A method of quantifying user

uncertainty in FDS by using

Monte Carlo analysis

Tor Lindström
David Lund

Department of Fire Safety Engineering

and Systems Safety
Lund University, Sweden

Brandteknik och Riskhantering

Lunds tekniska högskola
Lunds universitet

Report 5309, Lund 2009

A method of quantifying user
uncertainty in FDS by using
Monte Carlo analysis

Tor Lindström
David Lund

Lund 2009
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

Tor Lindström, David Lund

Report 5309
ISSN: 1402-3504
ISRN: LUTVDG/TVBB--5309--SE

Number of pages: 61
Illustrations: Tor Lindström, David Lund

Keywords:
Computational fluid dynamics, CFD, Monte Carlo, Fire Dynamics Simulator, FDS,
parallel computing, sensitivity analysis, uncertainty, distributions

Sökord:
Fluiddynamik, CFD, Monte Carlo, Fire Dynamics Simulator, FDS, parallella
datorberäkningar, känslighetsanalys, osäkerhet, fördelningar

Abstract
Several hundred FDS simulations have been run using Monte Carlo analysis and
probability distributions of input variables. The simulations have been run to investigate
the propagation of uncertainty from model inputs to outputs in order to quantify
uncertainties that could arise due to the choices of an FDS user. A method has also been
presented for performing FDS Monte Carlo analyses. The results show that a large degree
of uncertainty can arise in output data due to the user and that this uncertainty can be
quantified.

© Copyright: Brandteknik och Riskhantering, Lunds tekniska högskola, Lunds

universitet, Lund 2009

Författarna svarar för innehållet i rapporten

Brandteknik och Riskhantering Department of Fire Safety Engineering

Lunds tekniska högskola and Systems Safety
Lunds universitet Lund University
Box 118 P.O. Box 118
221 00 Lund SE-221 00 Lund
Sweden

[email protected] [email protected]
http://www.brand.lth.se http://www.brand.lth.se/english

Telefon: 046 - 222 73 60 Telephone: +46 46 222 73 60

Telefax: 046 - 222 46 12 Fax: +46 46 222 46 12
 Summary

Summary
When using FDS for fire modelling there is a degree of uncertainty concerning
input variables. When obtaining results from a simulation, there is consequently
a degree of uncertainty in the results introduced by the user. In this thesis, several
hundred FDS simulations of the same model have been run, with three input
variables, being soot yield, mass extinction coefficient and heat release rate,
independently varied by use of Monte Carlo analysis. These three variables were
selected by using a sensitivity analysis. Finally, simulations have also been run
where all three variables were varied simultaneously. The output quantity
investigated in all simulations was visibility. Distributions of the input variables
have been created based on a literature study and qualitative reasoning. Random
samples from these distributions have then been used in the simulations.

In order to create the vast number of input files needed, custom software has
been written for this purpose. In order to handle the massive amount of output
data, custom software was written for this purpose as the total amount of output
data comprises more than 285 million values.

The results from the simulations show that there is a large degree of uncertainty
in the output data when using Monte Carlo simulations in FDS given the chosen
distributions of input variables. If a user simply chooses a static value of any of
the input parameters in this thesis, there is a high probability that the results
could vary outside accepted criteria. The results of the subject simulations show
that at a height of 1.4 m, for example, 65 percent of the simulations yield non-
acceptable results. However, this value is only applicable to the circumstances in
this thesis but clearly show that there is a risk that the simulation yields
unacceptable results.

The method of using Monte Carlo analysis directly on FDS simulations has not
been attempted before, and this thesis represents a first step in managing
uncertainty in FDS by connecting it to risk analysis. The method could be used
by any FDS user to perform multiple Monte Carlo-based FDS simulations, thus
managing the uncertainty that often arises due to user uncertainty.

i
 Sammanfattning

Sammanfattning
När FDS används för brandmodellering så finns det en viss grad av osäkerhet vad
gäller indata. Således finns det också en grad av osäkerhet i utdata som har
introducerats av användaren. I detta examensarbete har flera hundra FDS-
simuleringar av samma modell utförts. Tre olika indatavariabler (soot yield, mass
extinction coefficient och heat release rate) har tagits fram genom en
känslighetsanalys och sedan varierats oberoende av varandra genom en Monte
Carlo analys. Slutligen har också simuleringar körts där alla tre variabler
varierades samtidigt. Utdata som undersöktes som undersöktes i alla simuleringar
var sikt. Fördelningar över indatavariablerna har skapats baserat på
litteraturstudier och kvalitativt resonemang. Slumpmässiga stickprov från dessa
fördelningar har sedan använts i simuleringarna.

För att kunna skapa det stora antal indatafiler som krävdes har egna program
skrivits. För att kunna hantera den enorma mängden utdata så skrevs även egna
program för detta ändamål, då den totala mängden utdata bestod av fler än 285
miljoner värden.

Resultatet från simuleringarna visar att det finns en stor grad av osäkerhet i
utdata när man använder Monte Carlo-simuleringar i FDS baserat på de valda
indatafördelningarna. Om en användare enbart använder ett deterministiskt
värde på någon av indatavariablerna som undersökts i detta arbete så finns det en
hög sannolikhet att resultaten skulle kunna hamna utanför valda
acceptanskriterier. Resultaten från simuleringarna visar till exempel att på en
höjd av 1.4 meter ovanför golvet så ger 65 procent av simuleringarna icke-
acceptabla resultat. Det skall understrykas att detta värde endast är giltigt för
omständigheterna som gäller i detta arbete, men att resultaten tydligt visar att det
finns en risk att simuleringen ger oacceptabla resultat.

Att använda Monte Carlo-analys direkt på FDS-simuleringar har inte gjorts

tidigare, och detta examensarbete är ett första steg i att hantera osäkerhet i FDS
genom att koppla det till riskanalys. Metoden kan användas av alla FDS-
användare för att utföra flera Monte Carlo-baserade FDS-simuleringar och
därmed hantera osäkerheten som ofta uppstår på grund av användarosäkerhet.

iii
 Preface

Preface
The authors would like to thank the following persons for their help during the
work on this thesis:

Håkan Frantzich Supervisor, Lund University

Patrick van Hees Supervisor, Lund University

Daniel Nilsson Lund University

v
Table of contents
1 Introduction 1
1.1 Background 1
1.2 Objectives 2
1.3 Limitations 3

2 Methodology 5

3 Fire Dynamics Simulator (FDS) 9

3.1 Brief history of CFD 9
3.2 Features of FDS 10
3.3 Validation and verification 13

4 Monte Carlo simulations in FDS 15

4.1 Probability and FDS 15
4.2 Latin Hypercube Sampling 15
4.3 Software 16

5 Simulated fire experiment and FDS setup 17

5.1 Materials 18
5.2 Meshes and parallel processing 19
5.3 Mesh cell size 19

6 Investigated output variable 21

6.1 Path 22
6.2 Width 23
6.3 Time 25

7 Selection of input variables for Monte Carlo simulations 27

7.1 Qualitative reasoning 27
7.2 Variables chosen for sensitivity study 30
7.3 Sensitivity study 30

vii
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

8 Distributions of input variables 37

8.1 Soot yield 37
8.2 Mass extinction coefficient 38
8.3 Heat Release Rate 40
8.4 Monte Carlo sampling 41

9 Results from simulations 43

9.1 Independent variable simulations 44
9.2 Simultaneous variable simulations 49
9.3 Summary of results 50

10 Discussion and conclusion 51

10.1 Conclusion 51
10.2 Discussion of assumptions and method 52
10.3 Discussion of results 54
10.4 Errors 56
10.5 Recommendations to FDS users 57
10.6 Further research 58

11 References 59

Appendix A – Parallel computing with FDS 63

Appendix B – Sampled values 65

Appendix C – Input files 69

Appendix D – Selected figures 79

Appendix E – Output summary 93

viii
1 Introduction

This master thesis is written at the Department of Fire Safety Engineering and
Systems Safety at Lund University during the fall of 2009. The objective is to
investigate how a user’s choice of input values variables may influence the
resulting output data when using the computational fluid dynamics (CFD) code
1
Fire Dynamics Simulator (FDS) and also to provide a method of quantifying
the uncertainty.

1.1 Background
The Swedish building code [1] allows a proprietor of a building to adopt
performance based design as opposed to prescriptive design. This allows for
analytical fire design solutions to fulfil the requirements of the building code
instead of following the prescriptive guidelines.

The legislative responsibility to ensure the safety of individuals using any

building lies upon the proprietor according to the Swedish building code
(SFS 1987:10). In order to do this the proprietor usually hires a fire engineering
consultant, many of whom use CFD codes in order to simulate the spread of
smoke. The outcome of this simulation is used as a basis for the design of fire
mitigating measures. However, the outcome of any fire simulation is directly
linked to the choice of input variables and the choice of values for these variables.
Any changes to these values will affect the outcome, and it is not certain that the
choice of values is always justified. It is also possible that some FDS users are not
aware of the number of variables that are possible to change and how these
choices affect the outcome. In addition, since a consultant is a private actor on
the market, he or she is driven by economic interest but also by the need to
deliver a solution to a client within a reasonable timeframe. This means that
there is a necessity to be time and cost efficient. The usage of CFD in this

1
For more on CFD and FDS, see Section 3.

1
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

context could mean that some decisions about the design are made in a hastily
order, without conducting a thorough sensitivity analysis of choices made.

Since 2000 the CFD code FDS is available free of charge and is being actively
developed by The National Institute of Standards and Technology (NIST) in the
USA. This has resulted in FDS5 (the current version at the time of writing)
being the perhaps most readily used CFD code today. At the same time the
computational power of computers has increased radically. These two factors
combined has allowed for more frequent use of FDS as a simulation tool. Since
the outcome of any FDS simulation is the direct result of the input variables and
their respective values a user chooses, the quality of simulations are bound to
differ.

This thesis seeks to investigate the uncertainty of input variables chosen by an

FDS user, and how these choices affect the results.

There has been some previous work in the area of probability and CFD
modelling. Hostikka [2] has created a model that combines Monte Carlo analysis
with CFD modelling. That approach is not the same as the approach used in this
thesis. Hostikka combined the results of fast but approximate two-zone models
with more accurate but computationally demanding CFD models.

Najm [3] used probabilistic uncertainty quantification methods to investigate the

uncertainty that propagates from input to output in a number of different CFD
models when the input has been characterized probabilistically. The CFD
models considered were not of fire design character but rather more general.

1.2 Objectives
The objective of this thesis is to make a connection between FDS and risk
analysis by demonstrating a method of doing a Monte Carlo analysis of an FDS
simulation. This will be done by showing how the choice of values of input
variables affects the result of an FDS simulation, from a statistical perspective.

The thesis also intends to provide basic data on levels of uncertainty when using
FDS by applying the method to a fire scenario.

In order to achieve this, the following questions will need to be answered:

 Which input variables are the most sensitive in an FDS simulation,

based on the chosen output data?
 To what degree does the input data affect the output data?

2
 Introduction

 How uncertain is the output of an FDS simulation given a variation in

the input data?
 What recommendations can be made to FDS users?

1.3 Limitations
Certain limitations have been made. These limitations are listed below.

 Only one CFD code was used, namely FDS5

 Only the selected input variables were further investigated
 The input variables were assumed not to be correlated
 Only one scenario was used for comparison
 Only one output quantity was investigated, being visibility along a path
 Visibility was measured in cells, not along a path
 Placement of mesh boundaries was not considered in the analysis
 No investigation of parallel versus serial simulations was conducted
 The heat release rate in any given scenario was constant
 All simulations were well-ventilated in FDS

Many of these limitations have been made in order to limit the number of
simulations that need to be conducted.

The output quantity, visibility, was measured in cells. This means that the
measurements are point measurements, as opposed to measurements along a
path. This is mainly due to technical limitations. This issue is discussed further
in Section 6.

The simulations use several meshes and parallel processing, but the placement of
the meshes was not investigated in any manner. The mesh placement has been
based on general recommendations [4,5], and the placement was identical in all
simulations. There has been no investigation of errors due to running the
simulations on parallel processors. Any errors that may have been introduced
from doing so will have been introduced in all simulations and thus would have
had no or little effect on the results from comparing simulations of different
values of input variables [5].

Since the input variables act within given intervals, the smoke layer is bound to
vary within certain heights as well. It would be possible to record FDS
measurements at the height required by the building code, but it is possible that
measurements at this height do not vary that much. Because of this it may not be
possible to take current fire design legislation into account. For the purpose of

3
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

this thesis it is more interesting to show differences in measurements, and

because of this a measurement height will be chosen where there is a larger
variation in measurements.

4
2 Methodology

To handle the uncertainty introduced by a user in FDS by connecting it to risk

analysis it is necessary to develop a method. The method needs to take into
account the inherent uncertainty of the input variables by some means. In order
to produce meaningful results from a large array of possible input variables, it is
also necessary to use some sort of statistical method to perform multiple
simulations in FDS. In this thesis, multiple simulations were performed by way
of Monte Carlo modelling. This section details the methodology used in this
thesis in applying this model.

In the initial part of this thesis, a scenario was chosen to use in all simulations,
where the input parameters (materials, building dimensions etc.) were
representative of a real building. This scenario has then acted as a foundation for
all simulations.

Finding relevant input variables has been done in two ways; partly through
qualitative reasoning and partly through a sensitivity study where one variable at
a time was varied to see how that variation affected the output.

The next step was to find and decide the distributions of the input variables. This
was done mostly in a qualitative manner.

Once the distributions were chosen, the variables were sampled randomly
according to their previously determined distributions. To achieve a large enough
base for the Monte Carlo simulations but still weigh it against the computational
load of running CFD simulations, each variable was sampled 200 times. These
samples of input variables have then been simulated in FDS5, independent of
each other. There have also been simulations where all chosen input variables
were varied simultaneously according to their previously determined
distributions.

As an output quantity, this thesis only looks at visibility along a path. Visibility is
a common criterion in fire safety engineering, together with temperature and

5
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

incident radiation [6]. However, due to the amount of data it was deemed
necessary to limit the scope of this thesis to only one output quantity.

The methodology described above is visualised in the flowchart in Figure 2.1 and
is represented by straight lines.

The “user risk flow” (perhaps more correctly termed the user uncertainty flow) is
also shown in the flowchart, as shown by the dotted lines, and originates with the
user. User risk flow, in this context, means how the uncertainty essentially
“created” by the user propagates to the output that FDS produces. The input to
FDS is coupled with a certain uncertainty that in term could be represented by
probability distributions. When running FDS, these uncertainties are propagated
to the output data produced by FDS and in turn yields probability distributions
of that output data. An FDS user would not, however, typically use probability
distributions when choosing input data. Rather, the input data would consist of a
best guess (maybe based on an assumed worst-case scenario) or perhaps just
common practice for that user. This thesis intends to investigate how different
input values, based on probability distributions, affect the output and also
demonstrate a method for doing this.

6
 Methodology

FIGURE 2.1 Flowchart showing the workflow methodology for this thesis

7
3 Fire Dynamics Simulator (FDS)

Computational Fluid Dynamics (CFD) is a branch of fluid dynamics that uses

numerical methods to solve the governing equations of fluid flows using
computers. CFD models are sometimes termed field models.

The National Institute of Standards and Technology (NIST) has for several years
developed a CFD code for fire dynamics applications. Their code, Fire Dynamics
Simulator (FDS), solves a form of the Navier-Stokes equations appropriate for
fire-driven fluid flow with an emphasis on smoke and heat transport from
fires [4]. The first public version of FDS was released in 2000.

3.1 Brief history of CFD

The Navier–Stokes equations, named after Claude-Louis Navier and George
Gabriel Stokes, describe the motion of fluid substances; that is substances which
can flow. Their form of the differential mathematical equations, proposed nearly
200 years ago, are the basis of the modern day computational fluid dynamics
applications. The Navier-Stokes equations contain expressions for the
conservation of mass, momentum, pressure, species and turbulence. The
equations are so closely coupled and difficult to solve that it was not until the
advent of modern computers in the 1960s that they could be resolved
numerically for real flow problems within reasonable timeframes [7].

The use of computers as a tool in fire protection engineering has increased with
the development of more powerful computers. Historically, two-zone models
were the first type of models to be widely used and accepted. These models
assume two gas layers in an enclosure, one hot gas layer and one cold gas layer.
As computational power increased, the possibility of CFD modelling became
possible [8].

CFD modelling is a much more complex way of modelling a fire than two zone
modelling. The CFD programme numerically solves the Navier-Stokes equations

9
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

in each of the cells in the computational domain. Because of the size of the
computational domain in a simulation, powerful computers are needed.

There are three general approaches to solving the turbulence within the Navier-
Stokes equations, Direct Numerical Solution (DNS), Reynolds Averaged Navier
Stokes (RANS) and Large Eddy Simulation (LES). The RANS and LES
approaches solve the equations with different turbulence models while DNS
solves the Navier-Stokes equations numerically without any turbulence model, as
it is not needed. A DNS solution requires the size of the individual cells to be at
-6
the Kolmogorov micro scale, or 10 m of length [9], which is not realistic for
larger models considering the computational power of modern computers.

RANS averages instantaneous values in time. Due to this, this approach is suited
for steady state type problems and solutions. A programme that uses RANS code
is SOFIE (Simulation of Fires in Enclosures). As the aim of this thesis is to
investigate uncertainties in FDS, SOFIE is not further discussed [8].

LES assumes that all turbulent energy is preserved in the largest scale. This means
that nothing that occurs below this scale is calculated. Any phenomena that
occur are instead modelled by sub grid models. As opposed to the RANS
approach, LES is transient, which means that all calculations use output from the
previous time step as input to the next time-step. Because LES is transient as
opposed to RANS, it requires much more time to calculate all time steps. FDS is
probably the most used LES programme for fire engineering applications today.

