481

A publication of

CHEMICAL ENGINEERING TRANSACTIONS

The Italian Association
of Chemical Engineering
Online at: www.aidic.it/cet

VOL. 31, 2013

Guest Editors: Eddy De Rademaeker, Bruno Fabiano, Simberto Senni Buratti
Copyright 2013, AIDIC Servizi S.r.l.,
ISBN 978-88-95608-22-8; ISSN 1974-9791

Analysis of a Propane Sphere BLEVE

Albrecht Michael Birka, Gilles Dusserreb, Frederic Heymesb
a

Department of Mechanical and Materials Engineering, Queens University, Kingston, Ontario, Canada
Ecole des Mines DAles, Institute of Risk Science, Ales, France
[email protected]
b

The events in March of 2011 in Japan have provided us with an example of what can happen when large LPG
spheres are exposed to severe fire engulfment. This paper presents an approximate failure analysis of a
sphere exposed to a severe fire engulfment scenario. It estimates the expected failure time of the sphere
based on predicted wall temperatures and internal pressure buildup. The prediction of the pressurization rate
is based on a two-zone, thermally stratified liquid model. It shows that the pressurization rate depends strongly
on the initial fill level. At high fill levels, the sphere can pressurize much faster then would be expected if the
liquid were well mixed and isothermal. This rapid pressurization can lead to early failure.
The analysis considered two fire heat flux conditions and shows how prediction of failure time can be affected
by high temperature stress rupture. The analysis also considers various initial fill conditions to predict BLEVE
hazards at failure. The results are in reasonable agreement with limited observations and data from the March
2011 incident in Japan.
Introduction
In March of 2011 a magnitude 9.0 earthquake and subsequent tsunami devastated the Tohoku area of Japan.
This disaster included some of the largest BLEVE type failures of LPG spheres ever recorded, at the Chiba
refinery near Tokyo (Koseki, 2011). Fireball diameters of greater than 200 m were recorded on video by
helicopter based news cameras. This paper presents an analysis of a large propane sphere exposed to partial
fire engulfment to give some idea of expected failure times and hazard potential.
Time to Failure for Spheres Exposed to Fire
The time to failure of a pressure vessel in a fire is driven by the fire heat flux, the fire exposure area, and the
vessel fill level. The failure is caused by a combination of high internal pressure and wall degradation due to
high wall temperatures. To get rapid pressurization the fire must impinge the liquid wetted wall. For high wall
temperatures, the fire must impinge the vapour wetted wall well above the liquid level. These processes have
been seen in many fire tests of pressure vessels exposed to fire - see for example (Townsend et al., 1974,
Balke et al., 1999, Birk et al., 1997, Moodie et al., 1988, Droste and Schoen, 1988, Appleyard, 1980)). All of
these tests involved cylinders of various L/D ratios. For spheres the process will be the same but there are
some differences such as:
i) Scale large cylinders may have diameters of 3-5 m and L/D ratios of 6 or greater, where large
spheres may have diameters of 20 m or larger. Wall thicknesses on large spheres can exceed
70 mm. This means longer time scales for heating of the wall and lading (see for example (Birk,
1995)).
ii) Different volume to surface area ratio (ratio of total sphere volume to the thermal boundary layer
volume) for sphere vs cylinder.
iii) Lower stress (sphere vs cylinder).
The different scale and volume to surface area ratio will affect how the vessel pressurizes when exposed to
fire. The large wall thicknesses will affect the time it takes to heat the steel wall to dangerous temperatures.
Fire Conditions
From the limited available photographs it appears the sphere BLEVE in Japan in March 2011 was partially
engulfed by a massive jetting liquid propane fire from a failed pipeline beside the spheres. The fraction of

482

engulfment was estimated to be approximately 20 % based on the photographs posted on the internet. The
engulfment appeared to cover part of the bottom and top of the sphere.
2
API 521 states that jet fires can have heat fluxes from 100-400 kW/m and large unconfined pool fires can
2
have heat fluxes of 100-250 kW/m . For the present analysis we will assume modest heat fluxes of 87 - 181
2
kW/m . API 521 recommends the following fire heating conditions for sizing PRVs for large spheres.
i) Fire heat input based on liquid wetted wall area A up to the equator of the sphere.
0.82
for bare steel without adequate drainage and no firefighting
ii) Heat transfer to liquid Q = 70.9 A
equipment where Q is in kW and A is in m2.
3
For a 15.6 m diameter sphere (2000 m ) we calculate a heat transfer rate to the liquid of 9325 kW based on
this API 521 approach. We have also estimated the heat transfer as a function of sphere fill level based on the
following assumptions:
i) Heat transfer to liquid based on liquid wetted area
2
ii) Average heat flux to liquid wall 150 kW/m
Table 1 gives a summary of this calculation. As can be seen in the table the estimated heat into the sphere
liquid exceeded 9325 kW for all fill levels above 30%. If the PRVs were sized based on the API 521 formula it
is possible the PRVs were undersized for the accident.

