A New Procedure To Estimate BLEVE Overpressure
A New Procedure To Estimate BLEVE Overpressure
A New Procedure To Estimate BLEVE Overpressure
Abstract
Keywords
1. Introduction
Boiling Liquid Expanding Vapour Explosions (BLEVEs), a major accident which can have
severe consequences, occur from time to time, both in fixed plants and in the transportation
of hazardous materials. Overpressure and ejection of vessel fragments are the common
effects of such an explosion; these can be followed by a fireball if the substance is flammable.
Corresponding author. Tel.: +34 93 401 17 36.
E-mail address: eulalia.planas@upc.edu (E. Planas).
1
When a vessel undergoes a BLEVE, part of the released mechanical energy is converted into
overpressure. There are different methodologies to calculate this mechanical energy, based
on diverse thermodynamic assumptions (Planas and Casal, 2015):
Real gas behaviour and adiabatic irreversible expansion (Planas et al., 2004; Casal
and Salla, 2006)
Comparative analysis show that all methodologies tend to provide conservative (i.e. high)
results, except those based on real gas behaviour and adiabatic irreversible expansion, which
give values that are less conservative but more realistic (Bubbico and Marchini, 2008;
Laboureur et al., 2014; Hemmatian et al., 2017). Most of these methods, however, are
somewhat cumbersome to be applied and require many thermodynamic data of the substance
involved. As for the one based on the superheating energy (Casal and Salla, 2006), although
it is much easier to apply, it does not take into account the contribution of the previously
existing vapour, what in some cases –a vessel with low filling degree– could imply a non-
negligible error.
For that reason, a research was performed to provide a new methodology to calculate the
mechanical energy released during a BLEVE phenomenon, easy and fast to implement and,
at least, as reliable and precise as the currently existing ones. In this paper a new procedure
is therefore presented, which is based on the thermodynamic assumption of real gas
behaviour and adiabatic irreversible expansion but that only requires as input data the vessel
filling degree and the temperature at failure. The simplicity of the equations provided allows
a fast and accurate estimation of the energy released in the BLEVE of the most common
substances undergoing this phenomenon.
When the influence of the diverse thermodynamic assumptions on the calculation of the
mechanical energy is analysed, something quite interesting is observed. This is the fact that
the model based on real gas behaviour and adiabatic irreversible expansion shows an almost
2
linear variation of the energy released as a function of the temperature at the moment of the
explosion; this can be seen in Fig. 1 for five substances: propane, butane, methane, water and
vinyl chloride. A linearity was also observed at any vessel filling degree (FD); as an example,
this is shown in Fig. 2 for the same substances. Here, “filling degree” refers to the liquid
filling level at the beginning of the heating process. If there is a loss of containment through,
for example, a safety relief valve or a broken pipe, then the filling degree (at initial conditions
of pressure and temperature) must be estimated taking into account the mass of material lost
during the loss of containment.
This behaviour was found with all the substances investigated, which were –according to a
historical analysis (Hemmatian et al., 2015)– the ones more frequently involved in BLEVE
accidents.
12
Propane
10
Energy (MJ/m3)
0
300 310 320 330 340 350 360
Temperature (K)
10
0
283 303 323 343 363 383 403
Temperature (K)
3
8
7 Methane
6
Energy (MJ/m3)
5
4
3
2
1
0
120 130 140 150 160 170 180 190
Temperature (K)
40
30
20
10
0
383 433 483 533 583 633
Temperature (K)
12
Vinyl chloride
Initial FD=70 % Initial FD=90 %
Energy (MJ/m3)
10
0
276 296 316 336 356 376 396 416
Temperature (K)
4
Fig. 1 - Mechanical energy released (per m3 of vessel volume) as a function of the temperature in the vessel at
the moment of the explosion, at different filling degrees, based on the real gas behaviour and adiabatic
irreversible expansion assumptions.
