Download as pdf or txt
Download as pdf or txt
You are on page 1of 63

Frédéric Conte

CFD simulations of smoke detection


in rooms with high ceilings
SP AR 2002:30
Brandteknik
Borås 2002

Ecole Nationale Supérieure


d'Ingénieurs de Mécanique
Energétique de Valenciennes
2

Abstract
In industry, there is usually a detection system placed to initiate the release of extinguishing
media or send an alarm to the fire brigade in case of a fire. Fires in industrial buildings can
generally cause serious damage so an early detection system is needed. Heat and smoke detectors
are the most common. The detection time depends on the detector and the fire scenario.

This report is included in a larger project "Early detection in rooms with high ceilings". This
project consists of the study of smoke production, smoke detectors, and two problems detecting
fires and will be finished by full-scale experiments. The objective of this report was to examine
several scenarios which could prevent an early detection of smoke in rooms with high ceilings.
Two different problems were studied: the effects of the existence of a temperature gradient on the
smoke spread in a room and the effects of a ventilation system on the smoke spread in big
industrial buildings.

During the five-month placement, several simulations have been performed with the CFD
software SOFIE. Three stages can be distinguished: determining the best model for the smoke
source, the simulations of the temperature gradient and the ventilation system. A lot of time was
spent to determine the best model for the smoke source doing experiments in a laboratory, and
carrying out a parametric study with SOFIE to fit the experiments. The work for the simulations
of the temperature gradient and the ventilation system was to find the manner to simulate a
temperature gradient in SOFIE and to reach the best model for these two problems.

The three different parts of this study are presented in this report.

Key words: Smoke, detection, CFD, ventilation, temperature gradient.

SP Sveriges Provnings- och SP Swedish National Testing and


Forskningsinstitut Research Institute

SP AR 2002:30 SP AR 2002:30
Borås 2002

Postal address:
Box 857,
SE-501 15 BORÅS, Sweden
Telephone: +46 33 16 50 00
Telex: 36252 Testing S
Telefax: +46 33 13 55 02
E-mail: info@sp.se
3

Preface
This report is the result of a five-month placement carried out at Brandteknik, the department of
fire technology of SP (Swedish national testing and research institute). This placement was a part
of my engineer formation in France at the ENSIMEV (High institute of mechanical engineering
of Valenciennes). This study was done with the collaboration of my supervisor at SP Petra
Andersson who gave good advice and suggestions during the placement.
4

Table of contents

Abstract 2

Preface 3

Table of contents 4

A - Problem position 7

1 Problems with detecting fires 7

2 Smoke source 8
2.1 Pyrolysis 8
2.2 Smouldering combustion 8

3 Presentation of SOFIE 9
3.1 CFD 9
3.2 SOFIE 10

B - Preparatory study 12

4 Parametric study 12
4.1 Reference case 12
4.2 The different changes and effects 12
4.3 Conclusion 12

5 Experiments 13
5.1 Experimental protocol 13
5.2 Results 15

6 Tuning in the smoke source 19


6.1 Basic script 19
6.2 Results 19
6.3 Boundary values 21
6.4 Geometry 21
6.5 Model 22
6.6 Conclusion 23

C - Smoke spread in a room with a temperature gradient 24

7 The model in SOFIE 24


7.1 Geometry 24
7.2 Models and constants 27
7.4 Interior values 28
7.5 Solving 29

8 Results 30
8.1 Temperature gradient in the small room 30
8.2 Temperature gradient in the large room 34

9 Conclusion 39
5

D - Smoke spread in a room with a ventilation system 40

10 The model 40
10.1 Geometry 40
10.2 Models and constants 44
10.3 Boundary 44
10.4 Solving 46

11 Results 47
11.1 Smoke source placed in the middle of four inflows. 47
11.2 Smoke source placed just under an inflow 49

12 Conclusion 51

13 Summary 52

14 References 53

15 Acknowledgements 54

Appendix 55

A1 Results of the experiments 55

A2 Command files used in SOFIE 57


A2.1 Temperature gradient 57
A2.1 Ventilation system 61
6
7

A - Problem position
1 Problems with detecting fires
Usually in industry, there are some detectors placed to initiate the release of extinguishing media
or send an alarm to the fire brigade in case of a fire. Heat and smoke detectors, are the most
common. The detection time depends on the detector and the fire scenario.

Early detection in rooms with ventilation and/or high ceilings is a difficult task. A simple fire can
cause severe damage in many cases and therefore smoke detectors should be placed at locations
where the smoke reaches the detector at an early stage of the fire. In order to calculate when a
detector will be activated one should know the smoke production from the fire. Further, we need
to know how the smoke is transported and the sensitivity of the detector.

Fires in electrical equipment and packaging materials are relatively common in industries [1].
The smoke production from these is usually not known, especially when in the beginning of the
fire there is a smouldering combustion. There is data available in the literature on smoke
production from mainly pure fuels and usually from flaming combustion.

Usually when doing smoke spread calculations a uniform temperature profile in the room is
assumed together with no airflow before the fire starts. This is not the case in the reality, in rooms
with high ceiling there is usually a significant temperature gradient and the airflow in the room
can be rather large due to the ventilation system [2]. There is also the problem of other heating
sources in the room that causes airflows. These problems cannot be modelled in, for instance, two
zone models. In general they require CFD type of simulations. Only a very limited number of
simulations of this kind have been published.

Working with a CFD model, several cases can be simulated. This type of calculation is the
cheapest way to undertake investigations in the smoke spread domain. In the last ten years,
calculations using CFD have known a big development. The computers powerful increasing, the
CFD calculations have got good results and are used nowadays in many cases. A final experiment
can be carried out to conclude a study, to check and develop rules and advice.
8

2 Smoke source
Different things can be at the origin of the smoke production. Smoke sources without flames,
studied in this report, often existing before the real fire, can product a lot of smoke. Most
common materials give less smoke with flaming combustion [3]. The smoke production often
comes from some packaging materials and electrical equipment, in industrial buildings.

2.1 Pyrolysis
The different kind of plastic used for packaging materials or electrical equipment can be an
important smoke source when they are heated. When heating plastic, the plastic reaches high
temperatures, melts, evaporates and so some gases are produced. In this case, the smoke consists
of gases from chemical decomposition of the plastics; there are very few oxidations. The
pyrolysis can be the origin of real fire. The pyrolysis is, by definition [4], an irreversible chemical
decomposition of a material under the only action of the heat. It is produced by a rise of
temperature without reaction with the oxygen. It should be distinguished from the combustion,
which requires a contribution of oxygen and releases a lot of heat.

2.2 Smouldering combustion


Only porous materials which form a solid carbonaceous char when heated can undergo self-
sustained smouldering combustion. Included are a wide range of materials of vegetable origin
such as paper, cellulosic fabrics, sawdust, fibreboard and latex rubber, as well as some
thermosetting plastics in expanded form [3]. A smouldering combustion is combustion without
flames, by oxidation, with smoke and char formation often occurring before fire.

Figure 1 Photo of a polyethylene sample pyrolysis.


9

3 Presentation of SOFIE
There has been considerable growth in the development and application of Computational Fluid
Dynamics to all aspects of fluid dynamics and heat transfer. CFD has made an increasingly
significant contribution to many branches of engineering since its emergence in the mid 1970’s as
a practical design and analysis tool. As part of the research, design and development process,
CFD programs are now considered to be standard numerical tools, widely utilised within
industry.

Several different CFD models exist today, general purpose codes like FLUENT and PHOENICS,
or fire purpose codes such as JASMINE or SOFIE.

SOFIE (Simulation Of Fires In Enclosures) is a field model using CFD written in Fortan and in
C. SOFIE 3 has been written by Pr Rubini at Cranfield University. It has been developed under
the sponsorship of a Consortium comprising a number of European fire research laboratories;
FRS, HSE, Cranfield University and Home Office (UK); SP Borås and Lund University
(Sweden); VTT (Finland); and CSTB (France).

The major interest of the simulation is its low use cost. But it cannot replace experiments; it is
necessary to compare the two ways to obtain the best results.

3.1 CFD
CFD codes are structured around the numerical algorithms that can tackle fluid flow problems.
All the field models using CFD use the same method. They first define a domain in space, the
computational domain, where the simulation will be carried out. This domain is divided into a
large number of small control volumes called cells, which can define a solid or a certain volume
of fluid. The problem geometry and boundary conditions can be defined. Then, CFD technique is
applied in order to solve a set of non-linear partial differential equations derived from basic laws
of nature. In order to make the calculation possible, complex real phenomena are modelled. In the
case of fire, a combustion model is used to simulate the course of combustion, a turbulence model
is required and usually a buoyancy modified k-ε model is used to predict the turbulent flow.
There are also others models for the radiation, the soot or the fire spread.

