You are on page 1of 24

International Journal on Engineering Performance-Based Fire Codes, Volume 9, Number 1, p.

7-30, 2007

FURTHER VALIDATION OF FIRE DYNAMICS SIMULATOR USING


SMOKE MANAGEMENT STUDIES

P. Coyle and V. Novozhilov


The Institute for Fire Safety Engineering Research and Technology, Faculty of Engineering
University of Ulster, United Kingdom

(Received 7 July 2006; Accepted 13 December 2006)

ABSTRACT

Further validation of Fire Dynamics Simulator (FDS) developed by NIST (USA) is performed using four smoke
filling scenarios reported in the literature. Careful comparison is made to experimental data available for those
scenarios.

Performance of the code was found to vary considerably with complexity of scenario (e.g. geometry). While
giving very reasonable results for a number of cases, average deviation from experimental values in smoke-
filling rates and temperature predictions were above limits claimed by developers.

The study emphasizes need for further development and extensive validation of CFD codes used by fire
engineering practitioners.

1. INTRODUCTION The objective of the present study is to validate the


FDS code further using a number of well-
Fire Dynamics Simulator is an increasingly popular documented scenarios for comparison.
CFD model choice for fire engineers and academia
researchers [e.g. 1,2]. The code has been developed Fire Dynamics Simulator (FDS) Version 4.0.5 [4]
at the National Institute of Standards and is used for the study.
Technology, USA.

The underlying motivation for the present study is 2. RESULTS AND DISCUSSION
assessment of accuracy of currently available
computer models for the purpose of their This section is segmented into four sub-sections,
integration into Engineering Performance-Based namely scenarios A, B, C and D, which correspond
Fire Codes (EPBFC). The importance of ongoing respectively to each selected physical experiment
validation and standardization work in fire-related under investigation.
CFD is well recognised [3].
For comparison with the experiments, some
The most interesting property of the FDS code is technical features of FDS 4.0.5 should be noted.
that it uses Large Eddy Simulation (LES) approach,
as opposed to many other (primarily commercial) The first one refers to smoke layer thickness
codes used by fire safety consultants. The latter use calculation in FDS. Relatively simple zone models
Reynolds-averaged (RANS) governing equations. compute this quantity directly, along with the
Use of LES is still a hot topic in CFD community, average temperature of the upper and lower layers.
and many important fundamental issues related to In a computational fluid dynamics (CFD) model
this approach are not completely resolved. As like FDS, there are generally no distinct zones, but
examples, one could point out to importance of rather a continuous profile of temperature.
different filtering procedures, influence of mesh Nevertheless, the methods can be developed to
refinement on solution, etc. Since this innovative estimate layer height and average temperatures
technology is penetrating into practical design, it is from a continuous vertical profile of temperature.
important that practitioners in the area are kept
informed of possible limitations and shortcomings The method employed by FDS is as follows.
of the methods. It should be kept in mind that many Consider a continuous function T(z) defining
of those using modern CFD software may not have temperature T as a function of height above the
sufficient background in CFD or combustion floor z, where z = 0 is the floor and z = H is the
fundamentals, therefore they need be warned of ceiling. Define Tu as the upper layer temperature, Tl
potential misuse of computational tools. as the lower layer temperature, and zint as the

7
International Journal on Engineering Performance-Based Fire Codes

interface height. Conservation of energy dictates strategy the smoke layer was just let descend to the
that floor without any additional smoke venting means.
This is considered as a starting point for the
investigation, since if the smoke filling rate cannot
be accurately predicted here, then there would be
no justification for analysing more complex
and conservation of mass (assuming an ideal gas) strategies.
dictates that
Hagglund et al. [5] conducted several experiments
in a room having a floor area of 5.62 m x 5.62 m
with a height of 6.15 m (Fig. 1). The room was
closed except for a 0.25 m high by 0.35 m wide
leakage opening near the floor. The fire source was
Solving for zint gives: placed 0.2 m above the floor and incorporated a
kerosene pan of 0.5 x 0.5 m, with a total steady
state heat release rate of 186 kW. The experimental
boundaries, i.e. the walls, floor and the ceiling of
the experiment were made up of concrete.

Letting Tl be the temperature in the lowest grid cell Karlsson and Quintiree [6] used this experiment to
and using Simpson’s Rule to perform the numerical validate a simple mathematical expression,
integration of I1 and I2, Tu can be defined as the developed by Zukoski and known as the Zukoski
average upper layer temperature via Smoke Filling Model. The results of this model
show an overestimation of the smoke-filling rate.
This was due to the delay time in the experiment
for the HRR to reach steady state, in addition to the
time taken for the plume to reach the ceiling.
Another note is on boundary conditions in FDS, Another important factor was the ignorance of heat
relevant for the cases considered in the present losses to the boundaries of the experiment.
paper. These are imposed in the following manner
[4]. FDS should not have these aforementioned
shortcomings as this model tracks flows and heat
By default, a partial-slip condition is employed at losses by conduction, convection, and radiation to
solid walls. This means that the velocity at the wall all surrounding boundaries over time.
is a fraction of its value in the grid cell adjacent to
the wall. This is a very crude analogue of a wall The input/output parameters for Scenario A are
function, which is required if boundary layer summarised in Table 1.
cannot be resolved. For the present study, the
details of boundary layers are not important, so that Fig. 2 depicts the averaged layer height predicted
the above solid boundary condition is deemed to be by FDS for nine various runs, which were subject
satisfactory. to change in order to highlight sensitive input
parameters. Overall, good agreement was observed
Fire with the fixed HRR (Heat Release Rate) is also between the FDS V 4.0.5 predictions and the
treated as a boundary condition, effectively physical experiment conducted by Hagglund.
modelling combustion as injection of pyrolysed
fuel through the solid surface with subsequent It can be seen from Fig. 2 that the smoke took
burning according to fast chemistry combustion approximately 4 s to reach the ceiling before
model [4]. starting its rapid descent. This is demonstrated by
the straight line at 6.15 m height at the top left of
At external boundaries (vents) a pressure-like the chart.
condition is prescribed either for natural ventilation,
or a forced flow extraction [4]. In the latter case, For the sensitivity analysis of this scenario, a
the volumetric extraction rate is prescribed. smaller HRR of 168 kW was intentionally chosen
during selected runs, primarily to investigate how
2.1 Scenario A − Smoke Filling in a Room sensitive FDS was to this HRR fluctuation.
Environment

In this first scenario the smoke-filling rate was


analysed within a single room environment. In this

8
International Journal on Engineering Performance-Based Fire Codes

CONRETE BOUINDARY WALL

KEROSENE FIRE SOURCE


186 (kW)

