You are on page 1of 10

ARTICLE IN PRESS

Fire Safety Journal 44 (2009) 415424

Contents lists available at ScienceDirect

Fire Safety Journal


journal homepage: www.elsevier.com/locate/firesaf

Assessment of a model development for window glass breakage due to re


exposure in a eld model
Kai Kang 
Jacobs, 260 Madison Avenue, New York, NY 10016, USA

a r t i c l e in fo

abstract

Article history:
Received 1 August 2008
Received in revised form
27 August 2008
Accepted 5 September 2008
Available online 15 October 2008

This paper presents a model development for the prediction of window glass breakage and fallout in a
eld model. Glass breakage is based on the temperature difference and the allowable glass breaking
stress; and glass fallout is determined by a preset number of successive breakages. As a validation,
generally good agreements are obtained between the numerical predictions and the data from a
compartment re experiment. The predicted glass surface temperature and the adjacent gas
temperature are within 1025% of the test data. For re sizes of 170, 280 and 390 kW, the time of
initial occurrence of glass breakage are shown within reasonable range of the experimental results. For
the 680 kW re case, the model shows an earlier glass fallout time, however, the predicted glass
temperature at fallout is around 450 1C and is consistent with previous experiments. Further research
to improve the model is discussed such as on radiation modeling and the criteria of glass breakage
and fallout.
& 2008 Elsevier Ltd. All rights reserved.

Keywords:
Glass breakage
CFD
Compartment re

1. Introduction
In a built environment, window glass breaking during a re
poses potential danger to occupants safe evacuation, and
challenges re ghting activities and the protection of structural
integrity. As window glass breakage and fallout create additional
vent openings, this essentially increases air and oxygen supply to
the re inside the enclosure, causing such phenomenon as back
draft [1]. For a ventilation-controlled re, the result would be a
sudden jump in heat release rate and a rapid spread of re when
fuel vapor mixes with incoming oxygen at the right temperature.
The research work on glass breakage due to re started with
the experiments in Harvard University [2]. The theoretical
analyses by [3,4] suggested the temperature difference as the
cause of glass breakage (crack) and subsequent fallout. Joshi and
Pagni [5] derived an analytical model for glass breakage, the
BREAK1 algorithm [6]. They also quantied through experiments
that the glass breaking stress follows a Weibull distribution. This
analytical model was later extended to double panes [7]. In the
mean time, the experiments by Skelly et al. [8] demonstrated the
effects of window assembly on glass breakage during a controlled
compartment re. They tested both framed (edge protected) and
unframed (edge unprotected) glass and the results showed that
the unframed glass could sustain a higher temperature increase

 Tel.: +1 212 481 9460; fax: +1 212 481 9484.

E-mail address: kai.kang@jacobs.com


0379-7112/$ - see front matter & 2008 Elsevier Ltd. All rights reserved.
doi:10.1016/j.resaf.2008.09.002

than what the framed glass does. Another study by Strege et al. [9]
examined re induced failure of polycarbonate windows. In a
recent experiment [10], the performance of multi-pane glazing
due to radiant heat ux has been tested at small, medium and
large-scale samples, and the radiant transmittance through the
glazing assembly was examined.
Along with experiments and theoretical analyses, numerical
modeling of glass breakage has been explored extensively for both
zone models and eld models. Sincaglia and Barnett [11]
presented a study on the implementation of BREAK1 glass
breakage model into a zone model. They paid special attention
to the dependence of radiative heat absorption coefcient on
wavelength, using a three-band radiation model. Parry et al. [12]
veried the three-band radiation model for the zone model
BRANZFIRE. This was largely based on the radiative heat transfer
model of [11] and the glass failure criterion of [5]. The latest
development in numerical modeling of glass breakage is by [13].
They used the glass failure temperature to determine the time
of window fallout and assumed that this failure temperature
follows a Gaussian probability distribution. The other study [14]
discussed the use of advanced radiation modeling in the
prediction of glass breakage.
One of the observations by [5] and many others is that in a re
environment, a uniform approximation to glass temperature
distribution with depth would not be valid. This is due to the
presence of the large temperature difference between the inner
and the outer surface. Therefore, the thickness of the glass should
be taken into account. It should also be noted that the interaction