Governing equations

All the models solve the conservation equations for mass, movement, energy, and evaluate the
species in each control volume [5]. All the conservation equations follow the same formula:


( ρφ ) + div ( ρ u φ ) = div ( Γ grad (φ )) + S φ
∂t
I II III IV

Where φ is an arbitrary dependent variable,


I is the rate of change of φ in the control volume,
II is the change of φ due to convection,
III is the change of φ due to diffusion,
IV is a source term.
10

In CFD calculations about 10–100 000 control volumes are used with the finite volume method
and it is also often possible to take advantages of symmetries in the room, using symmetry
conditions, so that only a part of it needs to be simulated.

The governing equations for 3D fluid flow are treated in the same way as above except that the
integration is carried out over a three-dimensional control volume. The result is a set of algebraic
equations, which in SOFIE is solved with for example a TDMA solver (line-by-line tridiagonal
matrix algorithm) [6].

3.2 SOFIE
SOFIE solves the Reynolds averaged Navier-Stokes equation using a finite volume procedure
like the other CFD models.

3.2.3 The interface

Figure 2 Main menu of SOFIE.

SOFIE uses a text interface pre-processor [7]. It uses five main menus: file where data exchanges
(to export or to import the solutions) are dealt with, setup where the model is defined, run where
the problem is solved, print where solutions can be visualized, and control where all the solver
parameters are accessible. The databases are in specific files, and contain information on material
and species properties that are required in order to solve the problem.

Usually one writes a command file. The required type of solution is then specified. The different
models for combustion, radiations, turbulence et cetera, are chosen. Then the geometry is defined.
It can be imported from different software, like AC3D, where it can be defined easily. The
boundary types and values are specified at the end.

The simulation can be run in steady state or in transient mode. The results from the simulation
can be exported to visualisation software like Fieldview or Smokeview.

3.2.2 The k-ε turbulence model


The k-ε turbulence model, the main model used in SOFIE for this report, is brievly presented in
the following paragraph.

The k-ε turbulence model is a two equation model implying that two additional transport
equations, partial differential equation, are employed to determine the local turbulent viscosity,
µ(x, y, z, t). Depending on the flow, different transport equations have to be used. Therefore, the
k-ε model can be divided into high Reynolds number and low Reynolds number k-ε model.
11

The k-ε model focuses on the mechanisms that affect the turbulent kinetic energy. k is the
turbulent kinetic energy and ε is the rate of dissipation of turbulent kinetic energy [8].

In words the equations are:

Rate of change Transport Transport Rate of Rate of


+ of k or ε by = of k or ε by + production - destruction
of k or ε
convection diffusion of k or ε of k or ε

The equation contain five adjustable constants Cµ, σk, σε, C1ε and C2ε.
The values generally taken are:
Cµ = 0.09; σk = 1.00; σε = 1.30; C1ε = 1.44; C2ε = 1.92.

Advantages:

• Simplest turbulence model for which only initial and/or boundary conditions need to be
supplied
• Excellent performance for many industrially relevant flows
• Well established; the most widely validated turbulence model

Disadvantages:

• More expensive to implement than mixing length model


• Poor performance in some cases:
- Some unconfined flows
- Flows with large extra strains
- Rotating flows
- Fully developed flows in non-circular ducts
12

B - Preparatory study
4 Parametric study
To begin this preparatory study, a parametric study was carried out in order to determine the
important parameters in the simulation of smoke spread, and to give an idea of the values of these
parameters. The study consisted of trying different means of letting the smoke in as a conserved
scalar and comparing them. The different changes were carried out on the parameters that can
affect the results, like smoke velocity, surface temperature and area of the smoke generator. The
parametric study was performed by changing one parameter at a time from the “reference case”
i.e. the first setup of parameters.

4.1 Reference case


The smoke generator was placed in the middle of a closed room 4 m high, 10 m long and 6 m
wide. The simulation was done on a quart of this room with SOFIE.

A box modelled the smoke source, and the smoke rose through the square top surface (20 cm x 20
cm by default). The smoke source represented a pyrolysis or smouldering combustion as
explained preciously. An inflow modelled the source with a certain velocity (0.5m/s), a certain
temperature at the surface (900K) and passive scalars represented the smoke. To simulate the
airflow in SOFIE, the turbulent model “high-Re k-e” was chosen. There were no combustion, no
radiation, only heat transfer.

Only a quart of the room was simulated in SOFIE. Mirrors in SOFIE can simulate the symmetry
conditions in order to save computing time.

4.2 The different changes and effects


The smoke spread was simulated for a surface temperature of 600, 750, 900, and 1200 K. The
surface temperature (t_f) affected the smoke spread a lot. Indeed, the smoke spread was faster and
bigger when the temperature was low, but the temperature of the room was not affected.
Concerning the velocity, the values used were 0.3, 0.5 and 0.7 m/s. The smoke spread was
logically faster for a high velocity, and also, the temperature of the room increased with higher
velocity.

Then the value of the generator surface was studied. When the smoke was generated by a larger
surface, the temperature of the room increased a lot and the smoke spread is much faster. A
heating part around the source, modelling the top face of the heating cone used in the next
experiments called “ovan”, was tested. This element had almost no effects on the smoke spread,
except to make the smoke plume a bit wider.

4.3 Conclusion
So the important parameters to simulate a smoke spread in a room are the surface temperature,
the inlet velocity, and the area of the smoke source. The values of the temperature and the
velocity have to be known by doing experiments for example. The area is to some extent known
by the geometry of the source.
13

5 Experiments
The objective of the experiments was to measure the temperature and the velocity of the smoke
above the source. These measurements were carried out in order to determine the boundary
values to be used in the simulations.

5.1 Experimental protocol


The sample used for the experiments was a square polyethylene sample, 10 cm wide. A heating
cone formed by a resistance placed 2.5 cm above the sample heated it. The heating cone produced
an even thermal radiation on the sample surface. It was the same type of cone calorimeter used in
general in fire technology. The sample was placed, below the cone, on the top face of a metal box
(figure3) containing a thermal insulator and an aluminium sheet. So, heating the plastic sample,
some smoke was produced and rose through the hole of the cone.

Figure 3 Plastic sample in the metal box.

The temperature was measured at 6 different heights at the centre of the plume, by K
thermocouples. The velocity was measured at 3 different heights, using a Pitot tube.
A sampler collected all the data and transmitted them to the computer.

Six tests were carried out:

- Test 1: checking test


- Test 2: velocity measures at 66.5 cm
- Test 3: velocity measures at 66.5 cm
- Test 4: velocity measures at 46.5 cm
- Test 5: velocity measures at 26.5 cm
- Test 6: velocity measures at 26.5 cm

The first test was done to check if all the instruments worked. The plastic sample was put in place
below the cone after a one-minute pre-measuring time, and then the surface temperature
increased quickly. The temperature was measured at all the heights during all the tests. The
temperature was measured at 1.065m, 86.5cm, 66.5cm, 46.5cm, 26.5cm, and at the sample
surface.
14

Experimental device

Figure 4 Photo of the experimental device.

The photo of the experimental device is explained on the drawing below:

Figure 5 Experimental device.


15

5.2 Results

Figure 6 Smoke spread in the experiments.

The experimental measurements for the temperature and the velocity were studied to extract some
average values, which were used to compare with the simulations. Thus, getting the right
parameters, the simulations could fit the experiments.

5.2.1 Temperature measures


The temperature was measured at six heights during five tests. The ambient temperature was
about 17 ºC. The results of the test 6 are displayed below:
16

Test 6
surface
450

26.5 cm
400

350
46.5 cm

300 66.5 cm
temperature (C)

250 106.5 cm

200 86.5 cm

150

100

50

0
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020
Time (s)

Figure 7 Temperature profiles of the test 6.

The results obtained were similar for all the tests (the results of all the tests are presented in
appendix). The values extracted below, from measures, are averages of all the tests.

Surface temperature

When the plastic sample was put in place under the cone after one minute, the surface
temperature increased quickly. After four minutes, the temperature rose quite steadily during six
minutes, and then reached a maximum of about 440 ºC, when almost the entire sample was
consumed. However, it was very difficult to measure the surface temperature, so, these results
have to be used carefully.

At a height of 26.5 cm

All the temperatures measured by thermocouples above the source fluctuated a lot. At 26.5 cm,
on average, the temperature increased and reached, after 8 minutes, a level of 160ºC that was
maintained. But, this value has to be used carefully because the thermocouple was close to the
source and the cone could heat it.

For the others thermocouples, the temperature measurements had the same profile, they increased
and reached, after 8 or 9 minutes a level that was maintained. The average reached temperatures,
for each height, are summed up in the following table:
17

Height (cm) Temperature (ºC)


0 440

26.5 160

46.5 60

66.5 45

86.5 40

106.5 35

5.2.2 Velocity measures


The velocity was measured at 3 heights with a Pitot tube:
- At 66.5 cm: test 3
- At 46.5 cm: test 4
- At 26.5 cm: test 5 and 6

The Pitot tube measures a pressure difference ∆P. The Bernoulli theorem is applied on the Pitot
2 ⋅ ∆P
tube placed at the centre of the plume, so the velocity is: V = . The density ρ depends
ρ
P
on the temperature. The law of Mariotte can be applied to the smoke, ρ ⋅ T = , and as the
r
ρ 0 ⋅ T0
pressure was almost constant, ρ = (with T0 = 273.15K and ρ0 = 1.293 kg/m3). So the
T
2 ⋅ ∆P ⋅ T
velocity was calculated using V =
ρ 0 ⋅ T0
It was very difficult to measure velocity because the values fluctuated a lot, especially at low
velocities there were a lot of errors. The wanted value was the average of the reached velocity in
the permanent state.

