Cent. Eur. J. Eng.

• 2(2) • 2012 • 279-288

DOI: 10.2478/s13531-011-0064-z

Central European Journal of Engineering

CFD modeling of Dalmarnock uncontrolled fire test

Research article

Sergey Mijorski1 , Peter Stankov2

1 Department of Hydroaerodynamics and Hydraulic Machines, Technical University of Sofia,

Boulevard Kliment Ohridski 8, ZIP-code 1000, Sofia, Bulgaria
2 Centre for Research and Design in Human Comfort, Energy and Environment, Technical University of Sofia,
Boulevard Kliment Ohridski 8, ZIP-code 1000, Sofia, Bulgaria

Received 01 July 2011; accepted 19 December 2011

Abstract: The study describes a CFD based modeling of a fire in multi-storey apartments building. The model parameters
are taken from a fire experiment with uncontrolled ventilation conditions. The aim of the current work is to describe
some features and difficulties of fire CFD modeling in close compartments. The instrument used for CFD modeling
was Fire Dynamic Simulator. The analysis is made in terms of fire development in an apartment and influence of
wind velocity changes on temperature fields
Keywords: Computational fluid dynamics • Field model • Fire modeling • Fire safety • Fire dynamic simulator
© Versita sp. z o.o.

1. Introduction Computational Fluid Dynamics (CFD) based software ap-

Fire safety is one of the main characteristics of buildings,
because of confined compartment construction and fire
specifics. Fire regulations require safety measures for
preventing fires and reducing consequences caused by
them on buildings and occupant’s health. Each public
building and construction is subject to strict control and
high regulation requirements for safety.
Last decades with numerous incidents and civil victims
proved that the process of fire inside of confined compart-
ment needs more detailed investigation. A lot of tests and
experiments were performed aiming measuring of different
fire parameters and their influence on building construc-
tions, evacuation procedures and human health as shown
in Yeoh and Yuen (2009). Complex physical processes,
expensive equipment and preparing of real size models
significantly embarrass performing of experiments. Test-
ing of each possible scenario seems to be impossible and
extremely expensive.
plication comes in help of engineers. Conventional CFD
applications show very good results for modeling of differ-
ent turbulent flows, but for strongly non-isothermal flows
they need to be modified. National Institute of Standards
and Technology (NIST) developed Fire Dynamic Simulator
(FDS), which makes good representation of low-speed,
thermally-driven flow with an accent on smoke and heat
transport from fires, as mentioned in McDermott et al.
(2010).
The CFD models applied and modified for fire predictions
are called Field models, as mentioned in Yeoh and Yuen
(2009). There are many examples for wide application
of Field models. Some of the works based on numerical
predictions worth to be mentioned, such as the investigation
of compartment fires by Rein et al. (2007), Jahn et al. (2007),
Gutiérrez-Montes et al. (2009), Mijorski and Stankov (2011)
and tunnels fires modeling by Miles (2008) and Kit et al.
(2009). Also there are good examples for analysis of real
fire accidents, such as World Trade Center by McGratten

279
 CFD modeling of Dalmarnock uncontrolled fire test

Figure 1. Compartment geometry.

