Ciambelli P.(1), Bucciero A.(1), Maremonti M.(1), Salzano E.(2), Masellis M.(3)

(1)Dipartimento Ingegneria Chimica e Alimentare, Università di Salerno, Via Ponte Don Melillo, 84084 Fisciano (SA), Italy
(2)CNR-GNDRCIE, P.le V.Tecchio 80, 80125 Naples
(3)Divisione di Chirurgia Plastica e Terapia delle Ustioni, Ospedale Civico, Palermo

SUMMARY. Boiling liquid expanding vapour explosion (BLEVE) is regarded as a major risk in the storage and transportation of hazardous materials. Nearly all the cases reported in the literature refer to open environments while BLEVEs in confined or congested areas are very uncommon. This paper presents the case history of a BLEVE of a tank truck transporting LPG that occurred in a highway tunnel near Palermo (Italy). The truck was involved in a car crash that gave rise to the release of propane gas through a crack that formed on the top of the vessel. The ignition of the gas cloud that was created caused critical burns in 25 persons. The subsequent tank truck BLEVE caused five deaths. The accident was described by modelling separately the deflagration of the gas cloud and the BLEVE. The evolution in time and space of the deflagration was modelled with a computational fluid dynamics technique, using a commercial code, while a beat transfer model was specifically developed in order to simulate the pressure increase inside the vessel until the tank collapsed and the BLEVE occurred.

The evaluation of the risks involved in the transport of dangerous goods on road and rail is of primary importance because of the different substances and scenarios involved. Typical situations of concern are related, for instance, to accidental releases in urban areas, where the consequences of an accident can be catastrophic.
Liquefied petroleum gas (LPG) is particularly exposed to the risk of serious accidents such as BLEVE (Boiling Liquid Expanding Vapour Explosion), i.e. explosion of a tank due to rapid evaporation of the liquid contained caused either by a depressurization wave generated by a hole accidentally formed in the tank shell or by an intense heat radiation load due to an external fire. As a consequence of the tank failure, the tank is flattened to the ground, giving rise to a strong blast wave, a destructive fire-ball, and in some cases the formation of numerous fragments that can be propelled significant distances. The characteristic time duration of a BLEVE usually ranges from 10 to 30 min, in the case of full fire-engulfment. I The time can become significantly shorter in tunnels.
Several examples of BLEVEs are reported in the literature but they always refer to open environment scenarios, while descriptions of accidents occurring in congested areas are unavailable. Obstructions and confinement are nevertheless recognized as significant factors affecting fire and explosion strength: road and rail tunnels must therefore be considered sites of major risk.
This paper describes the deflagration of a gas cloud that formed in a road tunnel near Palermo (Italy), followed by the BLEVE of a tank truck transporting LPG. The whole phenomenon is modelled by means of techniques of computational fluid dynamics (CFD) and of a heat-transfer model that describes the pressure build-up of a tank due to external fire.Special emphasis is given to the analysis of the time of occurrence of the BLEVE in a confined environment, which is of great importance for emergency operation planning.

On 18 March 1996, a tank truck was involved in a car crash in a highway tunnel near Palermo, Italy. The tunnel is 148 m long, 9.5 m wide, and 6.5 m high. It was a rainy day with a strong wind (43 m/s) and the ambient temperature was about 10 'C.
The collision gave rise to the escape of propane gas through a crack that formed on the top of the vessel. The ignition of this gas cloud and the subsequent BLEVE of the tank caused fatal injuries to five persons. A further twenty-five persons were critically burned by heat radiation.
Eye-witness accounts allowed the following reconstruction of the accident (Fig. I):

The rumbling noise and the hot wind causing burns to the persons in the tunnel were due to the deflagration of the propane cloud that formed as a result of the spill from the tank crack. Ignition occurred when the gas reached car N' 3, which was burning.
The hot burned gas produced by the cloud explosion and the jet-fire formed after the tank leakage provided an intense heat radiation load to the vessel, causing the internal pressure to build up to conditions of structural failure which normally occur at about 20 bar.' Knowledge of the leakage area is of primary importance for prediction of the quantity of gas discharged during the first minute.
The post-incident analysis confirmed the presence, in the tank shell, of a crack measuring about 20 sq cm. This area was probably enlarged by the jet-fire and the final BLEVE. The crack area was evaluated by means of a beattransfer model (see below) simulating the characteristic failure times for a totally fire-engulfed tank scenario.