At a height of 26.5 cm

The results of the test 6 with the velocity measurements at 26.5 cm are displayed below:
18

test 6
1.6

1.4

1.2

1
velocty (m/s)

0.8

0.6

0.4

0.2

0
0 60 120 180 240 300 360 420 480 540 600 660 720 780 840 900 960 1020
Time (s)

Figure 8 Velocity profiles of the test 6.

Two tests were carried out at this height. The velocity increased and reached on average 0.95 m/s
and stayed on this value.

At the two other heights, the velocity had the same profile. The average reached velocities, for
each height, are summed up in the following table:

Height (cm) Velocity (m/s)


26.5 0.95

46.5 0.90

66.5 0.80

The results of all the tests are joined in appendix.


19

6 Tuning in the smoke source


In order to find the best way to simulate the smoke source, a parametric study was performed.
The study was done modifying parameters one after one. The different changes were carried out
on the boundary values, the geometry, and the model. Each of them was discussed in section 6.3,
6.4 and 6.5 below.

6.1 Basic script


The simulation was done on a quart of the room at the beginning. Indeed, the symmetry
conditions can be used to build mirrors in SOFIE to save computing time. To simulate the airflow
in SOFIE, the turbulent model “high-Re k-e” was chosen. There was no combustion, only heat
transfer. An inflow modelled the source with a certain velocity, a certain temperature at the
surface and passive scalars represent the smoke (scalar1_f = 1). The ambient temperature used
was 290K like in the experiments.

6.2 Results
Many simulations were carried out for this parametric study. The main results are displayed
below to show the different changes. The different values were measured in the centre of the
plume. There are two graphs, for the temperature and the velocity, and the simulations are briefly
presented by the main parameters.

a - 12cmx12cm; 723K; 0.15m/s

440 b - 8cmx8cm; 723K; 0.15m/s

430 c - 8cmx8cm; 723K; 0.25m/s


420 d - 8cmx8cm; 723K; 0.25m/s in the whole room
410
e - 8cmx8cm; 823K; 0.25m/s
400
f - laminar model; 723K; 0.15m/s
390
g - 8cmx8cm; 773K; 0.25m/s in the whole room; prandtl
temperature (K)

380 number +10% for enthalpy


experiments
370
360
350
340
330
320
310
300
290
0 0.2 0.4 0.6 0.8 1 1.2 1.4 1.6
height (m)

Figure 9 Temperature profiles from the simulations.


20

a - 12cmx12cm; 723K; 0.15m/s

b - 8cmx8cm; 723K; 0.15m/s


1.5
c - 8cmx8cm; 723K; 0.25m/s
1.4
d - 8cmx8cm; 723K; 0.25m/s in the whole room
1.3
e - 8cmx8cm; 823K; 0.25m/s
1.2
f - laminar model; 723K; 0.15m/s
1.1
g - 8cmx8cm; 773K; 0.25m/s in the whole room; prandtl number
1 +10% for enthalpy
experiments
Velocity (m/s)

0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
height (m)

Figure 10 Temperature profile from the simulations.

The main changes, which were carried out to reach a better agreement, are summed up in the
following table. The different effects are quantified, a percentage is indicated to give an idea, but
this value depends on all the parameters.

Changes
A B C D E F G H I
Prandtl
Increasing Increasing Decreasing In the cµ=0.18 number
Tke=10 Laminar With
T_f V_f area whole cbuoy=0.8 + 10%
Ted=0.05 model ovan
(+25%) (+67%) (-55%) room 5 for
enthalpy

T + ++ - + -- - + +++ O
values (+4%) (+13%) (-2%) (+2%) (-6%) (-3%) (+3%) (+20%)

T
Effects

O O X O O X O XXX O
profile

V + ++ -- - -- -- O +++ +
values (+5%) (+29%) (-21%) (-4%) (-14%) (-27%) (+5%)

V O O X O XX XX O XX O
profile

It affected
V
only the
increased
Remarks centre of
on all the
the plume
height

Profile: O No significant effect Values: O No significant effect


X Small effect + (-) increased (decreased) a bit the values
XX remarkable effect ++ (--) increased (decreased) remarkably the values
XXX Strong effect +++ (---) increased (decreased) strongly the values
21

6.3 Boundary values


The first parameters which could be changed according to the experiment results, were the
boundary values. First, the inlet velocity and the surface temperature were studied. The surface
temperature measured in the experiments was about 450°C, but it was difficult to measure this
temperature with a good accuracy, so this result had to be used carefully (The pyrolysis
temperature of polyethylene, is reported in literature as 505 °C). A first observation was that
when increasing the surface temperature (t_f) in the simulation, the temperature and the velocity
of the plume increased a little. Two simulations were done for example with 723K and 823K (e
and c).

It was more difficult to choose the inlet velocity because it was not possible to measure it
experimentally. The maximum velocity observed was 0.95m/s. So the inflow velocity could be
chosen changing it until the predicted maximum velocity reached the expected value of 0.95m/s.
The smoke velocity increased obviously, increasing the inlet velocity (b and c). But, the
temperature increased too. Thus, the temperature and the velocity of the plume were linked, and
so these parameters had to be chosen together to fit the right values. But to have high
temperatures with a low velocity like in the experiments, other things had to be changed.

Other parameters could be studied more precisely like tke (turbulent kinetic energy) and the ted
(rate of dissipation per unit turbulent kinetic energy). For the source, the chosen values were tke =
10 and ted = 0.05. The temperature and the velocity of the plume obtained were compared in the
previous table and the results were interesting, the temperature and the velocity were enough low
compared with other values tried for tke and ted.

6.4 Geometry
Some things about the geometry could be changed to match better the experiments. Indeed, like
in the parametric study, the area of the surface of the smoke generator affected a lot the results. In
the experiments, the used sample had an area of 100 cm2, and the smoke raised through the hole
of the cone which had an area almost three time smaller.
Decreasing this area, the temperature in the plume decreased a bit and the velocity decreased a
lot. A quite good result was obtained with a square 8 cm wide (64 cm2).

A heating part around the source, modelling the top face of the heating cone in the experiment,
that was called “ovan” in the script, was tested in the simulations. But it did not affect so much
the results, so it was not chosen.

The simulations were carried out on a quart of the room and the different values were measured
in the middle of the plume, so on the mirror. But, there was a difference about these values when
the simulation was carried out on the whole room. This difference was due to an interpolation
error. So when the interest was the values in the middle, it was better to carry out the simulation
on the whole room.

The mesh of the room was modified to get a better agreement. Building a finer mesh, the results
of the simulation are more accurate and can match better the experiments. First of all, it was
important to avoid the errors which could be created by a bad mesh. So the grid was taken finer
above the smoke source on the width, but also on the height because the cells had to have, as
mush as it was possible, a square shape. The ratio between the height and the width has to be
22

inferior to 10. There must not be also a too big difference between the sizes of the cells, so there
can be some difficulties in big rooms.

Figure 11 The used vertical mesh Figure 12 The vertical mesh built by the
SOFIE’s grid generator

Figures 11 and 12 present the difference between the default mesh in the SOFIE and the mesh
chosen for the simulations above the smoke source.

6.5 Model
Finally, some parameters concerning the model could be modified to get a better accuracy.

To try to get high temperatures with low velocities in the plume, the laminar model was tested
instead of the turbulent one, without success. The results were very different; the velocity for
example increased on all the height.

A study on numerical simulation of thermal plumes done by Nam & Bill in 1992 has shown the
used model can be modified, changing constants (Cµ and C buoy) responsible for the turbulent
velocity in the energy equation, to get a better accuracy [9]. The advised values for these
constants, in this study, were here Cµ = 0.18 and C buoy = 0.85. Indeed, This change made the
velocity decrease a lot, and the temperature did not decrease so much.

A last parameter was modified to try to improve the accuracy of the simulation. Following a
peace of advise from Pr Rubini (author of SOFIE 3), the Prandtl number of the enthalpy was
increased 10%. The result was better because there were no effects on the velocity but the
temperature was a bit higher.
23

6.6 Conclusion
Several simulations were done to determine the right parameters which had to be used to model
the smoke source. This study was done comparing the results with the measurements of the
experiments. The different changes were carried out on the boundary values, the geometry, and
the model. The question of the mesh was also investigated to have more accurate results. This
study allowed getting a good agreement with the experiments.