2
AREA 31.4 M

INLET VENT

Fig. 1: Plan of experimental set-up for Scenario A

SMOKE FILLING IN ROOM ENVIRONMENT

6.50
6.00
ROOM BOUNDARY
5.50
5.00
GRID
4.50
HEIGHT DEVIATION 16%
4.00
HEIGHT (m)

3.50
3.00
TIME 55s, HEIGHT
2.50
FIRE
FIRE
2.00
1.50 INLET
LAYER HEIGHTDEVIATION
LAYER HEIGHT DEVIATION
1.00
0.50
0.00
0.00 25.00 50.00 75.00 100.00 125.00 150.00 175.00 200.00 225.00 250.00 275.00
TIME (s)

FDS LAYER HEIGHT RUN 1 (200X200X205)(168kW) EXPERIMENT BY HAGGLUND et al. (1985)


FDS LAYER HEIGHT RUN 2 (140X140X140)(168kW) FDS LAYER HEIGHT RUN 4 (140X140X140)(186kW)
FDS LAYER HEIGHT RUN 5 (100X100X100)(168kW) FDS LAYER HEIGHT RUN 6 (100X100X100)(186kW)
FDS RUN 7 LAYER HEIGHT (100X100X100)(186)(Soot Yield 2%) FDS RUN 8 LAYER HEIGHT (50X50X50)(186kW)
FDS RUN 3 (140x140x140) DT 0.005 FDS LAYER HEIGHT RUN 3A (100X100X100)(DT 0.01)

Fig. 2: Smoke layer time history plot. Scenario A

9
Table 1: Summary of input / output parameters. Scenario A

10
Scenario A Smoke filling in room environment

Input Output

Fire
Grid cell size Fire area HRRPUA Initial Soot yield Smagorinsky Simulation Q* [-] Fire resolve. Converge
Run No. Total cells Fire source output
(XYZ) (mm) (m2) (kW/m2) time step. (%) constant time (s) (0.1<Q*<2.5) (Dx) / (∂x) speed [s-1]
(kW)

Run 1 200x200x200 23520 0.25 168 672 0.128 0.042 0.2 250 0.65 1.45 1.64

Run 2 140x140x140 72000 0.25 168 672 0.09 0.042 0.2 250 0.65 2.04 0.39
International Journal on Engineering Performance-Based Fire Codes

Run 3 140X140X140 72000 0.25 186 744 0.005 0.042 0.2 150 0.72 2.18 0.6
Kerosene C_14 H_30,
Run 3A 100x100x100 174960 MW_FUEL=198.0 0.25 186 744 0.01 0.042 0.2 150 0.72 3.05 0.25
NU_O2=21.5
NU_CO2=14.0
Run 4 140x140x140 72000 NU_H2O=15.0 0.25 186 744 0.09 0.042 0.2 250 0.72 2.18 0.39
EPUMO2=12700.
CO_YIELD=0.012
Run 5 100x100x100 174960 SOOT_YIELD=0.042 0.25 168 672 0.064 0.042 0.2 250 0.65 2.85 0.24
SFPE Handbook
Run 6 100x100x100 174960 0.25 186 744 0.064 0.042 0.2 250 0.72 3.05 0.14

Run 7 100x100x100 174960 0.25 186 744 0.064 0.02 0.2 250 0.72 3.05 0.14

Run 8 50x50x50 1458000 0.25 186 744 0.032 0.042 0.2 200 0.72 6.12 0.15
International Journal on Engineering Performance-Based Fire Codes

For the run (1), i.e. the first simulation of this (7). This run also behaved slightly better than all
scenario, it can be clearly seen that after other runs in predicting long-term (asymptotic)
approximately eight seconds FDS began to behavior of the layer. Other runs tended not to go
dramatically over predict the smoke filling rate below 0.5 m in all simulations.
within the enclosure. This behaviour was not
expected while applying a lower HRR than that in Run (3) was the penultimate run of scenario A.
the experiment. Twelve seconds later FDS This run reverted back to a 0.14 m grid cell size,
predicted a layer interface height of 2.75 m, while similar to run (4) as it showed good promise in
the physical experiment gave a height of 4.6 m, a predicting the physical experiment. The time step
difference of 1.85 m. Thereafter, the models was reduced to 0.005 s in this simulation in a bid to
smoke filling predictions began to slow down, and highlight its effect compared to the grid size effect.
started gradually to move closer to the The reduced time step of 0.005 produced the trend
experimental result. In the physical experiment this similar to the run (4).
slowing up of the layer interface was attributed to
the fact that the rate of air entrainment from the The final simulation within this scenario, the run
lower layer decreases as the smoke layer descends (3A) simulated a 0.1 m grid cell size and increased
towards the floor. the time step to 0.01 s. It can be seen from the layer
height chart, Fig. 2, that the run (3A) was
For the run (2), the only change was in the grid cell comparable to the runs (6) and (7). This indicates
size from 0.2 m to 0.14 m. The smoke filling rate little effect of changing time in this scenario.
reflected this change remarkably, indicating that
grid size is a sensitive input parameter. In summary, the results demonstrated that the
model does not necessarily give better predictions
Run (3) was the final simulation (rerun), and will with decrease in the grid cell size. This can be seen
be discussed later. Run (4) incorporated the correct upon comparison of the Run (4) (grid cell size of
HRR of 186 kW and a grid cell size of 0.14 m. This 140 mm) with the runs (6) and (7) (finer grid cell
run gave the same trend as the run (2), but the sizes of 100 mm). The latter two runs do not really
filling rate was a little higher initially due to the give any better predictions in comparison with the
higher HRR. run (4).

Run (5) reverted back to a HRR of 168 kW, but For the majority of the runs, FDS V 4.0.5 has given
decreased the grid cell size further, from 0.14 m to smoke filling rate predictions within 20% of the
0.10 m. This was carried out in a bid to improve the experimental curve. This is in line with the claim of
filling rate from the ceiling to a height of 2.5 m. its developers.
However, this run gave a closer prediction from the
ceiling to approximately 3.0 m, but resulted in a Temperature comparisons were not attempted in
larger error thereafter. Run (6) went back to the the scenario A.
correct HRR of 186 kW and applied the same grid
cell size of 0.1 m. The result gave a slight 2.2 Scenario B − Smoke Filling in a Hanger
improvement, but still held an overall height Environment
deviation of 16% from 5.50 m to 2.75 m. Run (7)
confirmed that soot yield did not play any role in In this scenario the smoke filling rate, the effects of
layer height prediction, when it was halved from natural vents and the performance of mechanical
4% to 2%. It confirms that FDS uses average extraction were analysed. Having evidence from
temperatures from a continuous vertical the first scenario that FDS can predict smoke filling
temperature profile to predict the layer height. in a single room enclosure quite well, our initial
task in this scenario is to see how well the model
Run (8), incorporated a grid cell size of 0.05 m. It can predict smoke filling when applied to a larger
took a considerable amount of computational time, enclosure. Then FDS is given a task of simulating
approximately 72 hours. It was a disappointing natural smoke venting, followed by mechanical
result, as it goes against the models main extraction at various volume fluxes [m3s-1] within
underlining assumption, which is, “the accuracy of the same enclosure.
the results is a function of the fidelity of the
numerical solution, which is mainly dependent on Yamana and Tanaka [7] performed a series of full-
the size of the computational grid”, as stated in the scale experiments using the BRI fire test facility to
FDS technical reference guide [4]. This run from investigate the smoke filling behaviours in large-
the chart shows greater over predictions in the layer scale spaces under various smoke control
height from 4.52 m to 2.0 m, compared to all other conditions. These experiment (Part 2 of the their
runs with the exclusion of the run (1). Thereafter it study) also served as the validation exercise for the
begins to follow the trend of the physical simple analytical theories that presented in Part 1
experiment slightly better than the runs (5), (6) and [8]. As a result, the predictive capabilities of the