3.2 Features of FDS

This section summarises the main features in FDS. For more detailed
information, refer to the FDS user guide and technical guide [4,10].

HYDRODYNAMIC MODEL
FDS solves a form of the Navier-Stokes equations appropriate for low-speed,
thermally-driven flow. Turbulence is treated by means of Large Eddy Simulation
using a Smagorinsky sub-model [4].

COMBUSTION MODEL
For most applications, FDS uses a single step chemical reaction whose products
are tracked via a two-parameter mixture fraction model. The mixture fraction is a
conserved scalar quantity that represents the mass fraction of one or more
components of the gas at a given point in the flow field. By default, two
components of the mixture fraction are explicitly computed. The first is the mass

10
 Fire Dynamics Simulator (FDS)

fraction of unburned fuel and the second is the mass fraction of burned fuel (i.e.
the mass of the combustion products that originated as fuel) [4].

RADIATION TRANSPORT
Radiative heat transfer is included in the model via the solution of the radiation
transport equation for a grey gas, and in some limited cases using a wide band
model. The grey gas model assumes uniform radiation over all frequencies, while
the wide band model divides the radiation frequencies in sex different bands. The
radiation equation is solved using a technique similar to a finite volume method
for convective transport coupled with the absorption coefficient of the gas
mixture [4,10].

MESHES AND PARALLEL PROCESSING

FDS approximates the governing equations on a rectilinear mesh. Any
rectangular obstructions are forced to conform to the underlying mesh. Since its
first public release, the ability to use several meshes has been introduced. This
feature is crucial for larger simulations where the computational domain is not
easily embedded within a single mesh.

It is possible to run an FDS calculation on more than one computer using

parallel processing. This is discussed in more detail in Appendix A.

The computational mesh that FDS uses to perform its calculations on can exist
of several different rectilinear meshes, all of whom are connected to each other.
Meshes can also be placed without abutting, in which case a separate calculation
is essentially carried out on each zone of meshes without any communication
with other meshes. This latter approach is not employed in this thesis. When
using multiple meshes, they can (by default) be set to be synchronized with each
other, meaning that the mesh with the smallest time step in each iteration will
control the time step of all other meshes [4]. All models in this thesis are
simulated using synchronized meshes

MESH CELL SIZE

In any CFD calculation (not only related to fire applications) all calculations take
place in several cells (essentially control volumes) on a larger computational
mesh. The size and geometry of these cells is perhaps the single most important
factor for attaining good results from a CFD calculation [4]. If the cells are too
large, the model cannot accurately capture the fluid dynamics of the problem.
On the other hand, if the cells are too small, there is no way to run the model
with today’s computers.

11
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

For fire applications, it is crucial to properly resolve the actual fire, as the fire is
what drives the buoyant flow. Entrainment of air is highly dependent on cell size
and is the most important factor in smoke production from any fire. It has been
found [11] that a critical parameter for an FDS model is the non-dimensional
ratio 𝐷𝐷∗ /𝛿𝛿𝛿𝛿 where D* is the characteristic fire diameter and δx is the mesh cell
size across the fire according to Equation 3.1.
2/5
∗
𝑄𝑄̇
𝐷𝐷 = � � (3.1)
𝜌𝜌∞ 𝑐𝑐𝑝𝑝 𝑇𝑇∞ �𝑔𝑔

If this parameter is sufficiently large, the fire can be considered well-resolved

[11]. Studies have shown that values of 10 or more are required to resolve most
fires well and attain adequate flame temperatures [12,13].

To have a grid resolution such that 𝐷𝐷∗ /𝛿𝛿𝛿𝛿 ≥ 10 with a fire size of 300 kW
(which is the steady fire size of the target experiment) a cell size (using cubic
cells) of roughly 0.05 m would be needed.

BOUNDARY CONDITIONS
All solid surfaces (termed obstructions in FDS) are assigned thermal boundary
conditions and if necessary information about the burning behaviour of the
material. Heat and mass transfer to and from solid surfaces is handled with
empirical correlations using 1-D heat transfer.

GEOMETRY
As previously mentioned, FDS employs a rectilinear computational mesh. Any
obstructions introduced into the model need to conform to this rectilinear mesh.
A non-rectangular geometry needs to be approximated using stair-stepping
methods, as is visualized in Figure 3.1.

12
 Fire Dynamics Simulator (FDS)

FIGURE 3.1 Schematic example of stair-stepping of sloping object.

3.3 Validation and verification

FDS has been extensively validated against experimental data. This is partly
summarised in [11]. The numerical and mathematical models, which are
presented in [10], have also been the subject of verification and robustness tests
[14]. Verification and sensitivity analysis of the mathematical and numerical
model are not subjects of this thesis.

13
4 Monte Carlo simulations in FDS

Monte Carlo simulation is a method of performing multiple experiments using

random sampling of variables that belong to a certain probability distribution. As
Monte Carlo methods rely on repeated computation and random numbers, they
are most suited for use with computers and when it is unfeasible or impossible to
compute an exact result with a deterministic algorithm [15].

4.1 Probability and FDS

In FDS, any outcome, such as the temperature in a certain location or the
visibility along a path, is a function of all possible input variables having an effect
on that outcome. For example, if we define an event E as the visibility along a
path being above a certain value s, the probability of E happening depends on a
number of random variables, each having a probability distribution. These
random variables can be denoted by a vector 𝑿𝑿 = [𝑋𝑋1 , 𝑋𝑋2 , … , 𝑋𝑋𝑛𝑛 ]𝑇𝑇 . The
probability that E happens is then a function of X and time, as FDS is a transient
or time-dependent model:

𝑃𝑃𝑃𝑃𝐸𝐸 (𝑆𝑆 ≥ 𝑠𝑠) = 𝑔𝑔(𝑿𝑿, 𝑡𝑡) (4.1)

PrE belongs to a random distribution and is in this thesis calculated using Monte
Carlo simulations where the input variables are sampled randomly from their
respective distributions. The usability of a Monte Carlo simulation, however,
depends on the number of random samples and thus the number of simulations
that needs to be made.

4.2 Latin Hypercube Sampling

The sampling simulations are made using a sampling scheme called Latin
Hypercube sampling (LHS). LHS uses a technique called stratified sampling
without replacement [16]. When using LHS, the probability distribution, also

15
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

called the cumulative distribution function F(x), is split into n intervals of equal
probability, where n is the number of iterations that are performed. In the first
iteration, one of these intervals is selected using a random number. Another
random number is subsequently generated; deciding where within that interval
F(x) should lie. A sampled value is then calculated for that value of F(x). The
process is repeated for the number of desired iterations, but for each iteration the
interval used is marked and will not be selected again. As the number of
iterations is the same as the number of intervals, each interval will only have been
sampled once and the distribution will be reproduced with predictable
uniformity over the F(x) range [17].

4.3 Software
For the purpose of this thesis, different pieces of software were created. All
software written could be customised for other applications in this area.

4.3.1 Creating input files

In order to run several hundred simulations with FDS, a separate input file needs
to be created for each simulation. In each of these files, the values of the chosen
variables need to be changed to the values from the Monte Carlo sampling
process. For this purpose, an application was written to create these input files
and insert a list of sampled values into each file. The software is customizable,
and depends on user input for several parameters. To create the FDS input files,
desired filenames (and FDS headers), number of files to create and desired
location of the created files is entered by the user. For use on LUNARC (see
Appendix A), or any cluster using the same queue management system, script
files are also created. The user can customize the desired number of CPU:s, wall
time, e-mail address for notifications and the path to the FDS executable.

4.3.2 Managing output data

In order to interpret the considerable amount of data created from the
simulations, software also needed to be written for this purpose. The software
takes as input the path where FDS output files are stored. The software reads the
comma separated device output files that FDS creates and prints time-averaged
values from these files based on desired quantities.

16
5 Simulated fire experiment and FDS setup

As a base for the models in this thesis, a well-documented fire experiment was
used. The experiment was conducted by the Centre for Environmental Safety
and Risk Engineering at Victoria University in Australia [18]. The specific
experiment was set in a multi-room building containing three rooms and one
corridor and is shown in Figure 5.1. It was deemed suitable to use as a model in
this thesis because of its layout, being similar to part of a school, hospital, student
housing or such. Except for the location of the fire source, geometry and basic
buildings materials, no other specifics of the experiment have been used for the
simulated models.

FIGURE 5.1 Layout of the experiment; grey areas indicate where the experiment took
place, all other areas were sealed off.

17
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

The cell size of the mesh affects the size of any obstructions in the model.
Obstructions cannot have dimensions smaller than the mesh cell size. The
general geometry of the model has been faithfully reproduced according to
Figure 5.1. It should also be noted that FDS can calculate heat transfer through
obstructions regardless of their geometrical size on the mesh. The desired
thickness and material composition of obstructions is entered separately, and is
used by the heat transfer code in lieu of the geometrical size. Figure 5.2 shows
the FDS model as seen in Smokeview. General information on the model can be
found in Table 5.1.

FIGURE 5.2 The FDS model as seen in Smokeview.

TABLE 5.1 General information on the FDS model

Parameter Value
Number of meshes 14
Mesh cell size 0.1 m (cubic cells)
Radiative fraction 35 %
Ambient temperature 20 ºC

5.1 Materials
The materials used in the model were made to resemble those of the experiment
to the extent possible. The walls were made of concrete and gypsum, with no
material properties given in the experiment. For this reason, common values of
respective material properties [19] are given in Table 5.2 below.

18
 Simulated fire experiment and FDS setup

TABLE 5.2 Properties of the materials used in the model

Material Density (kg/m3) cp (kJ/kg K) k (W/m K)
Gypsum 1,440 0.84 0.48
Concrete 2,100 0.88 1.1

5.2 Meshes and parallel processing

For the purposes of parallel processing, any FDS model must be divided into two
or more meshes. Each processor in any parallel computation needs at least one
mesh to work with. Due to the large number of simulations needed for this
thesis, using parallel processing was a necessity to be able to run them within a
reasonable timeframe. For this purpose, the subject models were divided into a
total of 14 meshes.

5.3 Mesh cell size

The simulations in this thesis are run with a resolution of 0.1 m, even if it had
been desirable to run the simulations with a cell size of 0.05 m, as stated in
Section 3.2. Due to the number of simulations however, this was not possible.
Still, it is assumed that any error that may be introduced from running
simulations with a cell size of 0.1 m is equally introduced in all simulations,
consequently not affecting any comparison between simulations.

19
6 Investigated output variable

As an output quantity, this thesis will only look at visibility along a path.
Visibility is a common criterion in fire safety engineering, together with
temperature and incident radiation [6]. However, due to the amount of data it
was deemed necessary to limit the scope of this thesis to only one output
quantity. Also, the purpose of this thesis is to demonstrate a method, and for that
reason one output quantity is adequate.

In order to further limit the amount of data, it was decided to only take
measurements of visibility in the corridor. The corridor was chosen partly
because it is more interesting from a fire design perspective since it may be used
to evacuate. The other reason for choosing the corridor was that the visibility of
the other rooms was quickly reduced to zero in most simulations.

Visibility is measured in single cells at different heights. However, since visibility

is a quantity that typically needs to be evaluated along a path, it is necessary to
investigate if the point measurements are representative of measurements along a
path. It is also necessary to evaluate if the measurements used in the thesis, taken
in the middle of the corridor, differ from measurements taken in adjacent cells.
For these purposes, a study has been performed to find any differences in
measuring visibility in a single cell, compared to measuring over a path and
width, respectively. The measurement points can be seen in Figure 6.1 below.
The measurements used in the thesis correspond to location 3 (the middle of the
corridor) in position 1.

It was also deemed necessary to find if there were any differences in

measurements over time.

21
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

FIGURE 6.1 The corridor showing thermocouple locations used in the study.

6.1 Path
To investigate what potential errors point measurements may introduce,
simulations have been conducted where the visibility is measured at several
positions, at the same height throughout the corridor of the geometry, to try and
emulate a sight path. As shown in the Figure 6.2 below, the measurements closer
to the corridor exit indicate a slightly better visibility at lower heights, compared
to the other measurement positions. This is not entirely unexpected since the
smoke layer slopes in the corridor.

As the simulations show, measuring visibility in single cells is not an optimal

solution, but as the differences are small, and the purpose is to compare different
simulations against each other, this method has been employed in this thesis.

22
 Investigated output variable

Tree 1
2,5
Tree 2
Tree 3
2
HEIGHT (m)

1,5

0,5

0
0 5 10 15 20 25 30 35
VISIBILITY (m)

FIGURE 6.2 Differences in visibility at different location in the corridor.

6.2 Width
The values of visibility in the measured cells will also be compared to the values
of visibility in the connecting cells, to find any differences over the width of the
corridor. This is done simply by running a simulation with points of
measurement added to the adjacent cells, at a total of three positions in the
corridor, with each position having five locations across the width of the corridor
as is visualised in Figure 6.1.

Figure 6.3 to Figure 6.5 below shows the results at steady state for all
measurement locations at heights from floor to ceiling. As can be seen, there are
only slight differences across the width of the corridor.

23
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

3
Location 1 (Leftmost)

2,5 Location 2
Location 3 (Middle)
2
Location 4
HEIGHT (m)

1,5 Location 5 (Rightmost)

0,5

0
0 5 10 15 20 25 30 35

VISIBILITY (m)

FIGURE 6.3 Measurements at position 1 in the corridor.

3
Location 1 (Leftmost)

2,5 Location 2
Location 3 (Middle)
2 Location 4
HEIGHT (m)

Location 5 (Rightmost)
1,5

0,5

0
0 5 10 15 20 25 30 35
VISIBILITY (m)

FIGURE 6.4 Measurements at position 2 in the corridor.

24
 Investigated output variable

3
Location 1 (Leftmost)

2,5 Location 2
Location 3 (Middle)
2 Location 4
HEIGHT (m)

Location 5 (Rightmost)
1,5

0,5

0
0 5 10 15 20 25 30 35
VISIBILITY (m)

FIGURE 6.5 Measurements at position 3 in the corridor.

6.3 Time
Visibility fluctuates over time with smoke movement in the enclosure. In order
to have viable measurements it is necessary that the visibility at the measured
locations does not fluctuate too much. It was established that visibility had
reached steady state in the last 200 seconds, as shown in the diagram below. Due
to instantaneous fluctuations, all diagrams have been time averaged over that
time.

25
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

35 Corridor 0.7 m
Corridor 1.3 m
30
Corridor 1.4 m
25 Corridor 1.7 m
VISIBILITY (m)

20 Corridor 2.1 m

15

10

0
0 100 200 300 400 500 600 700 800 900 1000

TIME (s)

FIGURE 6.6 Visibility over time at different heights at position 3. Steady state at t > 750 s.

All diagrams showing the visibility at any location have been time averaged over
the last 200 seconds in order to show steady state conditions.

26
7 Selection of input variables for Monte Carlo
simulations

In order to analyze how output data from FDS varies a number of different input
variables need to be considered. Qualitative reasoning combined with a
sensitivity analysis has been used to find the most relevant variables to use.

7.1 Qualitative reasoning

7.1.1 Factors affecting visibility
There are several factors affecting visibility in any real or simulated fire scenario.
At its core, it is the smoke particulate that limits visibility. In the absence of
particles, visibility is, in some sense, “infinite”. The characteristics of fire smoke
include the composition, shape and size of the smoke particles, which in turn
depend on the combustible materials involved and the conditions of combustion.
These characteristics are also highly dependent on the surrounding flow and
temperature field and vary with time [20]. In FDS, the amount of fire smoke
created is based on the soot yield. This is simply the fraction of fuel mass that is
directly converted into smoke particulate.

There are two reasons for the decrease in visibility through smoke. Assuming an
observer looking at an exit sign through smoke, the reflected light from the sign
and its background is interrupted by smoke particles which reduce the intensity
of the light as it reaches the observer. Furthermore, the reflected light from the
general lighting of corridors or rooms is scattered by the smoke particles, further
impacting visibility [20]. There has been some research on smoke movement
taking into account direct illumination and indirect illumination from surfaces
and particulate scattering [21]. This thesis does not take this into account.

The most useful quantity for assessing visibility is the light extinction
coefficient [20]. The intensity of monochromatic light passing a distance L

27
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

through smoke (as described in the previous paragraph) is attenuated according

to

𝐼𝐼
= 𝑒𝑒 −𝐾𝐾𝐾𝐾 (7.1)
𝐼𝐼0
Where K is the light extinction coefficient [4], L is the path length and I and I0 is
the attenuated intensity and the source intensity, respectively. The light
extinction coefficient is a product of the density of smoke particulate, 𝜌𝜌𝑌𝑌𝑠𝑠 , and a
mass extinction coefficient that is fuel dependent

𝐾𝐾 = 𝐾𝐾𝑚𝑚 𝜌𝜌𝑌𝑌𝑠𝑠 (7.2)

Estimates of visibility through smoke can then be made using equation (7.3) [4],
[20]

𝐶𝐶
𝑆𝑆 = (7.3)
𝐾𝐾
where C is a non-dimensional constant that depends on the object being viewed
through the smoke. This constant is based on a correlation between visibility of
test subjects and the extinction coefficient which found that the visibility of light
emitting signs were two to four times greater than light reflecting signs [22].
Values of C = 8 for light-emitting signs and C = 3 for light-reflecting signs were
found to correlate well with the data.