To conclude, according to this study, the best parameters, to fit the experiments, could be chosen.

These parameters are displayed below:

- Surface temperature t_f = 773 K


- Inlet velocity V_f = 0.25 m/s
- Area of the source of 64 cm2
- Scalar1_f = 1
- tke = 10 and ted = 0.05
- Cµ = 0.18 and C buoy = 0.85
- The prandtl number of the enthalpy is increased 10%

These parameters were used for all the simulations to model the smoke source.
24

C - Smoke spread in a room with a temperature


gradient
Several simulations were carried out in order to simulate the smoke spread in a room with a
temperature gradient. Two different sizes of rooms were tested. For each room, two different
values of the temperature gradient were used: 0.5°C/m and 1°C/m.

7 The model in SOFIE


The problem was modelled in SOFIE. A lot of simulations were carried out to reach good results,
and the chosen model is displayed below.

7.1 Geometry
The first case was a small room 4 m high, 6 m wide, and 10 m long. The geometry of the room
was simple; there was nothing in the room, except the smoke source. There was only a small
opening of 10 cm on the bottom of two opposed walls, to avoid an overpressure. The smoke
source was modelled for the geometry by a simple square box 8 cm wide and 2 cm high, placed
in the centre of the room.

The smoke source

4m

6m
10 m

Figure 13 Picture of the small room in AC3D.

The computational domain had to be larger than the room in order to allow air inflow and outflow
to the room. The computational domain was closed by a fixed pressure boundary simulating
atmospheric conditions. The second case was a large room 10 m high, 15 m wide, 20m long. The
room configuration was the same as the small room, and the same smoke source was used.
25

The smoke source

10 m

15 m

20 m

Figure 14 Picture of the large room in AC3D.

The question of the mesh

SOFIE uses the finite volume method to solve the equations, so the computational domain was
divided in small cells. According to the simple geometry, the chosen mesh should be the same
everywhere. But as the smoke source had a small size compared with the room, the mesh had to
be finer to improve the accuracy of the results increasing the number of calculations above the
source where there were important physical phenomenon. Moreover, the number of cells had to
be limited to avoid a too long time of calculation.

So, for big rooms like the one, the cells had to be rather large in the room in general, but small
around the source. But this difference of size could be at the origin of problems. Indeed, there
could be some errors, like discontinuities. So to decrease the difference, there should be a
transition, a space with cells which had a medium size. It was difficult to find a mesh with which
the results were good. The shape of the smoke spread could be quite strange; it took square
shapes due to the mesh SOFIE´s Cartesian grid generator was used but a considerable amount of
time was spent on constructing the grid in order to achieve reasonable accuracy of the main flows
without too many cells.

Figure 15 shows the chosen mesh for the small room. The outlines of the walls and the
computational domain are displayed in grey. The mesh was finer in the middle of the room where
the smoke source was, and it was less fine outside, where nothing important happened.
26

Figure 15 The mesh in the small room.

The size of the different cells is exposed in the figure 16. The mesh is seen from a plan at a
constant height. The mesh is the same on all heights. The cells have a height of 11 cm. The
number of cells used in the computational domain for the small room is 145 728.

22.3 cm

3 cells 2.7 cm wide

Cell 9.6 cm wide

Figure 16 The horizontally mesh in the small room.


27

In the large room, the grid has the same structure; only the size of the cells changes. They have a
height of 20 cm in the room, and the other dimensions are shown in the following picture. The
number of cells used in this case is 342 240.

32.5 cm

3 cells 2.7 cm wide

Cell 12 cm wide

Figure 17 The horizontally mesh in the large room.

7.2 Models and constants


In this problem, only heat transfers were considered. The radiation phenomenon was not taken
into account because the smoke production occured without fire and flames. In addition no
combustion model was used, only heat transfer. Likewise, the heat losses trough the walls were
neglected because the room did not reach a high temperature compared with the ambient
temperature. This means that all the walls and the ceiling were adiabatic and declared as inactive.
The used turbulence model was the “high-Re k-e model” which is the most common. The
problem used the transient solution type and the buoyancy correction for the turbulence, and the
smoke was modelled by passive scalars. According to the preparatory study (“ tuning in the
smoke source”) which was done to improve the accuracy of the simulation, some constants
concerning the models had to be modified: Cµ = 0.18 and C buoy = 0.85 for the turbulence, and the
Prandtl number for the enthalpy had to be increased 10% (σ = 0.77).

7.3 Boundary
There were two boundaries to close the computational domain. The first boundary was the smoke
source which was simulated by a box with an inflow. Indeed, the smoke was simulated by an
inflow with a velocity of 0.25 m/s and a temperature of 773 K. So, the smoke source was defined
as a box which blew a hot smoke (simulated by passive scalars with a boundary values scalar1_f
= 1). The turbulence parameters were defined at the boundary with the following values: tke_f =
10 and ted_f = 0.05.
28

The outside boundary was on top of the computational domain. This boundary was modelled by a
surface declared as static pressure called static pressure boundary. Only one surface like this was
necessary. Static pressure boundaries provided an unspecified flow direction and could act as
either mass sources or sinks; they simulated ambient conditions.

The following picture shows the pressure boundary above the room:

Pressure boundary

Figure 18 The static pressure boundary in AC3D.

7.4 Interior values


The final solution of the governing transport equations only depends upon the boundary values
specified. However, the rate at which a solution may be obtained, or even if one can be obtained
at all, can be influenced by initial estimates for the value of solved variables within the interior of
the computational domain.

This SOFIE´s menu allows interior values for all solved and derived variables to be specified.
That is why in general a component of the velocity (here v) is estimated to make easier the
calculations. The value has no importance. So here, a vertical velocity v = 0.1 m/s was set for all
the cells in the fluid domain.

Likewise, the temperature gradient could be built in this menu. Indeed, a certain temperature
could be specified for each cell. So to define a temperature gradient, a certain temperature had to
be set for all the cells which were at the same height and so on for each level. In the case of the
small room 4 m high, there were 41 cells which had all the same height except the first four cells.
For the temperature gradient of 0.5°C/m and 1°C/m, the differences of temperature over the
height were 2°C and 4°C. A simulation was run for a problem of several minutes, with only the
temperature gradient, without smoke production, and the temperature gradient was maintained.
29

The used script in SOFIE, for a level, is displayed below:

The type of variable

The type of cell

The number of cells for each directions x, y, z

The temperature

Figure 19 Extract from a script file for SOFIE.

The picture 20 shows the temperature gradient of 0.5°C/m made with SOFIE in the small room.

Figure 20 Temperature gradient in the small room.

7.5 Solving
All the simulations were performed with a transient solution type. The solutions were exported
each 20 seconds, and 10 minutes of the problem was simulated. A timestep of 1 second was
chosen to have a good accuracy and a good convergence. The iterations for one timestep were
considered as converging when the mass residual was less than 0.1% or when 150 iterations were
completed. If the mass residual criterion was reached before 30 iterations the solver was forced to
continue until 30 iterations. The mass residual was defined as the average mass error over the
whole solution domain, normalised by the total mass flow into the domain. To get quicker
convergence, the relax values were taken for all the variables equal to 0.2.
30

8 Results
8.1 Temperature gradient in the small room
Two simulations were carried out in the small room with a temperature gradient of 0.5°C/m and
1°C/m. The results are described and analysed in this part. The two cases are treated in two
different paragraphs.

8.1.1 Temperature gradient of 0.5°C/m


There was a temperature gradient of 0.5°C/m in the room. The smoke spread from the source in
the centre of the room. Three different isosurfaces of smoke after 3 minutes (scalar1 = 0.002,
0.003, 0.004) are displayed in the figure 21. This picture shows the smoke spreading in a column
and reaching the ceiling.

Figure 21 Smoke spread

Figure 22 shows the smoke fields on a surface in the centre of the plume after 3 minutes; the
colour scale is displayed on the side.

Figure 22 Smoke fields in the small room after 3 minutes.


31

The smoke reached the ceiling and then followed the walls downwards. The smoke went down
until a certain height and then, by the airflow, filled the room area at that height. After 3 minutes
the smoke reached all the walls. Then the smoke layer depth increased. Figure 23 displays the
outlines of the walls and the boundaries of the computational domain, a vertical surface showing
the temperature gradient, and an isosurface of the smoke (scalar1 = 0.001) in grey after
10 minutes.

Figure 23 Smoke spread in the small room after 10 minutes.

The isosurface of smoke had a horizontal shape, and the temperature gradient was maintained
except in the plume. So the smoke spread with this temperature gradient in this room, was not so
different in the form than without temperature gradient.
32

Figure 24 Smoke fields after 10 minutes.

Figure 24 shows the smoke fields after 10 minutes. The smoke stayed mainly below the ceiling
near the walls.

8.1.2 Temperature gradient of 1°C/m


The simulation was carried out with a temperature gradient of 1°C/m. First the smoke spread like
with the other temperature gradient: it raised and reached the ceiling after 1 minute (figure 25).