12
Propane
10
Energy (MJ/m3)
0
10 20 30 40 50 60 70 80 90
Initial filling degree (%)
10
8
6
4
2
0
10 20 30 40 50 60 70 80 90
Initial filling degree (%)
5
4
3
2
1
0
10 20 30 40 50 60 70 80 90
Initial filling degree (%)
5
50
45 Water
40
Energy (MJ/m3)
35
30
25
20
15
10
5
0
10 20 30 40 50 60 70 80 90
Initial filling degree (%)
0
10 20 30 40 50 60 70 80 90
Initial filling degree (%)
Fig. 2 - Mechanical energy released (per m3 of vessel volume) by the explosion as a function of the filling
degree, at different temperatures, based on the real gas behaviour and adiabatic irreversible expansion
assumptions, for five different substances.
This linear relationship relating the energy, the temperature and the degree of filling indicates
a way to calculate the energy released in a BLEVE and, consequently, the overpressure
generated by the explosion. In the following sections, a deeper analysis of this linear
behaviour is performed for the substances included in Fig. 1 and Fig. 2 and also for other five
involved with a certain frequency in BLEVE accidents.
Initially, a set of 2713 scenarios for a 1 m3 vessel (used as a basis for all calculations),
covering both different filling degrees (from 1% to 99%) and temperatures at the moment of
explosion (from storage temperature to the critical one), were defined for the ten substances
6
included in Table 1, which are the ones most frequently involved in BLEVE accidents
according to a historical analysis (Hemmatian et al., 2015). For all the scenarios, the
mechanical energy per m3 of vessel volume was determined by assuming real gas behaviour
and adiabatic irreversible expansion, according to the methodology proposed by Planas et al.
(2004). The required thermodynamic data were obtained from NIST Reference Fluid
Properties, Version 9.1 (Lemmon et al., 2007). A dataset for each substance was therefore
prepared with the values of the mechanical energy recorded, together with the final
temperature and related filling degree, for each scenario.
However, it should be noticed that some scenarios could not be considered, because the
required physical condition was not fulfilled. For example, a container initially filled up to
90% with liquefied propane at 300 K could reach its maximum filling degree (100%) at a
temperature of 326.3 K, before the temperature reached the propane critical one (369.9 K).
This phenomenon is due to the variation of liquid and gas densities as a function of
temperature, according to which, at a certain moment, the decreasing gas volume collapses
(Casal, 2008) and the vessel becomes completely full of liquid. Therefore, taking this into
account, the number of scenarios was finally reduced to 2034 (Table 1).
Table 1. Scenarios used to calculate the mechanical energy for the ten selected substances.
7
1, 5, 10, 15, 20, 25, 30, 35, 40, 45, 50,
250, 260, 270, 280, 290, 300, 310, 320, 330,
Ammonia 55, 60, 65, 70, 75, 80, 85, 90, 95, 97,
340, 350, 360, 370, 380, 390, 400
98, 99
While the “best” equations (i.e., those keeping a relatively simple expression) were found by
using polynomial expressions, it was necessary to check how they achieved a good fit. The
visual examination or a graphical method was the first basic applied approach to see how the
surfaces were close to the calculated data and where potential deviations occurred;
afterwards, a statistical method was also used to evaluate the goodness-of-fit for the proposed
equations.
In this analysis, the filling degree (FD) and the temperature (T) were considered as input
variables, and the related mechanical energy was considered as an output one. The Curve
Fitting Toolbox provided different polynomials of the two input variables.
The multiple fits tested were compared through the aforementioned parameters. Table 2
summarizes the mean goodness-of-fit results for the different substances. In this Table, Poly
11 means first degree polynomial for both variables, Poly 12 means first degree polynomial
for FD and second degree for temperature, etc. According to these results, the suggested
surface model based on Poly13 showed the best performance, as it gives smaller values of
the Root Mean Square Error (RMSE) and Sum of Square Error (SSE).