Heat-transfer model
The tank truck had a total volume of 10 cub m (diameter = 2.05 m, length = 3.6 m), with a 6-trun shell thickness. At the moment of the accident the tank contained 2500 kg of LPG (filling percentage = 50%). No relief valve was fitted.
The BLEVE models available in the literature` are generally very sophisticated. These models provide temperature profiles in the different tank zones (liquid, vapour, walls) for different external fire scenarios and different regimes of thermal exchange to the liquid zone (nucleate and film boiling regime). This detailed description is beyond the scope of the present paper, which aims at the prediction of global characteristic BLEVE properties, such as internal pressure build-up and tank failure time.

• The main assumptions of the model are as follows:

• liquid and vapour phases are well mixed
• liquid is in saturation condition whereas vapour is superheated

• wet and dry walls have uniform temperature along their thickness

On the basis of these assumptions, the equations of conservation of energy for the liquid and the vapour phase can be written as follows:

where t is the time, M_liq, M_vap, C_vliq and C_vap are the masses and specific heats of liquid and vapour, T_hq, T_vap, T_wall,vap and T_ext are the liquid, vapour, dry wall and external temperatures, DH_ev, and DH_sur, are the latent heat of evaporation and the over-heating heat of vapour, A_vap and A_liq are the surfaces of walls enclosing vapour and liquid, A_liq is the surface of liquid-vapour interface, and h_r,liq is the radiative heat transfer coefficient from the dry wall to the liquid. U_liq and Uvap are the overall heat transfer coefficients from the external fire to the liquid and vapour respectively, defined as:

where K is the thermal conductivity of the steel and h_c,liq,ext , h_c,vap,ext , H_r,vap,ext , are the convective and radiative heat transfer coefficients from external fire to the wet and dry surfaces, and h_c,liq,int, h_c,vap,int and h_r,vap,int, are the convective and radiative heat transfer coefficients of the liquid phase and the gas phase. All coefficients were computed according to Aydemir et al. and Rohsenow. Finally, a mass balance which takes into account the mass spilled from the crack is included:

where m₀ is the initial amount of LPG, m_tot, is the mass at time t, and Q is the flow rate, modelled as sonic flow through an orifice:

where C° is the discharge coefficient,  g is the specific heat ratio, A_crack is the area of the hole, M is the molecular weight of the escaping gas, R is the ideal gas constant, and P is the internal pressure. Tank failure is assumed to occur at 20 bar, which is generally reported as the maximum permitted pressure for LPG tanks exposed to fire.

CFD modelling of the vapour cloud explosion
The deflagration of propane gas released from the tank was simulated during the first minute by means of a CFD code known as AutoReagas, developed by TNO (Netherlands) and Century Dynamics Ltd (UK). The code solves the conservation equations of mass, energy and momentum, using a finite volume method." The k-F, model" is included to describe turbulence effects. The combustion is modelled as a single-step reaction from reactants to products.
The accident scenario was reproduced using the CAD purchased with the code.

Pressure build-up inside the tank
Tle system of equations (1), (2) and was (5) solved using a Runge-Kutta method QV order), yielding T_liq, T_vap and %, as functions of time. Internal pressure was computed from T_liq.
Fig. 2 shows the model results, in terms of pressure increase in time, for crack areas ranging from 5 to 50 sq cm. For each computation run, a delay time of one min was assumed before full fire-engulfment condition were established. The case of a completely closed vessel has also reported, for thye sake of comparison.

As expected, the greater the crack area, the longer the time necessary to reach failure conditions (20 bar). The model produced a good prediction of failure time, as reported by witnesses, assuming a crack area of 10 sq cm. The results of the model also show that, in the absence of leakage, critical conditions are reached almost immediately.
The mass flow rate through the leakage is reported in Fig. 3, as a function of time, for the area considered. A total mass of about 50 kg can be calculated as having flowed through the crack in the first minute, i.e. prior to ignition. This amount of propane was therefore used in the CFD simulation of the gas cloud explosion.