Figure 25 Smoke field after 1 minute.

After 2 minutes (figure 26), the smoke stopped to spread horizontally and went down until a
certain height. Then the smoke spread just under the ceiling became slower and the smoke spread
at this height. Indeed, until it reached the ceiling, the smoke followed the airflow and the
temperature from the source. Then the smoke reached the high temperature of top of the room
due to the temperature gradient. The smoke temperature was less high, so the smoke went down
towards lower temperatures.
33

Figure 26 Smoke field after 2 minutes.

The smoke spread on this height during a certain time. Then, the smoke spread at the same time
under the ceiling and reached the walls (figure 30).

Figure 27 Smoke field after 10 minutes.

After 10 minutes (figure 27), the smoke reached all the walls and the entire ceiling. It did not stay
at a certain height and filled all the space up to the ceiling, like with the other temperature
gradient, but the smoke layer was less low.

Figure 28 Smoke spread in the small room after 3 minutes.


34

Figure 29 Smoke spread in the small room after 5 minutes.

Figure 30 Smoke spread in the small room after 10 minutes.

The three previous pictures show the smoke spread in the room during 10 minutes. The pictures
content a vertical surface showing the temperature, and an isosurface of smoke (scalar1 = 0.001).
After 3 minutes (figure 28), the isosurface reached the ceiling went down and spread at a height
of 3.15 m during 4 minutes. It began to fill all the space when it almost reached the farthest walls
after 5 minutes (figure 29). Then the smoke reached the walls and the smoke layer went down
regularly (figure 30). The smoke spread was slower than with the other temperature gradient. The
pictures show that the temperature field changed a bit in 10 minutes. Indeed, the smoke source
heated the room, and after several minutes the high temperatures on top of the room went down
with the smoke.

8.2 Temperature gradient in the large room


Two simulations were carried out in the large room with a temperature gradient of 0.5°C/m and
1°C/m. The results are described and analysed together in this part, because the two cases are
similar and it is easier to compare them.
35

Figure 31 Smoke spread with the temperature gradient of 1°C/m after 3 minutes.

There was a temperature gradient in the large room of 0.5°C/m or 1°C/m, so there was a
difference of 5°C or 10°C over the height. The smoke spread from the same source as in the small
room.

Figure 32 illustrates a surface with the temperature gradient and the same isosurface of smoke
(scalar1 = 0.001) as displayed for the small room, after 3 minutes. The smoke spread normally in
the plume. When it reached a certain height, the smoke started to spread downwards or stopped.
The smoke field had a shape of a mushroom; figure 32 shows clearly that the smoke did not reach
the ceiling after 3 minutes. As the smoke reached a too high ambient temperature, it stopped.

Figure 32 Smoke spread after 3 minutes with 0.5°C/ m Figure 33 Velocity pattern of the
plume.
36

The velocity pattern of the plume is displayed in figure 33 together with the smoke field. The
velocity vectors show that the smoke went a bit down after it reached the maximum point. This
type of picture gives an idea about the airflow and the circulation around the plume; the air
around the source followed the inflow of the source and went down on the side.

Figure 34 Smoke spread after 3 minutes with 1°C/m.

Figure 34 illustrates the isosurface in the same conditions as with 0.5°C/m after 3 minutes, so the
two smoke spreads are comparable. With a temperature gradient of 1°C/m, the isosurface had not
exactly the same shape, but it was the same phenomenon. After 3 minutes, the smoke did not
reach as high as in the 0.5° case, but the smoke spread more horizontally. The temperature
gradient is visible on the picture but the colour scale is not the same as in the 0.5° case.
37

6.35 m
4.75 m

Figure 35 Smoke spread with the temperature gradient of 0.5°C/m after 10 minutes.

Figure 35 still illustrates the same isosurface (scalar1 = 0.001). After 10 minutes, the maximum
height reached by the smoke did not changed compared with after 3 minutes: it was about 6.35 m.
The figure shows that the smoke stayed at a height of 4.75 m, after it reached the maximum point.
These values explain why there was not this phenomenon in the small room with the same
temperature gradient. Indeed, the height of the small room was 4 m, so the temperatures were not
too high on the top of the room and the smoke could reach the ceiling.

3.75 m 5m

Figure 36 Smoke spread with the temperature gradient of 1°C/m after 10 minutes.
38

With a temperature gradient of 1°C/m, the smoke reached a maximum height of 5 m; with 0.5°
the maximum height was 6.35 m. The smoke spread horizontally at an average height of 3.75m.
The smoke spread is bigger for the 1° case when looking at the isosurface (scalar1 = 0.001). The
heights marked in the figure explain why the phenomenon did not happen totally in the small
room, the smoke could find enough high temperature at 3.75 m, but the smoke reached the ceiling
because the height of the room was 4 m.

Figure 37 Smoke spread after 10 minutes. Figure 38 Smoke field after 10 minutes
For 0.5 °C/m For 0.5°C/m

Figure 39 Smoke spread after 10 minutes. Figure 40 Smoke field after 10 minutes.
For 1 °C/m For 1°C/m

Figure 37 and 38 show the isosurface of smoke with scalar1 = 0.0001. With this isosurface, the
smoke almost reached the walls. We can see that the smoke spread between two different heights,
in a specific range, until it reached the walls. Figures 39 and 40 illustrate the smoke field after 10
minutes. The figures clearly show that the smoke spread out at certain heights.

The smoke spread was smaller in height with a larger temperature gradient. It can be explained by
the fact that the temperature increased two times faster with the height, so the smoke stayed in a
smaller range. The high densities of smoke spread slower with 1°C/m compared with 0.5°. But
for the low densities the spread was similar for the two cases.
39

9 Conclusion
Different cases were simulated to show the effects of a temperature gradient on a smoke spread in
a room. Two different rooms were tested, a small room 4 m high and a large room 10 m high. The
chosen mesh for each room gave good results. Two values of the temperature gradient were
chosen: 0.5°C/m and 1°C/m. The smoke source was modelled in order to get the best results
matching the experiments. The simulations took several days, for the large room almost one
week.

The different simulations showed that the smoke spread could be affected by the existence of a
temperature gradient in the room. In rooms with high ceilings or large temperature gradient, the
temperature just under the ceiling was superior to the smoke temperature and therefore could the
smoke not reach the ceiling. Therefore, the smoke stayed at a certain height (4.75 m for a
temperature gradient of 0.5°C/m and 3.75 m for 1°C/m). The smoke spread had a mushroom
shape, and the smoke almost reached the walls in the large room after 10 minutes. This
phenomenon happened only in the large room where the room was height enough.

This phenomenon can prevent the detectors from detecting the smoke. In general, the smoke
detectors are placed in the ceiling in industrial buildings, so it can be difficult, when there is a
temperature gradient in the room, in the summer for example, to detect the smoke early. It is an
important problem because small smoke sources like smouldering combustion can be the origin
of a real fire and an early detection of this type of smoke production can avoid a disaster.
40

D - Smoke spread in a room with a ventilation system


The smoke spread was simulated in a room with a ventilation system. Two cases were modelled
in a big building ventilated by a system of small inflows. The objective of these simulations was
to see the effects of the ventilation system on the smoke spread, especially if the smoke did reach
the ceiling. In the first case, the smoke source was placed in the middle of four inflows. In the
second case, the source was placed just under a inflow.

10 The model
The simulations were simulated with SOFIE and the chosen model is explained below.

10.1 Geometry
The room, which was studied here, was a very big building 12 m high, 165 m wide and 200 m
long. This room had the same features as the room where the experiments will take place in the
future. The ventilation system of this room was composed of 68 vertical fans which were placed
at a height of 10.3 m with a flow of 178 l/s per inflow, and an outflow placed in one of the corner
7.5 m above the floor with a flow of 12 500 l/s. In the first case, the smoke source was placed in
the middle of the room in the middle of four inflows (figure 21). According to the huge size of
the building, only a quart of the room was simulated to save computing time. However, there
were no symmetry conditions because there was only one outflow in a corner but despite this fact,
one outflow was taken for each quart of room with a flow four times lower. Indeed the simulation
of one quart of the room was the only way to get a result without spending to much computing
time.

100 m
82.5 m

12 m

Figure 41 Geometry of the quart of the room used in the first case.

Figure 42 shows the distribution of the inflows in the room which was chosen to be simple and
well spread out in the room. So there were 17 inflows on the quart of the room.
41

16 m 8m
Smoke source

24 m

Inflow

22.5 m

Outflow

Figure 42 Distribution of the inflows.

Inflows have a circular area in reality, but square areas 22 cm wide were used in the simulation
since these are easier to construct using Cartesian coordinates. The outflow taken into account in
the quart of room had the same surface as the real outflow but it had a velocity four times lower.
The smoke source was exactly the same as used with the temperature gradient in the room.

22 cm
22 cm

2.3 m
2.7 m

5 cm

1.8 m

Figure 43 An inflow. Figure 44 The outflow in the corner.

