter
the large atrium typical of cruise liners (IMO, 2015a, 2015b).
(Themelis and Spyrou, 2012). If uncertainty is ignored and
The contemporary fire safety design of ships is governed by the
average values (point estimates) of such parameters are se-
regulations set out in Chapter II/2 of the SOLAS Convention, lected, the fire modelling predictions may vary from one de-
1974 as amended in 2002 and further amended in 2010. signer to another, and depending on the selected point esti-
Chapter II/2 consists of 20 regulations covered in 7 parts, 19 of mates, the output design could be either overly-safe (expensive
them prescribe in detail the fire safety provisions for all type of solution) or unsafe (cheap solution). Therefore, designers
ships. Regulation 17 of this chapter allows the designers to should utilise more advanced techniques to quantify un-
certainties in order to obtain more accurate results (i.e., safe
n
Corresponding address: Department of Marine Engineering, Faculty of Maritime
designs) (Kong et al., 2014).
Studies, King Abdulaziz University, Jeddah, Saudi Arabia. There are several techniques that can be followed to address
E-mail address: [email protected] uncertainty, but the most common, reliable and cost-effective

http://dx.doi.org/10.1016/j.oceaneng.2016.03.050
0029-8018/& 2016 Elsevier Ltd. All rights reserved.
412 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 4. Key stages of fire development in compartment fires.

Fig. 1. Basic idea of combining Monte Carlo Simulation with fire simulation.
properties of construction materials used to construct the com-
partments involved in the four case studies have been collected
from the deliverables of previous international projects, which
considered the fire safety design of both ro-ro/passenger ships and
cruise liners.

2. Methodology

As consequence calculations of fire scenarios use complex

deterministic fire models that need several deterministic as
Fig. 2. Illustration of the timeline approach. well as random input parameters, and in order to explore the
effect of the propagation of uncertainty of the random inputs
Table 1 in the predicted quantities of interest from fire models, several
K and D values for different activity levels methods are available for this purpose. Examples of such
methods include: response surface methodology (Albrecht and
Activity K D
Hosser, 2011), fuzzy logic approach (Kong et al., 2014), grey
At rest 40
system theory (Julong, 1989), Monte Carlo Simulation (Helton
2.81945 × 10−4
Light work 30 and Davis, 2003), among others. Due to its wide use and being
8.2925 × 10−4
Heavy work 1.6585 × 10−3 20 the best approach (more effective and robust) to analyse the
uncertainty (Wang et al., 2013; Kong et al., 2013), the author
decided to combine Monte Carlo Simulation with a fire simu-
lation model in order to address the propagation of un-
certainties while assessing the life safety of crew/passenger
aboard ships in case of fire. Fig. 1 illustrates the basic idea of
the approach adopted. The following sub-sections highlight
some of the basic concepts related to fire modelling and its use
in assessing life safety of people in case of fire, as well as the
basic idea of Monte Carlo Simulation and how to specify the
appropriate probability density function (PDF) of a random
variable.

Fig. 3. Example of inputs and outputs of a fire model.

2.1. Life safety assessment in case of fire

In the event of a fire in one of ship's compartments, the crew/

technique is the Monte Carlo Simulation (Helton and Davis, 2003; passenger on board may be subjected to untenable conditions that
Wang et al., 2013; Kong et al., 2013). The work presented in this may lead to their death. Untenable conditions could be defined as
paper introduces a methodology that combines the Monte Carlo the environmental conditions in which human life is not sustain-
Simulation technique with a fire simulation model to address the able (ABCB, 2005). Untenable conditions during a fire may result
propagation of uncertainties while assessing the life safety of from exposure to radiant and convective heat, inhalation of toxic
crew/passenger in case of fire outbreaks aboard ships. A computer gases (such as, carbon monoxide, carbon dioxide, hydrogen cya-
code is written for the sake of doing the merge and simplifying the nide and oxygen depletion) and visual impairment due to smoke
input/output processes. The paper presents the results obtained obscuration (SFPE, 2005).
from applying this code to four case studies that involve typical In modern risk-based fire safety design of ships, the life safety
layouts in the accommodation spaces aboard ro-ro/passenger of the crew/passenger on board during fire could be assessed by
ships and cruise liners. This paper considers for the first time using the timeline approach (see Fig. 2). This approach compares
enclosures that represent typical atrium aboard modern cruise two timelines, namely, the available safe egress time (ASET) and
liners as well as typical cabin-corridor-stairwell-corridor combi- the required safe egress time (RSET). On the one hand, ASET is the
nation aboard ro-ro/passenger ships. All dimensions and time interval between fire ignition and the development of
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 413

Table 2
Values of fire growth parameter with real fire examples.

Growth type Characteristic time (s) Fire growth parameter ( α, kW/s2 ) Example from real fire

Slow 600 0.00293 Floor coverings

Medium 300 0.01172 Solid wooden furniture, e.g. office furniture, shop counters
Fast 150 0.04689 Light wooden furniture, e.g. plywood wardrobes, bedding, displays
Ultra-fast 75 0.18756 Upholstered furniture, lightweight furnishing

Fig. 5. Typical HRR–time curve represents the four types of t-squared fire growth
rates.

untenable conditions. It represents the maximum time available

for people to evacuate safely from the fire scene. On the other
hand, RSET is the sum of three components: detection time, re-
sponse time and movement time. It represents the time required
to evacuate safely from the compartment of fire origin and the
adjacent compartments (Tosolini et al., 2012). The basic aim of the
timeline approach is to show that ASET is greater than RSET with
sufficient safety margin, and hence, the crew/passenger aboard the
ship are considered safe. It should be noted that, ASET is predicted
through fire simulation models, while RSET is predicted through
evacuation simulation models.
The ASET may be determined by establishing tenability criteria.
The literature presents three tenability criteria to estimate in-
capacitation due to heat and toxicity and visual impairment due to
smoke obscuration. People in close proximity to the fire or under
or within the smoke layer are exposed to thermal effects and may
be subjected to hyperthermia, skin burns, and respiratory tract
burns. Exposure to toxic gases impair person’s ability to self-
evacuate by decreasing the amount of oxygen available, causing
disorientation and possibly unconsciousness. Vision impairment
due to smoke obscuration could affect person's ability to see and
allocate escape routes efficiently.
Purser (2002), outlines the fractional effective dose (FED) and
the fractional effective concentration (FEC) approaches that de-
termine the accumulating effect of a time-changing exposure. The
fundamental concept of the FED approach is that when the sum-
mation of the proportional fractions of doses of heat or toxicants
that would cause an effect equals unity, the effect is expected to
occur. While the fundamental concept of the FEC approach is that
when the FEC of smoke equals unity, it represents a smoke con-
centration capable of seriously limiting escape capability. Purser
(2002) introduced three models to determine the FEDs of toxicity
and heat and the FEC of smoke. The following is a brief description
of these models. Fig. 6. Flowchart of the computer programme used to combine Monte Carlo Si-
Purser (2002) suggests a model to assess the potential harm mulation with the fire model “CFAST”.
414 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 7. Compartment configurations typically exist in accommodation spaces aboard ro-ro/passenger ships and cruise liners.