7.1.2 Other factors

Several other factors have a potential effect on visibility, including temperatures
and flow characteristics. The most prominent factor that effectively decides the
rate at which energy, and thus also smoke, is released into the model is the heat
release rate (HRR). In a field model, the HRR is the source term in the energy
equation, hence its importance [10]. In FDS, a fire is typically specified using a
boundary condition of a specific area where heat is released according to a
specific heat release rate per unit area (HRRPUA). It is thus possible to achieve
the same total HRR by using different areas and heat release rate densities of the
boundary condition. Some research has been done showing that the heat release
rate density in FDS has an effect on flame temperatures [12].

In FDS there must by necessity be a gas phase reaction. In effect, the user
specifies a gas phase reaction that acts as a surrogate for all potential fuel sources

28
 Selection of input variables for Monte Carlo simulations

[4]. Even if no reaction is specified, FDS uses the default reaction which is that
of propane.

In order to further investigate the effect temperature has on smoke production

and thus visibility, independent of the HRR, material characteristics
(conductivity, density and specific heat) and radiative fraction will also be
considered. Lower gas temperatures mean lower density and thus a lower gas
layer height, possibly affecting visibility at that height.

Heat loss through walls has an effect on the temperature of the hot gases in the
room. If the heat loss is greater than the gas temperature will be lower, and vice
versa. FDS uses a one-dimensional equation in order to determine this heat loss.
The one-dimensional heat transfer caused by conductivity through walls and
ceiling of a scenario are directly linked to the material characteristics, thermal
conductivity (k), density (ρ) and specific heat capacity (c). For the materials used
in the subject model, the conductivity is the dominant factor of these three. For
the purposes of this sensitivity analysis, the product of all three values has been
changed together.

The radiative fraction represents the fraction of energy released from the fire as
thermal radiation. Variations in this input variable will consequently mean
different amounts of energy released as convective energy as well. The effect the
radiative fraction has on temperature may affect the visibility.

29
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

7.2 Variables chosen for sensitivity study

Based on Section 7.1, the following variables have been chosen to be further
investigated in the sensitivity study.
TABLE 7.1 Variables chosen for sensitivity study
Variable FDS Name Default value
Mass extinction coefficient MASS_EXTINCTION_COEFFICIENT 8,700 m2/kg
a
Fire Area/HRRPUA (HRRPUA) N/A
Soot yield SOOT_YIELD 0.01 g/g
Heat Release Rate (HRRPUA)b N/A
c
Reaction (fuel) (REAC) Propane
Radiative fraction RADIATIVE_FRACTION 0.35
Conductivity (kρc) (CONDUCTIVITY)d N/A
Cell size (MESH)e N/A
a
Fire area is defined as the size of the boundary condition used as the fire source
b
Heat Release Rate is the product of the fire area and the HRRPUA
c
The reaction group consists of several variables
d
Conductivity is defined as kρc, and is user-defined
e
Meshes are user-defined

7.3 Sensitivity study

A sensitivity study was conducted with the intention to establish which of the
variables chosen in Section 7.2 had the largest effect on visibility. The study was
performed by assigning chosen variables an estimated value. These variables were
then generally varied by a factor of approximately 0.5 in order to determine the
difference in visibility. Exceptions apply to the fuel variable where different fuels
were chosen, conductivity where the product of kρc was varied as a whole as
opposed to the different variables individually, and the radiative fraction which
was only decreased to 10 % (see Table 7.2).

30
 Selection of input variables for Monte Carlo simulations

TABLE 7.2 Variation of values of selected input variables in the sensitivity study
Variable Estimated value Value -50 % Value +50 %
Mass extinction
8,700 m2/kg 4,350 m2/kg 13,050 m2/kg
coefficient
Fire Area/HRRPUAa 1.28 m2 0.72 m2 2 m2
Soot yield 0.1 g/g 0.05 g/g 0.15 g/g
HRR 400 kW 200 kW 600 kW
Reaction (fuel) Propane Ethanol Polyurethane
b
Radiative fraction 0.35 0.1 N/A
580 (Gypsum) 285 (Gypsum) 870 (Gypsum)
Conductivity (kρc)
2,033 (Concrete) 1,001 (Concrete) 3,060 (Concrete)
Cell size 0.1 0.05 N/A
a
Due to limitations in FDS, the fire area has not been changed by exactly 50 %
b
Radiative fraction has not been reduced with 50 %

The results from the simulations are shown in Figure 7.1 to Figure 7.7.

FIGURE 7.1 Variations in visibility at different heights as a result of changes made to the
mass extinction coefficient.

31
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

0.72 [m^2]
2,5
Height above floor [m]

2 [m^2]
2

1,5

0,5

0
0 5 10 15 20 25 30 35
Visibility [m]

FIGURE 7.2 Variations in visibility at different heights as a result of changes made to the
fire area.

0.05 [g/g]
2,5
Height above floor [m]

0.15 [g/g]
2

1,5

0,5

0
0 5 10 15 20 25 30 35
Visibility [m]

FIGURE 7.3 Variations in visibility at different heights as a result of changes made to soot
yield.

32
 Selection of input variables for Monte Carlo simulations

200 [kW]
2,5
600 [kW]
Height above floor [m]

1,5

0,5

0
0 5 10 15 20 25 30 35
Visibility [m]

FIGURE 7.4 Variations in visibility at different heights as a result of changes made to the
HRR.

Ethanol
2,5
Height above floor [m]

Polyurethane
2

1,5

0,5

0
0 5 10 15 20 25 30 35
Visibility [m]

FIGURE 7.5 Variations in visibility at different heights as a result of changes made to the
fuel.

33
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

0.1
2,5

0.35
Height above floor [m]

1,5

0,5

0
0 5 10 15 20 25 30 35
Visibility [m]

FIGURE 7.6 Variations in visibility at different heights as a result of changes made to the
radiative fraction (the two graphs superimpose).

285 (Gypsum) 870 (Concrete) [(W^2*s)/(m^4*K^2)]

2,5
HEIGHT ABOVE FLOOR (m)

1,001 (Gypsum) 3060 (Concrete) [(W^2*s)/(m^4*K^2)]

2

1,5

0,5

0
0 5 10 15 20 25 30 35
VISIBILITY (m)

FIGURE 7.7 Variations in visibility at different heights as a result of changes made to kρc.

34
 Selection of input variables for Monte Carlo simulations

2,5
0.05 m
HEIGHT ABOVE FLOOR (m)

2 0.1 m

1,5

0,5

0
0 5 10 15 20 25 30 35
VISIBILITY (m)

FIGURE 7.8 Variations in visibility at different heights as a result of changes made to cell
size.

The figures show how visibility changes at different heights as a result of varying
one input variable at a time. Table 7.3 below shows the ratio of the visibility for
each of the variables, at the given height.
TABLE 7.3 Results from sensitivity analysis
Variable Ratio of visibility at a height of 1.4 m
Mass extinction coefficient 2.24
Fire Area/HRRPUA 1.03
Soot yield 3.24
HRR 2.35
Reaction (fuel) 1.54
Radiation 1.06
kρc 1.07
Cell size 2.33

The four variables that had the largest ratio of visibility were:

35
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

 Mass extinction coefficient

 Soot yield
 HRR
 Cell size

However, the ratio of visibility regarding the cell size variable varies considerably
depending on the height at which one compares, as can be seen in Figure 7.8.
Furthermore, it would not be possible to conduct 200 simulations where the cell
size is varied each time. For these reasons the cell size variable is disregarded for
further study and only the following three variables are selected:

 Mass extinction coefficient

 Soot yield
 HRR

36
8 Distributions of input variables

This section details the process of selecting statistical distributions of the input
data, to attempt to account for their inherent uncertainty or statistical nature.

8.1 Soot yield

Soot particles are produced as a result of incomplete combustion and vary
depending on the type of combustible material as well as other factors such as
ventilation conditions. In FDS, the user specifies the desired soot yield as a
fraction of fuel mass that is then converted into smoke particulate.

In a fire, the probability distribution of the soot yield depends on the fuel. It is
therefore necessary to know the rough distribution of combustibles in the type of
setting being modelled. Due to the uncertainty of this, a triangular distribution
has been deemed reasonable, owing to the fact that reasonable assumptions can
be made of an expected value together with a minimum and maximum value.

In an office type setting, the fire load is typically dominated by paper, plastics
and wood [18]. For the purpose of this thesis, the proportions have been
assumed to be 60 % (paper), 30 % (wood) and 10 % (plastics). It should be
noted that these proportions can vary, but it is not the aim of this thesis to
absolutely represent reality, but rather to show a method and the sensitivity due
to user input.

By weighting soot yields found in the literature [23] and presented in Table 8.1
of materials commonly present in office type settings together with the
percentages in the above paragraph, the triangular distribution in Figure 8.1 was
created.

37
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

TABLE 8.1 Experimental soot yields for different materials

Material Soot Yield (g/g)
Wood 0.015
Polyethylene 0.06
Polystyrene 0.164
Polyurethane (rigid) 0.1
Polyurethane (flexible) 0.227

180

160

140
PROBABILITY DENSITY

120

100

80

60

40

20

0
0 0,05 0,1 0,15 0,2 0,25
SOOT YIELD [g/g]

FIGURE 8.1 Triangular probability distribution of soot yield.

8.2 Mass extinction coefficient

The mass extinction coefficient is a measurement of how much a substance
scatters light at a given wavelength (𝜆𝜆 = 633 𝑛𝑛𝑛𝑛). It affects visibility via the
following relationship:
𝐶𝐶
𝑆𝑆 = (8.1)
𝐾𝐾𝑚𝑚 ∙ 𝜌𝜌𝑌𝑌𝑠𝑠

38
 Distributions of input variables

Where S is visibility, C is a non-dimensional constant characteristic of the object

being viewed, Km is the mass extinction coefficient, and ρYs is the smoke
particulate [4].

Experimental data [24] reveals values of mass extinction coefficients from some
types of wood and plastics. There are however quite large experimental
uncertainties [24] and the values differ between experiments. These values are
given in Table 8.2.
TABLE 8.2 Experimental mass extinction coefficients for wood
Material Mean (m2/kg) ± (m2/kg)
Wood 10,300 -
Wood crib 8,500 1,000
Oak 7,600 2,400
Average 8,800 1,700

TABLE 8.3 Experimental mass extinction coefficients for plastic materials

Material Mean (m2/kg) ± (m2/kg)
Polystyrene (test 1) 10,000 -
Polystyrene (test 2) 9,600 -
Polycarbons (test 1) 7,600 1,000
Polycarbons (test 2) 10,200 -
Polyurethane crib 8,100 1,100
Average 9,100 1,050

With the office type setting and proportions of materials mentioned in

2 2
Section 8.1, the weighted average mean will be 8,830 m /kg ± 1,635 m /kg. Due
to the significant experimental uncertainties, as well as the weighted average
being reasonably similar to the recommended value, it was decided to go forth
with the recommended value of 8,700 ± 1,100 [4,24]. These values are suggested
specifically for flaming combustion of wood and plastics. For the purpose of this
thesis, a uniform distribution between 7,600 and 9,800 will be used for this
variable.

39
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

0,001

0,0008
PROBABILITY DENSITY

0,0006

0,0004

0,0002

0
7000 7500 8000 8500 9000 9500 10000 10500
MASS EXTINCTION COEFFICIENT (m2/kg)

FIGURE 8.2 Uniform distribution of mass extinction coefficient.

8.3 Heat Release Rate

A distribution of the HRR is very subjective since it is entirely up to the FDS
user to assess what may be involved in a fire in any given scenario. It is still
needed in order to achieve the aim of this thesis. The maximum ventilation
controlled fire size in the fire-room is 3.4 MW based on simple calculations on
maximum oxygen mass flow into the room. This however is a very large heat
release rate for the present setting; also the HRR is the single most controlling
variable in any FDS simulation, as it is the source term in the energy equation. It
is for this reason necessary to choose an HRR distribution that does not
dominate the other selected variables.

Bearing this in mind, the distribution of HRR will be set to a relatively low
interval, with a minimum of 100 kW ranging up to a maximum of 900 kW.
Since nothing is known about the distribution function, it is assumed to be
uniform.

40
 Distributions of input variables

0,003

0,0025
PROBABILITY DENSITY

0,002

0,0015

0,001

0,0005

0
0 200 400 600 800 1000
HEAT RELEASE RATE (kW)

FIGURE 8.3 Uniform distribution of Heat Release Rate.

8.4 Monte Carlo sampling

In order to sample variables from the distributions a software system called
@RISK was used. @RISK can be integrated into Microsoft Excel for the analysis
of technical situations impacted by risk, via Monte Carlo simulations [25]. The
program allows a user, among other things, to sample random numbers, based
on the given probability distribution of a variable. The distributions are listed in
Table 8.4.
TABLE 8.4 Distributions with attributes of input variables
Variable Unit Distribution Expected value Min. Max.
Mass extinction
m2/kg Uniform 8,700 7,600 9,800
coefficient
Soot yield g/g Triangular 0.027 0.015 0.227
HRR kW Uniform 500 100 900

In this thesis, 200 random numbers were sampled from each distribution
function belonging to a variable. The sampled values can be found in
Appendix B.

41
9 Results from simulations

In this chapter the results from the simulations are shown, where each variable
was varied one at a time. The figures in this chapter are given at a height (1.4 m)
where there is more variation in visibility in order to show the uncertainty, as
well as the height at which most humans would be able to look at (1.8 m).
Figures from selected heights are found in Appendix D

Also given for each of the figures is the probability that a simulation results in a
visibility less than 10 meters, i.e. Pr(S < 10). A complete table of this probability
can be found in Section 9.3.

43
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

9.1 Independent variable simulations

In this section, figures are presented for the simulations where only one variable
at a time was varied.

9.1.1 Soot yield

The results of the simulations are shown below in Figure 9.1 and Figure 9.2.

40

35
NUMBER OF SIMULATIONS

30

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE 9.1 Distribution of the visibility in the soot yield simulations at a height of
1.4 meters. Y-axis is the number of simulations, X-axis is the visibility Pr(x<10) = 0.745.

44
 Results from simulations

160

140
NUMBER OF SIMULATIONS

120

100

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE 9.2 Distribution of the visibility in the soot yield simulations at a height of 1.8
meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 1.

45
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

9.1.2 Mass extinction coefficient

The results of the simulations are shown below in Figure 9.3 to Figure 9.4.

70

60
NUMBER OF SIMULATIONS

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE 9.3 Distribution of the visibility in the mass extinction coefficient simulations at a
height of 1.4 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.

180

160
NUMBER OF SIMULATIONS

140

120

100

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
VISIBILITY (m)

FIGURE 9.4 Distribution of the visibility in the mass extinction coefficient simulations at a
height of 1.8 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 1.

46
 Results from simulations

As the results show, there is quite little variation in the visibility at any given
height.

47
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

9.1.3 HRR
The results of the simulations are shown below in Figure 9.5 and Figure 9.6.

35

30
NUMBER OF SIMULATIONS

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE 9.5 Distribution of the visibility in the heat release rate simulations at a height of
1.4 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.24.

120

100
NUMBER OF SIMULATIONS

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE 9.6 Distribution of the visibility in the heat release rate simulations at a height of
1.8 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 1.

48
 Results from simulations

9.2 Simultaneous variable simulations

In this section, figures are presented for the simulations where all variables were
varied at the same time.
30

25
NUMBER OF SIMULATIONS

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
VISIBILITY (m)

FIGURE 9.7 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.4 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.65.

160

140
NUMBER OF SIMULATIONS

120

100

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30
VISIBILITY (m)

FIGURE 9.8 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.8 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 1.

49
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

9.3 Summary of results

In Table 9.1 below the results are summarized by presenting the probability that
the visibility is less than 10 meters, for all input variables at all heights. For
example, between 0.1 and 1.1 meters, the visibility in all simulations is always
above 10 meters, i.e. the probability that the visibility is greater than 10 meters is
1 for all those heights.
TABLE 9.1 Probability of visibility not exceeding 10 m for all heights and simulations
Height (m) Soot yield Mass ext. coeff. HRR Simultaneous
0.1 0 0 0 0
0.2 0 0 0 0
0.3 0 0 0 0
0.4 0 0 0 0
0.5 0 0 0 0
0.6 0 0 0 0
0.7 0 0 0 0
0.8 0 0 0 0
0.9 0 0 0 0
1.0 0 0 0 0
1.1 0 0 0 0
1.2 0 0 0 0.075
1.3 0.195 0 0 0.24
1.4 0.745 0 0.24 0.65
1.5 0.99 1 0.7 0.955
1.6 1 1 0.965 0.995
1.7 1 1 1 1
1.8 1 1 1 1
1.9 1 1 1 1
2.0 1 1 1 1
2.1 1 1 1 1
2.2 1 1 1 1
2.3 1 1 1 1
2.4 1 1 1 1
2.5 1 1 1 1

50
10 Discussion and conclusion

10.1 Conclusion
The objectives of this thesis were to investigate a number of issues listed below in
order to develop a method of connecting FDS with risk analysis.

 Which input variables are the most sensitive in an FDS simulation,

based on the chosen output data?
 To what degree does the input data affect the output data?
 How uncertain is the output of an FDS simulation given a variation in
the input data?
 What recommendations can be made to FDS users?

Based on the chosen output quantity visibility, a sensitivity study was conducted
based on a selected number of possible input variables having an effect on
visibility. While it is possible that other variables could have had an impact on
visibility, the selected variables were chosen based on qualitative reasoning as is
detailed in Section 7. The sensitivity study showed that three of the variables had
the largest impact on the output, these variables being soot yield, mass extinction
coefficient and heat release rate. A user interested in another output quantity
such as temperature or radiation would need to conduct a similar study in order
to determine the most sensitive input variables for that specific output and
scenario.