In the second case, the smoke source was placed just under an inflow. In order to save computing
time, the chosen geometry was not the same. Indeed, the same distribution of inflow was taken
but with only four inflows in the centre of the room. So a whole room was taken into account but
with only 4 inflows and one outflow. The room 12 m high, 48 m long and 32 m wide could
simulate a similar case than with a smoke source under a inflow in a corner of the real room. The
42

outflow taken for this case had the same velocity as in the first case, but it had a smaller area, i.e.
the length was only 54 cm.
Outflow

32 m
Inflow

16 m 24 m 12 m

48 m

Smoke source

Figure 45 Geometry of the room in the second case.

The question of the mesh

The mesh building was a difficult task in these two cases. Indeed, the studied room had huge
dimensions compared with the smoke source size or the inflow size. So a compromise had to be
found between the big cells which had to structure the room and the small cells which had to
describe the small phenomenon near the inflow and the smoke source.

First, SOFIE´s Cartesian grid generator was used to build a first grid with a big number of cells.
The same method as presented in the previous chapter was used to make a transition between the
fine and coarse grid. A grid of 9 cells had to be used to define the surface of the smoke source
like with the temperature gradient, despite the fact that the cells had a very small width, because
with 4 cells the shape of the plume was too bad. A finer mesh was used at the height of the
outflow too.
43

Figure 46 The two different meshes used in the first case.

Two simulations were done in order to check if the results were independent to the grid. Figure
46 shows the two different meshes used in the first case in order to compare the results from two
different meshes, i.e. a coarse grid at the top of the figure and a finer grid below. The coarse mesh
used 352 408 cells to simulate a quart of this big room. The biggest cells, which were the more
numerous, were cubes 1.1 m wide. The finer mesh used 629 442 cells, so almost twice as many
cells, the biggest cells were cells 90 cm wide, 90 cm long and 66 cm high; this mesh was finer
than the other one near the smoke source. The simulations showed that the results from the two
meshes were similar, and so the doarse mesh brought good results compared with the computing
time for each simulation. However, the results presented in the next part come from the finer
mesh. However, if other simulations are carried out in the future, the coarse mesh can be used
bringing good results.

Figure 47 The coarse mesh on a quart of the room.


44

Figure 48 shows the mesh used in the second case. It is the same type of mesh as the coarse mesh
used in the case 1. About 170 000 cells were used.

Figure 48 The mesh chosen in the second case.

10.2 Models and constants


The same model was used for the two cases studied. The same model as used in the temperature
gradient case was used for both cases studied with the ventilation system. Only heat transfers
were considered, no radiation phenomenon and no combustion. Likewise, the heat losses trough
the walls were neglected so all the walls and the ceiling were adiabatic and declared as inactive.
The used turbulence model was the “high-Re k-e model” with the buoyancy correction. The
problem used the transient solution type and the smoke was modelled by passive scalars.

According to the preparatory study (“ tuning in the smoke source”) which was done to get a better
agreement with the experiments, some constants concerning the models had to be modified: Cµ =
0.18 and C buoy = 0.85 for the turbulence, and the Prandtl number for the enthalpy had to be
increased 10% (σ = 0.77).

10.3 Boundary
The boundary conditions for the two cases were almost the same, so they are treated together.
There were a lot of boundaries to close the computational domain. For each case, the computation
domain was extended in order to allow air inflow and outflow. A first boundary, like with the
temperature gradient was the smoke source which was simulated by a box with an inflow with a
velocity of 0.25 m/s and a temperature of 773 K. The turbulence parameters were defined at the
boundary with the following values: tke_f = 10 and ted_f = 0.05.
45

Figure 49 The static pressure boundary in the second case.

The outside boundary was on top of the computational domain. This boundary was modelled by a
surface declared as static pressure called static pressure boundary. Only one surface like this was
necessary. Static pressure boundaries provide an unspecified flow direction and can act as either
mass sources or sinks; they simulate ambient conditions.

The different element of the ventilation system are the others boundaries. In the first case where
one quart of the room was simulated, there were 17 inflows with a flow of 178 l/s, so they had a
velocity of –3.678 m/s for the vertical component. The inflow temperature was 15 °C. The
outflow was simulated in SOFIE as an extract boundary. Indeed, the air went out with a certain
velocity which was fixed at 0.755. The flow was four times lower than the real one (12500 l/s),
because this case simulated four outflows for the whole room.

The second case had the same boundary types. There were four inflows with still the same
velocity (-3.678 m/s). The outflow velocity was maintained as in the previous case but the area
was decreased to match the inflow. The temperature of the air inflow was 15 °C.

Figure 50 exposes the velocity pattern in the first case with the inflows in green and the outflow
in yellow.

Smoke source

Figure 50 Velocity pattern in the first case.


46

10.4 Solving
For the interior values, the same technique as in the previous chapter was used, i.e. all velocities
in the V direction were set to -0.1 m/s in order to improve the solution. All the simulations were
performed with a transient solution type. The solutions were exported each 20 seconds, and 15
minutes of the problem was simulated. A timestep of 1 second was chosen to have a good
accuracy and a good convergence. The iterations for one timestep were considered as converging
when the mass residual was less than 0.1% or when 150 iterations were completed. If the mass
residual criterion was reached before 30 iterations the solver was forced to continue until 30
iterations. These were the same conditions as with the temperature gradient. The relax values
were taken to 0.2 for all variables equal. In the first case, with the finer mesh, one iteration took
22 s, and with the other mesh, 12 s. For the second case, it was about 10 s.
47

11 Results
11.1 Smoke source placed in the middle of four inflows.
This paragraph presents the results for the first studied case where the smoke source was placed
in the middle of four inflows. These results come from the simulations done with the finer mesh
presented previously.

Figure 51 The smoke spread in the first case after 3 minutes.

Figure 51 exposes two different isosurfaces of smoke (scalar1 = 1.10 and 5.10 ), a surface with
-4 -4

the vertical velocity (v) fields for the inflows with the colour scale and with the u (perpendicular
to the outflow) velocity fields for the outflow. The velocity fields show the high velocity under
the inflows and the air which went out by the outflow, so the ventilation system worked. The
smoke spread normally and reached the ceiling after 3 minutes, at least for scalar1 = 1.10 .
-4

Indeed, the smoke source was simulated between four inflows which were quite far (8 and 12 m),
so the smoke was not hampered by the air flow.

The same isosurface of smoke (scalar1 = 1.10 ) is displayed in the figure 52 at two different
-4

moments. The grey surface shows the smoke after 10 minutes, and the blue one after 15 minutes.
For this value of scalar1 (1.10 ) the smoke stayed above the inflows. The blue surface (15
-4

minutes) was smaller than the grey the one. There was a phenomenon which happened after 10
minutes above the fans. The higher density of smoke spread above the inflows so they should not
be affected by the ventilation system. Nevertheless, after 15 minutes, the area reached by the
isosurfaces became smaller. The smoke was affected by the inflow because of the under pressure
created by the inflow.
48

Figure 52 Isosurfaces of smoke after 10 and 15 minutes.

Figure 53 shows an isosurface of (scalar1 = 1.10 ) smoke and a surface with smoke fields to see
-5

the different densities of smoke. For this value of scalar1, the smoke was blown by the nearest
inflows and so went down to the floor after 15 minutes. When the smoke reached the nearest
inflow, the smoke reached quickly the floor and spread on the floor. The smoke fields show that
the higher density of smoke stays above the inflows, just under the ceiling.

Figure 53 Smoke spread after 15 minutes in the first case.


49

The room studied was very big so the smoke did not reach all the inflows and the walls after 15
minutes. But the smoke reached a big part of the ceiling.

11.2 Smoke source placed just under an inflow


This paragraph presents the results for the second studied case where the smoke source was
placed just under an inflow. The simulation was carried out in a special “small room” with only
four inflows.

Figure 54 Small spread in the second case after 3 minutes.

Figure 54 shows two isosurfaces of smoke (scalar1 = 2.5.10 and 4.10 ) and the velocity fields
-4 -4

for two inflows (with the colour scale) and the outflow. The velocity fields illustrate that the
ventilation system worked. The inflow on the right was better described than the other in flow
because it used the same fine mesh as the smoke source. After 3 minutes, the smoke stayed on the
floor, blown by the inflow. More precisely, the smoke spread on the floor and began to rise on the
sides. The smoke began to rise on all the sides in the same manner. No direction was chosen.

After 15 minutes, the smoke managed to rise on the sides and reached the ceiling. Reaching the
ceiling, a big quantity of smoke was released.

An isosurface of smoke (scalar1 = 1.10 ) and a surface with the smoke fields are displayed in the
-4

figure 55. The smoke reached the ceiling rising in a column on a side. The smoke reached several
walls and a big part of the ceiling. The smoke fields show that the higher densities of smoke were
above the inflows, just under the ceiling. There was also some smoke on the floor under each fan.
50

Figure 55 Small spread in the second case after 15 minutes.