threshold value represents the time available for escape relative to

Table 3
Physical and thermal properties of the boundary materials (Themelis et al. 2011;
chosen safety criteria (FED value of 1.0 is associated with sub-le-
Rockwool 2012). thal effects that would make occupants of average susceptibility
incapable of completing their own escape).
Material Thickness Density (kg/m3) Thermal con- Specific heat
(mm) ductivity (W/m K) (J/kg:K)
FED Toxicity = FEDCO × VCO2 + FEDO2 (1)
Steel plate 7 7850 51.90 483
Rockwool 75 45 0.034 840
t2
K × [CO]1.036
FEDCO = ∑ Δt
t1
D (2)
resulting from exposure to toxic fire effluents by determining the
FED for each asphyxiant at each discrete increment of time, Δt. The
time at which their accumulated sum exceeds a specified
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 415

Fig. 9. The fitted Normal distribution of the time to reach untenable condition due
Fig. 8. The selected lognormal distribution of the fire growth parameter. to smoke obscuration.

exp (0. 1903 × %CO2 +2. 0004) where qrad is the radiant heat flux in kW/m2, T is temperature in
VCO2 =
7. 1 (3) °C.
When FEDHeat reaches unity, a tenability limit is predicted.
Furthermore, Purser (2002) has suggested a model to assess the
t2 visual obscuration effects of smoke based on the concept of FEC. In
1
FEDO2 = ∑ Δt this model, the smoke concentration is expressed as a fraction of
t1 (
exp 8.13 − 0.54 ( 20.9% − %O2 ) ) (4)
the concentration considered to significantly affecting the escape
where, [CO] is the average concentration of CO (ppm) over the efficiency. If the FECSmoke reaches unity, then it is predicted that
time increment, ∆t , K and D are constants depend on the activity the level of visual obscuration would be sufficient to seriously
of the person (see Table 1), %CO2 is the carbon dioxide con- affect escape attempts.
centration, and (20. 9% − %O2 ) is the percent of oxygen vitiation FECSmoke = (OD/0. 2) for smallenclosures (9)
over the time increment.
Moreover, Purser (2002) has introduced a model similar to that
used with toxic gases to assess the life threat due to exposure to FECSmoke = (OD/0. 08) for large enclosures (10)
radiant and convective heats in fires. The total FED of heat ac-
1
quired during an exposure can be calculated as follows: where OD is the smoke optical density in m .
Accordingly, the ASET may be taken as the interval between
t2
onset of ignition and the exposure time required for either
FEDHeat = ∑ ( 1/tIrad + 1/tIconv ) Δt
t1 (5) FEDToxicity, FEDHeat, or FECSmoke to reach unity. Since the use of fire
simulation models is necessary to predict the ASET, the following
−4/3 section is dedicated to give brief description of different fire
tIrad = 1. 333 ( qrad ) (6) modelling techniques.

( )
tIconv = 4. 1 × 108 T−3.61 for fully clothed subjects (7) 2.2. Fire models for ASET prediction

Fire simulation models have been developed and tested to

( )
tIconv = 5 × 107 T−3.4 for lightly clothed subjects (8) be useful in predicting the ASET from ship's compartment fires
416 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 10. The fitted normal distribution of the time to reach untenable condition due Fig. 11. The fitted lognormal distribution of the time to reach untenable condition
to thermal and radiant heats. due to toxicity of fire effluents.

Table 4
(Galea et al., 2003; Salem, 2007; Hakkarainen et al., 2009; Summary statistics for the three times to reach untenable conditions in the single
cabin fire scenario (all times are measured in seconds).
Tuovinen and Hertzberg, 2009; Salem, 2010, 2013; Salem and
Leheta, 2011; Azzi et al., 2011; Pawling et al., 2012; Wang et al., Cabin
2013; Salem et al., 2015). There are two different techniques of
tSmoke tHeat tToxicity
fire modelling that are commonly used within the community
of fire protection engineering, namely zone modelling and field Minimum 42.0 28.0 170.0
modelling. The concept of zone modelling divides the hy- Maximum 53.3 43.4 231.3
pothetical burning compartment into a limited number of Range 11.3 15.4 61.3
Mean 47.2 35.9 200.3
control volumes or zones. The most common type is the ‘two-
SD 0.99 1.54 5.06
zone model’ where the compartment is divided into an upper (5%) 45.6 33.4 192.0
hot zone and a lower cold zone. Zone models are faster and (95%) 48.9 38.5 208.7
their accuracy are acceptable but their major disadvantage is
the generality of their results (Averill, 1998). Examples of zone
fire models include CFAST (Peacock, 2013), BRANZFIRE (Wade,
2004), Ozone (Cadorin et al., 2001) and Räume (Shigunov, parameters, such as dimensions of enclosure(s), dimensions and
2005). On the other hand, field modelling presents a more status of vents (doors and windows), specifications of ventilation
scientifically accurate approach. This approach is based on di- system (volume flow rate, duct shape and size), fire source (fire
viding the domain of interest into a large number of control growth parameter, peak of heat release rate (HRR), fire source
volumes. Field models predict smoke movement caused by fire, location, etc.), and the thermal properties of the lining materials
ventilation system, and other factors in three dimensions. (thermal conductivity, density, emissivity, etc.). The predicted
The most significant limitations of field models are the cost and outputs of the fire model are, for instance, the time history of
time (Shorab et al., 2011). Examples of field fire models temperature, smoke layer height, species concentrations of toxic
include FDS (McGrattan et al., 2013) and SMARTFIRE (Ewer gases, radiant heat flux and optical density (see Fig. 3). These
et al., 2013). outputs are the basic components, which used to estimate the
Regardless of the type of fire modelling being used to predict ASET timeline.
the ASET timeline, it will asks for user-specified set of input Generally, ASET is a function of several deterministic as well as
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 417

Fig. 12. CDFs of the single cabin fire scenario.