The degree to which the input variables affect the output data is dependent on
the location where a user is interested in the output data. In this thesis for
example, the visibility varies very little close to the floor and ceiling, whereas the
variation is significant at around 1.4 meters above the floor. This is connected to
the uncertainty of the output data given a variation in the input data. The
simulations conducted in this thesis show that when all input variables are varied
simultaneously there is a considerable spread in output data. For example, at a
height 1.4 meters above the floor, 35 % of the 200 simulations achieved the
criterion of visibility being equal to or exceeding 10 meters.

51
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

With a probability of achieving an acceptable visibility of 35 %, for example, the

major recommendation to FDS users is to be very conscious about any decisions
made concerning input variables. Due to the time consuming task of applying
the method presented in this thesis to larger buildings, it is unreasonable to
expect fire engineers to conduct similar analyses at present. A strong
recommendation to FDS users is to conduct informed sensitivity analyses where
not only the HRR is varied, but any variables that may affect the output quantity
in question.

10.2 Discussion of assumptions and method

Fire scenarios are dynamic phenomena and the scenario used in this thesis is not
necessarily representative of other fire design settings encountered by fire
engineers. The results presented in this thesis are heavily dependent on the
chosen scenario and the static parameters attached to it. The uncertainties
presented should therefore only be used as indicative data, and not an absolute
truth. The aim of this thesis has been to use a scenario as realistic as possible, but
still manageable due to the immense workload that is connected to running
several hundred CFD simulations.

While this thesis does not present an absolute truth to user uncertainty when
using FDS, it does provide a method for connecting FDS with statistical analysis
to deal with the inherent uncertainty of CFD modelling in general and fire
engineering in particular.

Only visibility has been used as an output quantity, as has been presented
previously in Section 6. The method, however, could be used with any other
desired output from an FDS simulation, such as temperature, toxicity, incident
radiation or any other quantity available. In FDS, visibility is output as point
measurements. This puts some responsibility in the hands of the user, as visibility
needs to be evaluated along a path, all the way to the desired target, for example
an exit or an exit sign. In this thesis, the point measurements provided by FDS
have been used, but only after a study to determine if those measurements are
representative of path measurements. The study found that the point
measurements, in this case, were indeed representative.

Crucial to any FDS simulation, regardless if doing only a few simulations or

several hundred, are the input quantities that can be changed by the user. In this
case, the quantities that were deemed to have the most impact on visibility were
selected in a qualitative manner, and then included in a sensitivity analysis. This
resulted in the three selected input quantities, being soot yield, mass extinction

52
 Discussion and conclusion

coefficient and heat release rate (HRR). It should be noted that it is possible that
other input quantities were overlooked and that they could have an effect on
visibility. An optimal solution to this would be to include all possible input
parameters in a sensitivity study. This was not done in this thesis due to time
constraints and the vast amount of possible options in FDS. The selected
variables were not correlated in the Monte Carlo analysis. While it is reasonable
to assume there is some correlation between heat release rate, soot yield and mass
extinction coefficient, this was not considered in this thesis. Correlating the
variables could possibly yield different results, something that should be taken
into consideration in future research.

Mesh placement in all models have been the same, but it is possible that mesh
placement does have an effect on the results. The placement of mesh boundaries
(where meshes abut) have been chosen based on existing recommendations and
as far away from areas of increased flow activity as possible. This includes areas
around the fire source, door openings etc. In order to take advantage of parallel
computing and to reduce the workload, it was deemed necessary to use several
meshes.

Cell size is an important parameter in a CFD simulation and has been considered
in this thesis. Decreasing the cell size by half results in a theoretical 16-fold
increase in time (two times for each spatial dimension and two for the temporal).
This means that the choice of cell size had to be based not only on what was
desired, but also what was possible to do. The simulations in this thesis took
approximately two months to finish. This was due to both human limitations
and limitations in the computer cluster used. Halving the cell size would have
radically increased this time. However, to investigate the choice of cell sizes, a
sensitivity study was performed. This study showed minor differences between
the chosen cell size of 0.1 m (cubic cells) and a cell size of 0.05 m. It is
considered that the cell size used is reasonable and that the results are not overly
sensitive as a result of that choice. Any FDS user needs to consider the cell size, it
cannot be said that 0.1 m is an adequate cell size for any simulation in the fire
engineering field.

There is some randomness built into FDS when the flow field is initialized. This
results in fire plumes not being absolutely symmetric (as would be expected in
reality) and that one simulation could, at least in the near-field, slightly differ
from another. It is not expected that this randomness has any large effect in the
far-field.

Selecting the distribution for the HRR was especially difficult due to the
profound impact it has on the results. Ultimately, it was decided to use an

53
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

interval with relatively low values, so as to not make the results totally dependent
on the HRR alone. This is most likely the parameter that, in this thesis, is least
representative of how a fire engineer would choose it in reality.

In all simulations in this thesis, steady state conditions based on visibility were
achieved. It was a conscious choice to evaluate all results after steady state was
achieved, but it should be noted that this method would work equally well at any
desired time in a simulation, even though steady state has not been achieved.

10.3 Discussion of results

Not counting the models included in the sensitivity studies, a total of 800
simulations act as a foundation for the results in this thesis. When using visibility
as a criterion in fire safety engineering, it is relevant to evaluate it at some height,
usually what is called “head height” or the height at which a human head is
expected to be exposed to smoke. This height is usually, with some safety margin,
at roughly 1.8 to 2 m from the floor depending on the ceiling height. For the
specific setting used in the subject model, this height would, according to
guidelines in the Swedish building code, be roughly 1.8 m [1]. In all simulations
in this thesis, it was not possible to fulfil this criterion, meaning that acceptable
visibility was never achieved at around this height. In the subject simulations, the
height at which visibility varied was found to be at around 1.4 m from the floor.
This height would not normally be used in a fire engineering design, except
perhaps if people would be expected to be able to crawl in very special
circumstances such as airplanes or otherwise. But to obtain meaningful
distributions without introducing mitigating measures such as smoke ventilation,
output data was evaluated at around this height.

10.3.1 Soot yield

The results from the simulations where soot yield was varied yields a cumulative
distribution with visibility plotted against number of simulations. We define an
event E as the visibility S being less than 10 m at a height H m above the floor in
the specific scenarios used in this thesis The probability of E happening at 1.4 m
is PrE(H=1.4) = 0.745. This means that if an FDS user would select a value of
soot yield randomly within the range of soot yields used in the subject
simulations, the simulation would have a probability of actually achieving an
acceptable visibility at that height of less than 26 percent. At 1.3 m
PrE(1.3)=0.195 and at 1.5 m PrE(1.5)=0.99.

Soot yield is clearly a sensitive variable connected to visibility. Changing the soot
yield within the given distribution has a large effect on the results. Selecting a

54
 Discussion and conclusion

static value of soot yield would give a certain result, but changing this value
somewhat has a large impact on results as has been shown in this thesis.

10.3.2 Mass extinction coefficient

The results from the simulations where the mass extinction coefficient was varied
yields a cumulative distribution with visibility plotted against number of
simulations. At 1.4 m PrE(H=1.4) = 0. At 1.3 m PrE(1.3)=0 and at 1.5 m
PrE(1.5)=1. This shows that despite the initial sensitivity analysis, visibility is not
overly sensitive to changes in the mass extinction coefficient within the chosen
range. With a larger range of values of the mass extinction coefficient, the
resulting visibility distribution would most likely have looked different.
However, the chosen range of input values were selected based on an extensive
literature study and it would be unreasonable to expect values significantly higher
2
or lower than those used. The sensitivity study used a base value of 8,700 m /kg
and then used values of ± 50 percent. While this yielded noticeable differences in
visibility, clearly, the values used in the distribution functions did not have that
large an effect. This distribution was uniform with an expected value of 8,700
and a minimum and maximum value of about ± 13 percent.

10.3.3 Heat release rate

The results from the simulations where the HRR was varied yields a cumulative
distribution with visibility plotted against number of simulations. At 1.4 m
PrE(H=1.4) = 0.24. This means that if an FDS user would select a value of HRR
randomly within the range of heat release rates used in the subject simulations,
the simulation would have a probability of actually achieving an acceptable
visibility at that height of more than 75 percent. At 1.3 m PrE(1.3)=0 and at
1.5 m PrE(1.5)=0.7.

Within the rather moderate range in which the HRR has been varied, output is
still greatly affected. This was expected, as it is a very dominant variable. It
however shows that even small changes in the heat release rate give large effects
on the results.

10.3.4 Simultaneous variable simulations

The results from the simulations where all variables were varied yields a
cumulative distribution with visibility plotted against number of simulations. At
1.4 m PrE(H=1.4) = 0.65. This means that if an FDS user would select a value of
HRR, soot yield and mass extinction coefficient randomly within the ranges used
in the subject simulations, the simulation would have a probability of actually

55
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

achieving an acceptable visibility at that height of about 35 percent. At 1.3 m

PrE(1.3)= 0.24 and at 1.5 m PrE(1.5)=0.955.

Given the assumptions that the chosen variables are uncorrelated, that they are
randomly chosen and that their respective distributions are representative of
reality, the simultaneous variable simulations provide a probabilistic result of
how visibility varies at any given height. An FDS user might use a certain value
for each input variable and yield results that are, say, acceptable at a height of
1.4 m. The results of the subject simulations show, that at that height, there is
only a 35 percent probability that these results are indeed acceptable. That means
that there is a 65 percent probability that the results would not be acceptable.

Using safety margins is an approach that could possibly, in some way, ascertain
that results from a fire simulation are conservative enough. However, that could
create problems in the other end, meaning that a design is ultimately more
conservative than it needs to be, resulting in excessive costs for contractors. The
very purpose of fire engineering would also be somewhat moot if a fire engineer
always simply used the most conservative approach to ensure the safety of a
building. Clearly, using risk assessments is a way of ensuring adequate,
probabilistic, safety while still delivering reasonable cost-effective solutions.

10.3.5 Summary of results

Naturally, a typical FDS user would not normally select values of soot yield
randomly. However, he would most likely select a more or less static value and
use that in all simulations. In reality, however, soot yield is not necessarily static,
but follows a distribution based on type of setting and what combustibles exist
within that setting. A resulting fire safety design could then both overestimate
the height at which visibility fulfils the criterion, or underestimate it. This could
either create an unsafe design, or a design that is overly conservative, perhaps
resulting in excessive cost for fire mitigating measures or otherwise.

10.4 Errors
As with all CFD simulations, there are many uncertain parameters and potential
errors that can be introduced by the user. The use of parallel computing was
necessary to be able to conduct all simulations within a reasonable timeframe.
While it is possible, it is not expected that this introduced any large errors into
the simulations, as previous research has shown. Even if errors were introduced,
all simulations where simulated in the same computer setting and in the exact
same manner, therefore any errors are expected to be introduced equally into all
simulations.

56
 Discussion and conclusion

Visibility, as it is defined in fire safety engineering, is a somewhat subjective

quantity that to some extent is based on experiments with humans. While this is
something that a fire engineer or any FDS user needs to know, it does not have
an impact on any of the results in this thesis.

The chosen input variables were selected for a number of reasons, but it is noted
that other potential parameters exist and could equally well have been used. It
could have been possible to conduct a survey of how fire engineers would have
chosen the design fire, soot yield or any other parameter to try and obtain a
distribution. This was deemed as unrealistic due to the nature of the simulations.
The survey subjects would have needed to be thoroughly introduced into the
purpose of the simulations in order to give reasonable answers. Ultimately, it was
decided to use distributions of the selected input variables based on literature and
reasonable assumptions.

The assumption that all input variables are uncorrelated is not necessarily correct.
However, the literature study showed that it was very difficult to obtain any
meaningful data on any correlation between the chosen variables. It was therefore
decided to not attempt to correlate the variables.

10.5 Recommendations to FDS users

This thesis proposed a method of using Monte Carlo analysis with FDS. Using
FDS to simulate fire scenarios is computationally intensive, and as such the
authors acknowledge that it is not feasible to conduct hundreds of simulations of
large buildings that is often the case in fire engineering design today. Given the
advance in computers in the last 10-20 years and the assumed continued
advances, it is important to develop and improve on methods such as the one
presented in this thesis in order to increase the level of accuracy and certainty in
fire models.

Presently, it would be quite possible to use the proposed method for more
reasonably sized buildings, given that the user has access to parallel computing
networks. For larger buildings today, the feasible approach would be to conduct
a smaller scale analysis, not necessarily utilizing Monte Carlo, but rather an
informed sensitivity analysis, preferably changing more than just the heat release
rate. Changing more than just the heat release rate, but also any other input
variable that could have an effect on a given output quantity, ascertains that an
FDS user acknowledges the dynamic nature of fire modelling.

57
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

10.6 Further research

This thesis represents, to the authors’ best knowledge, a first step in connecting
risk analysis with FDS usage for fire engineering purposes. Apart from advances
in computational power more research could be conducted in order to further
simplify and develop the concept.

Finding reasonable distributions for selected input variables is time consuming

and ultimately somewhat qualitative. An effort could be made to find reasonable
distributions of key variables for different settings, much like fire growth
numbers has been researched and recommendations have been published for
different types of occupancies [26].

It is also reasonable that similar studies be conducted that focuses on other

output quantities such as temperature, toxicity and radiation.

58
11 References

[1] Boverket, Boverkets byggregler Kap. 5, Boverket, 2006.

[2] Hostikka, S., Development of fire simulation models for radiative heat transfer
and probabilistic risk assessment, VTT PUBLICATIONS 683, Helsinki
University of Technology, Espoo, Finland, 2008.

[3] Najm, N. Habib., Uncertainty Quantification and Polynomial Chaos

Techniques in Computational Fluid Dynamics, Sandia National Laboratories,
Livermore, California, USA, 2009.

[4] McGrattan, Kevin, et al., Fire Dynamics Simulator (Version 5) User's guide,
NIST Special Publication 1019-5, NIST Building and Fire Research Laboratory,
Maryland, 2009.

[5] Vestin, Gunnar, Wahlqvist, Jonathan, CFD-beräkningar på parallella

processorer, Lunds Tekniska Högskola, Lund, 2008.

[6] Björklund, Anders, Risks in using CFD codes for analytical fire-based design in
buildings with a focus on FDS:s handling of under-ventilated fires, Lunds Tekniska
Högskola, Lund, 2009.

[7] A Brief History of Computational Fluid Dynamics (CFD). Fluent. [Online]

[June 24 2009] http://www.fluent.com/about/cfdhistory.htm.

[8] Rubini, Dr. P., An introduction to CFD, Department of Fire and Safety
Engineering, Lund University, Lund, 2009.

[9] Rubini, Dr. P., An Introduction to Numerical Methods for Fire Simulation.
Department of Fire and Safety Engineering, Lund University, Lund, 2009.

[10] McGrattan, Kevin, et al., Fire Dynamics Simulator (Version 5) Technical

Reference Guide - Volume 1: Mathematical Model, NIST Special Publication
1018-5, NIST, Maryland, 2009.

59
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

[11] McGrattan, Kevin, et al., Fire Dynamics Simulator (Version 5) Technical

Reference Guide - Volume 3: Validation, NIST Special Publication 1018-5, NIST,
Maryland, 2009.

[12] Schulz, Judith, Case Study: Parameters influencing the Flame & Plume
temperatures in FDS, Holmes Fire & Safety, Auckland, 2007.

[13] U.S. Nuclear Regulatory Commission, Verification and Validation of Selected

Fire Models for Nuclear Power Plant Applications: Fire Dynamics Simulator (FDS),
NUREG-1824, Volume 7, U.S. Nuclear Regulatory Commission, 2007.

[14] McDermott, R., McGrattan, K., Hostikka, S., & Floyd, J., Fire Dynamics
Simulator (Version 5) Technical Reference Guide - Volume 2: Verification, NIST
Special Publication 1018-5, NIST, Maryland, 2009.

[15] Hubbard, Douglas, How to Measure Anything: Finding the Value of

Intangibles in Business, John Wiley & Sons, 2007.

[16] Iman, R. L., Davenport, J. M., and Zeigler, D. K., Latin Hypercube
Sampling (A Program User's Guide), Technical Report SAND79-1473, Sandia
Laboratories, Albuquerque, 1980.

[17] Vose Software, Help file for ModelRisk, Vose Software, 2007.

[18] Mingchun, Leo, Beck, Vaughan, The fire-environment in a multi-room

building - comparison of predicted and experimental results. Fire Safety Journal.
1994, Vol. 23, pp. 413-438.

[19] Drysdale, Dougal, An introduction to Fire Dynamics, 2nd edition, John Wiley
and Sons, 2005.

[20] Jin, Tadahisa, Visibility and Human Behavior in Fire Smoke. SFPE
Handbook of Fire Protection Engineering 3rd edition, National Fire Protection
Association, Quincy, Massachusetts, 2002.

[21] Rubini, P. A., Zhang, Q., Simulation of visibility in smoke laden

environments, 5th International Seminar on Fire and Explosion Hazards, School
of Engineering, Cranfield University, Edinburgh, 2007.

[22] Jin, T., Visibility Through Fire Smoke. 9, 1978, Journal of Fire and
Flammability, pp. 135-155.

60
 References

[23] Tewarson, A., Generation of Heat and Chemical Compounds in Fires.

SFPE Handbook of Fire Protection Engineering 3rd edition, National Fire
Protection Association, Quincy, Massachusetts, 2002.

[24] Mulholland, G.W., Croarkin, C., Specific Extinction Coefficient of Flame

Generated Smoke. Fire and Materials. 2000, Vol. 24, pp. 227–230.

[25] Palisade Corporation, Guide to using @RISK Version 5.0, Palisade

Corporation, Ithaca, USA, 2008.

[26] Karlsson, B., Quintiere, J. G., Enclosure Fire Dynamics, CRC Press,
Washington D.C., USA, 2000.