ARTICLE IN PRESS
416

K. Kang / Fire Safety Journal 44 (2009) 415424

Nomenclature

t
x

b
E
T
h
C
cp
L
l
I
Re
Pr
q
S

dimensionless factor
coefcient of thermal expansion (1/1C)
Youngs modulus (Pa)
temperature (K)
coefcient of convective heat transfer (w/m2 K)
constant
specic heat (kJ/kg K)
local characteristic length scale (m)
radiative decay length in glass (1/m)
incident radiative heat ux (W/m)
Reynolds number
Prandtl number
total heat ux (W/m2)
glass pane thickness (m)

r
e
s
Subscript
a
c
f
g
d

between re development and enclosure connement is important, especially when window glass breakage could promote the
re. Otherwise, an iterative approach would have to be taken, and
this might be impossible for a re in an enclosed environment
that has many windows of different geometric and material
congurations. Therefore, the prediction of either the glazing
behavior or the re development necessitates both be modeled
together. Using the re dynamics simulator (FDS), the objective of
this study is to integrate the predictive capability of window glass
breakage with re modeling. In this paper, the model development is rst described and a validation study with data from a
compartment re experiment [15] is discussed.

2. Description of model development


2.1. Criterion of window glass breakage and fallout
The criterion of window glass breakage follows the same of [5]
and is the same as that used in BREAK1. As shown in Eq. (1), it
assumes that glass breakage occurs when the re heat induced
thermal stress exceeds the allowable glass breaking stress, s 4 sb.

s f bEDT

time (second)
coordinate
difference
density (kg/m3)
surface emissivity
StefanBoltzmann constant (5.67  108 J/m2/s/K4)

(1)

where the factor f is to maintain over-all force balance in the


pane and takes into account glass half height and glass thickness.
If it is assumed that the shaded edge of the glass remains near the
initial temperature, the temperature change DT can be calculated as the increase in the average temperature of the pane. This
was implemented in BREAK1 [6].
As glass breakage (crack) does not necessarily mean glass
fallout, additional criterion would be needed. For window glass
fallout, this study uses the same approach as proposed by [16],
which is based on the number of successive breakages. In a
probabilistic analysis of glass behavior due to re exposure, [16]
prescribed a xed number of ve breakages as generally indicative
of the time of glass fallout. This observation was interpreted from
the reported test data such as [15,17,18].
2.2. Implementation of window glass breakage
FDS is a special-purpose computational uid dynamics (CFD)
program for re-driven buoyant ows and incorporates mostly
used re protection methods such as the modeling of heat and
smoke detectors and sprinklers. FDS is limited to structured grids
and the numerical method of nite volume technique uses second

non-re side
convective
re side
glass
solid boundary

order spatial and temporal discretization. Turbulence ow can be


modeled by either Large Eddy Simulation (LES) or Direct
Numerical Simulation (DNS), and the re is simulated using the
mixture fraction approach.
The wall heat transfer in FDS for thermally thick boundaries
takes into account both radiative and convective heat ux [18].
The coefcient of convective heat transfer to the wall is calculated
as the maximum of natural and forced convection,


1 k
4
1
h max C DT 3 ; 0:037 Re5 Pr3
(2)
L
The constant C takes a value of 1.31 for vertical and 1.52 for
horizontal surfaces. DT is the temperature difference between the
wall cell and the neighboring gas cell. The Reynolds number, Re, is
based on the local characteristic length scale, L, which is 1 m by
default in FDS [18].
The convective heat ux is taken as the maximum between
convective and that based on temperature gradient using the
following:



DT
qc max hDT; k
(3)
DX
The boundary conditions on the re and the non-re side of
window panes are


qT
hf T f t  T0; t f sT 4f t  sT 4 0; t q0; t
(4)
k
qx x0

k


qT
ha T a t  TS; t a sT 4a t  sT 4 S; t qS; t
qx xS

(5)

The one-dimensional heat conduction, including a radiative decay


term through the depth of the glass pane, is modeled as [5]


qT
q
qT
ex=l
k
It
(6)
rg cp
qt qx
qx
l
where in the source term, I(t) is the incident heat ux and l is the
decay length in the glass pane. The heat conduction through other
solid interface is solved as follows:


qT
q
qT
k
(7)
rd cp
qt qx
qx
Numerically, the above heat conduction equations are solved
using implicit CrankNicholson nite difference method over 20
grid points across the glass panes and other solid interface. This
discretization is based on the actual dimension of the specied
thickness and is calculated separately from the model grid [18]. In

ARTICLE IN PRESS
K. Kang / Fire Safety Journal 44 (2009) 415424

addition, the following modications and functions were implemented as part of the glass breakage and fallout algorithm:

 A special type of model boundary object is programmed to

represent the window glass. This boundary object allows the


user to specify each individual window pane directly. Such
specication needs geometric input info including window
coordinates, pane orientation, window glass half height and
thickness.
Either the default material setting, or specic thermal physical
properties for each window glass can be used, such as Youngs
modulus, glass failure stress, decay length (or radiation
absorption coefcient), specic heat, thermal conductivity
and density.
The glass breakage criterion is established as temperature rise,
DT in Eq. (1) at the beginning of the simulation. During a
simulation, the glass temperature is averaged across the pane
over the front and the back side similarly to BREAK1 [6], and
the difference between the averaged temperature and the
ambient temperature is compared to DT to determine the
occurrence of glass breakage. DT is also used to determine
the time of subsequent glass breakages, for instance, 2DT for
the second and 3DT for the third breakage.
During a simulation, the number and the location of each glass
breakages is tracked. With each glass breakage, the solid cell
with the maximum temperature is identied and removed.
This creates a small opening and approximates the leakage due
to glass fracture.
The entire glass is removed when the number of glass
breakages exceeds a pre-set threshold, which is also a user
input parameter. The model default number is set as three and
used in this study.
The program saves to a le tabulating the history of the
average temperature of the front (reside) and the backside of
each window glass, as well as a time record of each breakage
and fallout.

3. Verication using a compartment re test


3.1. Description of experiment
The compartment re experiment used pool res at either the
center or the corner of the oor [15,17]. As shown in Fig. 1, the
compartment is 2.4 m wide, 2.4 m high and 3.6 m long, and has a
single door opening of 2  0.4 m. The compartment is constructed
of 100 mm thick concrete blocks. The walls include 50 mm thick
thermal insulation quilt with 12.5 mm thick plasterboard. The

2400

2400

2000

400

417

oor has a 50 mm thick thermal quilt and 6 mm thick superlux


board. The ceiling of the compartment is supported by steel
beams with 25 mm sand cement and 6 mm superlux board.
For the compartment center re, four re heat release rates
(HRRs) of 170, 280, 390, and 680 kW were examined. The re
HRRs correspond to square pan sizes of 600, 700, 800 and
900 mm. Mineralized methylated spirits (MMS), consists primarily of ethanol (C2H6O) and a small percentage of methanol
(CH4O), is used for the pool re. The smoke yield is estimated to
be 0.0072, primarily based on that of ethanol. The measurement
of HRR was based on fuel mass loss rate and oxygen depletion
calorimetry [15].
On one of the sidewalls, three pieces of window glass (Fig. 2)
are installed. The two small windows are identical and each is
845 mm wide and 845 mm high, with pane 1 at the top and pane 2
at the bottom. The other large window (pane 3) is 1897 mm high
and 845 mm wide. All glass panes are ordinary 6 mm thick oat
glass with 3 mm clearance around the frame to accommodate
expansion. The glass surface temperature, the adjacent gas
temperature, and heat ux were measured at the perimeter and
the center of each window. Fig. 2 indicates the approximate
locations of thermocouples and ux meters, with each thermocouple location numbered as A to Q and each ux meter
from 1 to 6.
3.2. Scenarios and CFD model
Fig. 3 shows the re curves, which are re-produced as model
input from the experiment measurements. The duration of each
test case varies from 18 to 35 min. In the model, the re is
considered as a pre-dened surface ux from the pan. The model
consists of compartment interior and a small extension of the
space outside the door and the windows. Given the multi-layered
construction of the walls, ceiling and oor of the compartment,
the material properties such as thermal conductivity and specic
heat capacity were mass weighted for an equivalent total
thickness of 100 mm. Typical thermal properties of plasterboard,
thermal insulation (glass ber), concrete, cement and glass were
taken from data in the literature. The aggregated values of
specic heat, density and thermal conductivity are 0.84 kJ/kg K,
2049 kg/m3, 0.064 W/m K for the walls; 0.84 kJ/kg K, 1987 kg/m3,
0.066 W/m K for the oor; and 0.83 kJ/kg K, 2295 kg/m3,
0.38 W/m K for the ceiling.
Thermocouples are used to monitor the surface temperature
and the gas temperature. The monitoring locations are estimated
from the experiment (Fig. 2). A number of different grid size were
tested at the beginning of the study, varying from 28,800 to
110,592 cells. The results presented are based on 6 cm grid size or
about 100,000 cells. Initial results showed that further decrease in
grid size had little improvement in comparison to the experimental results. In addition, the parameters of the radiation model
such as the number of solid angles, were varied and the results
showed that the default setting was sufcient. Based on nite
volume method, the radiation model by default uses 100 discrete
solid angles, and the radiation calculation is performed every
three time steps during a simulation. The computational effort of
radiation modeling has been estimated to be about 15% of the
total calculation time [18].