et al. (2005), Howard Street tunnel fire by McGrattan and

Table 1. Doors and windows conditions.
Hamins (2002), Caldecott tunnel fire by McGrattan (2005),
Mont Blanc tunnel by Kumar et al. (2004) and others. Boundaries Annotation Status
The present study focuses on modeling of a fire in multi- Window – Test Compartment W1 Closed
Window – Kitchen W2 Ajar
storey apartment building by applying FDS. Accent is
Window – Room 1 W3 Closed
given to some features and modeling difficulties of com- Window – Room 2 W4 Opened
partment fires. The influence of different wind velocity Balcony Door – Room 2 BD Opened
profiles on the fire development and progression in an Door – Kitchen –Test Comp. DR1 Removed
apartment is investigated. A grid independent solution Door – Inner Corridor – Test Comp. DR2 Opened
Door – Inner Corridor – Room 1 DR3 Opened
is performed in terms of determining the mesh influence
Door – Inner Corridor – Room 2 DR4 Opened
on numerical results. Data from a fire test were applied Door – Corridor – Outside Corridor DR5 Ajar
for the aim of the study. The experiment was conducted
in multi-storey building in Dalmarnock, Glasgow, Scot-
land by BRE Centre for Fire Safety Engineering at the tree. This gives a good picture for the heat processes
University of Edinburgh. inside the TC. With total number of 240 thermocouples the
presentation of temperature distribution is correct enough.
The experimental study shows three distinctive phases
2. Physical model during 1320 seconds of fire development in the uncontrolled
fire test, as shown in Empis et al. (2007):
Data from uncontrolled Dalmarnock fire test as shown in
Empis et al. (2007) were used in the present study. The • A fire growth period (fire development from the ig-
apartment was situated on the fourth floor in 23-storey nition to fully developed fire);
building. Drawing of compartment situation is illustrated
• A fire flashover period (the moment of fire emergence
on Fig. 1.
in whole confined compartment);
The test was conducted on June 25 2006, start time 12:23,
end time 12:45. The fire ignition was in the Test Com- • A fire post-flashover period (the phase after
partment (TC) by inflaming a waste bin full with paper. flashover, characterized with low oxygen concen-
The paper was poured with 500 ml heptanes fuel. The fire trations in confined compartment).
sources were concentrated in the TC room of the apartment
as shown in Fig 1. Doors and windows conditions during
the fire test are listed in Table 1. Full list of the fire
sources in the TC can be found in Empis et al. (2007). 3. Field model
The temperature development in the TC volume is of inter-
est for the current study. For temperature measurement The Field model software FDS, version 5.5.3 was applied
in the volume during the experiment were used 20 ther- for the study. FDS is among the best tools for analysing
mocouple trees with 12 thermocouples attached to each of fires and fire safety in confined compartments.

280
S. Mijorski, P. Stankov
Figure 2. Meshes geometry for grid independent study.
The FDS is based on two alternative methods for numerical
simulations: Smagorinsky form of Large Eddy Simulation
(LES) and Direct Numerical Simulation (DNS).
The FDS model is based on the system of coupled partial
differential equations describing conservation of mass, mo-
mentum, energy and scalar properties. The Navier-Stokes
equations appropriate for low-speed, thermally-driven flow
with an accent on smoke and heat transport from fires
are solved. The core algorithm is based on an explicit
predictor-corrector scheme, second order accurate in space
and time. The combustion model is based on single step,
two-steps or multiple-steps finite rate reaction. Radiation
heat transfer is included in the model via the solution of the
radiation transport equation for a gray gas, and in some
limited cases using a wide band model. Full model descrip-
tion and governing equations can be found in McGrattan
et al. (2010a).
The main assumptions in FDS Field model are: low Max
number, infinitely fast chemical reactions and materials
approximation.
The FDS model was validated and verified numerous times.
Several examples for verifications and validation are given
in McDermott et al. (2010) and McGrattan et al. (2010b).
For simplicity only a rectilinear mesh can be applied with
a possibility for including several meshes with different
cell size. Rectangular obstructions and objects are forced
to conform with the underlying mesh. In the current numer-
ical study LES method and single step chemical reaction
combustions model with oxygen depletion is applied. The
combustion model is with fully prescribed Heat Release
Rate (HRR) approach. Heptane was used as a base fuel
for combustion reaction.
4. Numerical predictions
4.1. Grid independent study
The grid independent study is an important part of numer-
ical predictions.
In the current grid independent study the domain was
simplified by eliminating two adjacent rooms R1 and R2
from experimental apartment and the Outside Corridor
(OC), shown on Fig. 1. The generated domain is shown on
Fig. 2. During the grid independent study four different
mesh resolutions were investigated. Details for each of
the investigated cases are given in Table 2.
Simplification of the computational task is needed for get-
ting faster numerical solutions for refined meshes. Except
reduction of the computational domain also material prop-
erties of all objects and walls were ignored and wall was
accepted by default Inert properties in FDS. This precon-
ditions faster solution for different grids resolutions. The
HRR applied for the study is the same prescribed as shown
on Fig. 7. The fire sources approximations are identical
to the simulation scenario in the study of wind influence.
Time duration for the simulations was reduced to 500 sec-
ond, which corresponds to 200 second after flashover period
of the fire experiment. Thus the grid independent study in-
cludes the flashover period, which represents the greatest
flow disturbances and temperature gradients.
Fig. 3 to 5 show numerical results for different resolutions.
They represent average parameters quantities for Mesh 1
shown on Fig. 2. Fig. 3 illustrates clearly that in fire grow-
ing phase there is no difference between grid resolutions.
However, the differences arise in flashover period. It can
281
 CFD modeling of Dalmarnock uncontrolled fire test