[27] Almasi, G. S., Gottlieb, A., Highly parallel computing, IBM Systems Journal,
Vol. 29, pp. 165-166, Benjamin-Cummings publishers, Redwood, CA, 1990.

61
Appendix A – Parallel computing with FDS

A.1 Parallel computing

Parallel computing is a form of computation where many calculations are carried
out simultaneously by different computers or processors (also called Central
Processing Unit, CPU), taking advantage of the fact that large problems often can
be divided into smaller ones which are then solved simultaneously or “in parallel”
[7].

In order for different computers or CPUs to communicate with each other, some
sort of standardised interface is required. Message Passing Interface (MPI) is a
communications protocol that has become the de facto standard for parallel
computing.

A.2 LUNARC
Lund University Numeric Intensive Computation Application Research Center
(LUNARC) is a scientific centre for technical and scientific computing started in
1986 at Lund University. The parallel computations in this thesis have been
performed on one of LUNARC’s homogenous clusters called Milleotto.
Milleotto consists of 252 nodes with two Dual Core processors in each node,
resulting in a total of 1008 CPUs. The nodes are connected by two independent
Gigabit networks, one handling the MPI communication and one handling the
rest of the data traffic. Specific node information can be seen in Table A.1 below.
TABLE A.1 Node configuration
Node configuration on Milleotto
CPU 2 Intel Xeon 5160 (3.0 GHz dual core)
RAM 4 GB
Operating system Linux CentOS 5.3 x86_64 (RHEL4 compatible)

63
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

A.3 Compiling FDS

FDS is made available by NIST both as source code and as pre-compiled

executable files for a limited number of operating systems. In order to run FDS
on LUNARC, it needed to be compiled specifically for this cluster and the
operating system used. Compiling, in this context, means converting human
readable source code from one programming language to another. This is usually
in a binary format known as machine code. The machine code are a series of
instructions that are read directly by the CPU and thus needs to be compatible
with both the operating system and the specific type of CPU on the target
system.

For all simulations in this thesis, the same version of FDS was used (see
Table A.2 below). The version used was the latest stable version released at the
time the simulations in this thesis were started. All changes made to FDS by the
developers at NIST are saved and tracked by a version system software called
2
Subversion (SVN) . That means any previous version can be downloaded and
compiled if required.
TABLE A.2 FDS version information
FDS version used
Version FDS 5.3.1 Parallel (No OpenMP-version)
SVN Revision No. 3729
Release date April 8 2009

A.3 Running simulations on LUNARC

In order for several different users to run computing jobs on the LUNARC
cluster, resource management and job scheduling software is used. LUNARC
3 4
uses a resource manager called TORQUE and a cluster scheduler called Maui .
In order to submit computing jobs to LUNARC a script file is used. The script
file contains information about the maximum amount of time the job is allowed
to run, the number of CPUs to be used, an e-mail address to send information
on when the job starts and finishes and the path to the executable file that is to
run the job as well as any input file the particular executable needs.

2
See http://subversion.tigris.org/
3
See http://www.clusterresources.com/products/torque-resource-manager.php
4
See http://www.clusterresources.com/products/maui-cluster-scheduler.php

64
Appendix B – Sampled values

In this appendix the values of the input variables used for the simulations are
presented. One value corresponds to one simulation except for the simultaneous
simulations where the first, then second, then third etc., values from each table
respectively was used at the same time.

65
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

B.1 Soot yield

Sampled values of soot yield (g/g).
0.036 0.1327 0.1622 0.0543 0.0508 0.1075
0.1155 0.0776 0.0617 0.0262 0.0563 0.0679
0.1418 0.0948 0.074 0.0763 0.0414 0.0512
0.06 0.0873 0.0316 0.0802 0.0959 0.1038
0.0592 0.0298 0.1069 0.0399 0.0217 0.1201
0.1557 0.0917 0.0754 0.0665 0.0391 0.0271
0.0458 0.0377 0.0994 0.0452 0.1466 0.0865
0.0765 0.0404 0.0649 0.1609 0.0352 0.0496
0.0573 0.1886 0.0789 0.0991 0.127 0.1347
0.081 0.1122 0.0318 0.0502 0.1092 0.126
0.1945 0.0201 0.1198 0.1002 0.0468 0.1439
0.093 0.0963 0.0709 0.128 0.0432 0.1081
0.1224 0.0331 0.1916 0.0448 0.0341 0.0311
0.2135 0.1099 0.1483 0.0559 0.0779 0.0384
0.1292 0.0644 0.0222 0.1163 0.1247 0.1239
0.0591 0.1453 0.0584 0.1011 0.1589 0.164
0.0231 0.0722 0.0794 0.1706 0.1741 0.0529
0.1139 0.0719 0.0898 0.1055 0.0249 0.1511
0.063 0.1212 0.0969 0.0441 0.0686 0.067
0.029 0.083 0.1391 0.0425 0.0908 0.2022
0.053 0.0327 0.1294 0.0816 0.0371 0.0241
0.0284 0.0518 0.1523 0.0839 0.1707 0.0927
0.0858 0.0936 0.1877 0.0883 0.0264 0.1032
0.1401 0.1124 0.0287 0.1378 0.0484 0.2006
0.0488 0.1673 0.1565 0.1315 0.118 0.041
0.2089 0.1577 0.0476 0.1843 0.0879 0.0257
0.0891 0.1792 0.1663 0.1492 0.0548 0.1316
0.0334 0.1763 0.0695 0.057 0.0616 0.0662
0.119 0.0276 0.181 0.0849 0.0624 0.0439
0.0305 0.0606 0.0736 0.1143 0.0419 0.1786
0.0192 0.0824 0.0361 0.0474 0.0637
0.0347 0.0377 0.1049 0.1531 0.1361
0.0175 0.0707 0.1113 0.0748 0.0982
0.1427 0.1025 0.0536 0.1358 0.0696

66
 Appendix B

B.2 Mass extinction coefficient

2
Sampled values of mass extinction coefficient (m /kg).

8588 8223 8834 9615 9777 7676

8083 9280 8414 9532 8805 8473
8498 9557 7730 9048 9115 9258
8354 7741 8018 8637 9483 9190
8231 9497 8376 9727 7785 8390
7715 8533 8748 9086 9755 9299
9224 7889 7838 9068 8310 9405
8505 7974 7762 8884 9653 8634
8561 7772 7666 8599 7859 8090
7940 8440 8149 9364 7920 8576
8616 7881 8913 8737 8976 9322
8482 8158 9735 8686 9249 9588
9335 7693 7615 8046 8949 9631
9390 7798 7649 8986 9379 8361
8299 8394 8522 8548 9027 9052
8027 8245 8119 8927 8458 8262
9460 9178 9313 8717 8176 8191
7789 8054 8659 7826 9575 9676
8343 9401 8135 9509 8859 7955
9139 9159 8870 9439 9077 8164
9538 9290 7866 8955 9231 7989
7907 7946 9418 8429 9607 8792
8287 9593 8786 9341 7635 9036
7705 8197 9010 8966 9662 8730
7908 8421 7746 8692 8845 9785
8063 9356 8771 9102 9003 9698
8205 7604 7631 9208 8096 8939
8822 7809 7687 7977 9147 8281
9680 9520 9641 8538 9472 8031
9165 9797 9704 9431 8110 8905
8258 9568 8701 9451 8001
9199 8675 9261 8320 8817
8897 9122 8463 8756 9757
8602 7845 8329 8649 9715

67
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

B.3 HRR
2
Sampled values of HRR (kW). Divide with 0.25 m to get HRRPUA as used in
the FDS input files.

648.5 223.5 451.5 661.2 441.9 263.6

623.1 699 836.6 632.8 458.2 385
575 786.7 135.2 158.3 518.3 835.8
131.2 682.3 798.5 274.1 225.4 615.3
852.1 324.1 755.2 687.7 624.4 741.5
877.6 243.2 376.5 738.4 231.5 581.9
393.6 148.3 717.4 317.7 734.4 332.5
721.9 116.6 803.6 312.5 778.6 674
161.7 600.9 640.2 339.5 348.4 366.8
453.2 341.7 544.6 424.3 658.2 169.9
533.9 561.5 477.2 619.6 588 247.9
191.7 499 406.4 126.2 408.4 305.8
809.6 599.5 579.8 258.2 505 704.3
825.9 831.8 693.2 725.2 587.6 432
862.5 794 268.9 898.9 120.7 397.1
181.8 155.5 413.1 552.8 702.1 880.4
668.5 164.4 439.7 100.5 196.6 470.8
422.6 297 277.9 689.2 630.7 359
531 890.3 509.2 486.4 200.4 676.2
264.4 346.8 288 303.7 840.3 523
819.6 329.6 466 646.4 652.4 251.7
764.2 515.1 282.4 773.5 403.8 570.3
488.5 419.9 526.8 895.5 715.9 558.2
805.7 147.7 383.6 354.1 372.8 846.4
289.3 788.4 368.7 756.3 542.1 761.6
749.6 744.9 207.5 252.2 768.1 849.6
606.5 431.9 311.4 873.4 857.2 495.1
294.5 822.8 865.4 814.3 115.2 111.7
710.1 500.7 391.7 537.8 444.6 609.8
195.3 143.2 363.3 887.4 137.7 215
871.8 666.8 730.7 481.2 323.4
187.6 567.6 462.9 237.5 105.2
172.3 176 639 216.5 233.4
211.7 475.1 780.7 594 551

68
Appendix C – Input files

The input files used for the simulations are presented below, the highlighted
sections correspond to the variables that needed to be changed between each
simulation.

C.1 FDS input file

The template file used for the actual simulation in FDS is presented below.
Highlighted parts indicate were variable values were used.
&HEAD CHID='sy101'/
&TIME T_END=1000/
&DUMP COLUMN_DUMP_LIMIT=.FALSE., DT_PL3D=2000, DT_SLCF=10.00, SMOKE3D=.FALSE./

&MESH ID='Fire2', IJK=13,25,25, XB=1.10,2.40,-0.1000,2.40,0.00,2.50/

&MESH ID='Mesh1', IJK=13,25,25, XB=2.40,3.70,-0.1000,2.40,0.00,2.50/
&MESH ID='Mesh3', IJK=15,25,25, XB=-0.4000,1.10,-0.1000,2.40,0.00,2.50/
&MESH ID='Mesh4', IJK=11,20,25, XB=-1.50,-0.4000,0.4000,2.40,0.00,2.50/
&MESH ID='Mesh5', IJK=11,20,25, XB=-2.60,-1.50,0.4000,2.40,0.00,2.50/
&MESH ID='Mesh7', IJK=13,24,25, XB=1.10,2.40,2.40,4.80,0.00,2.50/
&MESH ID='Mesh9', IJK=11,24,25, XB=-1.50,-0.4000,2.40,4.80,0.00,2.50/
&MESH ID='Mesh10', IJK=11,24,25, XB=-2.60,-1.50,2.40,4.80,0.00,2.50/
&MESH ID='Mesh8', IJK=15,24,25, XB=-0.4000,1.10,2.40,4.80,0.00,2.50/
&MESH ID='Mesh6', IJK=13,24,25, XB=2.40,3.70,2.40,4.80,0.00,2.50/
&MESH ID='Mesh11', IJK=22,18,25, XB=-2.60,-0.4000,4.80,6.60,0.00,2.50/
&MESH ID='Mesh12', IJK=22,18,25, XB=-0.4000,1.80,4.80,6.60,0.00,2.50/
&MESH ID='Mesh13', IJK=19,18,25, XB=1.80,3.70,4.80,6.60,0.00,2.50/
&MESH ID='Mesh14', IJK=33,12,27, XB=3.70,7.00,5.40,6.60,0.00,2.70/

&REAC ID='PROPANE',
C=3.00,
H=8.00,
O=0.00,
N=0.00,
SOOT_YIELD=0.036,
MASS_EXTINCTION_COEFFICIENT=8700/

&MATL ID='CONCRETE',
SPECIFIC_HEAT=0.88,
CONDUCTIVITY=1.10,
DENSITY=2.1000000E003/
&MATL ID='GYPSUM_PLASTER',
SPECIFIC_HEAT=0.84,
CONDUCTIVITY=0.4800,
DENSITY=1.4400000E003/

69
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

&SURF ID='CONCRETE_SLAB',
COLOR='GRAY 60',
MATL_ID(1,1)='CONCRETE',
MATL_MASS_FRACTION(1,1)=1.00,
THICKNESS(1)=0.2500/
&SURF ID='TEST_WALL',
RGB=146,202,166,
MATL_ID(1,1)='GYPSUM_PLASTER',
MATL_ID(2,1)='CONCRETE',
MATL_MASS_FRACTION(1,1)=1.00,
MATL_MASS_FRACTION(2,1)=1.00,
THICKNESS(1:2)=0.0160,0.1000/
&SURF ID='BURNER',
COLOR='RED',
HRRPUA=2000/

&DEVC ID='Door102_1', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,0.1000/

&DEVC ID='Door102_102', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,0.3000/
&DEVC ID='Door102_103', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,0.50/
&DEVC ID='Door102_104', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,0.70/
&DEVC ID='Door102_105', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,0.90/
&DEVC ID='Door102_106', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,1.10/
&DEVC ID='Door102_107', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,1.30/
&DEVC ID='Door102_108', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,1.50/
&DEVC ID='Door102_109', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,1.70/
&DEVC ID='Door102_110', QUANTITY='V-VELOCITY', XYZ=-1.30,5.40,1.90/
&DEVC ID='Door103_1', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,0.1000/
&DEVC ID='Door103_102', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,0.3000/
&DEVC ID='Door103_103', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,0.50/
&DEVC ID='Door103_104', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,0.70/
&DEVC ID='Door103_105', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,0.90/
&DEVC ID='Door103_106', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,1.10/
&DEVC ID='Door103_107', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,1.30/
&DEVC ID='Door103_108', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,1.50/
&DEVC ID='Door103_109', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,1.70/
&DEVC ID='Door103_110', QUANTITY='U-VELOCITY', XYZ=-0.1000,4.00,1.90/
&DEVC ID='Door104_1', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,0.1000/
&DEVC ID='Door104_102', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,0.3000/
&DEVC ID='Door104_103', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,0.50/
&DEVC ID='Door104_104', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,0.70/
&DEVC ID='Door104_105', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,0.90/
&DEVC ID='Door104_106', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,1.10/
&DEVC ID='Door104_107', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,1.30/
&DEVC ID='Door104_108', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,1.50/
&DEVC ID='Door104_109', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,1.70/
&DEVC ID='Door104_110', QUANTITY='U-VELOCITY', XYZ=-0.1000,1.20,1.90/
&DEVC ID='FlameTemp101', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.1000/
&DEVC ID='FlameTemp102', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.2000/
&DEVC ID='FlameTemp103', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.3000/
&DEVC ID='FlameTemp104', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.4000/
&DEVC ID='FlameTemp105', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.50/
&DEVC ID='FlameTemp106', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.60/
&DEVC ID='FlameTemp107', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.70/
&DEVC ID='FlameTemp108', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.80/
&DEVC ID='FlameTemp109', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,0.90/
&DEVC ID='FlameTemp110', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.00/
&DEVC ID='FlameTemp111', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.10/
&DEVC ID='FlameTemp112', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.20/
&DEVC ID='FlameTemp113', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.30/
&DEVC ID='FlameTemp114', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.40/
&DEVC ID='FlameTemp115', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.50/

70
 Appendix C

&DEVC ID='FlameTemp116', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.60/