8
Average Poly ( FD, T ) SSE R2 Adjusted R2 RMSE
Finally, a set of the best equations for predicting the mechanical energy per cubic meter of
total vessel volume (e) as a function of the filling degree (FD) and the temperature at the
moment of the explosion (T) were obtained by this procedure (Table 3).
Table 3. Mechanical energy released per cubic meter of vessel as a function of explosion temperature and initial
filling degree (expressed in parts per unit instead of percentage) for different substances.
9
A slightly better fitting could probably have been reached by using more complex polynomial
expressions; however, the objective was to find a methodology that, while being accurate,
was also simple and practical to be applied; these expressions fulfil both conditions.
To go from energy to overpressure, the total vessel volume has to be multiplied by “e” in
order to find the total amount of mechanical energy released by a given system. Then, the
total energy can be converted to the equivalent TNT mass (mTNT) and afterwards into
overpressure by means of the corresponding conversion graph. A factor = 0.4 can be
applied to take into account the fact that an important amount of energy will be devoted to
break the vessel (ductile failure) (Casal, 2008); this implies that only 40% of the energy
released is invested in creating the overpressure.
4. Comparative study
We checked the equations obtained (Table 3) by comparing them with two sets of
experimental data from Johnson et al. (1990) and Birk et al. (2006, 2007) (second column of
Table 4) (the directional effect at short distances (Birk and VanderSteen, 2006; Birk et al.,
2016) was not considered). We also add the comparison of the original real gas behaviour
and adiabatic irreversible expansion (RAIE) method proposed by Planas et al. (2004) (first
column of Table 4) with the same set of experimental values. The resulting overpressures at
different distances corresponding to each method were obtained from the TNT equivalent
mass and the well-known plot of the scaled distance vs. peak overpressure for TNT.
The root-mean-square deviation (RMSD) (Piñeiro et al., 2008) was used as a statistical
parameter to perform a comparative analysis between the two methods and the average values
of the aforementioned experimental data. As it is shown in Table 4, the new method here
proposed gave a good accuracy as compared to the experimental data from Johnson et al.
(1990) and Birk et al. (2006, 2007).
The new approach gives results similar to those from RAIE method (from which it has been
developed) and sometimes even better; for example, the RMSE value for the Birk
experiments is lower for the new approach than the RAIE value (Table 4). Actually, Table 4
shows that the approach based on the polynomial method has some degree of deviation from
the data set from which the equations were derived, because the fitting method passes a
surface from the minimum distance to a data point. This deviation could be larger in some
points based on the fitted surface and its distance to the data points and the degree of
polynomial. Theoretically, the new method and the RAIE approach should give the same
RMSD value. The difference shown in Table 4 is due to the partial non-linear behaviour (later
10
commented) shown in Fig. 3; in fact, it is this “error” what improves de value of RMSD for
the new method.
Table 4. Root Mean Square Deviation (RMSD) values for different methods based on their thermodynamic
assumptions
The nonlinear relation between temperature and filling degree (shown in Fig. 3) is the reason
for the observed deviation in the polynomial equations that, as seen in the comparative
analysis, remains in the range of the expected accuracy of this type of calculation and,
therefore, should be considered acceptable.
0.8
Fillinig degree
0.6
0.4
0.2
0
300 310 320 330 340 350 360 365
Temperature (K)
Fig. 3 - Non-linear behaviour of the relationship between filling degree and explosion temperature (propane,
real gas behaviour and adiabatic irreversible expansion assumption).
The reliability of the new approach was also studied by using a full scale case. Bubbico and
Marchini (2008) studied the explosion of a propane road tank (13 m3) during the transfer to
11
a fixed storage vessel. A spill of liquefied propane was ignited, the tank was engulfed by
flames and after 25 min it exploded. At the moment of the explosion FD = 8.4% (it had
decreased significantly due to the continuous release) and the temperature of the propane
liquid phase was 300 K. From the analysis of the accident damages, the peak overpressure
was determined at 20 and 30 m (Table 5). The study showed that the new approach also gives
a good level of performance for predicting this full scale real case.