Table 2. Grid independent study – meshes parameters.

Grid parameters Mesh 0.2 Mesh 0.1 Mesh 0.05 Mesh 0.04
Grid size X [m] 0.2 0.1 0.05 0.04
Grid size Y [m] 0.2 0.1 0.05 0.04
Grid size Z [m] 0.2 0.1 0.05 0.04
Number of Meshes – 3 3 3 3
Mesh 1 – 5184 41472 331776 648000
Mesh 2 – 3420 27360 218880 427500
Mesh 3 – 900 7200 57600 112500
Total number of cells – 9504 76032 608256 1188000

be determined as grid independent solution. However the

solution with 0.05 m grid size only slightly differs from the
solution with 0.04 m.
Last figure from grid independent study shows the average
visibility in TC.

4.2. Wind influence study

Figure 3. Average gases temperature in Mesh 1 for different grid
resolutions.
Figure 4. Average viscosity in Mesh 1 for different grid resolutions.
be seen that the grid refining leads to lowering the average
gas temperature in the TC. For the temperature analysis
the grid independent solutions can be selected at 0.05 m
resolution, as it does not differ significantly from 0.04 m
resolution.
Fig. 4 shows average viscosity for the different resolutions.
As it can be seen there are significant differences between
0.2, 0.1 and 0.05, 0.04 grids. The grid size of 0.05 can
Figure 5. Average visibility in Mesh 1 for different grid resolutions.
In the current numerical study underlying mesh is with cell
sizes ∆X = 0.1 m, ∆Y = 0.1 m, ∆Z = 0.1 m and dimensions
of whole domain are: X = 11.2 m, Y = 9.6 m, Z = 3.0 m.
Total number of cells was 322560 divided into two meshes
(Mesh 1 = 144000 cells, Mesh 2 = 178560). The use of
grid independent mesh solution of 0.05 m would require
enormous CPU time for such an extended study.
Fig. 6 shows the numerical grid, thermocouples trees and
fire sources location.
For the aim of the study approximation of fire sources is
based on four different group types. The following graphs
were used for prescribed HRR of each one.
Fire suppression mechanism was employed for the combus-
tions: 0.15 Oxygen concentration and critical temperature
1427°C.
In order to assess more precisely the role of outdoor at-
mospheric conditions the real facade of the investigated
apartment was surrounded numerically with a layer of
1 meter. This allows the input of the wind to be precisely
assessed. The simulation is performed introducing data for
the atmospheric conditions (temperature, humidity, atmo-
spheric pressure), measured in weather station located in
Glasgow airport (EGPF). In previous numerical studies of
Jahn et al. (2007) the outdoor conditions were represented
just by ventilation inlet to the window openings. In the
current study different wind profiles are applied by the
boundary conditions on 1 meter outside layer. Thus the
flow can develop outside the building before entering in
the compartment through windows openings. The weather
station is located several miles away from the experimental
location. As shown in Table 3, the wind profile varies dur-
ing the fire. Taking into account the fluctuation of the wind
and far location of the weather station the extensive study

282
S. Mijorski, P. Stankov

Figure 6. Numerical grid, fire sources and thermocouple trees.

Table 4. Wind profile parameters.