4. Results

360

Fig. 1. Description of compartment re test [15].

For each re HRR, simulation is rst conducted without the


glass breakage and fallout algorithm. The same case is then
repeated with the algorithm. This is to conrm that the numerical
results are consistent before and after the model implementation.

ARTICLE IN PRESS
418

K. Kang / Fire Safety Journal 44 (2009) 415424

3600
844

PANE 3

2400

1895

422

844

PANE 1

844

422

844

PANE 2

422

422

Fig. 2. Arrangement of windows and measurement locations of thermocouples (A to Q) and ux meters (1 to 6) [15,17].

300

140
120
100
80
60
40
20
0

Heat Release Rate (kW)

Heat Release Rate (kW)

160

250
200
150
100
50
0

10

15
25
20
Time (min)

30

35

40

450

800

400

700
Heat Release Rate (kW)

Heat Release Rate (kW)

180

350
300
250
200
150
100

10

15
25
20
Time (min)

30

40

35

600
500
400
300
200
100

50

0
0

10

15
20
Time (min)

25

30

8 10 12
Time (min)

14

16

18

20

Fig. 3. Fire heat release rate as a function of time [15] for (a) 600 mm pan (170 kW); (b) 700 mm pan (280 kW); (c) 800 mm pan (390 kW) and (d) 900 mm pan (680 kW).

ARTICLE IN PRESS
K. Kang / Fire Safety Journal 44 (2009) 415424

250

250
Location of Crack
Pane Center

200

FDS Predictions (deg-C)

FDS Predictions (deg-C)

419

150

100

50

Location of Crack
Pane Center

200

150

100

50

0
0

50
150
200
100
Experimental Measurements (deg-C)

250

50
100
150
200
Experimental Measurements (deg-C)

250

350
Location of Crack
Pane Center

Location of Crack
Pane Center

350
FDS Predictions (deg-C)

FDS Predictions (deg-C)

300
250
200
150
100
50

300
250
200
150
100
50
0

0
0

50

100

150

200

250

300

350

Experimental measurements (deg-C)

50

100

150

200

250

300

350

400

Experimental measurements (deg-C)

Fig. 4. Comparison of glass surface temperature between numerical and experimental results; shown with 710% and 725% reference lines for (a) 600 mm pan (170 kW);
(b) 700 mm pan (280 kW); (c) 800 mm pan (390 kW); and (d) 900 mm pan re (680 kW).

The following presents the numerical results and discusses the


comparison with the experimental data.
4.1. Predictions without glass breakage and fallout
Fig. 4 shows the predicted temperature at the location of
glass breakage and the pane center. The data points are plotted
against the experimental measurements [15]. The experimental
results are indicated by the horizontal axis and the numerical
results are by the vertical axis. Any point that falls on the diagonal
line means complete agreement with the experiment; any data
point in the upper triangle suggests over-prediction by the
numerical model and any in the lower suggests under-prediction.
As an indication on how well is the agreement, two control lines
with relative error of 710% and 725% are shown in each plot. The
time associated with each data points is not included but can be
found in [15].
A number of observations can be made for Fig. 4. The
numerical results at the pane center seems to be more consistent
with the experimental results, as they are less spread than those
at the edge of the window pane (location of crack). Secondly, the
temperature predictions at glass crack locations are generally
higher than the experiment measurements, as most data points
fall into the upper triangle. In addition, the discrepancy between

the numerical and the experimental results increases as HRR


increases from 170 to 680 kW. Thirdly, it shows that the
comparison at the lower temperature range of 100 1C or
below is not signicantly different from those at the higher
temperature range. However, the numerical results at the two
smaller HRRs are under-predicted, while those at the other two
higher HRRs are over-predicted. In general, the differences are
typically within 1025%, except slightly higher at the highest
HRR of 680 kW.
4.2. Predictions with glass breakage and fallout
Fig. 5 is a repeat of all the cases in Fig. 4 but with the glass
breakage algorithm. Note that the predicted temperatures at the
pane center are consistent with those in Fig. 4. This is as expected
since pane center would not be affected much by glass breakage.
For the 600 mm pan re, the predicted temperatures near 50 1C at
either the pan center or the location of crack are lower than the
experimental results; the agreement is slightly better for the
800 mm pan re than those in Fig. 4. For the 900 mm pan re,
the results are generally higher than the experiment. It should
also be noted that on average, the model predicted temperatures
are lower than those without the glass breakage algorithm (Fig. 4)
for all cases except the case of 900 mm pan re.