11
International Journal on Engineering Performance-Based Fire Codes

simple theories were proved to be fairly good. In experiments; a schematic of the facility is given in
this section, this standard filling equation, Fig. 3. The facility consisted of a large space; floor
developed by Yamana and Tanaka, is indicated by area equating to 720 m2, with maximum
the black doted lines on all layer height charts (Figs. dimensions of 30 m by 24 m. The ceiling height
6-7, 9-10) for comparative purpose. was 26.3 m. The experimental conditions selected
for this study were (1) Natural Filling, (2) Natural
The experiment facility was the BRI full-scale fire Venting and (3) Mechanical Venting. Table 2
test laboratory, of the Ministry of Construction, provides some finer details that pertained to each
Japan, which was used for this series of experiment.

Tree 2 Tree 1

Smoke filling set-up Natural smoke venting set-up Mechanical extraction set-up

Fig. 3: Plan of experimental set-up for Scenario B

12
International Journal on Engineering Performance-Based Fire Codes

Table 2: Experimental conditions for Scenario B

Mechanical Mechanical Mechanical


Test No. Natural
Filling rate extraction 6.0 extraction 4.5 extraction 3.2
venting
Condition m3.s-1 m3.s-1 m3.s-1
Initial temp °C 14.0 18.0 13.0 12.0 13.0
Lower opening
---- Prior to test Prior to test ---- ----
time [s]
Lower opening
---- 3.23 3.23 ---- ----
area [m2]
Upper Opening
---- Prior to test ---- ---- ----
Time [s]
Upper opening
---- 6.46 ---- ---- ----
[m2]
Mechanical
venting
---- ---- Prior to test Prior to test 120
activation time
[s]
Mechanical
exhaust rate ---- ---- 6.0 4.5 3.2
[m3.s-1]

The source of fire and smoke consisted of fifteen 2.2.1 Natural filling rate
methanol pans 450 mm square, which were put
Similar to scenario A, the smoke layer was just let
together to form an almost 1800 mm square fire
descend within the experimental hangar enclosure.
source, placed centrally on the floor of the facility.
This can be seen in Fig. 4 where the smoke layer is
A smoke candle was incorporated to generate
very quick to descend at first, and then it begins to
smoke that would then follow the hot gases being
slow down after approximately 120 s into the
produced by the fire. The burning of the source
experiment. The measurements recorded by the two
demonstrated an almost constant rate. The heat
thermocouple trees and by the photometer give
release rate was obtained by converting the average
good comparison with each other. Also the
mass-burning rate, and this resulted in a 1300 kW
standard filling equation developed by Yamana and
fire for all experiments.
Tanaka has given remarkable results. This standard
filling equation is shown throughout the remaining
Two trees equipped with Chromel–Alumel
scenarios i.e. natural and mechanical, as a guide to
thermocouples, measured the smoke layer
the performance of each strategy. Note that a time
temperature during each test. One of the trees near
lag of approximately 60 s (Fig. 4) is present in the
the east wall tree No. 1 had thermocouples placed
experiment, due to the time it took for the hot gas
with 1 m spacing. The other tree No. 2 positioned
to rise from the fire source, impinge on the ceiling
near the west wall had thermocouples spaced 2 m
and then spread laterally across it.
along its vertical axis, except for proximity of the
ceiling where the spacing was reduced to 1 m.
The run (1) incorporated a grid cell size of 500 mm
in the x, y, z directions. This equated to 144000
Additionally photocells were set near the two-
cells within the hangar. The initial ambient
thermocouple trees, which had spacing of 2 m from
the ceiling and 4 m from the floor for the temperature was set at 18 °C. The fire source was
measurement of optical density. The smoke layer mimicked within FDS using a Heat Release Rate
was also recorded by eye. For observations Per Unit Area (HRRPUA) of 401 kWm-2.
purposes a scale was placed on the north boundary Multiplying this (HRRPUA) by the fire area of
wall to locate the smoke layer interface boundary. 3.24 m2 gave a Heat Release Rate (HRR) of 1300
In summary, the smoke layer height was measured kW.
using three sources, namely: (a) temperature profile,
(b) optical smoke density and (c) by eye. By comparing the experimental results to the
predictions of FDS, it can be concluded that the
The rate of mechanical smoke extraction was model did not behave as well as in the room
measured using pitot tubes and anemometers environment. Time and memory constraints
mounted in the exhaust duct on the 7th floor. The prohibited grid refinement below 300 mm.
rate of natural airflow velocities was measured
using a hot wire anemometer. The run (1) gave the best result out of the three
simulated runs, although the filling rate was

13
International Journal on Engineering Performance-Based Fire Codes

overpredicted until the time of approximately 150 s Then the grid cell size of 750 mm (46080 cells) and
with a maximum error of 30%, then under an initial time step of 0.5 s were employed in the
predicted the filling rate for the remainder of the run (3). This resulted in an overprediction of the
simulation with a error of 50%. The smoke filling filling rate.
rate demonstrates erratic movement from a height
of 25 m to a height of 22 m. This is due to the The overall results on grid size effect on quality of
turbulence in the developing shallow smoke layer. predictions are summarized in Table 3.
The grid cell size was reduced to 300 mm in the
run (2) (720000 cells overall) with the expectation The analysis of temperature predictions was made
of better results from the model. However, while for the selected heights of 24 m, 16 m and 8 m.
the initial filling rate improved somewhat, a larger Generally FDS overpredicted the temperature in all
magnitude of error in the filling rate was evident runs. Typical predictions (run 1) are presented in
towards the remainder of the simulation. This result Fig. 5. Temperatures at 24 m in run (1) has large
was confusing as the CFD numerical model is error in the beginning (32% at 100 s), and finish up
generally expected to give better results on finer by getting smaller (8% at 500 s).
grids.