Case Wind Speed Direction X component Y component

[m/s] [m/s] [m/s]
1 0 – 0 0
2 1 ENE 0.38 0.92
3 1 East 0 1
4 2 ENE 0.77 1.85
5 2 East 0 2
Figure 7. Fire sources fully prescribed HRR for CFD model. 6 3 ENE 1.15 2.77
7 3 East 0 3
8 4 ENE 1.53 3.7
9 4 East 0 4
Table 3. Weather station Glasgow (EGPF), records 12:20 PM and 10 4.6 ENE 1.7 4.25
12:50 PM on June 25 2006. 11 5.1 East 0 5.1
Time (BST) Temp. Wind Wind Humidity Sea Level
Dir. Speed Pressure
[°C] [m/s] [%] [hPa]
thermocouple exactly correspond to their real coordinates
12:20 PM 15.0 East 5.1 67 1019
in the experimental study as described in Empis et al.
12:50 PM 16.0 ENE 4.6 63 1019
(2007). Table 6 and 7 give the locations of thermocouples
according the coordinate system in the North-West corner
were preformed for the following atmospheric boundary of the apartment on Fig. 6.
conditions: This improvement allows better correspondence between
the experimental and numerical results to be achieved.
• Temperature – 15.0°C; According to fire report by Empis et al. (2007), at 1140 sec-
onds from the start of the experiment firemen interrupted
• Humidity – 67 %;
the process and starts to extinguish the fire. This was
• Atmospheric pressure – 1019 hPa; the main reason for determining simulation duration of
1150 seconds.
• Wind velocities as shown on Table 4.

Materials shown on Table 5 were used for all objects and

walls inside the apartment.
Usually the previous numerical studies were based on
the assumption for an average temperature in the test
compartment volume. In order to improve the simulations,
numerical thermocouples were introduced. The locations of
5. Wind influence – numerical re-
sults
Results from current numerical study are compared with ex-
perimental results from fire test in Dalmarnock, described

283
 CFD modeling of Dalmarnock uncontrolled fire test

Table 5. Thermal properties of materials.

Material Density Specific heat Conductivity Emissivity Absorption coefficient

[kg/m3 ] [kJ/(kg · K)] [W/(m · K)] [1/m]
Concrete 2280 1.04 1.8 0.9 5.00E+04
Steel 7850 0.46 45.8 0.95 5.00E+04
Yellow pine 640 2.85 0.14 0.9 5.00E+04
Foam 28 1.7 0.05 0.9 5.00E+04
Gypsum plaster 1440 0.84 0.48 0.9 5.00E+04
Gypsum 930 1.09 0.17 0.9 5.00E+04
PVC 1380 Temp. dependent Temp. dependent 0.95 5.00E+04
Tile material 313 0.75 6.11E-02 0.9 5.00E+04
Calcium silicate 720 Temp. dependent 0.12 0.83 5.00E+04

Table 6. Numerical thermocouples trees locations, X-axis and Y-axis coordinates.

ThTr 1 2 3 4 5 6 7 8 9 10
Y [mm] 4430 4450 4395 4815 3805 3860 3060 2790 3225 2470
X [mm] 640 1740 2485 3610 845 2580 475 1855 3450 640
ThTr 11 12 13 14 15 16 17 18 19 20
Y [mm] 2315 2505 1805 1965 2110 710 950 560 590 1000
X [mm] 1220 2770 650 1900 3470 985 1775 2285 3090 3450

by Empis et al. (2007). The influence of different wind

velocity profiles on the temperature field in the TC is
shown on Fig. 8. The main discrepancy between predicted
and measured temperature appear within interval 460 to
850 second. Basically it is due to the different conditions
in supplying air. After 460 seconds the lack of air penetra-
tion through connections of the windows panes and frame
contributes for lower temperature than the experiment.
As can be seen the effect of the atmospheric conditions
also has high influence on the numerical results. Also at Figure 8. Average hot gases temperatures of numerical simulation for
800 seconds the sudden breakage of window pane results different wind profiles and fire experiment in Dalmarnock.