Fig. 14. The fitted lognormal distribution of the time to reach untenable condition
Fig. 13. The fitted normal distribution of the time to reach untenable condition due due to smoke obscuration in the corridor.
to smoke obscuration inside the cabin.
418 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 15. The fitted normal distribution of the time to reach untenable condition due Fig. 16. The fitted lognormal distribution of the time to reach untenable condition
to thermal and radiant heats inside the cabin. due to thermal and radiant heats in the corridor.

random input parameters. Deterministic parameters are easy to Modelling the actual growth is extremely difficult and re-
determine, such as enclosure dimensions, where distance can be mains an area of active research (Fleischmann, 2009). Several
measured with a bit high degree of certainty, while random methods are available for estimating the growth rate for a
parameters are not easily to determine, due to either lack of ex- particular design fire. The most frequently suggested method is
perimental data or complexity of the dynamic behaviour of fire, the t-squared fire growth method, in which the HRR is assumed
and may be subjected to uncertainty. to continuously grow quadratically as a function of time until
either the fuel is totally consumed or the HRR is assumed to
2.2.1. The heat release rate (HRR) have reached a peak value. Depending on the characteristic
Fire scenarios require a quantitative design fire for use in fire time required for the HRR to reach a value of 1055 kW, the fire
simulation. Design fires are described in terms of the fire size, the growth parameter could be slow, medium, fast or ultra-fast. A
species being produced, the smoke production rate, and the time slow fire reaches a HRR of 1055 kW in 600 s. A medium fire
history of the HRR from fire. A typical “heat release rate – time” takes 300 s, fast fire takes 150 s and ultra-fast takes only 75 s to
curve shows the four key stages of fire development in a com- reach 1055 kW. The HRR–time relationship can be written for
t-squared fires in formula form as follows:
partment. These stages are ignition, growth stage, fully-developed
stage and decay stage, respectively (see Fig. 4). ⎧ 2
⎪ αt , if t < tg

Ignition process can be either piloted (caused by flame, spark, Q (t ) = ⎨

⎪
⎩ Q max, if t ≥tg (11)
etc.) or spontaneous (due to accumulation of heat in the fuel).
Once ignition occurs, part of the solid fuel in the compartment is where, Q (t ) is the HRR (kW) at time t (s); α is the fire growth
pyrolysing, releasing gaseous volatiles, which burn as they mix parameter (kW/s2 ); Q max is the peak HRR (kW), and t g is the
with air. Following ignition, fire grows at a rate dependent upon time when HRR reaches its peak value (s). Table 2, lists the four
the type of fuel, access to oxygen, compartment configuration and types of fire growth, mentioned above, and the corresponding
other factors. At fully-developed stage, the HRR reaches its max- values of the fire growth parameter ( α ). A typical HRR–time
imum and remain constant, and the development of the fire is curve that follows Eq. (11) is shown in Fig. 5.
often limited by the availability of oxygen. During the decay per- Results of several sensitivity analyses carried out by many
iod, the heat release rate diminishes as the fuel becomes con- researchers (Salem and Leheta, 2011; Albrecht et al., 2010; Jahn
sumed (Novozhilov, 2001). et al., 2009), have revealed that the most important input
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 419

Fig. 18. The fitted lognormal distribution of the time to reach untenable condition
Fig. 17. The fitted normal distribution of the time to reach untenable condition due due to toxicity of fire effluents in the corridor.
to toxicity of fire effluents inside the cabin.

Table 5
Summary statistics for the three times to reach untenable conditions in the cabin-
parameters that affect the prediction of ASET are the fire corridor fire scenario (all times are measured in seconds).

growth parameter ( α ), and peak of the HRR (Qmax ). Moreover, it Cabin Corridor
is found that Q max has an almost neglected effect on the results
in small compartments and especially when the fire is of the tSmoke tHeat tToxicity tSmoke tHeat tToxicity
ventilation-controlled type, hence, only ( α ) has to be con-
Minimum 45.0 30.0 228.0 66.0 158.0 280.0
sidered. While, in large compartments, the amount of oxygen is
Maximum 55.3 43.4 279.3 81.3 207.2 332.3
enough to make the fire of the fuel-controlled type and hence, Range 10.3 13.4 51.3 15.3 49.2 52.3
the fire is able to grow, as there is enough fuel for that. Mean 49.7 35.9 247.6 72.7 179.3 300.4
Therefore, both ( α ) and Qmax have to be taken into considera- SD 1.14 1.54 5.18 1.74 5.70 5.23
(5%) 47.9 33.4 239.1 69.6 170.0 291.8
tion when dealing with fires in large compartments. It should
(95%) 51.6 38.5 256.1 75.6 188.7 309.0
be noted that both ( α ) and Qmax are uncertain parameters and
no one can be certain about the type and amount of material,
which is going to be involved in a certain fire scenario. These
uncertainties, if ignored, will be reflected on the outputs of the variable.
fire model, and hence the ASET timeline.
One way to describe uncertainty is to use the probability
density function (PDF) of the uncertain parameters instead of their 2.3. Basic idea of Monte Carlo Simulation
mean values. The PDF shows the values that the parameter can be
assigned and how often these values are to be expected. The actual In Monte Carlo Simulation, a random value is selected from
value of a parameter is chosen randomly from the distribution and each of the random input parameters based on their assigned
the parameter is therefore a random variable. There are many PDFs. An input file to the fire model is created with the selected
techniques that can be followed to sample random variables that values of random parameters. The fire model is called to per-
are governed by complicated PDFs, but the most common, reliable form deterministic computation based on the input data pro-
and cost-effective technique is the Monte Carlo Simulation. The vided and the outputs are recorded, and the process is re-
following section presents a brief description of the basic idea of peated. A typical Monte Carlo simulation repeats this process
this technique and how to specify the appropriate PDF of a random may be hundreds, thousands or tens of thousands of times,
420 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 19. CDFs of cabin-corridor fire scenario.