SMOKE FILLING RATE IN HANGAR ENVIRONMENT

30

Grid Representation
25

20

1300 (Kw) Methanol


HEIGHT (m)

Fire Source
15

10

0
0 100 200 300 400 500 600 700
TIME (s)
EXP.THERMOCOUPLE EXP.PHOTOMETER EXP.EYE
CALCULATED EQUATION FDS LAYER HEIGHT RUN 1 (500X500X500) FDS LAYER HEIGHT RUN 2 (300x300x300)
FDS LAYER HEIGHT RUN 3 (750X750X750)

Fig. 4: Smoke layer time history plot. Natural filling in a hangar environment. Scenario B

Table 3: Correlation of grid size with the quality of predictions for natural filling rate in a hanger
environment

Grid Cell Size Maximum deviation


Run Total number of cells
(XYZ) (mm) from test results (%)
1 500 x 500 x 500 144000 50
2 300 x 300 x 300 720000 55
3 750 x 750 x 750 46080 62.5

14
International Journal on Engineering Performance-Based Fire Codes

TEMP RUN 1 SMOKE FILLING IN HANGAR ENVIRONMENT

45

40

35

30
TEMP DEG C

25

20

15

10

0
0 100 200 300 400 500 600 700
TIME (s)

TEMP EXP 24 TEMP EXP 16 TEMP EXP 8 FDS TEMP 8 FDS TEMP 16 FDS TEMP 24

Fig. 5: Temperature history plot. Natural filling in a hangar environment. Scenario B

2.2.2 Natural venting


The effect of mechanical extraction is seen in Fig.
For this smoke control strategy, known as gravity
7 as both the experimental Thermocouple and Eye
venting in the UK, FDS gave much improved
curves are slightly lifted off the standard filling
results in layer height predictions compared to
equation line.
those of the natural filling rate scenario. Gravity
venting scenario includes two natural openings
Modeling gave contradictory results. Run (1A)
(windows) at the 7th floor of the facility.
produced rather scattered results at initial stage,
and predicted strong downward propagation of
The effect of the natural vents is quite apparent
smoke contradicting with common sense
(Fig. 6) in that the predicted layer trend lies above
expectations and experimental results. Run (1B)
the standard smoke filling equation. It can be seen
showed better trend, although it overpredicts smoke
that FDS and the photometer readings in the
descend rate initially and slightly underpredicts at
experiment are in a good agreement. The standard
later stages. The over prediction error is
filling equation and the thermocouples had a
approximately (12%) at 50 s. The under prediction
marginal error against FDS of approximately
error is approximately (35%) at 400 s. The
(12%). Runs (3) and (4) performed slightly better
temperature plot (Fig. 8, run (1B)) shows an
than run (1) due to better resolution.
overprediction. The maximum temperature error is
approximately (21%) at 100 s.
The final smoke layer height is predicted correctly
for this scenario, in contrast to Scenario A, where it
Run (2) incorporated a reduced volume flux
was consistently overpredicted.
extraction rate of 4.5 m3s-1. Grid was changed to
500mm cell size in this simulation, as simulation
2.2.3 Mechanical extraction (different
times where in the order of 35 hours. This change
volumetric rates)
served to check how much speed-up can be
The next analysis looks at three different types of achieved, and to what extent the resulting error
mechanical smoke extraction from the hangar would increase. Many designers may be tempted to
facility. In the first run the mechanical extraction do this in a bid to receive results quicker.
rate was documented as 6.0 m3s-1.

15
International Journal on Engineering Performance-Based Fire Codes

NATURAL SMOKE VENTING IN HANGAR ENVIRONMENT

30

Natural vents
25

20
HEIGHT (m)

15

1300 (kW) Fire


Source
Make up vent
10

0
0 100 200 300 400 500 600 700
TIME (s)
EXP.THERMOCOUPLE EXP.PHOTOMETER STANDARD FILLING EQUATION
Run 1(500x500x500) Run 2(500x500x500)(DT0.05) Run 3(400x400x400)(DT0.1)
Run 4 (300x300x300)

Fig. 6: Smoke layer time history plot. Natural venting in a hangar environment. Scenario B

SMOKE EXTRACTION 6.0m^3/s IN HANGAR ENVIRONMENT

30

25

20
HEIGHT (m)

15

10

0
0 100 200 300 400 500 600 700
TIME (s)
EXP.THERMOCOUPLE EXP. EYE STANDARD FILLING EQUATION
FDS RUN 1A (500X500X500) DT =0.1 FDS RUN 1B (400X400X400) DT =0.1

Fig. 7: Smoke layer time history plot. Smoke extraction of 6.0 m3s-1 in a hangar environment

16
International Journal on Engineering Performance-Based Fire Codes

TEMP RUN 1B MECH 6.0m^3/s SMOKE EXTRACTION IN HANGAR ENVIRONMENT

40

35

30

25
TEMP DEG C

20

15

10

0
0 100 200 300 400 500 600 700
TIME (s)

EXP TEMP 24 EXP TEMP 16 EXPT TEMP 8 FDS TEMP 8 FDS TEMP 16 FDS TEMP 24

Fig. 8: Temperature history plot. Smoke extraction of 6.0 m3s-1 in a hangar environment

Peculiar feature of 4.5 m3s-1 scenario (and the next Temperature predictions for this scenario were
one with 3.2 m3s-1) was that experiments were run slightly improved, with the maximum error being
with closed vent to investigate how well the (22%) for the 16 m high thermocouple within the
mechanical extract system would perform without facility. FDS temperature predictions for the other
replacement air. Mechanical or natural extraction two heights within the enclosure were below (22%).
methods always need replacement air to function.
However for the purposes of this validation study, The final simulation of this scenario analyses
we follow the experiment exactly, i.e. the vent was mechanical smoke extraction with a volume flux of
closed. 3.2 m3s-1. This scenario had all the same
characteristics of the previous except for a reduced
It is clear from the layer height chart (Fig. 9) that volume flux, which was activated 120 s into the
closing up of the inlet vent did not effect the smoke simulation. The filling rate takes a sudden turn at
extraction rate, as the experimental results lie approximately 130 s (Fig. 10), and was over
above the standard filling equation. However FDS predicted from a height of 26.3 m to a height of
did not perform really well, a brief look at the layer 12.5 m, the maximum error up to approximately
height chart for smoke extraction using a 4.5 m3s-1 40% at 68 s. Thereafter the effect of the extraction
volume flux reveals this. FDS over predicted the was evident, resulting in a final layer height error
filling rate from a height of 26.3 m to a height of 16 of 42% at a time of 400 s. Temperature predictions
m by approximately (8%); thereafter it under at heights 24 m, 16 m and 8 m were also poor. This
predicted the filling rate by approximately (34%). error was probably due to the relative coarse grid
applied.