&DEVC ID='FlameTemp117', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.70/
&DEVC ID='FlameTemp118', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.80/
&DEVC ID='FlameTemp119', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,1.90/
&DEVC ID='FlameTemp120', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,2.00/
&DEVC ID='FlameTemp121', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,2.10/
&DEVC ID='FlameTemp122', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,2.20/
&DEVC ID='FlameTemp123', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,2.30/
&DEVC ID='FlameTemp124', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,2.40/
&DEVC ID='FlameTemp125', QUANTITY='TEMPERATURE', XYZ=1.80,1.10,2.50/
&DEVC ID='Room101-101', QUANTITY='visibility', XYZ=-1.40,1.20,0.1000/
&DEVC ID='Room101-102', QUANTITY='visibility', XYZ=-1.40,1.20,0.2000/
&DEVC ID='Room101-103', QUANTITY='visibility', XYZ=-1.40,1.20,0.3000/
&DEVC ID='Room101-104', QUANTITY='visibility', XYZ=-1.40,1.20,0.4000/
&DEVC ID='Room101-105', QUANTITY='visibility', XYZ=-1.40,1.20,0.50/
&DEVC ID='Room101-106', QUANTITY='visibility', XYZ=-1.40,1.20,0.60/
&DEVC ID='Room101-107', QUANTITY='visibility', XYZ=-1.40,1.20,0.70/
&DEVC ID='Room101-108', QUANTITY='visibility', XYZ=-1.40,1.20,0.80/
&DEVC ID='Room101-109', QUANTITY='visibility', XYZ=-1.40,1.20,0.90/
&DEVC ID='Room101-110', QUANTITY='visibility', XYZ=-1.40,1.20,1.00/
&DEVC ID='Room101-111', QUANTITY='visibility', XYZ=-1.40,1.20,1.10/
&DEVC ID='Room101-112', QUANTITY='visibility', XYZ=-1.40,1.20,1.20/
&DEVC ID='Room101-113', QUANTITY='visibility', XYZ=-1.40,1.20,1.30/
&DEVC ID='Room101-114', QUANTITY='visibility', XYZ=-1.40,1.20,1.40/
&DEVC ID='Room101-115', QUANTITY='visibility', XYZ=-1.40,1.20,1.50/
&DEVC ID='Room101-116', QUANTITY='visibility', XYZ=-1.40,1.20,1.60/
&DEVC ID='Room101-117', QUANTITY='visibility', XYZ=-1.40,1.20,1.70/
&DEVC ID='Room101-118', QUANTITY='visibility', XYZ=-1.40,1.20,1.80/
&DEVC ID='Room101-119', QUANTITY='visibility', XYZ=-1.40,1.20,1.90/
&DEVC ID='Room101-120', QUANTITY='visibility', XYZ=-1.40,1.20,2.00/
&DEVC ID='Room101-121', QUANTITY='visibility', XYZ=-1.40,1.20,2.10/
&DEVC ID='Room101-122', QUANTITY='visibility', XYZ=-1.40,1.20,2.20/
&DEVC ID='Room101-123', QUANTITY='visibility', XYZ=-1.40,1.20,2.30/
&DEVC ID='Room101-124', QUANTITY='visibility', XYZ=-1.40,1.20,2.40/
&DEVC ID='Room101-125', QUANTITY='visibility', XYZ=-1.40,1.20,2.50/
&DEVC ID='Room101-201', QUANTITY='visibility', XYZ=-1.40,3.20,0.1000/
&DEVC ID='Room101-202', QUANTITY='visibility', XYZ=-1.40,3.20,0.2000/
&DEVC ID='Room101-203', QUANTITY='visibility', XYZ=-1.40,3.20,0.3000/
&DEVC ID='Room101-204', QUANTITY='visibility', XYZ=-1.40,3.20,0.4000/
&DEVC ID='Room101-205', QUANTITY='visibility', XYZ=-1.40,3.20,0.50/
&DEVC ID='Room101-206', QUANTITY='visibility', XYZ=-1.40,3.20,0.60/
&DEVC ID='Room101-207', QUANTITY='visibility', XYZ=-1.40,3.20,0.70/
&DEVC ID='Room101-208', QUANTITY='visibility', XYZ=-1.40,3.20,0.80/
&DEVC ID='Room101-209', QUANTITY='visibility', XYZ=-1.40,3.20,0.90/
&DEVC ID='Room101-210', QUANTITY='visibility', XYZ=-1.40,3.20,1.00/
&DEVC ID='Room101-211', QUANTITY='visibility', XYZ=-1.40,3.20,1.10/
&DEVC ID='Room101-212', QUANTITY='visibility', XYZ=-1.40,3.20,1.20/
&DEVC ID='Room101-213', QUANTITY='visibility', XYZ=-1.40,3.20,1.30/
&DEVC ID='Room101-214', QUANTITY='visibility', XYZ=-1.40,3.20,1.40/
&DEVC ID='Room101-215', QUANTITY='visibility', XYZ=-1.40,3.20,1.50/
&DEVC ID='Room101-216', QUANTITY='visibility', XYZ=-1.40,3.20,1.60/
&DEVC ID='Room101-217', QUANTITY='visibility', XYZ=-1.40,3.20,1.70/
&DEVC ID='Room101-218', QUANTITY='visibility', XYZ=-1.40,3.20,1.80/
&DEVC ID='Room101-219', QUANTITY='visibility', XYZ=-1.40,3.20,1.90/
&DEVC ID='Room101-220', QUANTITY='visibility', XYZ=-1.40,3.20,2.00/
&DEVC ID='Room101-221', QUANTITY='visibility', XYZ=-1.40,3.20,2.10/
&DEVC ID='Room101-222', QUANTITY='visibility', XYZ=-1.40,3.20,2.20/
&DEVC ID='Room101-223', QUANTITY='visibility', XYZ=-1.40,3.20,2.30/
&DEVC ID='Room101-224', QUANTITY='visibility', XYZ=-1.40,3.20,2.40/
&DEVC ID='Room101-225', QUANTITY='visibility', XYZ=-1.40,3.20,2.50/
&DEVC ID='Room101-301', QUANTITY='visibility', XYZ=-1.40,4.70,0.1000/
&DEVC ID='Room101-302', QUANTITY='visibility', XYZ=-1.40,4.70,0.2000/

71
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

&DEVC ID='Room101-303', QUANTITY='visibility', XYZ=-1.40,4.70,0.3000/

&DEVC ID='Room101-304', QUANTITY='visibility', XYZ=-1.40,4.70,0.4000/
&DEVC ID='Room101-305', QUANTITY='visibility', XYZ=-1.40,4.70,0.50/
&DEVC ID='Room101-306', QUANTITY='visibility', XYZ=-1.40,4.70,0.60/
&DEVC ID='Room101-307', QUANTITY='visibility', XYZ=-1.40,4.70,0.70/
&DEVC ID='Room101-308', QUANTITY='visibility', XYZ=-1.40,4.70,0.80/
&DEVC ID='Room101-309', QUANTITY='visibility', XYZ=-1.40,4.70,0.90/
&DEVC ID='Room101-310', QUANTITY='visibility', XYZ=-1.40,4.70,1.00/
&DEVC ID='Room101-311', QUANTITY='visibility', XYZ=-1.40,4.70,1.10/
&DEVC ID='Room101-312', QUANTITY='visibility', XYZ=-1.40,4.70,1.20/
&DEVC ID='Room101-313', QUANTITY='visibility', XYZ=-1.40,4.70,1.30/
&DEVC ID='Room101-314', QUANTITY='visibility', XYZ=-1.40,4.70,1.40/
&DEVC ID='Room101-315', QUANTITY='visibility', XYZ=-1.40,4.70,1.50/
&DEVC ID='Room101-316', QUANTITY='visibility', XYZ=-1.40,4.70,1.60/
&DEVC ID='Room101-317', QUANTITY='visibility', XYZ=-1.40,4.70,1.70/
&DEVC ID='Room101-318', QUANTITY='visibility', XYZ=-1.40,4.70,1.80/
&DEVC ID='Room101-319', QUANTITY='visibility', XYZ=-1.40,4.70,1.90/
&DEVC ID='Room101-320', QUANTITY='visibility', XYZ=-1.40,4.70,2.00/
&DEVC ID='Room101-321', QUANTITY='visibility', XYZ=-1.40,4.70,2.10/
&DEVC ID='Room101-322', QUANTITY='visibility', XYZ=-1.40,4.70,2.20/
&DEVC ID='Room101-323', QUANTITY='visibility', XYZ=-1.40,4.70,2.30/
&DEVC ID='Room101-324', QUANTITY='visibility', XYZ=-1.40,4.70,2.40/
&DEVC ID='Room101-325', QUANTITY='visibility', XYZ=-1.40,4.70,2.50/
&DEVC ID='Room103-101', QUANTITY='visibility', XYZ=2.60,4.70,0.1000/
&DEVC ID='Room103-102', QUANTITY='visibility', XYZ=2.60,4.70,0.2000/
&DEVC ID='Room103-103', QUANTITY='visibility', XYZ=2.60,4.70,0.3000/
&DEVC ID='Room103-104', QUANTITY='visibility', XYZ=2.60,4.70,0.4000/
&DEVC ID='Room103-105', QUANTITY='visibility', XYZ=2.60,4.70,0.50/
&DEVC ID='Room103-106', QUANTITY='visibility', XYZ=2.60,4.70,0.60/
&DEVC ID='Room103-107', QUANTITY='visibility', XYZ=2.60,4.70,0.70/
&DEVC ID='Room103-108', QUANTITY='visibility', XYZ=2.60,4.70,0.80/
&DEVC ID='Room103-109', QUANTITY='visibility', XYZ=2.60,4.70,0.90/
&DEVC ID='Room103-110', QUANTITY='visibility', XYZ=2.60,4.70,1.00/
&DEVC ID='Room103-111', QUANTITY='visibility', XYZ=2.60,4.70,1.10/
&DEVC ID='Room103-112', QUANTITY='visibility', XYZ=2.60,4.70,1.20/
&DEVC ID='Room103-113', QUANTITY='visibility', XYZ=2.60,4.70,1.30/
&DEVC ID='Room103-114', QUANTITY='visibility', XYZ=2.60,4.70,1.40/
&DEVC ID='Room103-115', QUANTITY='visibility', XYZ=2.60,4.70,1.50/
&DEVC ID='Room103-116', QUANTITY='visibility', XYZ=2.60,4.70,1.60/
&DEVC ID='Room103-117', QUANTITY='visibility', XYZ=2.60,4.70,1.70/
&DEVC ID='Room103-118', QUANTITY='visibility', XYZ=2.60,4.70,1.80/
&DEVC ID='Room103-119', QUANTITY='visibility', XYZ=2.60,4.70,1.90/
&DEVC ID='Room103-120', QUANTITY='visibility', XYZ=2.60,4.70,2.00/
&DEVC ID='Room103-121', QUANTITY='visibility', XYZ=2.60,4.70,2.10/
&DEVC ID='Room103-122', QUANTITY='visibility', XYZ=2.60,4.70,2.20/
&DEVC ID='Room103-123', QUANTITY='visibility', XYZ=2.60,4.70,2.30/
&DEVC ID='Room103-124', QUANTITY='visibility', XYZ=2.60,4.70,2.40/
&DEVC ID='Room103-125', QUANTITY='visibility', XYZ=2.60,4.70,2.50/
&DEVC ID='Room103-201', QUANTITY='visibility', XYZ=0.80,3.00,0.1000/
&DEVC ID='Room103-202', QUANTITY='visibility', XYZ=0.80,3.00,0.2000/
&DEVC ID='Room103-203', QUANTITY='visibility', XYZ=0.80,3.00,0.3000/
&DEVC ID='Room103-204', QUANTITY='visibility', XYZ=0.80,3.00,0.4000/
&DEVC ID='Room103-205', QUANTITY='visibility', XYZ=0.80,3.00,0.50/
&DEVC ID='Room103-206', QUANTITY='visibility', XYZ=0.80,3.00,0.60/
&DEVC ID='Room103-207', QUANTITY='visibility', XYZ=0.80,3.00,0.70/
&DEVC ID='Room103-208', QUANTITY='visibility', XYZ=0.80,3.00,0.80/
&DEVC ID='Room103-209', QUANTITY='visibility', XYZ=0.80,3.00,0.90/
&DEVC ID='Room103-210', QUANTITY='visibility', XYZ=0.80,3.00,1.00/
&DEVC ID='Room103-211', QUANTITY='visibility', XYZ=0.80,3.00,1.10/
&DEVC ID='Room103-212', QUANTITY='visibility', XYZ=0.80,3.00,1.20/
&DEVC ID='Room103-213', QUANTITY='visibility', XYZ=0.80,3.00,1.30/
&DEVC ID='Room103-214', QUANTITY='visibility', XYZ=0.80,3.00,1.40/

72
 Appendix C

&DEVC ID='Room103-215', QUANTITY='visibility', XYZ=0.80,3.00,1.50/

&DEVC ID='Room103-216', QUANTITY='visibility', XYZ=0.80,3.00,1.60/
&DEVC ID='Room103-217', QUANTITY='visibility', XYZ=0.80,3.00,1.70/
&DEVC ID='Room103-218', QUANTITY='visibility', XYZ=0.80,3.00,1.80/
&DEVC ID='Room103-219', QUANTITY='visibility', XYZ=0.80,3.00,1.90/
&DEVC ID='Room103-220', QUANTITY='visibility', XYZ=0.80,3.00,2.00/
&DEVC ID='Room103-221', QUANTITY='visibility', XYZ=0.80,3.00,2.10/
&DEVC ID='Room103-222', QUANTITY='visibility', XYZ=0.80,3.00,2.20/
&DEVC ID='Room103-223', QUANTITY='visibility', XYZ=0.80,3.00,2.30/
&DEVC ID='Room103-224', QUANTITY='visibility', XYZ=0.80,3.00,2.40/
&DEVC ID='Room103-225', QUANTITY='visibility', XYZ=0.80,3.00,2.50/
&DEVC ID='RoomCorr-001', QUANTITY='visibility', XYZ=-1.50,6.00,0.1000/
&DEVC ID='RoomCorr-002', QUANTITY='visibility', XYZ=-1.50,6.00,0.2000/
&DEVC ID='RoomCorr-003', QUANTITY='visibility', XYZ=-1.50,6.00,0.3000/
&DEVC ID='RoomCorr-004', QUANTITY='visibility', XYZ=-1.50,6.00,0.4000/
&DEVC ID='RoomCorr-005', QUANTITY='visibility', XYZ=-1.50,6.00,0.50/
&DEVC ID='RoomCorr-006', QUANTITY='visibility', XYZ=-1.50,6.00,0.60/
&DEVC ID='RoomCorr-007', QUANTITY='visibility', XYZ=-1.50,6.00,0.70/
&DEVC ID='RoomCorr-008', QUANTITY='visibility', XYZ=-1.50,6.00,0.80/
&DEVC ID='RoomCorr-009', QUANTITY='visibility', XYZ=-1.50,6.00,0.90/
&DEVC ID='RoomCorr-010', QUANTITY='visibility', XYZ=-1.50,6.00,1.00/
&DEVC ID='RoomCorr-011', QUANTITY='visibility', XYZ=-1.50,6.00,1.10/
&DEVC ID='RoomCorr-012', QUANTITY='visibility', XYZ=-1.50,6.00,1.20/
&DEVC ID='RoomCorr-013', QUANTITY='visibility', XYZ=-1.50,6.00,1.30/
&DEVC ID='RoomCorr-014', QUANTITY='visibility', XYZ=-1.50,6.00,1.40/
&DEVC ID='RoomCorr-015', QUANTITY='visibility', XYZ=-1.50,6.00,1.50/
&DEVC ID='RoomCorr-016', QUANTITY='visibility', XYZ=-1.50,6.00,1.60/
&DEVC ID='RoomCorr-017', QUANTITY='visibility', XYZ=-1.50,6.00,1.70/
&DEVC ID='RoomCorr-018', QUANTITY='visibility', XYZ=-1.50,6.00,1.80/
&DEVC ID='RoomCorr-019', QUANTITY='visibility', XYZ=-1.50,6.00,1.90/
&DEVC ID='RoomCorr-020', QUANTITY='visibility', XYZ=-1.50,6.00,2.00/
&DEVC ID='RoomCorr-021', QUANTITY='visibility', XYZ=-1.50,6.00,2.10/
&DEVC ID='RoomCorr-022', QUANTITY='visibility', XYZ=-1.50,6.00,2.20/
&DEVC ID='RoomCorr-023', QUANTITY='visibility', XYZ=-1.50,6.00,2.30/
&DEVC ID='RoomCorr-024', QUANTITY='visibility', XYZ=-1.50,6.00,2.40/
&DEVC ID='RoomCorr-025', QUANTITY='visibility', XYZ=-1.50,6.00,2.50/
&DEVC ID='RoomCorr-101', QUANTITY='visibility', XYZ=0.80,6.00,0.1000/
&DEVC ID='RoomCorr-102', QUANTITY='visibility', XYZ=0.80,6.00,0.2000/
&DEVC ID='RoomCorr-103', QUANTITY='visibility', XYZ=0.80,6.00,0.3000/
&DEVC ID='RoomCorr-104', QUANTITY='visibility', XYZ=0.80,6.00,0.4000/
&DEVC ID='RoomCorr-105', QUANTITY='visibility', XYZ=0.80,6.00,0.50/
&DEVC ID='RoomCorr-106', QUANTITY='visibility', XYZ=0.80,6.00,0.60/
&DEVC ID='RoomCorr-107', QUANTITY='visibility', XYZ=0.80,6.00,0.70/
&DEVC ID='RoomCorr-108', QUANTITY='visibility', XYZ=0.80,6.00,0.80/
&DEVC ID='RoomCorr-109', QUANTITY='visibility', XYZ=0.80,6.00,0.90/
&DEVC ID='RoomCorr-110', QUANTITY='visibility', XYZ=0.80,6.00,1.00/
&DEVC ID='RoomCorr-111', QUANTITY='visibility', XYZ=0.80,6.00,1.10/
&DEVC ID='RoomCorr-112', QUANTITY='visibility', XYZ=0.80,6.00,1.20/
&DEVC ID='RoomCorr-113', QUANTITY='visibility', XYZ=0.80,6.00,1.30/
&DEVC ID='RoomCorr-114', QUANTITY='visibility', XYZ=0.80,6.00,1.40/
&DEVC ID='RoomCorr-115', QUANTITY='visibility', XYZ=0.80,6.00,1.50/
&DEVC ID='RoomCorr-116', QUANTITY='visibility', XYZ=0.80,6.00,1.60/
&DEVC ID='RoomCorr-117', QUANTITY='visibility', XYZ=0.80,6.00,1.70/
&DEVC ID='RoomCorr-118', QUANTITY='visibility', XYZ=0.80,6.00,1.80/
&DEVC ID='RoomCorr-119', QUANTITY='visibility', XYZ=0.80,6.00,1.90/
&DEVC ID='RoomCorr-120', QUANTITY='visibility', XYZ=0.80,6.00,2.00/
&DEVC ID='RoomCorr-121', QUANTITY='visibility', XYZ=0.80,6.00,2.10/
&DEVC ID='RoomCorr-122', QUANTITY='visibility', XYZ=0.80,6.00,2.20/
&DEVC ID='RoomCorr-123', QUANTITY='visibility', XYZ=0.80,6.00,2.30/
&DEVC ID='RoomCorr-124', QUANTITY='visibility', XYZ=0.80,6.00,2.40/
&DEVC ID='RoomCorr-125', QUANTITY='visibility', XYZ=0.80,6.00,2.50/
&DEVC ID='RoomCorr-201', QUANTITY='visibility', XYZ=3.10,6.00,0.1000/