Fig. 20. The selected triangular distribution of the peak of the heat release rate. Fig. 21. The fitted lognormal distribution of the time to reach untenable condition
due to smoke obscuration in the 2nd deck.
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 421

Fig. 23. The fitted triangular distribution of the time to reach untenable condition
Fig. 22. The fitted lognormal distribution of the time to reach untenable condition due to thermal and radiant heats in the 2nd deck.
due to smoke obscuration in the 3rd deck.

parameters is the Normal distribution, while for other un-

each time using different randomly-selected input values. certain parameters, where large uncertainties are expected
When the simulation is finished, a large number of results from (such as the fire growth parameter), Lognormal distribution is
the fire model are recorded. These results are then presented in selected.
the form of probability distributions. In this way, Monte Carlo As mentioned in Section 2.2, the most important input
simulation provides a much more comprehensive view of the parameters into a fire model in a given fire scenario that may
predicted outputs from the fire model. It tells not only what is significantly affect the prediction of the ASET are the fire
the outcome from a fire scenario, but also how likely it is to growth parameter ( α ), and peak of the HRR (Q max ). Due to
happen (Wang et al., 2013; Palisade, 2015). uncertainty of these parameters, they should be assigned sui-
table probability density functions. Many researchers such as
2.3.1. Specifying probability distribution for uncertain parameters Frantzich (1998), Holborn et al. (2004), Kong et al. (2011),
The probability distributions are defined by the type (nor- Guanquan and Jinhui (2012), and Kong et al. (2013, 2014) have
mal, uniform, lognormal, triangular, etc.), the minimum, the pointed out that based on studies of several previous real fire
maximum, the mean “ μ ”, and the standard deviation “s” (if incidents, the fire growth parameter ( α ) is reasonably well
applicable). How a type of probability distribution should be approximated by the Lognormal distribution. Moreover, due to
chosen for each random input parameter is a question that lack of detailed data regarding the peak HRR (Q max ), Jannat and
many researchers have tried to answer during the last two Greenwood (2012) have stated that the triangular distribution
decades. Frantzich (1997) has pointed out that it is not that is commonly used in such situations to represent uncertainty.
important if the random input parameters have not assigned The triangular distribution can take on a variety of shapes and
the exact distribution type, which is usually impossible to requires three easy to estimate basic parameters, namely,
achieve in practice. He also added that the most important minimum, maximum, and most likely.
information of a random parameter are the minimum and
maximum values, mean, and standard deviation, which should
be chosen based on a combination of experimental data, sta- 3. The approach adopted (Monte Carlo Simulation with CFAST)
tistical data and expert judgment. Moreover, in a subsequent
study by Frantzich (1998), he highlighted that the common As mentioned above, this work adopts an approach that com-
type of distribution used for most of the random input bines the Monte Carlo Simulation technique with a fire simulation
422 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

model to address the propagation of uncertainties while assessing based on the randomly generated N-number of samples of the
the life safety of crew/passenger in case of fire outbreaks aboard uncertain parameters of concern;
ships. 2. Calling CFAST and runs it an N-number of times and col-
As large number of runs of the fire simulation model are ne- lecting the N-number of output files;
cessary to carry out the uncertainty analysis using Monte Carlo 3. Reading these N-number of output files and extracting the
simulation, and in order to keep the computational cost at a outputs of concern for estimating the ASET;
minimum, the author has decided to adopt the two-zone fire 4. Calling the ASET model and runs it an N-number of times to
model “CFAST” to model the fire and smoke development and to estimate the two FEDs of heat and toxicity, and the FEC of smoke
predict the outputs of concern that are required to estimate the and then predicting the time at which each one of these criterion
ASET. A computer programme is written for the sake of the reaches unity, and
5. Fitting the predicted times to reach the three untenable
following:
1. Preparing the required N-number of input files for CFAST conditions unto probability distributions and hence quantifying
the uncertainties.
Fig. 6 shows the flowchart of the written programme.
In the following section, four case studies will be considered
to illustrate the use of the written code in studying the pro-
pagation of uncertainties into the three times to reach unten-
able conditions from the fire growth parameter alone and
from both the fire growth parameter & the peak of the HRR
(depending on the size of the compartment(s) under
consideration).

4. Case studies

In order to demonstrate the utilisation of the approach adopted

and the written computer programme, four case studies that in-
volve enclosure arrangements, which are typical ship layouts in
accommodation spaces aboard ro-ro/passenger ships and cruise
liners are considered.
It should be noted here that both ro-ro/passenger ships and
cruise liners have been under investigation by many research
groups in several international projects such as SAFER EURORO
(1997), SAFEDOR (2005), FIREPROOF (2009), and FAROS (2012).
The common theme for these projects is “Design for Safety”, where
the fire safety design of these two types of ships has been studied.
Data collected from the deliverables of these projects make it
possible to examine real enclosures that are typically found on
board such ships.
It should be noted also here that in order to define the worst
possible fire scenarios, a semi-quantitative risk analysis method
called Preliminary Hazard Analysis (PHA) is usually used for this
purpose. This type of analysis is mainly used as a first step to
identify and estimate the possible hazards on a low-detailed
level in existing structures. According to a previous PHA carried
out by Andersson and Säterborn (2002), fire in a cabin with
possible cause of smoking in bed,and fire in a public space (such
as atrium) with possible cause of faulty electric equipment or
smoking have been identified as the worst possible fire case
Fig. 24. The fitted triangular distribution of the time to reach untenable condition scenarios that could occur in accommodation spaces aboard
due to thermal and radiant heats in the 3rd deck. passenger ships.

Table 6
Summary statistics for the three times to reach untenable conditions in the atrium fire scenario (all times are measured in seconds).