Predicting Time to Failure

To predict time to failure we need to predict the wall temperature in the vapour space and the sphere
pressure. With these we can determine the stresses and material strength. The pressure will rise at a different
rate than the wall temperature will. The failure will require high wall temperature and high pressure at the
same time. The pressure will rise to the PRV set pressure and then the PRV will open. The PRV full flow
capacity is reached at 120 % of the PRV set pressure. Here we have assumed a tank operating pressure of 2
MPa. This assumes the PRV has been properly sized for this fire condition.
Table 1 : Estimated Heat Transfer to Liquid as a function of fill level (assuming exposure is 20% and fire heat
2
flux is 150 kW/m ).
Fill Level

0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8

Heat In (kW)
20% exposure
fire Q = 150 kW/m2
6,720
8,291
9,484
10,529
11,515
12,501
13,546
14,739

Heat In (kW)
API 521 for PRV sizing

9,325

Peak Wall Temperature vs Time

The prediction of the vapour space wall temperature depends on the fill level and on the fire conditions. We
have assumed the following:
2
o
i) fire heat flux of 87 or 181 kW/m to a cool surface ( i.e. 871 and 1100 C black body fire)
ii) sphere liquid fill 50%
iii) surface emissivity = 0.9
iv) wall thickness 52 mm
v) internal convection coefficient 10 W/m2 K
vi) internal wall sees hot vapour space wall and cool liquid surface
With these assumptions we predict wall temperature rise rates as shown in Figure 1.
Tank Pressure vs Time
Predicting the tank pressure rise rate is complicated by the fact of liquid temperature stratification with high fill
levels. This stratification causes the pressure to rise much more rapidly than it would if the liquid was
isothermal (Birk and Cunningham, 1996). Here we have used a simplified liquid stratification model. The
following assumptions have been used.

483

Figure 1: Predicted Vapour Space Wall Temperature for a Sphere

i)

Lading is divided into two zones.

a. Saturated zone includes the liquid boundary layer heated by the fire exposed wall and the
vapour space.
b. Subcooled zone is the remaining liquid in the core
ii) The liquid boundary layer thickness is a constant fraction of the tank diameter
iii) All the fire heat goes into the boundary layer where the liquid wetted wall is exposed to fire. This heat
input increases the internal energy of the heated boundary liquid and vapour space
iv) The core liquid has no heat input and therefore it remains at the initial temperature
v) The 2 zones are heated at constant total volume which is equal to the vessel volume
The heat that enters the liquid boundary generates the vapour that pressurizes the vessel. In reality some heat
does mix into the core liquid by free convection, but this is relatively small when the PRV is closed (Birk,
1983). The mixing between the boundary layer and core increases after the PRV opens because boiling
causes strong buoyancy driven convection currents (Birk and Cunningham, 1996). The model used here is
only appropriate for the time the PRV is closed.
3
Figure 2 shows the predicted time to reach 2 MPa for a North American Rail Tank car (125 m ). This vessel
measures 3 m in diameter and approximately 18 m in length. The plot shows the time to reach 2 MPa for a
range of fill levels and fire exposure fractions (exposure = fraction of liquid wetted wall covered by fire = 0.1,
0.2, 0.5 and 0.7) for the case where the liquid is stratified assuming the boundary layer is 0.075 of the tank
diameter. Two curves are also shown assuming the liquid is isothermal for two fire exposure conditions of 0.1
and 0.7. We have included two data points from two full scale fire tests of rail tanks cars. The first is from
3
(Townsend et al., 1974) and involved an unprotected 125 m rail tank car fill to 95% full with propane. The
3
second is from (Balke et al., 1999) and involved a 45 m rail tank filled to 22% with liquid propane. Further
details can be found in Table 2. As can be seen both models do a reasonable job of predicting the
pressurization rate for the low fill level tank. This is because at low fills the boundary layer is most of the liquid
volume. However at high fill levels the unheated core liquid is the largest fraction of the liquid volume and
therefore stratification must be considered when calculating the pressurization rate.

Table 2: Full Scale Rail Tank Car Fire Tests, Observed Pressurization Times
Tank

Volume
3
(m )

Initial Fill

RAX 201
BAM 1999

125
45

95%
22

Failure
Pressure
(MPa)
2.5
2.5

Time to
PRV
activation
2 min
14 min

Time to
Failure

Peak Wall T
o
( C)

24 min
17

650
650

3
This same model was used for the 2,000 m sphere. The model was changed to account for the different
shapes of the tank. The results are shown in Figure 3. If we consider the case of a 50 % full sphere 20 %
2
exposed to an engulfing 87 kW/m fire the stratified model predicts the PRV will be activated after about 50
min.