ARTICLE IN PRESS
420

K. Kang / Fire Safety Journal 44 (2009) 415424

250
Location of Crack
Pane Center

200

FDS Predictions w/ Window


Break (deg-C)

FDS Predictions w/ Window


Break (deg-C)

250

150

100

50

150

100

50

0
0

50
100
150
200
250
Experimental Measurements (deg-C)

350
300

200
50
100
150
Experimental measurements (deg-C)

250

400

Location of Crack
Pane Center

FDS Predictions w/ Window


Break (deg-C)

FDS Predictions w/ Window


Break (deg-C)

Location of Crack
Pane Center

200

250
200
150
100
50
0

Location of Crack
Pane Center

350
300
250
200
150
100
50
0

50

100

150

200

250

300

350

50

Experimental measurements (deg-C)

100

150

200

250

300

350

400

Experimental measurements (deg-C)

Fig. 5. Comparison of glass surface temperature between numerical and experimental results; shown with 710% and 725% reference lines for: (a) 600 mm pan (170 kW);
(b) 700 mm pan (280 kW); (c) 800 mm pan (390 kW); and (d) 900 mm pan re (680 kW).

4.3. History of local glass and local gas temperature

350
Gas Temperature A
Gas Temperature K
Gas Temperature M

300
Temperature (C)

The predicted local glass and gas temperature are shown in


Figs. 69 for HRRs of 170 and 680 kW. For the 600 mm pan re,
Fig. 6 shows the exposed local gas temperature at locations A,
C, K, L, M and N, and Fig. 7 shows the exposed local glass
temperature. Figs. 8 and 9 shows the results for the 900 mm pan
re. The referenced location names are as indicated in Fig. 2. Note
that local glass temperature refers to the glass or solid surface
temperature as locally measured, local gas temperature
refers to the adjacent air temperature as locally measured, and
exposed refers to the re side as opposed to the non-re side.
Note that the selected locations A, C and N are in the upper
layer region and K, L and M are in the lower region of the
compartment.
The predicted local gas temperature history is in qualitatively
good agreement with the experimental data. Both the experimental [15] and the numerical results show that the peak
temperature is between 200 and 250 1C at the upper layer, and
is around 50 1C in the lower region. Note that the gas temperature
history follows closely the HRR curve (Fig. 3a). The local gas and
glass temperature are similar in trend but different in absolute
values. As shown, the glass surface temperature in the upper
region are generally lower than the gas temperature, whereas
those in the lower region are slightly higher. This agrees with
the experimental results and indicates that the predominate

Gas Temperature C
Gas Temperature L
Gas Temperature N

250
A
N
C

200
150
100

K
M
L

50
0
0

10

15

20

25

30

35

40

Time (min)
Fig. 6. Enclosure local gas temperature history at selected locations for 600 mm
pan re.

mechanism of heat transfer to the glass is different between the


upper and the lower region. Heat transfer to the glass is likely predominated by radiation in the lower region and that in the upper
by convection.

ARTICLE IN PRESS
K. Kang / Fire Safety Journal 44 (2009) 415424

350

700
Glass Temperature A

Glass Temperature K

Glass Temperature L

Glass Temperature M

Glass Temperature N

Glass Temperature C

Glass Temperature A

600

250
A
N
C

200
150
100

M
K
L

50

Glass Temperature M

Window 1 Fall Out

Temperature (C)

Temperature (C)

300

500

Glass Temperature N

A
N
C

400

Glass Temperature L
Glass Temperature C
Glass Temperature K

300
200

K
M

100

0
0

10

15

20
25
Time (min)

30

35

40

Fig. 7. Exposed local glass temperature history at selected locations for 600 mm
pan re.

Window 1 Fall Out

Gas Temperature N
Gas Temperature M

Gas Temperature L

Gas Temperature C

400

Gas Temperature K

300
200

K
M

100

8
10
Time (min)

12

14

16

18

Heat Flux #1
Heat Flux #2
Heat Flux #3
Heat Flux #4
Heat Flux #5
Heat Flux #6

5
Heat Flux (kW/m2)

500

Gas Temperature A

600

Fig. 9. Exposed local glass temperature history at selected locations for 900 mm
pan re.