1st deck 2nd deck 3rd deck

tSmoke tHeat tToxicity tSmoke tHeat tToxicity tSmoke tHeat tToxicity

Minimum Not reached 195.3 2486.5 Not reached 217.4 2801.7 Not reached
Maximum 368.5 6242.5 432.3 6251.6
Range 173.2 3756.0 214.9 3449.9
Mean 270.1 4051.7 310.2 4378.1
SD 20.11 797.96 24.94 711.98
(5%) 237.1 2910.0 269.3 3270.0
(95%) 303.3 5520.0 351.3 5640.0
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 423

Fig. 25. CDFs of atrium fire scenario.

Fig. 26. The fitted normal distribution of the time to reach untenable condition due Fig. 27. The fitted normal distribution of the time to reach untenable condition due
to smoke obscuration in the cabin. to smoke obscuration in the 1st corridor.
424 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 28. The fitted normal distribution of the time to reach untenable condition due Fig. 29. The fitted normal distribution of the time to reach untenable condition due
to smoke obscuration in the stairwell. to smoke obscuration in the 2nd corridor.

4.1. Compartment geometries and boundaries levels. The cabin and the 2 corridors have dimensions similar
to those in case 2, while the stairwell has a length of 4.0 m,
This work consider four compartment geometries: a single width of 4.0 m and height of 5.0 m. The four compartments
cabin; a cabin-corridor arrangement; an atrium, and a cabin-cor- (cabin, 1st corridor, stairwell and 2nd corridor) are connected
ridor-stairwell-corridor arrangement (two-level configuration). via 3 doors that have similar dimensions of 1.0 m width and
Schematic drawings of the four considered geometries are shown 2.0 m height (see Fig. 7d).
in Fig. 7a–d. In all considered cases, the boundaries forming the floor, ceiling
The single cabin geometry represents a large cabin typically and walls of each compartment are assumed to be constructed of
found aboard ro-ro/passenger ships. The cabin has a length of “A-60” class division. The “A-60” class division is assumed to be
5.0 m, width of 5.0 m, and height of 2.5 m and is connected to made of two-slab material (steel and insulation). Both physical and
outside via a door of 1.0 width and 2.0 m height (see Fig. 7a). thermal properties of each slab are listed in Table 3.
The cabin-corridor arrangement represents a cabin with same
dimensions as above connected to a corridor that has a length 4.2. Single cabin fire scenario
of 20.0 m, width of 1.0 m, and height of 2.5 m via a door of
1.0 m width and 2.0 m height (see Fig. 7b). The atrium config- In this scenario, a fire is assumed to occur in one of the
uration represents a three-storey atrium typically found upholster furniture items located at the middle of the floor of
aboard modern cruise liners. The atrium is consisted of three the cabin. The door is assumed to be closed to represent a
consecutive decks, each deck has a square area of critical case for the occupant of the cabin due to rapid accu-
40.0 m  40.0 m and a height of 3.0 m and connected to the mulation of dense smoke. The HRR from the fire source is as-
outside via two doors, each has a width of 4.0 m and height of sumed to follow a t-squared fire growth rate with maximum
2.0 m. The three decks are connected to each other to construct HRR of 1 MW, which is assumed to be relevant when con-
the atrium via two ceiling openings, each has an area of 400 m2 sidering the life safety assessment of the occupants of such
of square shape (see Fig. 7c). Finally, the cabin-corridor-stair- compartments as they should evacuate during the first
well-corridor configuration represents a two-level arrange- few minutes of fire development. Due to the nature of the
ment that typically found aboard ro-ro/passenger ships, which furniture materials, it is assumed that the probable fire growth
constitute the escape route of crew/passenger located in these rate may vary between a more than medium and less than
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 425

Fig. 30. The fitted normal distribution of the time to reach untenable condition due Fig. 31. The fitted normal distribution of the time to reach untenable condition due
to thermal and radiant heats in the cabin. to thermal and radiant heats in the 1st corridor.

ultra-fast. that is slightly affected by the uncertainty.

This scenario has been simulated for 20 min using CFAST 3. The occupants in the cabin (room of fire origin) have an
according to the approach adopted and with the help of the average of about 36.0 s to evacuate the cabin before the onset of
written programme. All input parameters are dealt with as untenable condition due to the effect of heat.
deterministic parameters except the fire growth parameter,
which is dealt with as a random variable. A lognormal dis- 4.3. Cabin-corridor fire scenario
tribution with mean of 0.046889 kW/s 2 and standard deviation
of 0.005 kW/s2 has been assigned for the fire growth parameter In this scenario, an exactly similar fire scenario to the previous
(see Fig. 8). A sample size of 5000 has been used. The results of case is assumed to occur but with the door connecting the cabin to
the simulation are shown in Figs. 9–11 and summarised in the corridor is assumed to remain open during the simulation, as
Table 4. Fig. 12 shows the Cumulative Density Functions (CDFs) this would fulfil the worst-case scenario in the corridor. Again, due
of the three predicted times to reach untenable conditions, to the uncertainty in defining the fire growth parameter of this
which are measured in seconds. fire, a Monte Carlo Simulation has been carried out by varying only
The results of the Monte Carlo simulation indicate the the fire growth parameter and keeping the other parameters
following: constant. The scenario has been simulated for 20 min using CFAST
according to the approach adopted and with the help of the
1. Both “tSmoke” and “tHeat” have PDF in the form of normal dis- written programme. The results of the simulation are shown in
tribution, while “tToxicity” has a PDF function in the form of Figs. 13–18 and summarised in Table 5. The CDFs of the three
lognormal distribution, which is similar to the distribution of predicted critical times inside the Cabin as well as the Corridor are
the fire growth parameter (the only stochastic input variable in shown in Fig. 19.
the scenario). The results of the Monte Carlo Simulation indicate the
following:
2. “tToxicity” is the output that is highly affected by the un-
certainty in the fire growth parameter, while “tSmoke” is the output 1. In the cabin of fire origin, the three critical times follow normal
426 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 33. The fitted triangular distribution of the time to reach untenable condition
Fig. 32. The fitted normal distribution of the time to reach untenable condition due due to thermal and radiant heats in the 2nd corridor.
to thermal and radiant heats in the stairwell.