17
International Journal on Engineering Performance-Based Fire Codes

SMOKE EXTRACTION 4.5m^3/s IN HANGAR ENVIRONMENT

30.00

25.00

20.00
HEIGHT (m)

15.00

10.00

5.00

0.00
0.00 100.00 200.00 300.00 400.00 500.00 600.00 700.00
TIME (s)

FDS RUN 2 (500X500X500)(DT=0.1) EXP THERMOCOUPLE EXPT BY EYE STANDARD FILLING EQUATION

Fig. 9: Smoke layer time history plot. Smoke extraction of 4.5 m3s-1 in a hangar environment

SMOKE EXTRACTION (120s) 3.2m^3/s IN HANGAR ENVIRONMENT

30

25

20
HEIGHT (m)

15

10

0
0 100 200 300 400 500 600 700
TIME (s)

EXP THERMOCOUPLE EXP BY EYE STANDARD FILLING EQUATION FDS RUN 3 (500X500X500) DT 0.5

Fig. 10: Smoke layer time history plot. Smoke extraction (120 s) of 3.2 m3s-1 in a hangar environment

18
International Journal on Engineering Performance-Based Fire Codes

2.3 Scenario C − Smoke Filling and CFAST Zone Model are kept in the presentation
Extraction in an Atrium Environment for comparative reasons.

The next scenario analyses the predictive capability The free NIST software, DXF2FDS, which helps
of FDS V 4.0.5 when applied to an atrium type with the set-up of complex geometries within FDS,
enclosure. Atrium buildings are becoming highly was employed in this scenario. It must be borne in
popular in modern built environment. Architects mind that the rectilinear approach on which FDS V
are required to house many different divisions 4.0.5 was developed is preserved after conversion.
within one building for accessibility and For instance, a curved roof will actually be
sustainability purposes. In order to supply light and modelled as a roof consisting of stair stepped
fresh air throughout such building architects like to rectangular obstructions. DXF2FDS creates a
use the atrium effect. The atria may also be pretty picture for the geometry under investigation,
designed for aesthetic reasons. However, smoke however this picture is not what is really being
spread in such buildings can pose various threats to modelled. No matter what a modeller specifies in a
its occupants if not properly addressed. Due to FDS input code, this is rearranged to suit the grid,
large voids between the floors smoke can readily which has been specified in that scenario. FDS
move quickly around the building, effecting areas users must be careful when determining vent area
quite remote from the fire source. Our main etc., because the requested and the actual can be
objective in this scenario is to challenge FDS to totally different.
determine the most important physical parameters
in the development of a smoke layer and the filling 2.3.1 Series 1
process within an atria environment. As can be seen from the accompanying layer height
chart (Fig. 12), FDS failed to predict accurately the
The experimental data used in the following study smoke-filling rate within the atrium environment.
was published by Chow in 1994 [9]; the As can be seen from the chart, a number of runs
experiments were previously conducted by were carried out in a bid to improve the FDS
Hagglund in 1985 [10]. The present analysis predictions.
incorporated experiments from test 1 of Chow’s
paper [9] only. Looking closely at the experiments trend lines i.e.
the smoke meter and the eye recordings, an
The enclosure under investigation consisted of two agreement between the two methods in the smoke
rooms (Fig. 11). A larger hall (Atrium) with decent rate is evident. FDS has to a certain extent
maximum dimensions of 5.6 m by 5.6 m by 6.1 m mimicked this trend, although the position of the
high and a smaller fire room constructed adjacent smoke interface was over predicted. The runs 2D,
to the atrium. The two rooms share a common 2EA, 2EB, 2F and 2F2 all show the same trend.
doorway with dimensions of 1m wide by 2 m high. The run 2D came closest to the experimental
Two fires with different HRR are considered in this prediction.
study. For simplicity the investigation is divided up
into series. Series (1) incorporates a fire size of 250 The error associated with the 2D run layer height
kW, while series (2) has a fire source of 560 kW. during the over prediction phase is approximately
Experimental conditions are summarised in Table 4. (52%) at 40 s, when compared to the eye trend, but
Numerous computational parameters for each is greater when related to the smoke meter trend.
series are reported in [11]. Run 2F1 had a Smagorinsky Constant of 0.15, but
shows some numerical error when compared to 2F2.
The inlet vent in the lower part of the atrium wall Run 2G employed finer mesh, taking
had dimensions of 0.8 m wide by 0.25 high and approximately 30 hours of computation time. Its
was open to the atmosphere. The walls, floors and results are quite surprising as being the poorest
ceilings consisted of concrete. The fire source simulation whilst incorporating the finest grid. This
comprised of kerosene pan situated centrally in the goes against the underpinning principle of
fire room floor. In the experiments, temperatures Computational Fluid Dynamics. This run also
were measured at heights 200 mm, 700 mm, 1200 shows twice the HRR specified in the input file
mm, 1700 mm, and 2200 mm above the floor in the (250 kW), Fig. 13. All other HRR predictions did
fire room. Temperature and smoke layer heights not have this problem.
were also measured in the atrium space. These
were positioned 500 mm, 1500 mm, 2500 mm, The CFAST zone model trend line is incorporated
3500 mm, 4500 mm and 5500 mm above the floor. in Fig. 12; it is evident that CFAST and FDS are in
These values were plotted by Chow [9] and agreement from 30 s to 80 s. Note that CFAST
compared to three zone models. The results of ignores the momentum equation and assumes
simple two-layer structure.

19
International Journal on Engineering Performance-Based Fire Codes

View from the fire room, looking into the atrium View from the atrium looking into the fire room

Atrium

Fire Room

Fig. 11: Plan of experimental set-up for Scenario C

20
International Journal on Engineering Performance-Based Fire Codes

SMOKE FILLING SERIES 1 IN ATRIUM ENVIRONMENT

7.00

ATRIUM
6.00
250 (kW) Fire

5.00

INLET VENT
HEIGHT (m)

4.00

3.00
e
DOOR

2.00

1.00

0.00
0 20 40 60 80 100 120 140 160 180
TIME (s)
Expt. By Eye Expt. By Smoke Meter
FDS LAYER HEIGHT 2F (100X100X100)(100X100X100) FDS LAYER HEIGHT 2D (200X200X200)(150X150X150)
FDS LAYER HEIGHT 2EA (100X100X100)(200X200X200) FDS LAYER HEIGHT 2F1(200X200X225)(200X200X200)(0.15)
FDS LAYER HEIGHT 2F2(200X200X180)(155X155X155)(0.15) FDS LAYER HEIGHT 2G (50X50X50)(155X155X155)(SMAG 0.1)
FDS LAYER HEIGHT 2EB (100x100x100)(200x200x200)(S2%) CFAST ZONE MODEL