Table 5. Polynomial prediction vs. real values for a given propane BLEVE accident (Bubbico and Marchini,
2008).
Estimated from
Distance (m) New approach (Polynomial) (kPa) accident damages
(kPa)
20 7.6 5-6
30 4.8 3
A cylindrical vessel with a volume of 80 m3, containing liquid propane at room temperature
(20 oC), undergoes a BLEVE due to fire engulfment; a loss of containment takes place
through a safety valve. At the burst moment, the content temperature is 50 oC and the filling
degree is 34%. Estimate the overpressure (∆P) at a distance of 100 m.
Solution:
FD = 0.34
T = 323 K
Using the propane equation in Table 3 to find the mechanical energy per cubic meter (e):
𝑒 = 43.97 − 213.9 ∙ 0.34 − 0.152 ∙ 323 + 1.349 ∙ 0.34 ∙ 323 − 0.0004361 ∙ 3232 − 0.002045 ∙ 0.34
∙ 3232 + 1.55 ∙ 10−6 ∙ 3233 = 4.5 MJ/m3
𝐸 ∗ = 𝑒 ∙ 𝑉𝑇 = 4.5 ∙ 80 = 360 MJ
12
Next, the scale distance for 𝑟 = 100 m is:
𝑟 100
𝑅̅ = 1 = 1⁄ = 31.9
(𝑚 𝑇𝑁𝑇 ) ⁄3 (30.8) 3
By using the TNT curve (Casal, 2008), ∆𝑃 at 100 m is 3.6 kPa (0.036 bar).
The prediction of the overpressure generated in a BLEVE explosion will always be subjected
to some uncertainty, essentially due to the fact that only a fraction of the mechanical energy
released in the explosion is invested in creating the overpressure. It can be assumed that
between 50% and 60% of the overall energy released is required to break the vessel (ductile
failure) and to eject the vessel fragments, but this percentage cannot be predicted in an
accurate way; it will depend on different aspects, such as, for example, the way in which the
vessel is heated, the influence of the welding, the condition (aging) of the vessel, etc.; the
value = 0.4 seems to be adequate, taking into account the relatively good agreement
obtained with the experimental values. Nevertheless, and even taking this into account, it is
obvious that some methods give better predictions –closer to the real values– than others.
Of course, there will also be always some uncertainty related to the filling degree, which will
depend on the time during which material will have been released through the pressure relief
valve, or to the liquid temperature at the moment of the explosion (a temperature near the
equilibrium with the set pressure of the pressure relief valve can be assumed). Nevertheless,
these circumstances will exist as well for any other method which could be applied, this
uncertainty being also found in many calculations of accidental effects when performing a
risk analysis.
The new method proposed, based on the almost linear relationship between the released
mechanical energy, the temperature at the moment of explosion and the filling degree, allows
obtaining fairly good values in a quick and simple way. It does not require the substance
thermodynamic properties (enthalpy, entropy, internal energy, etc.) and it only needs the
rupture temperature and the filling degree to calculate the BLEVE mechanical energy and
the resulting overpressure. The comparison of its predictions with the values corresponding
to experimental data gives very positive results, this validating the reliability of the method.
13
Acknowledgement
This work was supported by the Spanish Ministry of Economy and Competitiveness [Project
CTM2014-57448-R, funded with FEDER funds].
References
Birk, A. M., VanderSteen, J. D. J., 2006. On the transition from non-BLEVE to BLEVE
failure for a 1.8 m3 propane tank. J. Press. Vessel Technol, 128 (4), 648-655.
Birk, A. M., Davison, C., Cunningham, M., 2007. Blast overpressures from medium scale
BLEVE tests. J. Loss Prev. Process Ind., 20 (3), 194-206. doi:
dx.doi.org/10.1016/j.jlp.2007.03.001.
Birk, A.M., Heymes, F., Aprin, L., Slangen, P., Eysette, R., Lauret, P., 2016. Near field blast
effects from BLEVE. Chemical Engineering Transactions 48, 283-288. doi:
dx.doi.org/10.3303/CET1648048.