700

Temperature (C)

421

5
2
1
3

4
3

6
4

2
1

0
0

8
10
12
Time (min)

14

16

18

10

15

20

25

30

35

40

Time (min)

Fig. 8. Enclosure local gas temperature history at selected locations for 900 mm
pan re.

Fig. 10. History of predicted gauge heat ux for 600 mm pan re.

For the 900 mm pan re, the window fallout is predicted at


about 260 s after ignition, whereas it was at approximately 7.5 min
(450 s) in one of the tests [15]. The predicted temperature
development is consistent with the experiment before the
window fallout. For example, for location A, both the experimental and the numerical results show that the peak gas
temperature is approximately 550 1C at 4 min, and the peak glass
temperature is near 500 1C. Again in the lower region, the higher
glass temperature than the gas temperature indicates predominately radiative heat transfer to the glass. Note that the sudden
drop in temperature is most likely due to the sudden inow of
cool air when the window glass fallout. Despite the difference in
the time of window fallout, the temperature change after window
fallout is similar to the experiment, for example, both results
suggest that the glass temperature at the location A changed
from approximately 500 to 200 1C within 8 min.

pan re, the heat ux at all the locations except #5 converges after
the window fallout. Of all the locations, #4 and #6 consistently
have the lowest heat ux, which is the same as the experiment.
The location of the highest heat ux is, however, different from
the experiment, for example, the model result indicates location
#5 instead of #3 for the 600 mm pan re. The peak heat ux as
calculated is approximately 4.5 and 20 kW/m2, respectively.
In comparison, the experimental results showed the peak heat
ux as 10 and 50 kW/m2. This is probably because of the nite
spatial discretization used by the radiation model and the size
of the measuring point, which might have missed the point
evaluation.

4.4. History of heat ux


The prediction of wall heat ux at the prescribed locations (Fig.
2) is also evaluated. Figs. 10 and 11 show the results for the
600 mm and the 900 mm pan re, respectively. For the 900 mm

5. Time of glass breakage and fallout


Fig. 12 shows the model predictions of the rst three
occurrence of glass breakage, regardless of which window pane,
and compares with the experimental results. The experimental
data are presented as the average and one standard deviation of
multiple test results for each case. For the rst three cases, the
predicted times of glass breakage are within the range of the test
data; whereas in the case of 900 mm pan re, the times of glass

ARTICLE IN PRESS
422

K. Kang / Fire Safety Journal 44 (2009) 415424

breakage are generally earlier. As a result, the time of window


fallout is 259 s (window pane 1). The associated temperature with
window glass fallout is around 450500 1C (Fig. 9), and is
consistent with the previous experiments by [15] and others. A
summary of the window fallout time is shown in Table 1. In the
experiment, the earliest time is 398 s for pane 1 when more than
50% of the glass failed and is 232 s for pane 3.

6. Discussions
In this paper, a model to predict glass breakage and fallout due
to re exposure was developed within the FDS program. The
criteria of glass breakage [5] and window fallout [16] are
implemented, and the modication of FDS took into account
one-dimensional heat conduction and radiative decay through the
glass. As a validation exercise, the numerical results are compared
to the experimental data of a compartment re by [15] and four
cases are examined with peak re HRRs of 170, 280, 390 and
680 kW. In general, good agreement with the experiment has been
obtained: Quantitatively, the predicted glass surface temperatures

25

Heat Flux (kW/m2)

20

Heat Flux #1
Heat Flux #2
Heat Flux #3
Heat Flux #4
Heat Flux #5
Heat Flux #6

2
3
1
5

15

10

Table 1
Comparison of time of more than 50% window glass fallout

6
4

Experiment [15]

0
2

8
10
Time (min)

12

14

16

18

Model w/glass
breakage
a

Fig. 11. History of predicted gauge heat ux for 900 mm pan re.

900

Time of Crack (Seconds)

Time of Crack (Seconds)

Peak re HRR
(kW)a

Time
(s)

Window pane
no.

800  800
800  800
900  900
900  900
900  900
900  900
900  900
900  900

390
390
680
680
680
680
680
680

336
726
398
650
450
505
232
259

3
3
1
1
3
3
3
1

Time based on more than 50% of the glass fallout in the experiments.