Fig. 12: Smoke layer time history plot. Smoke filling, Series 1 in atrium environment

HHR 2G SMOKE FILLING IN ATRIUM ENVIRONMENT

650.0

600.0

550.0

500.0

450.0

400.0
HRR (kW)

350.0

300.0

250.0

200.0

150.0

100.0

50.0

0.0
0.00 20.00 40.00 60.00 80.00 100.00 120.00 140.00 160.00
TIME (s)

HRR (50X50X50)(155X155X155) kW

Fig. 13: HRR plot. HHR for the Run 2G. Smoke filling in atrium environment

21
International Journal on Engineering Performance-Based Fire Codes

Temperature predictions for the run 2D, Fig. 14, floor. The overall descent time was also quite well
show an over prediction when compared to predicted by CFAST.
measured values. The maximum error associated
with the run 2D at a height of 2500 mm is Variation in the time step within FDS in run 2 did
approximately 33%. This error goes up to 60% for not change the result. Maximum layer height errors
the lower readings within the atrium. Temperature when compared to experimental eye recordings
predictions in the runs 2F and 2F2 show the error were approximately 57% at 40 s. FDS did not
similar to the case 2D. The run 2G showed the provide accurate smoke filling predictions within
highest temperature errors as HRR was over an atrium environment incorporating a fire of 560
predicted by FDS for this case. kW.

2.3.2 Series 2 Temperatures in the fire room were over predicted.


The thermocouple at the height 2200 mm had a
For the Series 2 experiment smoke descended to
maximum error of approximately 46%. The lowest
floor at approximately 70 s, compared to 110 s in
thermocouple within the fire room at the height of
series 1. This is a result of the higher HRR in this
200 mm was also over predicted by FDS and had a
scenario. The same trend occurred here as in series
maximum error of approximately 77%.
1, i.e. FDS over predicted the filling rate (Fig. 15).
The FDS smoke filling prediction went only down
Predicted temperatures in the atrium space also
to 1 m above the floor. CFAST and FDS are once
demonstrated over predictions. The maximum error
again showing some agreement with each other
resulted from the thermocouple reading at the
until approximately 60 s. Apparently, the simple
height of 4500 mm (approximately 40% at 70 s).
CFAST model performed much better in this case,
since it was able to predict smoke descent to the

Table 4: Experimental conditions for Scenario C

Test No. 1 Fire size/m2 Fire HRR, kW Lower vertical vent area [m2] Ceiling vent area [m2]

Series 1 0.25 250 0.2 0

Series 2 0.56 560 0.2 0

TEMP RUN 2D SERIES 1 SMOKE FILLING IN ATRIUM ENVIRONMENT

80

70

60

50
TEMP DEG C

40

30

20

10

0
0 50 100 150 200 250 300 350
TIME (s)
EXP TEMP 3500 EXP TEMP 2500 EXP TEMP 1500 EXP TEMP 500 FDS TEMP 3500
FDS TEMP 2500 FDS TEMP 1500 FDS TEMP 500

Fig. 14: Temperature history plot. Run 2D, Series 1. Smoke filling in atrium environment

22
International Journal on Engineering Performance-Based Fire Codes

SMOKE FILLING SERIES 2 IN ATRIUM ENVIRONMENT

7.00

ATRIUM
6.00
560 (kW) Fire

5.00

INLET VENT
HEIGHT (m)

4.00

3.00
e
DOOR

2.00

1.00

0.00
0 20 40 60 80 100 120 140 160 180
TIME (s)

CFAST ZONE EXPT BY EYE


EXPT SMOKE METER FDS RUN 1 (200X200X200)(155X155X155)
FDS RUN 2 (200X200X200)(155X155X155)(0.05)

Fig. 15: Smoke layer time history plot. Smoke filling, Series 2 in atrium environment

2.4 Scenario D − Smoke Movement in a Traditionally smoke curtains have been used to
Corridor Environment contain smoke in the event of fire. The curtain may
be a part of either active (where by it drops to a
The final scenario in this study deals with smoke specified height on alarm activation) or the passive
filling and gas temperatures within a corridor fire protection system. In the present study the
environment. Smoke curtains will be used in this curtains will act as passive systems.
scenario as means of smoke control. This method
can be classed as compartmentation. Compart- Matsuyama and Wakamtsu [12] carried out a recent
mentation is probably the oldest and most basic study to test room and corridor smoke filling for
form of fire protection. use in calibration of zone and field models. Two
scenarios of their study are used to investigate the
Compartmentation consists of using partitions, accuracy of FDS in predicting smoke movement,
floors, and ceilings of a building to restrict the smoke filling and gas temperatures.
smoke spread. In the overall fire safety strategy for
any building, smoke and hot gases are recognised The fire source in the experiment was a
as the major hazard, and its control must be a polyurethane triangular shaped mattress. The base
primary concern for not only the client, but also for width was 600 mm, while the height of the triangle
the consulting engineer and equally the architect in was 900 mm. Using three load cells, the mass loss
charge of the design. rate was recorded continuously. The HRR was then
calculated by multiplying loss rate by the heat of
This type of smoke protection may form part of the combustion. By calculating the area of the triangle,
fire safety strategy within, for example, a nursing it was possible to model this experiment within
home. Such premises usually incorporate long FDS. The HRR in this experiment is not steady,
corridors, which link almost every part of the therefore the ramp file was set up to reflect the
building. With the incorporation of smoke curtains HRR obtained from the experiment within FDS.
at strategic locations, the spread of smoke and The fire in FDS can be ramped by specifying a
gases can be restricted, giving staff ample time to time and a fractional HRR at that specific time.
evacuate the building.

23
International Journal on Engineering Performance-Based Fire Codes

Thermocouple trees were situated within the only filling to approximately 600 mm above the
building to measure the smoke layer height and floor, this happened in all the runs A, A1 and B.
temperature. The thermocouples used in the
experiment were type K (0.3 mm-diameter). The corridor smoke layer in the experiment begins
Fourteen thermocouples were placed on each tree. to develop at position T2 at about 120 s. It can be
Using the N-percent method (N=10%) the smoke seen from the chart that position T5 is delayed by a
layer height and the average temperatures were time of approximately 30 s, this is due to the length
calculated for all the experimental results. of the corridor. FDS over predicts smoke
movement within the corridor at a time of 85 s,
In this scenario the two studies were performed. making a time discrepancy of 35 s.
The setup can be seen from Fig. 16. In the Study (A)
smoke curtain (B) (color green) was situated at the The effect of the smoke curtain is observed to delay
west end of the corridor and its role was to prevent the smoke propagation time down the corridor.
smoke from spilling into the lobby room at the end This observation is through the comparison
of the corridor. between the experimental chart results for the cases
A and B. In the study A the lobby smoke developed
The Study (B) used same conditions as (A), but had at approximately 150 s (EXP T6 curve), while in
a decreased HRR and utilised an additional smoke the study B it developed at 220 s (EXP T6). This
curtain (color green) situated more or less centrally difference is due to introduction of the second
across the corridor width. smoke curtain in the middle of the corridor.