Figure 56 shows the velocity pattern and the smoke field above the smoke source, the surface was
not taken in the middle of the plume but one metre ahead of the plume to avoid the big velocity
vectors of the fan on the figure. Figure 57 illustrates the two same isosurfaces as figure 13 but
after 15 minutes. After 10 minutes, the smoke chose a direction to rise and after 15 minutes, the
smoke rose in a column to reach the ceiling. The velocity pattern shows that the air blew the
smoke on a side and the smoke reached the ceiling. This figure illustrates the velocity directions
but does not explain why this direction for the smoke was chosen. Anyway in reality, the
geometry is not perfectly symmetric so the smoke chooses a direction to rise.

Figure 56 Velocity pattern and smoke fields Figure 57 Smoke spread after
15 minutes
51

12 Conclusion
Several simulations were carried out in order to study the effects of a ventilation system on the
smoke spread in a big industrial building. The goal of these simulations was to preplan future
full-scale experiments that will be carried out to determine the problems with smoke detection.
The building taken into account was the real building where the experiments will occur. The
ventilation system considered in this study was composed of 64 inflows and an outflow. A certain
distribution was chosen. Different cases were studied, when the smoke source was in the middle
of four in flows or when the smoke source was just under an inflow. A very fine computing grid
was considered but the other studied mesh could be sufficient. It is still the same smoke source
which was used.

In the first case, when the smoke source was in the middle of four fans the smoke spread and
could reach the ceiling. It took a certain time for the smoke to reach the inflows because of the
big distance between them. Then some smoke was blown down by the ventilation inflow but a
big part of the smoke spread above the inflows. In the second case, the smoke source was under
an inflow. So the smoke source stayed on the floor for a while, but could manage to reach the
ceiling. Despite the fact the smoke was blown by the inflows, an important part of smoke could
spread above the fans. So in the two cases the smoke managed to reach the ceiling and could so
be detected by the detectors in the ceiling. In big buildings with this type of ventilation system,
the detection of the smoke could be slowed, but not prevented. So at least for these two cases, the
smoke could be detected but time to detector activation is increased.
52

13 Summary
In many cases, small smoke sources like smouldering combustion or pyrolysis are origins of real
fires. Many industries have a detecting system in order to detect the small fires before they
became large. The detectors are usually placed in the ceiling. The objective of this report was to
study some cases where there could be some problems detecting the smoke in rooms with high
ceilings in order to preplan future full-scale experiments. To investigate these issues, the way of
CFD calculations was chosen to be able to examine several cases. The simulations were carried
out using the CFD-code SOFIE.

The smoke source was a smouldering or pyrolysis source. Some experiments were done to
determine the boundary values which had to be set in SOFIE in order to model the smoke source.
According to the results of the experiments, some parameters were adjusted. The geometry,
different constants of the model, and boundary values were modified to fit the experiments.

Two different problems were studied. First, the effects of the existence of a temperature gradient
on the smoke spread in a room were simulated. Two different values of the temperature gradient
(0.5°C/m and 1°C/m.) were studied in two types of room. The simulations showed that in rooms
with high ceiling or large temperature gradient, the temperature under the ceiling is superior to
the smoke temperature; the smoke cannot reach the ceiling. Therefore, the smoke stays at a
certain height (4.75 m for a temperature gradient of 0,5°C/m and 3.75 m for 1°C/m). This
phenomenon can prevent the detectors placed in the ceiling from detecting the smoke.

The other problem was the effect of a ventilation system on the smoke spread in big industrial
buildings. The room simulated was the real building where the experiments will occur. Different
cases were studied, when the smoke source was in the middle of four ventilation inflows or when
the smoke source was just under an inflow. In the two cases, the smoke managed to reach the
ceiling and a big quantity of smoke could spread above the inflows. So the detectors in the ceiling
could detect the smoke. At least for these two cases, the smoke could be detected but time to
detector activation is increased.

The study has shown that there can be some problem to detect the smoke when there is a
temperature gradient in a room and that the ventilation system studied does not prevent the smoke
detection but time to detector activation could be increased. Other simulation will be carried out
in the larger project once the geometry of the room for the temperature gradient and the
distribution of the inflows for the experiments will be known. Then, some full-scale experiments
for the temperature gradient and the ventilation system will be done. The final goal will be to
develop rules and advice on detector placement to facilitate early smoke detection in rooms with
high ceilings.
53

14 References

[1] Petra Andersson and Haukur Ingason, “Smoke production and detection”,
Arbetsrapport 2001:35, SP Fire technology Borås 2001, Sweden.

[2] Petra Andersson, “Evaluation and mitigation of industrial fire hazards”, Lund
university Department of fire safety engineering, report LUTVDG/(TVBB-1015),
1997, Sweden.

[3] Dougal Drysdale, “An introductiom to Fire Dynamics”, WILEY, 1999.

[4] Web page of “Le grand dictionnaire terminologique”, www.granddictionnaire.com.

[5] Jörgen Carlsson, “Fire modeling using CFD”, Lund University, Department of fire
safety engineering, report 5025, 1999, Sweden.

[6] Jesper Axelsson, “Computational study of the flow pattern in reaction to fire test for
cables”, Master thesis 99/3, Chalmers University of technology, Department of
thermo and fluid dynamics, 1999, Sweden.

[7] Phil. Rubini, “SOFIE-User guide”, School of Mechanical Engeneering, Cranfield


University, England.

[8] H K Versteeg and W Malalasekera, “An introduction to computational fluid


dynamics”, Prentice Hall, 1995.

[9] Soonil Nam and Robert G. Bill, Jr, “Numerical Simulation of Thermal Plumes”, Fire
Safety journal 21 (1993) 231-256.
54

15 Acknowledgements
First of all, I would like to thank Petra Andersson who guided me during the placement, for her
patience and her kindness.

I also thank Jesper Axelsson and Heimo Tuovinen for helping me to a better understanding of the
computer code SOFIE. I thank everybody at Fire Technology for being so welcome and helpful,
with a special attention to Mr Ulf Wickström, Mrs Margaret Simonson, Mr Patrick Van Hees, Mr
Haukur Ingason, Mr Henry Persson and Mrs Erika Hjelm.
55

Appendix
A1 Results of the experiments

Surface Surface
Test 2 Test 4
26.5 cm 26.5 cm
46.5 cm 46.5 cm
500
66.5 cm 66.5 cm
500
86.5 cm 86.5 cm
400 106.5 cm
106.5 cm
400
velocity (m/s)

velocity (m/s)
300
300

200
200

100 100

Figure 58 - Temperature profiles of the test 2 Figure 59 - Temperature


0 profiles of the test 4 0
0 120 240 360 480 600 720 840 960 1080 1200 0 120 240 360 480 600 720 840 960 1080
time (s) time (s)

Test 3 Test5
Surface Surface
26.5 cm 26.5 cm
500 500
46.5 cm 46.5 cm
66.5 cm 66.5 cm
400 86.5 cm 400 86.5 cm
106.5 cm 106.5 cm
velocity (m/s)
velocity (m/s)

300 300

200 200

100 100

0 0
0 120 240 360 480 600 720 840 960 0 120 240 360 480 600 720 840 960 1080
time (s)
time (s)

Figure 59 Temperature profiles of the test 3. Figure 60 Temperature profiles of the test 5.
56

Test 3

1.4

1.2

1
velocity

0.8

0.6

0.4

0.2

0
0 120 240 360 480 600 720 840 960
time (s)

Figure 61 Velocity profiles of the test 3.

Test 4

1.6
1.4
1.2
velocity (m/s)

1
0.8
0.6
0.4
0.2
0
0 120 240 360 480 600 720 840 960 1080
time (s)

Figure 62 Velocity profiles of the test 4.

test 5

1.2

1
velocity (m/s)

0.8

0.6

0.4

0.2

0
0 120 240 360 480 600 720 840 960 1080
time (s)

Figure 63 Velocity profiles of the test 5.