500

1000
Experiments
FDS

800
700
600
500
400
300

Experiments
FDS

450
400
350
300
250
200

200
0

2
3
Sequence of Crack

300

2
3
Sequence of Crack

160
Experiments
FDS

250

Time of Crack (Seconds)

Time of Crack (Seconds)

Pan size
(mm  mm)

200
150
100

Experiments
FDS

140
120
100
80
60

50
0

2
3
Sequence of Crack

2
3
Sequence of Crack

Fig. 12. Comparison of the rst three occurrence of glass breakage between numerical and experimental results; for (a) 600 mm pan (170 kW); (b) 700 mm pan (280 kW);
(c) 800 mm pan (390 kW); and (d) 900 mm pan re (680 kW).

ARTICLE IN PRESS
K. Kang / Fire Safety Journal 44 (2009) 415424

are largely within 1025% of the experimental data; and


qualitatively, the predicted gas and glass temperature proles
compared well with the test results. It is noted that the
discrepancies from the experimental results increases as re
HRR increases.
It should also be noted that glass is a (semi)-transparent
material for radiation and exhibits spectra-dependency on
radiative absorption coefcient, that is, from nearly transparent
up to wave length of 2.75 mm to almost completely opaque when
wave length is over 4.5 mm. In addition, as shown in the
experiment by [10], a 20 kW/m2 incident heat ux could result
in as high as 20% transmittance for a 305  305 mm pane of
double glazing. The transmittance would be higher for single
glazing. The current model simply considered the radiative decay
as a source term in the heat conduction equation (Eq. (6)), a more
appropriate treatment would have to couple radiative transmittance through the glass boundary with radiation modeling. This
has been implemented in the latest FDS program that was not
available at the time of this study. However, despite the
simplications in the current study, the model predictions are
generally within good agreement and reasonable range of the
experimental results, especially considering the many factors of
the experiment that were estimated such as the properties of
window glass, the exact monitor locations and the exact window
locations relative to the wall (Fig. 2).
The yield stress of glass is set at 47 MPa in this study. This is
probably the pre-dominate factor in the glass breakage criteria
(Eq. (1)) It has been reported that the glass breaking stress vary
from 40 MPa to more than 100 MPa from many studies. Based on
the tests data from 59 samples, Joshi and Pagni [19] reported that
the distribution of the breaking stress correlates well with a threeparameter cumulative Weibull distribution. Considering the
entire range of the suggested Weibull distribution, the relative
error due to breaking stress is estimated approximately 38%. It can
be estimated that this translates, along with errors from other
parameters, such as shading thickness and thermal expansion
coefcient, into a combined uncertainty of nearly 40% in the glass
breakage criterion. Consequently, the predicted time of glass
breakage (Fig. 12) should also have been associated with an error
bar similarly to that of the experiment. It is expected that the
uncertainty on glass breakage criteria alone may be equivalent
with or have exceeded the typical discrepancy between the
numerical and the experimental results (e.g., Fig. 5). Therefore,
further improvement to the numerical model would require much
more detailed and denitive information from the experiment,
especially on glass performance.
In practical applications of such model to an environment with
multiple windows, a randomized procedure may be used for glass
breaking stress, such as random sampling of a truncated Weibull
distribution, to determine the breaking temperature criterion. The
current model can be easily modied to incorporate such
technique. For glass fallout prediction, the approach is based on
the well-established theory of glass breakage, and the selection of
three breakages for glass fallout could be conservative. However,
there is limited number of test data for validation, especially on
the number of breakages. The test data of [8] provided the time of
crack initiation and catastrophic window failure, which would
be useful in further model validation. If catastrophic window
failure can be interpreted as glass fallout, the required number
of breakages for glass fallout may also be estimated using the
compartment gas temperature. Further research is still needed on
the mechanism and the model prediction of glass fallout. In
addition, since the initial model development was before the
release of the latest version of FDS, the effect of radiation heat
transfer through glass would need to be re-evaluated, especially
when re HRR is high, such as in the case of the 900 mm pan re.

423

Currently, the model is being ported into the latest version of FDS
and it is expected that the model will eventually be developed
more as an add-on module. For either probabilistic or
deterministic approaches, further studies also need to be undertaken for model validation.