The smoke curtain depths were 700 mm below the FDS simulations (runs A and A1) predict smoke in
ceiling for each study. The corridor had an inlet the lobby at a time of 150 s, which is correct, but
vent situated near the floor on the west wall. Layer then grossly over predicts the filling time within
height, smoke movement and gas temperature were the lobby by about 50 s. In the run B the lobby
analysed against experimental data, and also smoke developed at approximately 165 s, this is an
against each other, to show the effects, if any, of over prediction of 55 s.
the smoke curtain(s).
As a general tendency, the smoke layer descends
Summary of the computational parameters for both right after the fire source has decayed. In the two
runs is presented in Table 5. experiments, the corridor smoke layer has dropped
to a level lower than 1 m above the floor. This
From the layer height charts for the runs A and A1 means that the corridor is unsafe after
(Figs. 17 and 18), it is evident that FDS has once approximately 180 s at position T4.
again over predicted the smoke-filling rate within
the fire room. At approximately 60 s the smoke Comparing this to FDS predictions A, A1 and B
layer begins to develop in the fire room. According demonstrates that even though FDS start off by
to experimental data, smoke layer descends at an over predicting filling rate, its merges with the
almost constant rate. experimental data at position T4 in all the runs.
This is a fair prediction by the model from a
In contrast, FDS predictions show that the smoke designer’s perspective.
layer develops approximately 10 to 15 s earlier and
fills the room very quickly down to a height of In all the above runs, comparing the experimental
1200 mm above the floor (run A). This filling temperature to the FDS predicted temperatures it
slows down in the runs A1 and B, due to finer grid. can be concluded that the temperature near the fire
The results for the run B are presented in Fig. 19. source are all in good agreement (13% discrepancy)
The error in layer perdition between the experiment with the experiment until a maximum temperature
and FDS is approximately 42% at 50 s in the run A. observed in the experiment. After the maximum
This error tends to decrease with the incorporation point FDS predictions diverge from the
of the finer grid in the runs A1 and B. After decay experimental ones (Fig. 20), far field temperatures
of the fire in the experiment, the smoke layer in the tend to be over predicted in all the runs.
fire room descends to 200 mm from the floor at
position T0; FDS under predicts this situation by

24
Study A Study B

FIRE SOURCE
FOLLOWS
LOBBY ROOM
t2 GROWTH RATE
Area 10m2 FIRE ROOM

Area 25m2

Smoke curtain B
Active for runs A and B
CORRIDOR Area 28m2
Inlet vent
Smoke curtain A Active for run B only

Fig. 16: Plan of experimental set-up for Scenario D


International Journal on Engineering Performance-Based Fire Codes

25
26
Table 5: Summary of input/ output parameters for Scenario D

Scenario D Smoke filling and movement in corridor environment

Input Output

Fire Fire Initial Soot


Grid cell size Total Fire HRRPUA Smagorinsky Simulation Q* [-] Fire resolve. Converge
Run No. area output 2 time yield
(XYZ) (mm) cells source (kW/m ) constant time (s) (0.1<Q*<2.5) (Dx) / (∂x) speed [s-1]
International Journal on Engineering Performance-Based Fire Codes

(m2) (kW) step. (%)

Run A 4 x 200x200x200 20618 0.27 Ramped 789 - 0.042 0.2 245 0.75 1.65 -

Polyurethane
Run A1 4 x 100x100x100 155616 0.27 Ramped 789 - 0.042 0.2 245 0.75 1.65 -
Mattress

Run B 4 x 100x100x100 20618 0.27 Ramped 555 - 0.042 0.2 250 0.53 2.60 -
International Journal on Engineering Performance-Based Fire Codes

LAYER HEIGHT RUN A SMOKE MOVEMENT IN CORRIDOR ENVIRONMENT

2.5

2
HEIGHT (m)

1.5

0.5

0
0 50 100 150 200 250 300 350 400
TIME (s)

EXPT1 EXPT2 EXPT3 EXPT4 EXPT5 EXPT6 FDS T0


FDS T1 FDS T2 FDS T3 FDS T4 FDS T5 FDS T6 EXPT0

Fig. 17: Smoke layer time history plot. Run A. Smoke movement in corridor environment

LAYER HEIGHT RUN A1 SMOKE MOVEMENT IN CORRIDOR ENVIRONMENT

2.5

2
HEIGHT (m)

1.5

0.5

0
0 50 100 150 200 250 300 350 400
TIME (s)

EXPT1 EXPT2 EXPT3 EXPT4 EXPT5 EXPT6 EXPT0 FDS TO FDS T1


FDS T2 FDS T3 FDS T4 FDS T5 FDS T6

Fig. 18: Smoke layer time history plot. Run A1. Smoke movement in corridor environment

27
International Journal on Engineering Performance-Based Fire Codes

LAYER HEIGHT RUN B SMOKE MOVEMENT IN CORRIDOR EXPERIMENT

2.5

2
HEIGHT (m)

1.5

0.5

0
0 50 100 150 200 250 300 350 400
TIME (s)
EXPT TO EXPT T1 EXPT T2 EXPT T3 EXPT T4 EXP T5 EXPT T6
FDS T1 FDS T0 FDS T2 FDS T3 FDS T4 FDS T5 FDS T6

Fig. 19: Smoke layer time history plot. Run B. Smoke movement in corridor environment

TEMP RUN A SMOKE MOVEMENT IN CORRIDOR ENVIRONMENT

90

80

70
TEMP DEG. C

60

50

40

30

20
0 50 100 150 200 250 300 350
TIME (s)
EXPT0 EXPT1 EXPT2 EXPT3 EXPT4 EXPT5 FDS T0 FDS T3 FDS T1 FDS T2
FDS T4 FDS T5

Fig. 20: Smoke layer time history plot. Run A. Smoke movement in corridor environment

28
International Journal on Engineering Performance-Based Fire Codes

3. CONCLUSIONS closed and the result showed initially an over


prediction of 8%, and an under prediction of 34%
Significant, although limited due to a shortage of thereafter. Finally, a case was run with 3.2 m3s-1
experimental data number of CFD simulations have extraction rate, initiated at 120 s and also with no
been undertaken within this study in an attempt to inlet vent. This had an initial over prediction filling
evaluate smoke control strategies using the FDS V rate of 40%, which then ended up under predicting
4.0.5 computational fluid dynamics model. by 42%.

