Fire Engulfment of LPG Storage Tanks
Fire Engulfment of LPG Storage Tanks
Fire Engulfment of LPG Storage Tanks
INTRODUCTION
Whenever highly flammable fluids such as LPG are either stored or transported
in bulk there is the risk that any leak, may, if ignited, result in the
storage vessel becoming engulfed by fire. This, in the case of LPG, may have
particularly serious consequences because it is often stored as a pressurised
liquid at ambient temperature.
Thus if the vessel is heated externally,
causing a rise in internal pressure and a loss in the mechanical strength of
its walls due to the elevated temperatures, it may fail catastrophically and
suddenly release its contents. Such a release may propel fragments of the
vessel over considerable distances. In addition, significant amounts of fluid
may flash to vapour, which if ignited can lead to fire ball behaviour with
associated releases of thermal energy.
Pressure relief valves (PRVs) are therefore fitted to tanks to vent their
contents at a pressure above the maximum working pressure, and at a rate that
will prevent further pressure build up leading to loss of integrity.
The
valves are sized, according to most codes of practice, on the assumption of a
uniform heat flux from the engulfing fire of about 100 kW/m 2 (no account being
taken of non-uniform heating effects, nor of the maximum wall temperatures
likely to be attained).
In order to make realistic assessments of the potential hazards it is
necessary to know the thermal response of storage vessels in a fire engulfment.
This is currently being investigated at the Explosion and Flame
Laboratory of the Health and Safety Executive. Firstly, the development of
theoretical models is being funded which will predict vessel behaviour for a
range of fire engulfment scenarios, as described in (1). Secondly, an
*Health and Safety Executive, Explosion & Flame Laboratory, Harpur Hill,
Buxton, Derbyshire
87
EXPERIMENTAL FACILITIES
The fire engulfment trials were carried out by supporting the tanks ever a
bund and exposing them to pool fires of kerosene contained within the bund.
Five tests in total were undertaken, two with a 1/4 tonne tank and three with
a 1 tonne tank.
The tanks were re-used after each test, but only after
carefully checking that they had not been weakened excessively in the previous
fire.
In the second 1/4 tonne test the tank failed catastrophically as a
result of a PRV malfunction.
3
The bund, which was used for all of the tests, consisted of a firebrick
enclosure some 4m long by 2.4m wide by 0.6m deep. A lm high windbreak wall
surrounded the bund at a distance of lm from it. The windbreak ensured that
flame of at least 1 m thickness engulfed the tank under normal wind
conditions.
The test tanks were supported on firebrick piers so that the
bottom of a tank was flush with the top of the pool walls. When testing the 1
tonne tanks four load cells were also fitted so that the weight of the vessel
and its contents could be recorded continuously throughout a trial.
The burning rate of kerosene in the bund was approximately 0.64 1/s.
88
Hence
TANK INSTRUMENTATION
TEST PROCEDURE
The test procedure was to connect up the instrumentation and data logging
system, fit the PRV, fill the tank to the required level, fill the bund with
kerosene, start the data recording system then ignite the fire. Typically, a
fire took up to 11/2-2 minutes to become fully established, after which the
vessel contents would begin to heat up. The PRV's then opened within the
following 2-5 minutes, depending on vessel size, fill level and initial
temperature.
In most tests successive PRV venting then occurred until the
tanks were emptied of LPG and the vapour pressure fell below the relief
pressure.
6
89
Two tests were undertaken with this size of tank, both 40% full of propane
(200 1 ) , but only one test was completed without damaging the tank.
A
representative outer wall temperature for the vapour space and also one for
the liquid space outer wall are shown in Fig 3. The bulk internal temperatures recorded by the three internal thermocouples T,M and B are shown in Fig
4. The internal pressure is shown in Fig 5.
These results show that the average upper wall temperature rose almost
linearly, starting after 90 s from ignition, to reach a value of around 700C
at 8 minutes. The lower wall temperature, after a similar delay, rose in
180 s to a plateau value of 90C, and remained so for a further 120 s, before
rising linearly to 500C after 8 minutes. The bulk temperature of the liquid
propane as recorded by thermocouples M & B reached 47C after 180 s, and
remained more or less constant for a further 120 s, before rising at an
increasing rate to reach 450C after 8 minutes. The bulk vapour thermocouple
reached 100C in 180 s and thereafter rose to reach 150C after 300 s, and
520C after 8 minutes.
This sequence of variations is similar to those obtained for larger vessels as
reported in (2) and (3) and tend to confirm the proposition that the majority
of the heat is transferred to the liquid propane by nucleate boiling, both
before and after the PRV has opened. Thus from Figs 3 and 4 the estimated
inner temperature wall, allowing for the wall temperature gradient, can be
seen to be always higher than the bulk saturated liquid temperature (the
latter following closely the saturation curve for liquid propane). Thus at
the moment venting begins Tw-Tsat has a value of some 20C.