is assumed that the fire growth parameter follows the same dis-
distribution type of PDF, while in the corridor, the same three tribution as in the previous scenarios, while the peak of the HRR is
critical times follow lognormal distribution same as the dis- assumed to follow a triangular distribution with minimum of
tribution type of the only stochastic input variable in the 1000 kW, mode (most likely) of 2000 kW, and maximum of
scenario. 3000 kW (see Fig. 20). It should be noted here that the triangular
2. “tToxicity” in both cabin and corridor as well as “tHeat” in the distribution is commonly used in simulation projects to represent
corridor are the outputs that are highly affected by the un- probabilistic processes in absence of detailed data (Jannat and
certainty in the fire growth parameter, while “tSmoke” is the Greenwood, 2012).
output that is slightly affected by the uncertainty. A Monte Carlo simulation has been carried out to assess the
3. The occupants in the cabin of fire origin have an average of uncertainty in estimating the three mentioned critical times in
about 36.0 s to evacuate the cabin before the onset of untenable
each of the three decks. Both the fire growth parameter and the
condition due to the effect of heat, while the occupants in the
peak of the HRR are varied during the simulation while the
corridor have an average of about 73.0 s to evacuate before the
other parameters are kept constant. The scenario has been si-
onset of untenable condition due to the effect of smoke
mulated for 120 minutes using CFAST according to the ap-
obscuration.
proach adopted and with the help of the written programme.
The results of the simulation are shown in Figs. 21–24 and
4.4. Atrium fire scenario summarised in Table 6. Fig. 25 shows the CDFs of the predicted
times to reach untenable conditions inside the two upper
In this scenario, a fire is assumed to outbreak in a piece of decks of the atrium.
furniture located at one corner on the floor of the lowest deck. The The results of the Monte Carlo Simulation indicate the
doors connecting the atrium to outside are assumed to be opened following:
during the simulation to provide the atrium with the sufficient
oxygen for combustion. Due to the large size of the space, the 1. Due to the nature of the atrium configuration (large openings
uncertainty in defining both the fire growth parameter and the between decks), the location of the fire, and the nature of the
peak of the HRR of this fire has to be considered. In this scenario, it movement of hot gases, none of the three critical times has
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 427

Fig. 34. The fitted normal distribution of the time to reach untenable condition due Fig. 35. The fitted normal distribution of the time to reach untenable condition due
to toxicity of fire effluents in the cabin. to toxicity of fire effluents in the 1st corridor.

been reached until the end of the simulation (120 min) in the cabin. Due to the large total size of the connected compart-
lowest deck (the deck of fire origin). ments, the uncertainty in defining both the fire growth para-
2. The time to reach untenable condition due to toxicity of fire meter and the peak of the HRR has to be considered. Moreover,
effluent has not been reached until the end of the simulation in both the fire growth parameter and the peak of the HRR are
the two upper decks. assumed to follow the same distributions as in the atrium
3. It is also clear that “tSmoke” follows the same statistical dis- scenario.
tribution as the fire growth parameter, while “tHeat” follows the A Monte Carlo simulation has been conducted to assess the
same statistical distribution as the peak of the heat release rate. uncertainty in estimating the three previously mentioned
4. “tHeat” in each of the upper decks is heavily affected by the un- critical times. Both the fire growth parameter and the peak of
certainties in the two stochastic input variables, while “tSmoke” is the heat release rate are varied during the simulation while the
less affected by the uncertainties than “tHeat”. other parameters are kept constant. The scenario has been si-
5. The 1st deck is the place with minimum risk and the occupant of mulated for 20 minutes using CFAST according to the approach
this deck have enough time (120 min) to evacuate to a place of adopted and with the help of the written programme. The
safety, while the occupants of the 2nd and 3rd decks have only
results of the simulation are shown in Figs. 26–37 and
an average of about 270.0 s and 310.0 s, respectively, to evacuate
summarised in Table 7. The CDFs of the three predicted
to a place of refuge before the onset of untenable condition due
critical times inside each of the four rooms are shown in
to reduced visibility.
Fig. 38.
The results of the Monte Carlo Simulation indicate that:
4.5. Cabin-corridor-stairwell-corridor fire scenario
1. The three critical times in the four rooms follow normal dis-
In this scenario, a fire is assumed to outbreak in one of the tribution, except “tHeat” in the 2nd corridor and “tToxicity” in both
upholster furniture items located at the centre of floor of the the stairwell and the 2nd corridor follow triangular distribution,
428 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Fig. 36. The fitted triangular distribution of the time to reach untenable condition Fig. 37. The fitted triangular distribution of the time to reach untenable condition
due to toxicity of fire effluents in the stairwell. due to toxicity of fire effluents in the 2nd corridor.

Table 7
Summary statistics for the three times to reach untenable conditions in the four-room fire scenario (all times are measured in seconds).

Cabin 1st corridor Stairwell 2nd corridor

tSmoke tHeat tToxicity tSmoke tHeat tToxicity tSmoke tHeat tToxicity tSmoke tHeat tToxicity

Minimum 44.5 95.3 190.6 65.3 145.0 320.1 108.6 296.7 490.9 109.1 1093.2 654.9
Maximum 55.0 129.4 274.5 82.6 204.7 455.4 139.3 417.4 577.1 144.1 1884.0 685.1
Range 10.5 34.1 83.9 17.3 59.7 135.3 30.7 120.7 86.2 35 790.8 30.2
Mean 48.7 111.0 222.4 71.9 167.8 371.4 120.3 344.3 531.2 122.4 1411.0 607.9
SD 1.10 4.18 8.52 1.75 6.04 13.76 3.11 12.75 17.64 3.55 166.46 19.09
(5%) 46.9 104.1 208.4 69.0 157.9 348.7 115.2 323.3 503.6 116.6 1181.0 578.5
(95%) 50.5 117.8 236.4 74.8 177.8 394.0 125.4 365.2 562.6 128.2 1735.0 642.3

the same distribution type as the peak heat release rate. 122.0 s, respectively to evacuate to a place of safety before the
2. “tHeat” in the 2nd corridor is the output that is highly affected by onset of untenable condition due to the effect of smoke
the uncertainties in both the fire growth parameter and the obscuration.
peak of heat release rate, while “tSmoke” is the output that is
slightly affected by the uncertainties in the same input
parameters.
3. The farther the room from the room of fire origin the higher the 5. Conclusions
propagation of the uncertainties on all the outputs of concern.
4. The occupants in the cabin, 1st corridor, stairwell, and the 2nd The work presented in this paper highlighted the importance of
corridor have an average of about 49.0 s, 72.0 s, 120.0 s, and using the Monte Carlo Simulation technique in order to quantify
 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430 429