57

A2 Command files used in SOFIE


Two examples of command files are displayed below

A2.1 Temperature gradient


%------------------------------------------------------------
% TEMPERATURE GRADIENT OF 0,5 IN THE SMALL ROOM
%------------------------------------------------------------
file
option
export path
fieldview
C:\sofie3\simulation\gradient0,5_small
end
end
end
setup
%------------------------------------------------------------
solution type
%------------------------------------------------------------
heat transfer
turbulence
high-Re k-e
end
end
buoyancy
transient
passive scalar
add scalar1 scalar1_f
end
end
%------------------------------------------------------------
physical geometry
%------------------------------------------------------------
import vrml1
%------------------------------------------------------------
rok2_8x8.wrl
geometry tolerance 0.002
end
%------------------------------------------------------------
computational grid
%------------------------------------------------------------
cartesian
key points
add
x edge_x 1
z edge_z 1
x edge2_x -1
z edge2_z -1
end
end
auto mesh1d
Key point tolerance = 0.002
Total number of cells = 10000
Minimum cells per region = 2
modify mesh
y
yh_wall3 36 1
end
x
xh_smoke 3 1
edge_x 10 1
xl_smoke 10 1
xl_wall1 18 1
edge2_x 18 1
end
z
zh_smoke 3 1
edge_z 10 1
zl_smoke 10 1
zh_wall4 9 1
edge2_z 9 1
end
end
58

generate mesh3d
end
end
%------------------------------------------------------------
assign blockages
%------------------------------------------------------------
geometry
inactive smoke
inactive wall1
inactive wall2
inactive wall3
inactive wall4
inactive ceiling
end
end
%------------------------------------------------------------
surface groups
geometry
create rok smoke_f:top
create korr_exit pbound
end
end
%------------------------------------------------------------
boundary types
%------------------------------------------------------------
fluid staticp korr_exit
fluid inflow rok
end
%------------------------------------------------------------
boundary values
%------------------------------------------------------------
korr_exit
t_f = 298.0
tke_f = 1.0 % Intensity
ted_f = 0.1 % Length scale
end
rok
t_f = 773.0
scalar1_f = 1.0
v_f = 0.25
tke_f = 10
ted_f = 0.05
end
end
%------------------------------------------------------------
interior values
%------------------------------------------------------------
expert
solution
v
fluid
all
ok
0.1
derived
% 1-4
t fluid range 7 67 1 4 1 44 ok 298
derived
% 5
t fluid range 7 67 5 5 1 44 ok 298.05
derived
% 6
t fluid range 7 67 6 6 1 44 ok 298.1
derived
% 7
t fluid range 7 67 7 7 1 44 ok 298.15
derived
% 8
t fluid range 7 67 8 8 1 44 ok 298.20
derived
% 9
t fluid range 7 67 9 9 1 44 ok 298.25
derived
% 10
t fluid range 7 67 10 10 1 44 ok 298.30
derived
% 11
t fluid range 7 67 11 11 1 44 ok 298.35
derived
% 12
t fluid range 7 67 12 12 1 44 ok 298.40
59

derived
% 13
t fluid range 7 67 13 13 1 44 ok 298.45
derived
% 14
t fluid range 7 67 14 14 1 44 ok 298.50
derived
% 15
t fluid range 7 67 15 15 1 44 ok 298.55
derived
% 16
t fluid range 7 67 16 16 1 44 ok 298.60
derived
% 17
t fluid range 7 67 17 17 1 44 ok 298.65
derived
% 18
t fluid range 7 67 18 18 1 44 ok 298.70
derived
% 19
t fluid range 7 67 19 19 1 44 ok 298.75
derived
% 20
t fluid range 7 67 20 20 1 44 ok 298.80
derived
% 21
t fluid range 7 67 21 21 1 44 ok 298.85
derived
% 22
t fluid range 7 67 22 22 1 44 ok 298.90
derived
% 23
t fluid range 7 67 23 23 1 44 ok 298.95
derived
% 24
t fluid range 7 67 24 24 1 44 ok 299
derived
% 25
t fluid range 7 67 25 25 1 44 ok 299.05
derived
% 26
t fluid range7 67 26 26 1 44 ok 299.1
derived
% 27
t fluid range 7 67 27 27 1 44 ok 299.15
derived
% 28
t fluid range 7 67 28 28 1 44 ok 299.20
derived
% 29
t fluid range 7 67 29 29 1 44 ok 299.25
derived
% 30
t fluid range 7 67 30 30 1 44 ok 299.30
derived
% 31
t fluid range 7 67 31 31 1 44 ok 299.35
derived
% 32
t fluid range 7 67 32 32 1 44 ok 299.40
derived
% 33
t fluid range 7 67 33 33 1 44 ok 299.45
derived
% 34
t fluid range 7 67 34 34 1 44 ok 299.50
derived
% 35
t fluid range 7 67 35 35 1 44 ok 299.55
derived
% 36
t fluid range 7 67 36 36 1 44 ok 299.60
derived
% 37
t fluid range 7 67 37 37 1 44 ok 299.65
derived
% 38
t fluid range 7 67 38 38 1 44 ok 299.70
derived
% 39
t fluid range 7 67 39 39 1 44 ok 299.75
60

derived
% 40
t fluid range 7 67 40 40 1 44 ok 299.80
derived
% 41
t fluid range 7 67 41 41 1 44 ok 299.85
end
end
control
physical models
time step 1.0
ambient temperature = 298
reference temperature = 298
end
end
%------------------------------------------------------------
end % End of setup
%------------------------------------------------------------
control
solved variables
rel
u = 0.2
v = 0.2
w = 0.2
ted = 0.2
tke = 0.2
end
sigma
enth 0.77
end
end
transient control
minimum timestep iterations 30
maximum timestep iterations 300
fieldview frequency 20
end
expert
model constants
turbulence
cmu 0.18
c_buoy 0.85
end
end
end
end
run 600
61

A2.1 Ventilation system


%------------------------------------------------------------
% VENTILATION SYSTEM WHEN THE SMOKE SOURCE WAS IN THE MIDDLE OF FOUR INFLOWS
%
%------------------------------------------------------------
file
option
export path
fieldview
C:\sofie3\simulation\ventil
end
end
end
setup
%------------------------------------------------------------
solution type
%------------------------------------------------------------
heat transfer
turbulence
high-Re k-e
end
end
buoyancy
transient
passive scalar
add scalar1 scalar1_f
end
end
%------------------------------------------------------------
physical geometry
%------------------------------------------------------------
import vrml1
%------------------------------------------------------------
ventilation.wrl
geometry tolerance 0.002
end
%------------------------------------------------------------
computational grid
%------------------------------------------------------------
cartesian
key points
add
x edge_x 3
x edge2_x 0.5
z edge2_z 0.5
z edge_z 3
y edge_y 1
end
end
auto mesh1d
Key point tolerance = 0.002
Total number of cells = 200000
Minimum cells per region = 2
modify mesh
y
yl_outflow1 10 1
yh_outflow1 4 1
edge_y 4 1
end
x
xh_smoke 2 1
edge_x 10 1
edge2_x 5 1
end
z
zh_smoke 2 1
edge_z 10 1
edge2_z 5 1
zh_outflow1 4 1
end
end
generate mesh3d
end
end
%------------------------------------------------------------
assign blockages
%------------------------------------------------------------
geometry
62

inactive smoke
inactive wall1
inactive wall2
inactive ceiling
inactive outflow1
inactive fan1
inactive fan2
inactive fan3
inactive fan4
inactive fan5
inactive fan6
inactive fan7
inactive fan8
inactive fan9
inactive fan10
inactive fan11
inactive fan12
inactive fan13
inactive fan14
inactive fan15
inactive fan16
inactive fan17
end
end
%------------------------------------------------------------
surface groups
geometry
create rok smoke_f:top
create korr_exit pbound
create c1_outflow outflow1_f:top
create turb fan1_f:down
append turb fan2_f:down
append turb fan3_f:down
append turb fan4_f:down
append turb fan5_f:down
append turb fan6_f:down
append turb fan7_f:down
append turb fan8_f:down
append turb fan9_f:down
append turb fan10_f:down
append turb fan11_f:down
append turb fan12_f:down
append turb fan13_f:down
append turb fan14_f:down
append turb fan15_f:down
append turb fan16_f:down
append turb fan17_f:down
create walls wall1_i
append walls wall2_i
create ceiling ceiling_i
create floor floor
end
end
%------------------------------------------------------------
boundary types
%------------------------------------------------------------
fluid staticp korr_exit
fluid inflow rok
fluid inflow turb
fluid extract c1_outflow
expert
fluid mirror west range f+1 f+1 f l f l ok
63

fluid mirror bottom range f l f l f+1 f+1 ok


end
end
%------------------------------------------------------------
boundary values
%------------------------------------------------------------
korr_exit
t_f = 288.0
tke_f = 1.0 % Intensity
ted_f = 0.1 % Length scale
end
rok
t_f = 773.0
scalar1_f = 1.0
v_f = 0.25
tke_f = 10
ted_f = 0.05
end
turb
v_f = -3.678
t_f = 288.0
tke_f = 1.0
ted_f = 0.1
end
c1_outflow
u_f = 0.755
end
end
%------------------------------------------------------------
interior values
%------------------------------------------------------------
expert
solution
v
fluid
all
ok
0.1
end
end
control
physical models
time step 1.0
ambient temperature = 288
reference temperature = 288
end
end
%------------------------------------------------------------
end % End of setup
%------------------------------------------------------------
control
solved variables
rel
u = 0.2
v = 0.2
w = 0.2
ted = 0.2
tke = 0.2
end
sigma
enth 0.77
end
end
transient control
minimum timestep iterations 30
maximum timestep iterations 300
fieldview frequency 20
end
expert
model constants
turbulence
cmu 0.18
c_buoy 0.85
end
end
end
end
run 600

You might also like