The study investigated the capability of FDS V The third test, Scenario C ‘Smoke Filling in an
4.0.5 to accurately predict smoke control strategies Atrium Environment’, was divided into two series.
within a variety of selected enclosures. Series 2, which incorporated a larger fire source,
Computational results were compared to full-scale did not give satisfactorily results; five runs did
physical experiments involving Natural Smoke have the same trends over predicting the entire atria
Filling, Natural Smoke Venting, Mechanical smoke filling process by 52%. CFAST zone model
Smoke Extraction and Smoke Movement. performed better than FDS in this case, predicting
smoke descent down to the floor. This run also
These obviously do not cover all the situations a highlighted that the finer grid resulted in the
fire safety engineer may encounter, however, they poorest performance. HRR did not stay stable in
are all representative of real full-scale fire the run 2G, diverging from its prescribed value.
behaviour. Additionally, to the authors current Maximum over prediction in the Series 2 was 57%.
knowledge, none of the experimental data used in
the present paper have been employed earlier in The final case of this study, Scenario D
FDS evaluation studies. investigated ‘Smoke Filling and Movement in a
Corridor Environment’. FDS over predicts smoke
From the first case of the present study, Scenario A filling by 42%. FDS also over predicted smoke
‘Smoke Filling in a Room Environment’, it can be movement in the corridor by approximately 58%.
concluded that FDS V 4.0.5 is capable of However, the predictions converge back to
predicting the smoke filling rate within the limits experimental data by the time when untenable
specified by the models developers. An important conditions are achieved in the corridor.
discovery in this scenario was that, the
specification of finer grids does not always reap The principal conclusions of the study are as
better results. This is evident from the layer height follows:
chart representing the room scenario. The chart
shows an over prediction of 16% in layer height • Performance of FDS varies significantly
during initial smoke filling, followed by an under depending on particular smoke movement
prediction of 20% in the latter part of the scenario. Performance seems to deteriorate
simulation. In summary the results are satisfactory with increase in geometrical complexity of
for the smoke-filling rate in a room environment. simulated compartment/building.

In the second case, Scenario B ‘Smoke Filling and • FDS gave an overall average error of 32% in
Extraction in a Hangar Environment’ Natural the smoke-filling rate.
Filling, Natural Venting and Mechanical Extraction
strategies were investigated. For this scenario the • Temperature predictions for FDS throughout
area of the hangar was 24 m x 30 m with an overall the study were over predicted on average by
ceiling height greater than 26 m. The initial over 35%.
prediction for the smoke-filling rate increased to
30%, and the later filling rate had an under • The above figures suggest that the claim [4]
prediction error of approximately 50%. This results that FDS can predict temperatures within 5 to
are beyond the limits highlighted by the models 20% of the actual value is incorrect. Identical
developers, and therefore FDS did not predict claim with respect to flow velocities was not
accurately the smoke filling rate within the hangar. investigated in the present study. However, it
would be important to verify that claim as well.
FDS performed very well to predict Natural
Venting scenario and predicted values were well • Effects of grid are very important to solution.
below the error limit of 20%. Mechanical The specification of finer grids does not
Extraction scenario was considered for three always lead to better results. It has been
different extraction rates. For the volume extraction experienced in the present study that the
rate of 6.0 m3s-1 the layer height predictions solution does not always stabilise, as the grid
performed reasonably well, with an over prediction is refined. Moverover, the specification of
of 12% and a under prediction of 35%. In the next
case (4.5 m3s-1 extraction rate) the inlet vent was

29
International Journal on Engineering Performance-Based Fire Codes

finer grids in some cases may significantly control problems”, Fire Science and Technology,
worsen predictions. Vol. 5, No. 1 (1985).
9. W.K. Chow, “A short note on the simulation of the
This effect (observed also by other atrium smoke filling process using fire zone
investigators) requires careful investigation as models”, Department of Building Services
it contradicts (at least, at first glance) with Engineering, Hong Kong Polytechnic University,
general philosophy behind CFD, and LES in Hong Kong (1994).
particular. 10. B. Hagglund, Comparing fire models with
experimental data, FOA Report C20864-24,
• HRR is not necessarily maintained in National Defense Research Establishment,
computations, in contrast to what is claimed. Sundbyberg, Sweden (1985).
An example of this was observed in the run 2G, 11. P. Coyle, An evaluation of smoke control strategies
Scenario C were the predicted HRR was twice using Fire Dynamics Simulator (FDS), MSc Thesis,
as high as it should be. University of Ulster, UK (2005).
12. K. Matsuyama and T. Wakamtsu, Department of
• In some cases (such as smoke filling in the Architecture, Faculty of Science and Technology,
atrium) simple zone model can give better Science University of Tokyo, NISTIR 6588 (2000).
prediction than FDS.

• Time step is an important parameter in FDS, as


in any CFD code. Unfortunately, FDS does not
allow users to vary this parameter. The authors
believe that this flexibility is essential for any
credible CFD code, and should be provided by
developers.

• Smagorinsky constant did not have significant


effect in the problems considered in the
present study.

REFERENCES

1. G.W. Zou and W.K. Chow, “Numerical studies on


fire environment by a field model”, International
Journal on Engineering Performance-Based Fire
Codes, Vol. 7, No. 1, pp. 18-34 (2005).
2. C.W. Lautenberger, J.L. de Ris, N.A. Dembsey,
J.R. Barnett and H.R. Baum, “A simplified model
for soot formation and oxidation in CFD
simulation of non-premixed hydrocarbon flames”,
Fire Safety Journal, Vol. 40, pp. 141-176 (2005).
3. W.K. Mok and W.K. Chow, “Verification and
validation in modeling fire by computational fluid
dynamics”, International Journal on Architectural
Science, Vol. 5, No. 3, pp. 58-67 (2004).
4. K. McGrattan and G. Forney, Fire Dynamics
Simulator (Version 4) user’s guide, NIST Special
Publication 1019 (2005).
5. B. Hagglund, R. Jansson and K. Nireus, Smoke
filling experiments in 6 x 6 x 6 meter enclosure,
FOA Report C 20585-D6, National Defence
Research Establishment, Sweden (1985).
6. B. Karlsson and J.G. Quintiere, Enclosure fire
dynamics, CRC Press, Boca Raton (2000).
7. T. Yamana and T. Tanaka, Smoke control in large
scale spaces, Part 2, Building Research Institute,
Ministry of Construction, Japan (1985).
8. T. Tanaka and T. Yamana, “Smoke control in large
spaces, Part 1: Analytic theories for simple smoke

30

You might also like