484

Failure Prediction
With a pressure of 2 MPa the hoop stress and von Mises stress for the assumed sphere (D = 15.6 m, wall
thickness 52 mm) was approximately 150 MPa. With this level of stress it is possible to determine the time to
failure for various vapour space wall temperatures. Figure 4 shows example high temperature stress rupture
o
data (Birk and Yoon, 2006) for TC 128 tank car steel for the temperature range from 550 720 C. This data
was obtained for constant temperature and load tests.
The time to failure for the actual sphere is determined by the accumulated stress rupture damage as the wall
is heated and stressed. Here we will use a simplified approach using Figure 4. The time to failure is
approximated by the time it takes to reach the peak wall temperature or peak pressure, plus the time for stress
rupture failure at that temperature/pressure combination.
o
Table 3 gives a summary of the failure times taken from Figure 4. For the case with the 871 C fire we expect
o
to see wall temperatures of about 650 C in about 50 min. We do not expect failure at that time because the
stress is low. However, we expect high temperature stress rupture after about 17 minutes at that condition of
o
wall temperature and stress. This suggests a tank failure time of around 67 minutes. For the 1,100 C fire the
o
wall reaches very high temperatures (700 C) in around 12 min. The PRV is estimated to activate at around 10
minutes. These conditions combine to give an estimated failure time of 12 min. This is summarized in Table 4
Time to Reach PRV Ac va on vs Fill for Cylinder
87 kW/m^2 fire, 130 m^3 D = 3 m 2 MPa set Pressure
fire exposure = 0.7 liquid stra fied
0.2 strat
0.1 strat
0.5 strat
0.7 isotherm
0.1 isothermal
RAX 201
BAM 1999

Time to Reach PRV Set P (min)

1000

100

10

1
0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Volume Fill Frac on

Figure 2: Predicted Time to PRV Activation for a Cylindrical Rail Tank Car.

Figure 3: Predicted Time to PRV Activation for Sphere

The actual failure time from news reports was of the order of 60 minutes for the sphere BLEVE in Japan. This
o
is in line with the estimate for the 871 C fire case.
Sphere BLEVE Hazards
The hazards from the sphere BLEVE depend on the fill level and the pressure and lading temperature at
failure. Here we have assumed failure at 2 MPa with saturated conditions. The results are summarized in
Table 5.

485

Figure 4: High Temperature Stress Rupture Data for TC 128 Pressure Vessel Steel.
Table 3: Summary of time to Temperature and Time for Stress Rupture.
Wall Temperature and stress
rupture time at 150 MPa stress

Time for Wall to Reach

Temperature with 871 C fire

Time for Wall to Reach

Temperature with 1100 C fire

550 oC SR fail time, no fail

600 SR fail > 200 min
650 SR fail at 17 min
700 SR fail at 1 min

22 min
28
50
temperature not achieved

8 min
10
11
12

3
Table 4 : Summary of Failure Times for 2,000 m sphere, 50 % full, 20 % exposed to engulfing fire (D = 15.6
m, 52 mm wall thickness, 2 MPa pressure, TC 128 steel)

Wall T reaches
P reaches 2 MPa
Failure Time based on
Ult stress
Failure Time based on
Stress Rupture

Fire 87 kW/m2
650 C in 50 min
50 min
failure not indicated

Fire 181 kW/m2

700 C in 12 min
10 min
12 min

failure after 17 min of stress rupture

total time 67 min

12

Fire Ball and Blast Overpressure

The fireball size and duration can be estimated from very simple correlations (see for example (Birk, 1996)).
0.333
and t = 0.075D, where m = propane mass in kg, D = fireball
The following have been used here. D = 6m
diameter in m and t = duration in seconds. One video showing a sphere BLEVE from Tokyo was available on
the internet and was reviewed. The fireball duration was approximately 26 s. This suggests a propane mass of
3
approximately 250,000 kg. If this was a 2,000 m sphere is suggests it was about 15% full of liquid at failure.
At this fill level we would expect a fireball diameter of about 350 m.
The blast overpressure produced by a BLEVE is still a subject of active research. Some data suggests that the
shock produced by a BLEVE is due to the energy in the vapour space. The liquid phase change process may
be too slow to produce a shock. Data from small scale experiments have supported this (Baker et al., 1983).
In this analysis the blast was predicted using the methods described in (Birk et al., 2007). From the following
Table 5 it is clear that the fireball hazard reaches further than the BLEVE overpressure.