90
One test with a 40% fill provided the opportunity to study the consequences of
a catastrophic failure, because the PRV, after first opening correctly, failed
to do so subsequently. Consequently both the internal pressure and the vapour
space wall temperature rose uncontrollably. The fire extinguishing system was
91
where Pt> is the burst pressure, K the outside diameter/inside diameter ratio
of the cylinder and
are respectively the yield strength and ultimate
tensile strength of the vessel material at the elevated temperature T. Thus
taking the maximum wall temperature as the recorded value of 600C and substituting the appropriate strength properties, the calculated burst pressure is
38 bar. This is within 8% of that observed. Furthermore if this formulae is
taken as a criteria for vessel failure then it can be shown that all the other
tests either depressurised safely or were stopped before reaching unsafe
temperatures. Application of equation (1) to the data of (2) and (3) will
predict all of the burst pressures with less than a 18% error. The greater
inaccuracy is probably due either to doubt about the appropriate strength
properties or in some cases the point where rupture began.
6.3
1-TONNE TANK
Three trials were undertaken using fill ratios of 40%, 80% and 20% respectively. The same type of data was recorded as in the previous tests but in
addition the tank was weighed continuously.
The vapour and liquid wall
temperatures are shown in Figs 6 and 7 respectively for all three fills. The
bulk liquid and vapour temperatures are shown in Fig 8, the internal pressures
in Fig 9, and the weight losses in Fig 10.
The results from all three tests showed the same trends as the previous
tonne test. The fires again took
minutes to establish themselves,
after which both vapour and liquid wall temperatures rose as previously but at
slower rates. The upper surface temperatures rose at more or less the same
rates in the 40% and 20% fill tests to reach 600-700C in 12-15 minutes from
ignition. In the case of the 80% fill the rate of rise was slower and about
the same as that of the liquid space wall temperature. This was because the
thermocouples were close to the liquid level until after the PRV had vented
some of the contents. The lower surface temperatures, after a 60-90 s delay,
rose to their plateau temperatures after 1(20%), 11(40%), and 18(80%) minutes.
92
The tank weighed 0.99 tonnes and contained initially 870 , 420 and 160 kgs of
propane to correspond with the three fill levels of 80, 40 and 20% respectively. The load cell output, Fig 10, confirmed that the times for the PRV's
to first open were as obtained from the internal pressures. They also showed
the initial thrust of the jet flare which was different in all three cases and
possibly indicative of the different degrees of valve opening that may occur
with nominally similar valves. The slopes of the curves represent the mass
flow rates during discharge and show that in all cases the maximum discharge
rate occurs during the first opening of the valves. Maximum (initial) and
average valves for the three tests are also given in Table 2. These confirm
the suggestion of two phase discharge discussed previously in connection with
the 1/4 tonne results.
The actual PRV fitted to the tank had a flow rate of 1.16 kg/s measured with
air at 15C and 16.6 bar pressure. Thus the effective critical valve area,
A E , as determined from the choked mass flow equation for compressible flows is
given by:-
93
where m is the mass flow rate, R the gas constant, T 0 the initial temperature,
Z the compressibility factor, P 0 the initial pressure, and
the specific
heat ratio.
v
This gives a value for
of
for this particular valve.
Assuming that propane is vented at a similar pressure through the same
effective area, then from the bulk vapour temperatures immediately prior to
venting, the maximum propane vapour flow rates, (using a compressibility
factor of .75 as given by Sallet (10)) are for descending fill levels, 1.04,
0.88 and 0.97 kg/s. These may be compared with the maximum and average values
shown in Table 2. It will be observed that in all three tests there appears
to be some high quality two-phase flow or droplet carry over during the first
discharge of the PRV. This is not unexpected with the 80% fill, as the vessel
is almost liquid locked just prior to venting and the PRV is only just able to
cope with the boil off rate as the pressure trace shows.
It is perhaps
surprising in the 20% fill case, especially as the valve throat area to liquid
surface area is relatively small. One possible explanation may be that in
this case the valve lifts further initially giving a larger effective throat
area.
The average mass flow rates are slightly less than the choked flow values, and
this is probably due to the increasing vapour temperature not accounted for in
these comparisons. The results do however illustrate that the performance of
an actual valve can be predicted reasonably well from the standard
compressible flow equations, as used in reference (1).
7
CONCLUSIONS
The test data obtained from fire engulfment tests on two sizes of tanks
compared favourably with other fire engulfment data reported in the
literature and was shown (1) to agree with theoretical predictions.
The burst pressure of a 1/4 tonne tank was reasonably well predicted from
its material strength properties at elevated temperatures.
The two tank sizes and different initial fill levels all responded to the
fire engulfment in a similar manner, differing only in their time scales.
The maximum wall temperatures occurred in the vapour space walls and were
of the order of 600-800C when the tests were stopped.
94
REFERENCES
(1)
(2)
(3)
(4)
(5)
(6)
Venart, J E S, Sousa, A C M ,
Aydemir, N U "Transient thermal
stratification in heated partially filled horizontal cylindrical
tanks", ASME/AIchE Nat. Ht. Trans. Conf. Niagara Falls 1984.
(7)
(8)
(9)
(10)
(11)
(12)
ACKNOWLEDGEMENTS
The authors wish to thank the Director, Research and Laboratory Services
Division of the Health and Safety Executive for permission to publish.
(c) Crown Copyright
95
96
97
98
99
100
101
102
103
104
105
106