Fig. 38. CDFs of cabin-corridor-stairwell-corridor fire scenario.

the propagated uncertainties from the stochastic input parameters References

while predicting the ASET timeline used in analysing the con-
sequences of a given fire scenario on the life safety of crew/pas- ABCB, 2005. International Fire Engineering Guidelines. Australian Building Codes
Board, Canberra, Australia.
senger aboard ships. A computer programme was specifically Albrecht, C., Hosser, D., 2011. A response surface methodology for probabilistic life
written for this purpose and has been tested in four fire scenarios safety analysis using advanced fire engineering tools. Fire Saf. Sci. 10,
that involved typical ship layouts in accommodation spaces aboard 1059–1072 10.3801/IAFSS.FSS.10-1059.
Albrecht, C., Siemon, M., Hosser, D., 2010. Application of sensitivity analysis to
ro-ro/passenger ships and cruise liners. specific problems in Fire Protection Engineering to identify the most critical
The main conclusions drawn from the research presented in parameters and to reduce dimensionality. In: Gucma, L., Proske, D., van Gelder,
P. (Eds.), Proceedings of the 8th International Probabilistic Workshop in
this paper could be summarised as follows:
Szczecin, pp. 13–26.
Andersson, C., Säterborn, D., 2002. Smoke Control Systems Aboard – A Risk Analysis
1) In all fire scenarios considered, the three predicted times to Of Smoke Control Systems In Accommodation Spaces On Passenger Ships. Lund
reach untenable conditions due to fire toxicity, heat, and smoke University, Department of Fire Safety Engineering, Lund.
Averill, J.D., 1998. Performance-Based Codes: Economics Documentation, And De-
obscuration were always affected by the uncertainties propa- sign (M.Sc. thesis). Faculty of the Worcester Polytechnic Institute, Massachu-
gated from the stochastic input parameter(s) in each scenario – setts, USA.
with the time to reach untenable condition due to fire toxicity Azzi, C., Pennycott, A., Mermiris, G., Vassalos, D., 2011. Evacuation Simulation of
Shipboard Fire Scenarios. In: Proceedings of the Fire and Evacuation Modelling
being the most severely affected output in almost all examined Technical Conference; August 15–16, 2011, Baltimore, MD, USA.
cases. Cadorin, J.F., Pintea, D., Franssen, J.M., 2001. The Design Fire Tool Ozone V2.0 –
2) The more remote the compartment from the room of fire origin, Theoretical Description And Validation On Experimental Fire Tests. Rapport
interne SPEC/2001_01 University of Liege.
the more the quantified uncertainties in the predicted outputs.
Ewer, J., Jia, F., Grandison, A., Frost, I., Galea, E., Patel, M., 2013. SMARTFIRE v4. 3:
3) In most cases, the highly affected output parameters have fol- SMARTFIRE technical reference manual, user guide and release notes. Uni-
lowed the same type of distribution as that of the most influ- versity of Greenwich, Greenwich, London, UK.
encing stochastic input parameter. FAROS, 2012. FAROS (2012–2015). Human Factors in Risk-Based Design Methodol-
ogy. EC-FP7 project.
4) The most crucial time that firstly threaten the life safety of oc- FIREPROOF, 2009. FIREPROOF (2009–2011). Probabilistic Framework for On board
cupants was the time to reach untenable condition due to Fire Safety, EC-FP7 project.
smoke obscuration, especially in the scenarios that involved Fleischmann, C., 2009. Prescribing the input for the ASET versus RSET analysis: Is
this the way forward for performance based design?, Paper presented at the
large floor areas or large number of connected compartments. Fire Protection and Life Safety in Building and Transportation Systems Work-
5) In the scenarios that involved small number of compartments, shop, 15–17 October 2009, Santander, Spain.
the most crucial time that firstly threaten the life safety of oc- Frantzich, H., 1997. Fire Safety Risk Analysis of a Hotel. Report 3091. Department of
Fire Safety Engineering, Lund Institute of Technology, Lund University, Lund,
cupants inside the room of fire origin was the time to reach
Sweden.
untenable condition due to convective and radiant heats. Frantzich, H., 1998. Uncertainty and Risk Analysis in Fire Safety Engineering. Report
TVBB-1016. Department of Fire Safety Engineering, Lund Institute of Technol-
ogy, Lund University, Lund, Sweden.
Galea, E. R., Lawrence, P., Gwynne, S., Filippidis, L., Blackshields, D., Sharp, G., Hurst,
N., Wang, Z., Ewer, J., 2003. Simulating ship evacuation under fire conditions.
Acknowledgement In: Proceedings of the 2nd International Pedestrian and Evacuation Dynamics
Conference, CMS Press, ISBN 1904521088, Greenwich, UK.
Guanquan, C., Jinhui, W., 2012. Study on probability distribution of fire scenarios in
This work was funded by the Deanship of Scientific Research risk assessment to emergency evacuation. Reliab. Eng. Syst. Saf. 99, 24–32.
(DSR), King Abdulaziz University, Jeddah, under Grant no. 940- Hakkarainen, T., Hietaniemi, J., Hostikka, S., Karhula, T., Kling, T., Mangs, J., Mikkola,
E., Oksanen, T., 2009. Survivability for ships in case of fire. Final report of
410-D1435. The author, therefore, acknowledge with thanks DSR
SURSHIP-FIRE project. VTT Res. Notes, 2497.
technical and financial support. Helton, J.C., Davis, F.J., 2003. Latin hypercube sampling and the propagation of
uncertainty in analyses of complex systems. Reliab. Eng. Syst. Saf. 81 (1), 23–69.
430 Ahmed.M Salem / Ocean Engineering 117 (2016) 411–430