7. Conclusion
This paper describes a model development for window glass
breakage and fallout predictions in FDS. A specic object type was
developed for window glass and the associated heat transfer was
modeled to determine glass breakage that is similar to the
BREAK1 algorithm. The glass fallout is based on an assumed
number of three breakages. For model validation, the gas and the
glass temperature, and the heat ux in a compartment center re
are compared between the numerical and the experimental
results. Four cases are studied with different pan size of 600,
700, 800 and 900 mm, corresponding to the peak re size of 170,
280, 390 and 680 kW, respectively. In general, good agreements
have been demonstrated quantitatively and qualitatively. At
discrete monitoring locations around the window glass, the
predicted glass surface temperature and the adjacent gas
temperature are shown typically within 1025%, with the
discrepancy increasing slightly as the heat release rate increases.
The time histories of the gas and the glass surface temperature at
selected monitoring locations are compared well with the
experiment. It is shown that the rst three occurrences of glass
breakage are within reasonable ranges of the experimental
observation. The discrepancy in the predicted heat ux, especially
for the 680 kW case, may be a numerical issue; however, together
with the earlier predicted glass fallout would need further
research. The glass temperature at window fallout is around
450500 1C and is consistent with previous tests. Given the
uncertainty in the glass material properties and hence the
breaking criterion, detailed information on the performance of
window glass when exposed to re would be needed to improve
the model predictive capabilities. Currently this model is
being transferred to the latest version of FDS and developed as
an add-on module, with further validation and sensitivity
studies being undertaken.
References
[1] D. Dysdale, An Introduction to Fire Dynamics, second ed., Wiley, 2009.
[2] H.W. Emmons, Window glass breakage by re, Home Fire Project Technical
Report No. 77, Havard University, Cambridge, MA, 1988.
[3] O. Keski-Rahkonen, Breaking of window glass close to re, Fire Mater. 12
(1988) 6169.
[4] O. Keski-Rahkonen, Breaking of window glass close to re, II: circular panes,
Fire Mater. 15 (1991) 1116.
[5] A.A. Joshi, P.J. Pagni, Fire-induced thermal elds in window glass, I: theory,
Fire Saf. J. 22 (1994) 2543.
[6] A.A. Joshi, P.J. Pagni, Users Guide to BREAK1, The Berkeley Algorithm
for Breaking Window Glass in a Compartment Fire, NIST-GCR-91-596,
1991.
[7] B.R. Cuzzillo, P.J. Pagni, Thermal breakage of double-pane glazing by re,
J. Fire Prot. Eng. 9 (1) (1998) 111.
[8] M.J. Skelly, R.J. Roby, C.L. Beyler, An experimental investigation
of glass breakage in compartment res, J. Fire Prot. Eng. 3 (1) (1991)
2534.
[9] S. Strege, B.Y. Lattimer, C. Beyler, Fire induced failure of polycarbonate
windows in railcars, in: Proceedings of re and materials 2003, Interscience
Communications Ltd.,2003, pp. 269278.
[10] M.S. Klassen, J.A. Sutula, M.M. Holton, R.J. Roby, T. Izbicki, Transmission
through and breakage of multi-pane glazing due to radiant exposure, Fire
Technol. 42 (2006) 79107.
[11] P.E. Sincaglia, J.R. Barnett, Development of a glass window fracture
model for zone-type computer re codes, J. Fire Prot. Eng. 8 (3) (1997)
101118.
[12] R. Parry, C.A. Wade, M. Spearpoint, Implementing a glass fracture module in
the BRANZFIRE zone model, J. Fire Prot. Eng. 13 (2003) 157183.

ARTICLE IN PRESS
424

K. Kang / Fire Safety Journal 44 (2009) 415424

[13] N. Pope, C. Bailey, Development of a gaussian glass breakage model within a


re Field Model, Fire Saf. J., 2007, doi: 10.1016/j.resaf.2006.12.005.
[14] S. Dembele, R.A.F. Rosario, J.X. Wen, K.S. Varma, S. Dale, M. Wood, P.D. Warren,
Study of glazing behaviour in re conditions using advanced radiation heat
transfer model and computational uid dynamics, Presented at the 7th AsianOceania Symposium for Fire Safety Science & Technology, Hong Kong,
September 2007.
[15] T.J. Shields, G.W.H. Silcock, M.F. Flood, Performance of a single glazing assembly
exposed to a re in the centre of an enclosure, Fire Mater. 26 (2002) 5175.

[16] J. Hietaniemi, Probabilistic simulation of glass fracture and fallout in re, VTTWork-41, VTT Building and Transport, Finland, 2005.ISBN:951-38-6593-2
[17] T.J. Shields, G.W.H. Silcock, M.F. Flood, Performance of a single glazing
assembly exposed to enclosure corner res of increasing severity, Fire Mater.
25 (2001) 123152.
[18] K. McGrattan, Fire Dynamics Simulator Technical Reference Guide, v4, NIST
SP1018, March 2006.
[19] A.A. Joshi, P.J. Pagni, Fire-induced thermal elds in window glass, II:
experiments, Fire Saf. J. 22 (1994) 4565.

You might also like