486

Table 5: Summary of Calculated Hazards for 2,000 m3 Sphere BLEVE containing saturated propane (failure
pressure = 2 MPa).
Fill

mass
kg

vapour
Energy
kJ

total
energy
kJ

Distance
to 300
mbar
from
vapour
energy

Distance
to 300
mbar
from total
energy

Fireball D
(m)

Fireball
Time (s)

0.05
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8

137.E3
175.E3
250.E3
326.E3
402.E3
477.E3
553.E3
628.E3
704.E3

12.0E6
11.E6
10.E6
8.8E+06
7.6E+06
6.3E+06
5.0E+06
3.8E+06
2.5E+06

14.9E6
17.2E6
21.8E6
26.4E6
30.9E6
35.5E6
40.1E6
44.7E6
49.3E6

66.8
65.6
63.1
60.4
57.3
54.0
50.1
45.5
39.8

71.9
75.4
81.5
86.9
91.7
96.0
99.9
103.6
107.0

307.9
334
376.5
411.2
440.8
466.8
490.3
511.7
531.4

23.1
25.05
28.24
30.84
33.06
35.01
36.77
38.37
39.85

Conclusions
3
A failure analysis has been presented for a 2,000 m propane sphere. Two fire conditions were considered,
o
o
one very severe (i.e. 1,100 C fire) and another more modest (871 C). The estimated failure time for the
extreme fire was of the order of 12 min. The estimated failure time for the more modest fire was 67 min. The
actual failure time for the sphere BLEVE in Japan is believed to be of the order of one hour.
The sphere fill level for the Tokyo BLEVE was estimated to be about 15 % based on the fireball duration of 26
s. This would correlate with a fireball of about 350 m diameter. We have no data to show that the correlations
used for fireball size and duration apply to this scale of BLEVE.
References
APPLEYARD, R. D. 1980. Testing and Evaluation of the Explosafe Explosion Suppression System as a
Method of Controlling the Boiling Liquid Expanding Vapour Explosion. Transportation Development
Centre, Transport Canada.
BAKER, W. E., COX, P. A., WESTINE, P. S., KULESZ, J. J. & STREHLOW, R. A. 1983. Explosion Hazards
and Evaluation, Elsevier Scientific Publishing.
BALKE, C., HELLER, W., KONERSMANN, R. & LUDWIG, J. 1999. Study of the Failure Limits of a Tank Car
Filled with Liquefied Petroleum Gas Subjected to an Open Pool Fire Test. Berlin, Germany: Federal
Institute for Materials Research and Testing (BAM).
BIRK, A. M. 1983. Development and Validation of a Mathematical Model of a Rail Tank-Car Engulfed in Fire.
Department of Mechanical Engineering, Queen's University, Kingston, Ontario, Canada.
BIRK, A. M. 1995. Scale Effects with Fire Exposure of Pressure Liquefied Gas Tanks. Journal of Loss
Prevention in the Process Industries, 8, 275-290.
BIRK, A. M. 1996. Hazards from BLEVEs: An Update and Proposal for Emergency Responders. Journal of
Loss Prevention in the Process Industries, 9, 173-181.
BIRK, A. M. & CUNNINGHAM, M. H. 1996. Liquid Temperature Stratification and its Effect on BLEVEs and
their Hazards. Journal of Hazardous Materials, 48, 219-237.
BIRK, A. M., CUNNINGHAM, M. H., OSTIC, P. & HISCOKE, B. 1997. Fire Tests of Propane Tanks to Study
BLEVEs and Other Thermal Ruptures: Detailed Analysis of Medium Scale Test Results. Transport
Canada.
BIRK, A. M., DAVISON, C. & CUNNINGHAM, M. H. 2007. Blast Overpressures from Medium Scale BLEVE
Tests. Journal of Loss Prevention in the Process Industries, 20, 194-206.
BIRK, A. M. & YOON, K. T. 2006. High Temperature Stress-Rupture Data for the Analysis of Dangerous
Goods Tank-Cars Exposed to Fire. Journal of Loss Prevention in the Process Industries, 19, 442451.
DROSTE, B. & SCHOEN, W. 1988. Full Scale Fire Tests with Unprotected and Thermal Insulated LPG
Storage Tanks. Journal of Hazardous Materials, 20, 41-54.
KOSEKI, H. 2011. Current Large Petroleum Tank Fires by Earthquakes and Mutual Fire Fighting System.
TIEMS Workshop. Nimes, France.
MOODIE, K., COWLEY, L. T., DENNY, R. B., SMALL, L. M. & WILLIAMS, I. 1988. Fire Engulfment Test of a 5
Tonne LPG Tank. Journal of Hazardous Materials, 20, 55-71.
TOWNSEND, W., ANDERSON, C. E., ZOOK, J. & COWGILL, G. 1974. Comparison of Thermally Coated and
Uninsulated Rail Tank-Cars Filled with LPG Subjected to a Fire Environment. US DOT Repor