in surge changes of the thermal conditions in the TC due

to intensive income of fresh air in the zone of fire. In the
experiment this is not sudden; in fact an outdoor air supply
is still possible before breakage. In the present predictions
numerical breakage is instant. However, there is very good
qualitative agreement between the numerical and experi-
mental curves. Evidently the boundary conditions based
on record 12:20 PM in the Table 3, gives the most correct
numerical prediction compared to experimental results.
Fig. 9 and 10 illustrate realistic development of the fire
starting from its ignition till firemen intervention.
Temperature fields
Temperature field of case 5.1 m/s East is shown on Fig. 9.
The effect of fire suppression mechanism can be seen clearly
after flashover at 300 seconds. The effect of North pane
brakeage of Window 1 at 800 seconds after start can be
also seen on Fig. 8 as sudden increase of the temperature.
Up to 300 seconds the fire grows. The picture at 400 sec-
onds illustrates post-flashover. After the moment of win-
dow’s pane breakage at 801 seconds a fire grow appears
again and it continues till 900 seconds. The final period
shown on Fig. 8 after 900 second, describes the correla-
tion of continuous decay of HRR and decay of average
sensors temperature till South pane breakage of Window 1
at 1111 seconds. Evidently there is a correlation between
the temperature and the HRR as show on Fig. 8.
Visibility fields
Figure 10 illustrates the visibility during the fire develop-
ment. As it can be seen the visibility drastically decreases
at 270 seconds and totally disappears at 300. The visibility
occurs again after South pane breakage of Window 1 at
1111 seconds.
Velocity fields
Figures 11 and 12 show velocity field in two time instance
of the fire. The section 3.3 X and 4.6 Y are presented.

284
S. Mijorski, P. Stankov

Table 7. Numerical thermocouples Z-axis coordinates.

Thermocouple I II III IV V VI VII VIII IX X XI XII

Z [mm] 2500 2450 2400 2300 2200 2100 1900 1700 1500 1200 900 500

Figure 9. Temperature field on section 0.8 X – 9, 270, 300, 400, 800 and 900 seconds.

Evidently the numerical solution based on 0.05 m resolution

Table 8. Grid independent solution – meshes parameters.
is good prediction compared to test results. This is true
Grid parameters Mesh 1 Mesh 2 Mesh 3 Mesh 4 till post-flashover period, as shown on Fig. 13. Again the
Grid size X [m] 0.05 0.05 0.05 0.05 conclusions from the previous studies were proven. The
Grid size Y [m] 0.05 0.05 0.05 0.05 FDS simulator underpredicts the temperature fields in
Grid size Z [m] 0.05 0.05 0.05 0.05 the zone of fire during post-flashover. Better description
Total number of cells 844800 636480 230400 599040 of burning rate and boundary condition will improve the
results after flashover.

5.1. Grid independent solution

The grid independent solution is based on resolution of
0.05 m. The total number of cells in the domain is 2310720.
Table 8 gives details for the computational domain.
The division of the domain gives the possibility for parallel
processing of the numerical solution, which will reduce the
computational time. The fire sources approximation is the
same in wind influence study. Only some modifications are
done on the time functions of burning rate (HRR diagram,
Fig. 7).
6. Discussion of the numerical re-
sults
The simulation results show good agreement with the ex-
perimental results in the period before fire flashover and
after breakage of Window 1 north pane at 801 seconds. The
main difference between fully prescribed HRR approach
and fire comes after flashover period at 460 seconds from
the beginning of fire, when oxygen concentration was dras-
tically reduced and heat release was limited by the fire

285
 CFD modeling of Dalmarnock uncontrolled fire test

Figure 10. Visibility field on section 3.3 X - 9, 100, 200, 270, 300 and 1140 seconds.

Figure 11. Velocity vectors fields on section 3.3 X – a) 300 and b) 1140 seconds.

286
S. Mijorski, P. Stankov

Figure 12. Velocity vectors fields on section 4.6 Y – a) 300 and b) 1140 seconds.