Holborn, P.G., Nolan, P.F., Golt, J., 2004. An analysis of fire sizes, fire growth rates SAFER-EURORO, 1997. SAFER-EURORO I and II (1997–2005). Thematic Network,
and times between events using data from fire investigations. Fire Safety Design for Safety: An Integrated Approach for Safe European Ro-Ro Ferry De-
Journal 39 (6), 481–524. sign, Thematic Network funded by the European Community under the In-
IMO 2015a. Safety of ro-ro ferries, International Maritime Organization, 〈http:// dustrial and Materials Technologies (BRITE-EURAM III) Programme (1994–
www.imo.org/en/OurWork/Safety/Regulations/Pages/RO-ROFerries.aspx〉 (ac- 1998).
cessed 01.12.15). Salem, A., 2007. Risk-based design for fire safety of ro-ro/passenger ships. (Doctoral
IMO 2015b. Safety of ro-ro ferries, International Maritime Organization, 〈http:// dissertation, University of Strathclyde).
www.imo.org/en/OurWork/Safety/Regulations/Pages/PassengerShips.aspx〉 (ac- Salem, A., 2010. Fire engineering tools used in consequence analysis. Ships Offshore
cessed 01.12.15). Struct. 5 (2), 155–187.
Jahn, W., Rein, G., Torero, J., 2009. The effect of model parameters on the simulation Salem, A., 2013. Parametric analysis of a cabin fire using a zone fire model. Alex.
of fire dynamics. Fire Saf. Sci. 9, 1341–1352. Eng. J. 52 (4), 627–636.
Jannat, S., Greenwood, A. G., 2012. Estimating parameters of the triangular dis- Salem, A.M., Dabess, E.M., Banawan, A.A., Leheta, H.W., 2015. Fire safety design of
tribution using non-standard information. In: Proceedings of 2012 Winter Si- Nile-floating hotels. Ships Offshore Struct., 1–19, In press.
mulation Conference. Berlin, Germany. Salem A. M., Leheta H. W., 2011. Sensitivity analysis of a fire model used in fire
Julong, D., 1989. Introduction to grey system theory. J. Grey Syst. 1, 1–24. consequence calculations. In: Rizzuto E, Soares CG, editors. Sustainable Mar-
Kong, D.P., Lu, S.X., Feng, L., Xie, Q., 2011. Uncertainty and sensitivity analyses of itime Transportation and Exploitation of Sea Resources. In: Proceedings of the
heat fire detector model based on Monte Carlo simulation. J. Fire Sci. 29 (4), 14th International Congress of the International Maritime Association of the
317–337.
Mediterranean (IMAM). Genova, Italy: 13–16 September 2011.
Kong, D.P., Lu, S.X., Kang, Q.S., Lo, S.M., Xie, Q.M., 2014. Fuzzy risk assessment for life
SFPE, 2005. SFPE Engineering Guide To Application Of Risk Assessment In Fire
safety under building fires. Fire Technol. 50 (4), 977–991.
Protection Design. Society of Fire Protection Engineers, Bethesda: MD.
Kong, D.P., Johansson, N., van Hees, Patrick, Lu, S., Lo, S., 2013. A Monte Carlo
Shigunov, V., 2005. A zone model for fire development in multiple connected
analysis of the effect of heat release rate uncertainty on available safe egress
compartments. Fire Saf. J. 40 (6), 555–578.
time. J. Fire Prot. Eng. 23 (1), 5–29.
Shorab, J., Singh, A., Ansari, A., 2011. Performance-based fire safety design: pre-
McGrattan, K., Hostikka, S., McDermott, R., Floyd, J., Weinschenk, C., Overholt, K.,
diction of un-tenability conditions in assembly hall corridor using Zone and
2013. Fire Dynamics Simulator, User's Guide. National Institute of Standards
CFD Modelling. In: Proceedings of Conference on Fire Science & Technology–
and Technology (NIST) Special Publication 1019, Gaithersburg, Maryland, USA.
Novozhilov, V., 2001. Computational fluid dynamics modelling of compartment Research and its Implementations (FIRST 2011). India, pp. 277–289.
fires. Prog. Energy Combust. Sci. 27, 611–666. Themelis, N., Spyrou, K.J., 2012. Probabilistic fire safety assessment of passenger
Palisade 2015. Palisade Monte Carlo simulation products, 〈http://www.palisade. ships. J. Ship Res. 56 (4), 252–275.
com/risk/monte_carlo_simulation.asp〉 (accessed 10.02.15). Themelis, N., Niotis, S., Spyrou, K., 2011. Managing Uncertainty In Performance-
Pawling, R., Grandison, A., Lohrmann, P., Mermiris, G., Dias, C. P., 2012. The Devel- based Fire Safety Assessment Of Ships. International Maritime Association of
opment of Modelling Methods and Interface Tools Supporting a Risk Based the Mediterranean, Genova, Italy.
Approach to Fire Safety in Ship Design. In: Proceedings of the 11th International Tosolini, E., Grimaz, S., Pecile, L.C., Salzano, E., 2012. People evacuation: simplified
Conference on Computer Applications and Information Technology in the evaluation of available safe egress time (ASET) in enclosures. Chem. Eng. Trans.
Maritime Industries, Liege, Belgium, 16–18 April 2012. 26, 501–506. http://dx.doi.org/10.3303/CET1626084.
Peacock, R., 2013. CFAST-Consolidated Model Of Fire Growth And Smoke Transport Tuovinen, H., Hertzberg, T., 2009. Simulation of fires in a RoPax vessel (SP Report
(Version 6) User’s Guide. US Department of Commerce, National Institute of 2009:02). SP Technical Research Institute of Sweden, Borås, ISBN 978-91-
Standards and Technology. 85829-85-9.
Purser, D., 2002. Toxicity assessment of combustion products. SFPE Handbook of Wade, C., 2004. BRANZFIRE Technical Reference Guide, Study Report No.92.
Fire Protection Engineering, 3rd ed. NFPA, Quincy, MA. Building Research Association of New Zealand, Judgeford, Porirua City, New
Rockwool, 2012. Rockwool Technical Insulation, Marine and Offshore Insulation Zealand.
Worldwide. Rockwool, Denmark. Wang, J., Chu, G., Li, K., 2013. Study on the uncertainty of the available time under
SAFEDOR, 2005. SAFEDOR (2005–2009). Design, Operation and Regulation for ship fire based on Monte Carlo sampling method. China Ocean Eng. 27,
Safety, EC-FP6 project. 131–140.

View publication stats