73
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

&DEVC ID='RoomCorr-202', QUANTITY='visibility', XYZ=3.10,6.00,0.2000/

&DEVC ID='RoomCorr-203', QUANTITY='visibility', XYZ=3.10,6.00,0.3000/
&DEVC ID='RoomCorr-204', QUANTITY='visibility', XYZ=3.10,6.00,0.4000/
&DEVC ID='RoomCorr-205', QUANTITY='visibility', XYZ=3.10,6.00,0.50/
&DEVC ID='RoomCorr-206', QUANTITY='visibility', XYZ=3.10,6.00,0.60/
&DEVC ID='RoomCorr-207', QUANTITY='visibility', XYZ=3.10,6.00,0.70/
&DEVC ID='RoomCorr-208', QUANTITY='visibility', XYZ=3.10,6.00,0.80/
&DEVC ID='RoomCorr-209', QUANTITY='visibility', XYZ=3.10,6.00,0.90/
&DEVC ID='RoomCorr-210', QUANTITY='visibility', XYZ=3.10,6.00,1.00/
&DEVC ID='RoomCorr-211', QUANTITY='visibility', XYZ=3.10,6.00,1.10/
&DEVC ID='RoomCorr-212', QUANTITY='visibility', XYZ=3.10,6.00,1.20/
&DEVC ID='RoomCorr-213', QUANTITY='visibility', XYZ=3.10,6.00,1.30/
&DEVC ID='RoomCorr-214', QUANTITY='visibility', XYZ=3.10,6.00,1.40/
&DEVC ID='RoomCorr-215', QUANTITY='visibility', XYZ=3.10,6.00,1.50/
&DEVC ID='RoomCorr-216', QUANTITY='visibility', XYZ=3.10,6.00,1.60/
&DEVC ID='RoomCorr-217', QUANTITY='visibility', XYZ=3.10,6.00,1.70/
&DEVC ID='RoomCorr-218', QUANTITY='visibility', XYZ=3.10,6.00,1.80/
&DEVC ID='RoomCorr-219', QUANTITY='visibility', XYZ=3.10,6.00,1.90/
&DEVC ID='RoomCorr-220', QUANTITY='visibility', XYZ=3.10,6.00,2.00/
&DEVC ID='RoomCorr-221', QUANTITY='visibility', XYZ=3.10,6.00,2.10/
&DEVC ID='RoomCorr-222', QUANTITY='visibility', XYZ=3.10,6.00,2.20/
&DEVC ID='RoomCorr-223', QUANTITY='visibility', XYZ=3.10,6.00,2.30/
&DEVC ID='RoomCorr-224', QUANTITY='visibility', XYZ=3.10,6.00,2.40/
&DEVC ID='RoomCorr-225', QUANTITY='visibility', XYZ=3.10,6.00,2.50/
&DEVC ID='RoomCorr-301', QUANTITY='visibility', XYZ=5.40,6.00,0.1000/
&DEVC ID='RoomCorr-302', QUANTITY='visibility', XYZ=5.40,6.00,0.2000/
&DEVC ID='RoomCorr-303', QUANTITY='visibility', XYZ=5.40,6.00,0.3000/
&DEVC ID='RoomCorr-304', QUANTITY='visibility', XYZ=5.40,6.00,0.4000/
&DEVC ID='RoomCorr-305', QUANTITY='visibility', XYZ=5.40,6.00,0.50/
&DEVC ID='RoomCorr-306', QUANTITY='visibility', XYZ=5.40,6.00,0.60/
&DEVC ID='RoomCorr-307', QUANTITY='visibility', XYZ=5.40,6.00,0.70/
&DEVC ID='RoomCorr-308', QUANTITY='visibility', XYZ=5.40,6.00,0.80/
&DEVC ID='RoomCorr-309', QUANTITY='visibility', XYZ=5.40,6.00,0.90/
&DEVC ID='RoomCorr-310', QUANTITY='visibility', XYZ=5.40,6.00,1.00/
&DEVC ID='RoomCorr-311', QUANTITY='visibility', XYZ=5.40,6.00,1.10/
&DEVC ID='RoomCorr-312', QUANTITY='visibility', XYZ=5.40,6.00,1.20/
&DEVC ID='RoomCorr-313', QUANTITY='visibility', XYZ=5.40,6.00,1.30/
&DEVC ID='RoomCorr-314', QUANTITY='visibility', XYZ=5.40,6.00,1.40/
&DEVC ID='RoomCorr-315', QUANTITY='visibility', XYZ=5.40,6.00,1.50/
&DEVC ID='RoomCorr-316', QUANTITY='visibility', XYZ=5.40,6.00,1.60/
&DEVC ID='RoomCorr-317', QUANTITY='visibility', XYZ=5.40,6.00,1.70/
&DEVC ID='RoomCorr-318', QUANTITY='visibility', XYZ=5.40,6.00,1.80/
&DEVC ID='RoomCorr-319', QUANTITY='visibility', XYZ=5.40,6.00,1.90/
&DEVC ID='RoomCorr-320', QUANTITY='visibility', XYZ=5.40,6.00,2.00/
&DEVC ID='RoomCorr-321', QUANTITY='visibility', XYZ=5.40,6.00,2.10/
&DEVC ID='RoomCorr-322', QUANTITY='visibility', XYZ=5.40,6.00,2.20/
&DEVC ID='RoomCorr-323', QUANTITY='visibility', XYZ=5.40,6.00,2.30/
&DEVC ID='RoomCorr-324', QUANTITY='visibility', XYZ=5.40,6.00,2.40/
&DEVC ID='RoomCorr-325', QUANTITY='visibility', XYZ=5.40,6.00,2.50/
&DEVC ID='RoomCorrTemp-101', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.1000/
&DEVC ID='RoomCorrTemp-102', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.2000/
&DEVC ID='RoomCorrTemp-103', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.3000/
&DEVC ID='RoomCorrTemp-104', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.4000/
&DEVC ID='RoomCorrTemp-105', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.50/
&DEVC ID='RoomCorrTemp-106', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.60/
&DEVC ID='RoomCorrTemp-107', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.70/
&DEVC ID='RoomCorrTemp-108', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.80/
&DEVC ID='RoomCorrTemp-109', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,0.90/
&DEVC ID='RoomCorrTemp-110', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.00/
&DEVC ID='RoomCorrTemp-111', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.10/
&DEVC ID='RoomCorrTemp-112', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.20/
&DEVC ID='RoomCorrTemp-113', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.30/

74
 Appendix C

&DEVC ID='RoomCorrTemp-114', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.40/

&DEVC ID='RoomCorrTemp-115', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.50/
&DEVC ID='RoomCorrTemp-116', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.60/
&DEVC ID='RoomCorrTemp-117', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.70/
&DEVC ID='RoomCorrTemp-118', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.80/
&DEVC ID='RoomCorrTemp-119', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,1.90/
&DEVC ID='RoomCorrTemp-120', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,2.00/
&DEVC ID='RoomCorrTemp-121', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,2.10/
&DEVC ID='RoomCorrTemp-122', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,2.20/
&DEVC ID='RoomCorrTemp-123', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,2.30/
&DEVC ID='RoomCorrTemp-124', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,2.40/
&DEVC ID='RoomCorrTemp-125', QUANTITY='TEMPERATURE', XYZ=0.80,6.00,2.50/
&DEVC ID='RoomCorrTemp-201', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.1000/
&DEVC ID='RoomCorrTemp-202', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.2000/
&DEVC ID='RoomCorrTemp-203', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.3000/
&DEVC ID='RoomCorrTemp-204', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.4000/
&DEVC ID='RoomCorrTemp-205', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.50/
&DEVC ID='RoomCorrTemp-206', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.60/
&DEVC ID='RoomCorrTemp-207', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.70/
&DEVC ID='RoomCorrTemp-208', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.80/
&DEVC ID='RoomCorrTemp-209', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,0.90/
&DEVC ID='RoomCorrTemp-210', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.00/
&DEVC ID='RoomCorrTemp-211', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.10/
&DEVC ID='RoomCorrTemp-212', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.20/
&DEVC ID='RoomCorrTemp-213', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.30/
&DEVC ID='RoomCorrTemp-214', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.40/
&DEVC ID='RoomCorrTemp-215', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.50/
&DEVC ID='RoomCorrTemp-216', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.60/
&DEVC ID='RoomCorrTemp-217', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.70/
&DEVC ID='RoomCorrTemp-218', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.80/
&DEVC ID='RoomCorrTemp-219', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,1.90/
&DEVC ID='RoomCorrTemp-220', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,2.00/
&DEVC ID='RoomCorrTemp-221', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,2.10/
&DEVC ID='RoomCorrTemp-222', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,2.20/
&DEVC ID='RoomCorrTemp-223', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,2.30/
&DEVC ID='RoomCorrTemp-224', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,2.40/
&DEVC ID='RoomCorrTemp-225', QUANTITY='TEMPERATURE', XYZ=3.10,6.00,2.50/
&DEVC ID='RoomCorrTemp-301', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.1000/
&DEVC ID='RoomCorrTemp-302', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.2000/
&DEVC ID='RoomCorrTemp-303', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.3000/
&DEVC ID='RoomCorrTemp-304', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.4000/
&DEVC ID='RoomCorrTemp-305', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.50/
&DEVC ID='RoomCorrTemp-306', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.60/
&DEVC ID='RoomCorrTemp-307', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.70/
&DEVC ID='RoomCorrTemp-308', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.80/
&DEVC ID='RoomCorrTemp-309', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,0.90/
&DEVC ID='RoomCorrTemp-310', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.00/
&DEVC ID='RoomCorrTemp-311', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.10/
&DEVC ID='RoomCorrTemp-312', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.20/
&DEVC ID='RoomCorrTemp-313', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.30/
&DEVC ID='RoomCorrTemp-314', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.40/
&DEVC ID='RoomCorrTemp-315', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.50/
&DEVC ID='RoomCorrTemp-316', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.60/
&DEVC ID='RoomCorrTemp-317', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.70/
&DEVC ID='RoomCorrTemp-318', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.80/
&DEVC ID='RoomCorrTemp-319', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,1.90/
&DEVC ID='RoomCorrTemp-320', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,2.00/
&DEVC ID='RoomCorrTemp-321', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,2.10/
&DEVC ID='RoomCorrTemp-322', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,2.20/
&DEVC ID='RoomCorrTemp-323', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,2.30/
&DEVC ID='RoomCorrTemp-324', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,2.40/
&DEVC ID='RoomCorrTemp-325', QUANTITY='TEMPERATURE', XYZ=5.40,6.00,2.50/

75
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

&HOLE XB=-1.70,-0.90,5.30,5.60,0.00,2.00/ Door102

&HOLE XB=-0.2000,0.1000,3.60,4.40,0.00,2.00/ Door103
&HOLE XB=-0.2000,0.1000,0.80,1.60,0.00,2.00/ Door104

&OBST XB=0.00,3.60,0.00,2.40,-0.2000,0.00, COLOR='INVISIBLE', SURF_ID='CONCRETE_SLAB'/

Obstruction
&OBST XB=0.00,3.60,2.60,5.40,-0.2000,0.00, COLOR='INVISIBLE', SURF_ID='CONCRETE_SLAB'/
Obstruction
&OBST XB=3.70,6.10,0.00,5.40,-0.2000,0.00, COLOR='INVISIBLE', SURF_ID='CONCRETE_SLAB'/
Obstruction
&OBST XB=-2.50,-0.1000,0.50,5.40,-0.2000,3.6082248E-016, COLOR='INVISIBLE',
SURF_ID='CONCRETE_SLAB'/ Obstruction
&OBST XB=-2.50,6.60,5.50,6.50,-0.2000,0.00, COLOR='INVISIBLE', SURF_ID='CONCRETE_SLAB'/
Obstruction
&OBST XB=-2.60,6.20,2.60,6.60,2.50,2.60, COLOR='INVISIBLE', SURF_ID='TEST_WALL'/
Obstruction
&OBST XB=-2.60,-0.1000,0.4000,2.60,2.50,2.60, COLOR='INVISIBLE', SURF_ID='TEST_WALL'/
Obstruction
&OBST XB=3.70,6.20,-0.1000,2.60,2.50,2.60, COLOR='INVISIBLE', SURF_ID='TEST_WALL'/
Obstruction
&OBST XB=-0.1000,3.70,-0.1000,2.60,2.50,2.60, COLOR='INVISIBLE', SURF_ID='TEST_WALL'/
Fire room
&OBST XB=1.50,2.00,0.90,1.40,0.00,0.1000, SURF_ID='INERT'/ Box

&OBST XB=0.00,3.70,-0.1000,-1.3877788E-017,-0.2000,1.30, SURF_ID='TEST_WALL'/

Fireroom_bottom1
&OBST XB=3.60,3.70,0.00,2.60,-0.2000,1.30, SURF_ID='TEST_WALL'/ Fireroom_right1
&OBST XB=0.00,3.60,2.40,2.60,-0.2000,1.30, SURF_ID='TEST_WALL'/ Fireroom_top1
&OBST XB=-0.1000,0.00,-0.1000,2.60,-0.2000,1.30, SURF_ID='TEST_WALL'/ Fireroom_left1
&OBST XB=0.00,3.60,2.40,2.60,1.30,2.50, SURF_ID='TEST_WALL'/ Fireroom_top2
&OBST XB=3.60,3.70,0.00,2.60,1.30,2.50, SURF_ID='TEST_WALL'/ Fireroom_right2
&OBST XB=0.00,3.70,-0.1000,-1.3877788E-017,1.30,2.50, SURF_ID='TEST_WALL'/
Fireroom_bottom2
&OBST XB=-0.1000,0.00,-0.1000,2.60,1.30,2.50, SURF_ID='TEST_WALL'/ Fireroom_left2
&OBST XB=-2.50,6.10,5.40,5.50,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=6.10,6.20,-0.1000,5.50,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=-2.50,-0.1000,0.4000,0.50,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=-2.60,-2.50,0.4000,6.50,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=-2.60,6.20,6.50,6.60,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=3.60,3.70,2.60,5.40,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=3.70,6.20,-0.1000,-1.3877788E-017,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall
&OBST XB=-0.1000,0.00,2.60,5.40,-0.2000,2.50, SURF_ID='TEST_WALL'/ Wall

&VENT SURF_ID='OPEN', XB=7.00,7.00,5.40,6.60,0.00,2.70, COLOR='INVISIBLE'/ Vent Max X

for Mesh14
&VENT SURF_ID='OPEN', XB=6.20,7.00,5.40,5.40,0.00,2.70, COLOR='INVISIBLE'/ Vent Min Y
for Mesh14
&VENT SURF_ID='OPEN', XB=6.20,7.00,6.60,6.60,0.00,2.70, COLOR='INVISIBLE'/ Vent Max Y
for Mesh14
&VENT SURF_ID='OPEN', XB=3.70,7.00,5.40,6.60,2.70,2.70, COLOR='INVISIBLE'/ Vent Max Z
for Mesh14
&VENT SURF_ID='BURNER', XB=1.50,2.00,0.90,1.40,0.1000,0.1000/ Burner

&SLCF QUANTITY='TEMPERATURE', PBX=-1.30/

&SLCF QUANTITY='V-VELOCITY', VECTOR=.TRUE., PBX=-1.30/
&SLCF QUANTITY='visibility', PBY=6.00/
&SLCF QUANTITY='TEMPERATURE', PBZ=1.90/
&SLCF QUANTITY='visibility', PBX=-1.30/
&SLCF QUANTITY='visibility', PBY=4.00/
&SLCF QUANTITY='TEMPERATURE', PBX=1.80/
&SLCF QUANTITY='TEMPERATURE', PBY=1.20/
&SLCF QUANTITY='V-VELOCITY', VECTOR=.TRUE., PBY=1.20/

76
 Appendix C

&SLCF QUANTITY='TEMPERATURE', PBY=4.00/

&SLCF QUANTITY='U-VELOCITY', VECTOR=.TRUE., PBY=4.00/
&SLCF QUANTITY='U-VELOCITY', VECTOR=.TRUE., PBY=6.00/
&SLCF QUANTITY='TEMPERATURE', PBY=6.00/
&SLCF QUANTITY='TEMPERATURE', PBZ=1.50/

&TAIL /

C.2 Script file

The template script file below was used in order to submit computing jobs to
LUNARC.
#!/bin/sh
# Request number of nodes
#PBS -l nodes=14
# Request of wall-clock time
#PBS -l walltime=10:00:00
# regular output (stdout) and terminal output (stderr)
#PBS -o stdout.txt
#PBS -e stderr.txt
# Send notification when job starts, finishes and aborts.
#PBS -m bea
# Mail address to send notifications
#PBS -M [email protected]
cd $PBS_O_WORKDIR
# Enable modules and add software
. use_modules
module add intel/10.1
module add mpich-intel10/1.2.7p1
# Run on all nodes and create output for monitoring simulation progress
mpiexec /sw/pkg/brand/fds_5.3.1_lunarc_dt sy101.fds >regoutput.out
2>terminaloutput.err

77
Appendix D – Selected figures

The figures presented in this appendix are the selected variables in Table D.1
below. All other output distributions have a probability of visibility not
exceeding 10 meters of either 1 or 0 and are therefore not presented.
TABLE D.1 Probability of visibility not exceeding 10 m for selected heights
Height (m) Soot yield Mass ext. coeff. HRR Simultaneous
0 – 1.1 0 0 0 0
1.2 0 0 0 0.075
1.3 0.195 0 0 0.24
1.4 0.745 0 0.24 0.65
1.5 0.99 1 0.7 0.955
1.6 1 1 0.965 0.995
1.7 – 2.5 1 1 1 1

79
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

D.1 Soot yield

70

60
NUMBER OF SIMULATIONS

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.1 Distribution of the visibility in the soot yield simulations at a height of 1.2
meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.

20

18
NUMBER OF SIMULATIONS

16

14

12

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.2 Distribution of the visibility in the soot yield simulations at a height of 1.3
meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.195.