Fig. 3 - Mass flow rate history trhought 10sq cm orifice

CFD simulation of the propone cloud explosion
The three-dimensional CFD modelling of the deflagration was carried out by assuming a propane cloud at stoichiometric concentration (4%). The gas cloud was presurned to extend from the tank truck to car N° 1, where ignition occurred.
Haine propagation along the tunnel, reported in Fig. 4 at different times, reached its largest extent in about 1.7 seconds. Peak overpressures resulting from the model simulation at various locations in the tunnel were of the order of 2kPa, which is consistent with the "hot wind" felt by the persons present.

Fig. 4 - Modelled flame propagation in road tunnel

The description of the accident shows that the characteristic times of BLEVE in confined areas such as road and rail tunnels are significantly shorter than those in open environments. The rigid confinement provided by the tunnel walls prevents the hot burned gas from exiting from the crash area, causing rapid and complete fire-engulfment of the tank shell, thus accelerating the heating process.
It is worth pointing out that the ignition of the vapour cloud that initially formed and the subsequent fire should be regarded not only as precursors of the BLEVE but also as a providential warning for the people who succeeded in escaping.
Thus, in order to define emergency planning with regard to the possible consequences of the transportation of hazardous materials, the analysis must take into account the geometrical characteristics of the areas involved.

RESUME. La BLEVE (boiling liquid expanding vapour explosion: explosion de vapeur en expansion de liquide bouillant) constitue un important risque dans l'emmagasinage et le transport des matériaux dangereux. Presque tous les cas décrits dans la littérature traitent des environnements ouverts, tandis que les BLEVEs en environnement clos ou encombré sont peu communs. Les Auteurs décrivent la BLEVE d'un camion-citerne qui transportait du GPL (gaz de pétrole liquéfié), explosion qui s'est produite dans un tunnel de l'autoroute près de Palerme (Italie). Le camion avait été endommagé dans un accident de voiture qui a provoqué l'échappement de gaz de propane à travers une fissure qui s'est produite sur le dessus du citeme. Uignition du nuage de gaz qui a été créé a causé des brûlures critiques à 25 personnes. La BLEVE successive du citerne a causé la mort de cinq personnes. L'accident a été décrit en modelant séparément la déflagration du nuage de gaz et la BLEVE. L'évolution dans le temps et l'espace de la déflagration a été modelée avec une technique de la dynamique computationnelle des liquides, avec un code commercial, tandis qu'un modèle de transfert de chaleur a été développé expressément pour simuler l'augmentation de la pression à l'intérieur du citerne jusqu'à la BLEVE.

BIBLIOGRAPHY

1. Birk A.M.: Scale effects with fire exposure of pressure-liquefied gas tanks. J. Loss Prev. Process Ind., 8: 275-90, 1995.
2. Davenport J.A.: Hazards and protection of pressure storage and transport of LP gas. J. Haz. Mat., 20: 3-19, 1988.
3. Pugh R.W.: Quantify BLEVE hazards. Chem. Eng. Prog., 87: 66-72, 1991.
4. Lees RP: "Loss prevention in the process industries". Butterworth, Oxford, 1996.
5. Beynon G.V., Cowley L.T., Small L.M., Williams L: Fire engulfment of LPG tanks: Heat-up, a predictive model. J. Haz. Mat., 20: 227-38, 1988.
6. Aydemir N.U., Magapu V.K., Sousa A.C.M., Venart J.E.S.: Thermal response analysis of LPG tanks exposed to fire. J. Haz. Mat., 20: 239-62, 1988.
7. Hadjisophocleus G.V, Sousa A.C.M., Venart LE.S.: A study on the effect of the tank diameter on the thermal stratification in LPG tanks subjected to fire engulfment. J. Haz. Mat., 25: 19-31, 1990.
8. Ramskill PX.: A description of the computer codes-codes to model the thermal response of an LPG tank either fully or partially engulfed by fire. J. Haz. Mat., 20: 177-96, 1988.
9. Rohsenow W.M.: "Heat, mass, and momentum transfer". Prentice Hall, New Jersey, 196 1.
10. Patankar S.V.: "Numerical heat transfer and fluid flow". Taylor and Francis, Minnesota, 1980.
11. Launder B.E., Spalding D.B.: "Mathematical models of turbulence".Academic Press, London, 1972.