expensive and very complicated. Thus CFD becomes very

promising alternative for investigating fire development. It
should be mentioned that, when simulating fire develop-
ment the exact determination of burning rate and HRR
are of significant importance for numerical results and for
the selection of appropriate combustion model. Materials
properties and boundary conditions are also of significant
importance for the accuracy of the numerical predictions.
Figure 13. Average hot gases temperatures of numerical simulation
for 5.1 m/s wind velocity, direction East and data from fire
experiment in Dalmarnock.
References
suppression mechanism. Also there is an influence of ven-
tilation through apartment openings, which lowers the [1] Empis C.A., Cowlard A., Welch S., Torero J.L., Test
TC average temperatures. The change of the conditions one: the ‘uncontrolled’ fire, the dalmarnock fire tests:
during flashover, changes the fire inside of TC from “fuel experiments and modelling, University of Edinburgh,
controlled” to “ventilation controlled”. The fire conditions 2007
after breakage of north pane of Window 1 at 801 seconds [2] Gutiérrez-Montes C., Sanmiguel-Rojas E., Viedma A.,
reverse the fire from “ventilation controlled” to “fuel con- Rein G., Experimental data and numerical modelling of
trolled”. The FDS numerical predictions show a significant 1.3 and 2.3 MW fires in a 20 m cubic atrium, Building
underprediction of average TC temperatures during post- and Environment 44, 2009, pp. 1827–1839
flashover. By taking a required attention on the model [3] Jahn W., Rein G., Torero J.L., A posteriori modelling of
set up during post-flashover period and applying correct fire test one, the dalmarnock fire tests: experiments
boundary conditions there is a possibility for improving and modelling, published by the School of Engineering
the numerical results. and Electronics, University of Edinburgh, ISBN 978-
0-9557497-0-4, 2007, pp. 193–210
[4] Kit C.M., Spearpoint M., Fleischmann C., Assessment
7. Conclusions of vehicle fire development in road tunnel for smoke
control ventilation design, University of Canterbury,
The study performed confirms the advantages offered by the New Zealand, 2009
CFD based simulation of fire development. The physical [5] Kumar S., Brandt R., Riess I., Recent achievements
nature of the fire makes its experimental study extremely in modelling the transport of smoke and toxic gases

287

in tunnel fires, safe & reliable tunnels. Innovative Eu-
ropean Achievements, First International Symposium,
Prague 2004
[6] McDermott R., McGrattan K., Hostikka S., Floyd J.,
Volume 2: Verification, fire dynamics simulator (ver-
sion 5), Technical reference guide, NIST, 2010
[7] McGrattan K.B., Hamins A., Numerical simulation of
the howard street tunnel fire, NISTIR 6902, National
Institute of Standards and Technology, 2002
[8] McGrattan K.B., Numerical simulation of the caldecott
tunnel fire, Fire Research Division Building and Fire
Research Laboratory, NISTIR 7231, National Institute
of Standards and Technology, 2005
[9] McGrattan K.B., Bouldin C., Forney G.P., Federal
building and fire safety investigation of the World
Trade Center disaster: computer simulation of the
fires in the World Trade Center, National Institute
of Standards and Technology (US), NCSTAR 1–5F,
Gaithersburg, MD, 2005
[10] McGrattan K., Hostikka S., Floyd J., Baum H., et al.,
In Volume 1: Mathematical model, fire dynamics
simulator (version 5), Technical reference guide, NIST
Special Publication 1018-5, 2010
288
CFD modeling of Dalmarnock uncontrolled fire test
[11] McGrattan K., Hostikka S., Floyd J., McDermott R.,
In volume 3: Validation, fire dynamics simulator (ver-
sion 5), Technical reference guide, NIST Special Pub-
lication 1018-5, 2010
[12] Mijorski S., Stankov P., Field model in fire safety
engineering, Proceeding of VIII International Course
“Computational Engineering”, Kopaonik, Serbia, 2011
[13] Miles S., CFD modeling of tunnel fires, Work package
2, fire development and mitigation measures D214,
BRE 225475, 2008
[14] Rein G., Torero J.L., Jahn W., Stern-Gottfried J., et al.,
A priori modelling of fire test one, the Dalmarnock fire
tests: experiments and modelling, published by the
School of Engineering and Electronics, University of
Edinburgh, ISBN 978-0-9557497-0-4, 2007, pp. 173–
192
[15] Yeoh G.H., Yuen R.K.K., On computational fluid dy-
namics in fire engineering – theory, modeling and
practice, ISBN 978-0-7506-8589-4, Oxford, UK, 2009,
pp. 771–785