80
 Appendix D

40

35
NUMBER OF SIMULATIONS

30

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.3 Distribution of the visibility in the soot yield simulations at a height of 1.4
meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.745.

80

70
NUMBER OF SIMULATIONS

60

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.4 Distribution of the visibility in the soot yield simulations at a height of 1.5
meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.99.

81
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

120

100
NUMBER OF SIMULATIONS

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.5 Distribution of the visibility in the soot yield simulations at a height of 1.6
meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 1.

82
 Appendix D

D.2 Mass extinction coefficient

250
NUMBER OF SIMULATIONS

200

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.6 Distribution of the visibility in the mass extinction coefficient simulations at a
height of 1.2 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.

180

160
NUMBER OF SIMULATIONS

140

120

100

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.7 Distribution of the visibility in the mass extinction coefficient simulations at a
height of 1.3 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.

83
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

70

60
NUMBER OF SIMULATIONS

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.8 Distribution of the visibility in the mass extinction coefficient simulations at a
height of 1.4 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.

140

120
NUMBER OF SIMULATIONS

100

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.9 Distribution of the visibility in the mass extinction coefficient simulations at a
height of 1.5 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 1.

84
 Appendix D

180
NUMBER OF SIMULATIONS 160

140

120

100

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.10 Distribution of the visibility in the mass extinction coefficient simulations at
a height of 1.6 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 1.

85
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

D.3 HRR

250
NUMBER OF SIMULATIONS

200

150

100

50

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.11 Distribution of the visibility in the heat release rate simulations at a height
of 1.2 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.

120

100
NUMBER OF SIMULATIONS

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.12 Distribution of the visibility in the heat release rate simulations at a height
of 1.3 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.

86
 Appendix D

35

30
NUMBER OF SIMULATIONS

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.13 Distribution of the visibility in the heat release rate simulations at a height
of 1.4 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.24.

35

30
NUMBER OF SIMULATIONS

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.14 Distribution of the visibility in the heat release rate simulations at a height
of 1.5 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.7.

87
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

60

50
NUMBER OF SIMULATIONS

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.15 Distribution of the visibility in the heat release rate simulations at a height
of 1.6 meters. Y-axis is the number of simulations, X-axis is the visibility. Pr(x<10) = 0.965.

88
 Appendix D

D.4 Simultaneous variables

80

70
NUMBER OF SIMULATIONS

60

50

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.16 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.2 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.075.

35

30
NUMBER OF SIMULATIONS

25

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.17 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.3 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.24.

89
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

30

25
NUMBER OF SIMULATIONS

20

15

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.18 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.4 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.65.

60

50
NUMBER OF SIMULATIONS

40

30

20

10

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.19 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.5 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.955.

90
 Appendix D

120

100
NUMBER OF SIMULATIONS

80

60

40

20

0
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30

VISIBILITY (m)

FIGURE D.20 Distribution of the visibility in the simultaneous variable simulations at a

height of 1.6 meters. Y-axis is the number of simulations, X-axis is the visibility.
Pr(x<10) = 0.995.

91
Appendix E – Output summary

This appendix presents the number of simulations resulting in a certain visibility

at a given height.

E.1 Soot yield

H( m)a
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
V (m)b
1 0 0 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 0 0 0 0 0 0
7 0 0 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 0 0 0 0
13 0 0 0 0 0 0 0 0 0 0 0 3
14 0 0 0 0 0 0 0 0 0 0 0 2
15 0 0 0 0 0 0 0 0 0 0 0 7
16 0 0 0 0 0 0 0 0 0 0 0 12
17 0 0 0 0 0 0 0 0 0 0 0 4
18 0 0 0 0 0 0 0 0 0 0 0 10
19 0 0 0 0 0 0 0 0 0 0 1 6
20 0 0 0 0 0 0 0 0 0 0 3 6
21 0 0 0 0 0 0 0 0 0 0 4 11
22 0 0 0 0 0 0 0 0 0 0 10 7
23 0 0 0 0 0 0 0 0 0 2 8 9
24 0 0 0 0 0 0 0 0 0 4 9 9
25 0 0 0 0 0 0 0 0 2 9 5 8
26 0 0 0 0 0 0 0 1 5 8 17 7
27 0 0 0 0 0 0 1 4 9 14 8 10
28 0 0 0 0 0 0 5 8 13 14 17 13
29 0 0 0 0 1 6 14 19 20 21 18 11
30 200 200 200 200 199 194 180 168 151 128 100 65
a
Height above floor
b
Visibility

93
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

H( m)a
V (m)b 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5
1 0 0 43 113 138 148 155 161 166 170 171 175 175
2 0 22 67 54 49 44 40 36 32 29 28 25 25
3 0 35 32 24 12 8 5 3 2 1 1 0 0
4 0 23 22 6 1 0 0 0 0 0 0 0 0
5 0 21 13 3 0 0 0 0 0 0 0 0 0
6 5 17 10 0 0 0 0 0 0 0 0 0 0
7 10 12 6 0 0 0 0 0 0 0 0 0 0
8 12 11 2 0 0 0 0 0 0 0 0 0 0
9 12 8 3 0 0 0 0 0 0 0 0 0 0
10 8 6 1 0 0 0 0 0 0 0 0 0 0
11 12 7 1 0 0 0 0 0 0 0 0 0 0
12 12 4 0 0 0 0 0 0 0 0 0 0 0
13 6 7 0 0 0 0 0 0 0 0 0 0 0
14 4 4 0 0 0 0 0 0 0 0 0 0 0
15 10 3 0 0 0 0 0 0 0 0 0 0 0
16 9 4 0 0 0 0 0 0 0 0 0 0 0
17 5 2 0 0 0 0 0 0 0 0 0 0 0
18 9 6 0 0 0 0 0 0 0 0 0 0 0
19 6 3 0 0 0 0 0 0 0 0 0 0 0
20 6 0 0 0 0 0 0 0 0 0 0 0 0
21 8 0 0 0 0 0 0 0 0 0 0 0 0
22 3 3 0 0 0 0 0 0 0 0 0 0 0
23 5 0 0 0 0 0 0 0 0 0 0 0 0
24 5 1 0 0 0 0 0 0 0 0 0 0 0
25 7 1 0 0 0 0 0 0 0 0 0 0 0
26 7 0 0 0 0 0 0 0 0 0 0 0 0
27 5 0 0 0 0 0 0 0 0 0 0 0 0
28 7 0 0 0 0 0 0 0 0 0 0 0 0
29 8 0 0 0 0 0 0 0 0 0 0 0 0
30 19 0 0 0 0 0 0 0 0 0 0 0 0
a
Height above floor
b
Visibility

94
 Appendix E

E.2 Mass extinction coefficient

H( m)a
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
V (m)b
1 0 0 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 0 0 0 0 0 0
7 0 0 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 0 0 0 0
13 0 0 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0
15 0 0 0 0 0 0 0 0 0 0 0 0
16 0 0 0 0 0 0 0 0 0 0 0 0
17 0 0 0 0 0 0 0 0 0 0 0 0
18 0 0 0 0 0 0 0 0 0 0 0 0
19 0 0 0 0 0 0 0 0 0 0 0 0
20 0 0 0 0 0 0 0 0 0 0 0 0
21 0 0 0 0 0 0 0 0 0 0 0 0
22 0 0 0 0 0 0 0 0 0 0 0 0
23 0 0 0 0 0 0 0 0 0 0 0 0
24 0 0 0 0 0 0 0 0 0 0 0 0
25 0 0 0 0 0 0 0 0 0 0 0 0
26 0 0 0 0 0 0 0 0 0 0 0 0
27 0 0 0 0 0 0 0 0 0 0 0 0
28 0 0 0 0 0 0 0 0 0 0 0 0
29 0 0 0 0 0 0 0 0 0 0 0 0
30 200 200 200 200 200 200 200 200 200 200 200 200
a
Height above floor
b
Visibility

95
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

H( m)a
V (m)b 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5
1 0 0 0 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 64 154 200 200 200 200 200 200 200
3 0 0 0 155 136 46 0 0 0 0 0 0 0
4 0 0 0 45 0 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0 0 0 0 0 0
6 0 0 84 0 0 0 0 0 0 0 0 0 0
7 0 0 115 0 0 0 0 0 0 0 0 0 0
8 0 0 1 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 0 0 0 0 0
13 0 0 0 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0 0
15 0 39 0 0 0 0 0 0 0 0 0 0 0
16 0 60 0 0 0 0 0 0 0 0 0 0 0
17 0 54 0 0 0 0 0 0 0 0 0 0 0
18 0 47 0 0 0 0 0 0 0 0 0 0 0
19 0 0 0 0 0 0 0 0 0 0 0 0 0
20 0 0 0 0 0 0 0 0 0 0 0 0 0
21 0 0 0 0 0 0 0 0 0 0 0 0 0
22 0 0 0 0 0 0 0 0 0 0 0 0 0
23 0 0 0 0 0 0 0 0 0 0 0 0 0
24 0 0 0 0 0 0 0 0 0 0 0 0 0
25 0 0 0 0 0 0 0 0 0 0 0 0 0
26 0 0 0 0 0 0 0 0 0 0 0 0 0
27 0 0 0 0 0 0 0 0 0 0 0 0 0
28 0 0 0 0 0 0 0 0 0 0 0 0 0
29 47 0 0 0 0 0 0 0 0 0 0 0 0
30 153 0 0 0 0 0 0 0 0 0 0 0 0
a
Height above floor
b
Visibility

96
 Appendix E

E.3 HRR
H( m)a
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
V (m)b
1 0 0 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0 0 0 0 0
6 0 0 0 0 0 0 0 0 0 0 0 0
7 0 0 0 0 0 0 0 0 0 0 0 0
8 0 0 0 0 0 0 0 0 0 0 0 0
9 0 0 0 0 0 0 0 0 0 0 0 0
10 0 0 0 0 0 0 0 0 0 0 0 0
11 0 0 0 0 0 0 0 0 0 0 0 0
12 0 0 0 0 0 0 0 0 0 0 0 0
13 0 0 0 0 0 0 0 0 0 0 0 0
14 0 0 0 0 0 0 0 0 0 0 0 0
15 0 0 0 0 0 0 0 0 0 0 0 0
16 0 0 0 0 0 0 0 0 0 0 0 0
17 0 0 0 0 0 0 0 0 0 0 0 0
18 0 0 0 0 0 0 0 0 0 0 0 0
19 0 0 0 0 0 0 0 0 0 0 0 0
20 0 0 0 0 0 0 0 0 0 0 0 0
21 0 0 0 0 0 0 0 0 0 0 0 0
22 0 0 0 0 0 0 0 0 0 0 0 0
23 0 0 0 0 0 0 0 0 0 0 0 0
24 0 0 0 0 0 0 0 0 0 0 0 0
25 0 0 0 0 0 0 0 0 0 0 0 0
26 0 0 0 0 0 0 0 0 0 0 0 0
27 0 0 0 0 0 0 0 0 0 0 0 0
28 0 0 0 0 0 0 0 0 0 0 0 0
29 0 0 0 0 0 0 0 0 0 0 0 1
30 200 200 200 200 200 200 200 200 200 200 200 199
a
Height above floor
b
Visibility

97
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

H( m)a
V (m)b 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5
1 0 0 0 0 0 0 0 0 0 5 9 33 33
2 0 0 0 49 88 113 130 141 150 152 151 134 134
3 0 0 23 55 62 53 44 39 36 33 32 28 28
4 0 0 32 32 25 20 19 18 14 10 8 5 5
5 0 0 20 21 14 11 7 2 0 0 0 0 0
6 0 6 18 14 9 3 0 0 0 0 0 0 0
7 0 19 21 10 2 0 0 0 0 0 0 0 0
8 0 13 13 7 0 0 0 0 0 0 0 0 0
9 0 10 13 5 0 0 0 0 0 0 0 0 0
10 0 7 9 5 0 0 0 0 0 0 0 0 0
11 0 7 8 2 0 0 0 0 0 0 0 0 0
12 0 7 7 0 0 0 0 0 0 0 0 0 0
13 0 6 5 0 0 0 0 0 0 0 0 0 0
14 0 7 6 0 0 0 0 0 0 0 0 0 0
15 0 1 4 0 0 0 0 0 0 0 0 0 0
16 0 9 4 0 0 0 0 0 0 0 0 0 0
17 0 6 4 0 0 0 0 0 0 0 0 0 0
18 2 4 3 0 0 0 0 0 0 0 0 0 0
19 3 7 2 0 0 0 0 0 0 0 0 0 0
20 9 5 2 0 0 0 0 0 0 0 0 0 0
21 3 5 4 0 0 0 0 0 0 0 0 0 0
22 9 6 1 0 0 0 0 0 0 0 0 0 0
23 2 4 1 0 0 0 0 0 0 0 0 0 0
24 13 5 0 0 0 0 0 0 0 0 0 0 0
25 7 5 0 0 0 0 0 0 0 0 0 0 0
26 6 7 0 0 0 0 0 0 0 0 0 0 0
27 8 7 0 0 0 0 0 0 0 0 0 0 0
28 8 8 0 0 0 0 0 0 0 0 0 0 0
29 16 8 0 0 0 0 0 0 0 0 0 0 0
30 114 31 0 0 0 0 0 0 0 0 0 0 0
a
Height above floor
b
Visibility

98
 Appendix E

E.4 Simultaneous variables

H( m)a
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1 1.2
V (m)b
1 0 0 0 0 0 0 0 0 0 0 0 0
2 0 0 0 0 0 0 0 0 0 0 0 0
3 0 0 0 0 0 0 0 0 0 0 0 0
4 0 0 0 0 0 0 0 0 0 0 0 0
5 0 0 0 0 0 0 0 0 0 0 0 1
6 0 0 0 0 0 0 0 0 0 0 0 1
7 0 0 0 0 0 0 0 0 0 0 0 7
8 0 0 0 0 0 0 0 0 0 0 0 4
9 0 0 0 0 0 0 0 0 0 0 0 2
10 0 0 0 0 0 0 0 0 0 0 0 2
11 0 0 0 0 0 0 0 0 0 0 1 4
12 0 0 0 0 0 0 0 0 0 0 0 1
13 0 0 0 0 0 0 0 0 0 0 1 3
14 0 0 0 0 0 0 0 0 0 0 4 4
15 0 0 0 0 0 0 0 0 0 0 3 2
16 0 0 0 0 0 0 0 0 0 1 1 2
17 0 0 0 0 0 0 0 0 0 1 5 5
18 0 0 0 0 0 0 0 0 0 0 0 2
19 0 0 0 0 0 0 0 0 0 0 5 8
20 0 0 0 0 0 0 0 0 0 2 5 4
21 0 0 0 0 0 0 0 0 2 5 2 5
22 0 0 0 0 0 0 0 0 0 0 3 11
23 0 0 0 0 0 0 0 0 2 5 4 6
24 0 0 0 0 0 0 0 2 1 9 2 7
25 0 0 0 0 0 0 0 2 6 4 4 7
26 0 0 0 0 0 0 3 2 4 4 12 11
27 0 0 0 0 0 3 2 7 11 3 12 3
28 0 0 0 0 3 3 8 9 7 12 13 4
29 0 0 2 4 7 9 14 11 12 23 20 18
30 200 200 198 196 190 185 173 167 155 131 103 76
a
Height above floor
b
Visibility

99
A method of quantifying user uncertainty in FDS by using Monte Carlo analysis

H( m)a
V (m)b 1.3 1.4 1.5 1.6 1.7 1.8 1.9 2 2.1 2.2 2.3 2.4 2.5
1 0 15 49 103 123 136 146 149 159 160 160 163 163
2 12 21 54 53 52 48 43 41 33 33 34 32 32
3 6 27 24 19 17 11 7 9 7 6 5 4 4
4 3 15 25 15 4 4 3 0 0 0 1 1 1
5 4 17 13 2 3 0 0 1 1 1 0 0 0
6 5 10 10 4 0 1 1 0 0 0 0 0 0
7 4 9 2 3 1 0 0 0 0 0 0 0 0
8 6 8 9 0 0 0 0 0 0 0 0 0 0
9 8 8 5 0 0 0 0 0 0 0 0 0 0
10 9 10 1 0 0 0 0 0 0 0 0 0 0
11 10 4 1 0 0 0 0 0 0 0 0 0 0
12 5 8 1 1 0 0 0 0 0 0 0 0 0
13 5 4 1 0 0 0 0 0 0 0 0 0 0
14 5 9 1 0 0 0 0 0 0 0 0 0 0
15 7 3 2 0 0 0 0 0 0 0 0 0 0
16 5 1 1 0 0 0 0 0 0 0 0 0 0
17 6 0 0 0 0 0 0 0 0 0 0 0 0
18 3 3 0 0 0 0 0 0 0 0 0 0 0
19 4 3 0 0 0 0 0 0 0 0 0 0 0
20 5 1 0 0 0 0 0 0 0 0 0 0 0
21 5 0 0 0 0 0 0 0 0 0 0 0 0
22 5 4 0 0 0 0 0 0 0 0 0 0 0
23 4 3 0 0 0 0 0 0 0 0 0 0 0
24 8 5 1 0 0 0 0 0 0 0 0 0 0
25 6 0 0 0 0 0 0 0 0 0 0 0 0
26 6 3 0 0 0 0 0 0 0 0 0 0 0
27 5 1 0 0 0 0 0 0 0 0 0 0 0
28 9 1 0 0 0 0 0 0 0 0 0 0 0
29 9 2 0 0 0 0 0 0 0 0 0 0 0
30 31 5 0 0 0 0 0 0 0 0 0 0 0
a
Height above floor
b
Visibility

100