Brode, H. L., 1959. Blast wave from a spherical charge. The Physics of Fluids, 2 (2), 217–
229.
Bubbico, R., Marchini, M., 2008. Assessment of an explosive LPG release accident: A case
study. J. Hazard. Mater., 155 (3), 558-565.
doi:dx.doi.org/10.1016/j.jhazmat.2007.11.097.
Casal, J., Salla, J. M., 2006. Using liquid superheating energy for a quick estimation of
overpressure in BLEVEs and similar explosions. J. Hazard. Mater., 137 (3), 1321–1327.
Casal, J., 2008. Evaluation of the Effects and Consequences of Major Accidents in Industrial
Plants. Elsevier, Amsterdam.
CCPS, 2010. Guidelines for Vapor Cloud Explosion, Pressure Vessel Burst, BLEVE, and
Flash Fire Hazards, 2nd Edition. Wiley Subscription Services, Inc., A. Wiley Company,
New York.
Crowl, D. A., 1991. Using thermodynamic availability to determine the energy of explosion.
Plant/Operations Progress, 10 (3), 136-142. doi: 10.1002/prsb.720100306.
Crowl, D. A., 1992. Using thermodynamic availability to determine the energy of explosion
for compressed gases. Plant/Opererations Progr., 11 (2), 47-49. doi:
10.1002/prsb.720110206.
14
Hemmatian, B., Planas, E., Casal, J., 2015. Fire as a primary event of accident domino
sequences: the case of BLEVE. Reliability Eng. Syst. Safety, 139, 141-148.doi:
dx.doi.org/10.1016/j.ress.2015.03.021.
Hemmatian, B., Planas, E., Casal, J., 2017. Comparative analysis of BLEVE mechanical
energy and overpressure modelling. Process Safety and Environmental Protection, 106,
138-149. doi: dx.doi.org/10.1016/j.psep.2017.01.007.
Johnson, D.M. , Pritchard, J.M. , Wickens, M.J., 1990. Large catastrophic release of
flammable liquids. Comission of the European Communities Report, contract n.
EV4T.0014.UK, 1990.
Laboureur, D., Heymes, F., Lapebie, E., Buchlin, J., Rambaud, P., 2014. BLEVE
overpressure: multiscale comparison of blast wave modeling. Process Safety Progress,
33(3), 274-284.
Lemmon, E., McLinden, M., Huber, M., 2007. REFPROP: Reference fluid thermodynamic
and transport properties. NIST standard reference database, 23(8.0).
Piñeiro, G., Perelman, S., Guerschman, J. P., Paruelo, J. M., 2008. How to evaluate models:
Observed vs. predicted or predicted vs. observed? Ecological Mod., 216 (3–4), 316-322.
doi: doi: dx.doi.org/10.1016/j.ecolmodel.2008.05.006.
Planas Cuchi , E., Salla , J. M., Casal , J., 2004. Calculating overpressure from BLEVE
explosions. J. Loss Prev. Process Ind., 17 (6), 431-436. doi:
dx.doi.org/10.1016/j.jlp.2004.08.002.
Prugh, R. W., 1991. Quantitative Evaluation of “BLEVE” Hazards. J. Fire Prot. Eng., 3 (1),
9–24.
Smith, J. M., Van Ness, H. C., Abbott, M., 1996. Introduction to chemical engineering
thermodynamics (5th Ed.). McGraw-Hill, Inc., New York.
Nomenclature
𝑒 Explosion energy considering the expansion of the pre-existing vapour plus the vapour
generated in the flashing of the liquid, MJ / (vessel m3)
𝐸 ∗ Explosion energy considering the expansion of the pre-existing vapor plus the vapor
generated in the flashing of the liquid, J
15
FD Filling degree, %/100
𝑟 Distance between the center of the explosion and the point at which the overpressure
has to be estimated, m
Greek Letters
16