Investigation of an LPG accident with different mathematical model applications. Fausto Zenier and Franco Antonello

Transcription

1 340 Int. J. Risk Assessment and Management, Vol. 2, Nos. 3/4, 2001 Investigation of an LPG accident with different mathematical model applications Fausto Zenier and Franco Antonello ARTES srl, via C. Battisti 2/A Mirano (VE) Italy Fax: Fabio Dattilo Comando Provinciale Vigili del Fuoco di Rovigo, via Ippodromo , Rovigo, Italy L. Rosa Università degli Studi Di Padova, Dip. di Ingegneria Meccanica, via Gradenigo , Padova, Italy Abstract: An LPG (liquefied petroleum gas) storage accident was reconstructed through several simulation models, with the aims of the reported and filmed effects and damage. The accident happened in Paese, near Treviso (Italy) in March 1996 and was characterized by complex phenomena due to a choked release of LPG from a tanker. There was a dispersion of light winds and, after about 40 minutes, a flash-fire throughout followed by a transient jet release and BLEVE (boiling-liquid-expanding-vapour explosion). The evolution of these phenomena was studied by application of some available models. The primary purpose of these applications was to check the possible use and the efficiency of the simulation models using comparisons and evidence on the dynamics of the accident. Another purpose was to get more detailed information on the dynamics of the accident. The work was done in collaboration with the Fire Brigade of Rovigo, the University of Padova and ARTES (consulting on risk analysis Mirano Venezia). Complex situations, such as the dispersion of heavy gas in a semi-confined area, with light winds and variable atmospheric stabilities, prolonged time of release with different types of flow, various types of fire (flash fire, transient jet fire, BLEVE and fireball) were simulated with adaptations of models or input to obtain realistic results. Keywords: Liquefied petroleum gas; accident; simulation models; gas dispersion; BLEVE. Reference to this paper should be made as follows: Zenier, F., Antonello, F., Dattilo, F. and Rosa, L. (2001) Investigation of an LPG accident with different mathematical model applications, Int. J. Risk Assessment and Management, Vol. 2, Nos. 3/4, pp Biographical notes: From 1991 to 1994 Fausto Zenier carried out occasional jobs in the field of environmental pollution (emissions, writing of technical relations). In 1994 he began to collaborate with ARTES srl company (Risks Analysis & Ecology and Safety Technologies) developing or modernizing software programs for operability analysis and calculation models for risk Copyright 2001 Inderscience Enterprises Ltd.

2 Investigation of an LPG accident with different mathematical applications 341 analysis. From 1997 he began to work on risk and reliability analysis (Hazop, fault-trees and events-trees), continuing the development, verification and validations of mathematical models. Franco Antonello is presently the Safety Manager of ARTES srl, a consulting company which performs environmental and safety studies for Montedipe and other companies. He worked in the Safety Department of Montedipe from 1974 to 1989, performing risk assessment and reliability analysis on several plants and studying mathematical models to simulate the effects of accidents. Fabio Dattilo is the Commander of the Provincial Fire Brigade of Rovigo, Italy. Lorenzo Rosa is the associated professor at the University of Padova, Italy in the Department of Mechanical Engineering. 1 Chronological sequence of the events related to the accident At about 7 am the unloading of a 52m 3 LPG tanker started. Shortly afterwards, probably because of a reduced flow, the operator interfered with the truck valve to increase the flow rate. For reasons not yet clarified, a leakage took place around the valves of the tanker, inside the valve protecting box. A two-phase propane jet filled the valve box, overflowing and forming a small pool of sub-cooled liquid on the asphalt on the ground. While the surrounding area was immersed in a thick low cloud of vaporized propane, the operators responsible for carrying out the emergency plan, stopped the work, turned off the electrical system, evacuated the area and called the National Fire Brigade (NFB). Some minutes after the arrival of the NFB, at 7.40 am, a spark caused a flash fire, which caused damage to several different utilities. Meanwhile, an explosion took place in a building some 10m from the source of the plume, seriously damaging part of the building. A fire broke out in the pool of liquid propane and the flames wound round the rear of the truck. The increased internal pressure and the mechanical stress due to the high temperature, caused a wide rupture of the tanker wall with a release of gas, very similar to a fireball, followed by the combustion of the released gas like a flare. The initial flashfire also caused damage to the loading arm connected to a 15m 3 tanker nearby, containing LPG of no more than 10% of its volume. At 8.40 am the shell of this tanker, heated by the jet fire generated from the damaged loading arm, split and a BLEVE took place, followed by a fireball; the BLEVE destroyed the second tanker. From about 9.00 am the efforts of the Fire Brigade were rewarded, putting the situation under control, limiting the area of the fire and leaving the gas flowing out of the first tank to burn out. The fire was completely put out at 5 pm. 2 Adopted criteria for the study of the event The most important conditions and aspects related to the present study will be examined according to the chronological sequence of the events.

3 342 F. Zenier, F. Antonello, F. Dattilo and L. Rosa 2.1 Initial outflow The initial outflow can be considered as a two-phase flow in a half-enclosed environment with the starting point, the initial jet in the valve protecting box placed at the rear of the tanker. A dispersed spout created the presence of liquid propane in the box, (sub-cooled LPG due to isenthalpic flash) and spillage of liquid propane from the valve box to the asphalt on the ground. For model simulation purposes, the scenario described above can be considered as a typical one, as the common options for modelling refer either to a free two-phase jet or to a single phase flow. However, the calculation was performed taking into account the length of the pipe fixed in front of the valve (1m length and 25 mm diameter). 2.2 Gas dispersion With reference to the start of the events, the atmospheric conditions at the site were obtained from the data supplied by two different weather stations, located few kilometres from the accident area (Table 1). Table 1 Atmospheric conditions at the start of the event first station second station wind velocity 0.5 m/s 0.5 m/s temperature 5 C 6 C wind direction relative humidity 80% 85% The low wind velocity, compared to a standard altitude of 10m and therefore lower if referring to ground level, excludes the use of Gaussian models, which have some limits at low velocity, particularly lower than 1 m/s [1,2]. On the other hand these conditions cause some problems in using all the common simulations models, by the lack of data concerning the choice of atmospheric stability class. The first station recorded an sky overcast sky ratio of 5:8 7:8 with clouds above 4000 ft; the second station recorded a ratio of <4:8 with base clouds at 3000 ft. With reference to this situation and to the wind velocity, according to the Pasquill scheme [3], a stability class B can be considered, while referring to the standard deviation of the wind direction [4] and considering that no variation in the wind direction takes place at 0.2 m/s velocity at an altitude of 2m, a stability class F could be assumed. 2.3 Flash fire and fire The reconstruction of this phase was not closely examined, due to several difficulties in defining the initial steps of the flash fire and the geometry of the cloud formation. 2.4 Burst of the first tank A few minutes after the flash fire, an explosion wrecked the wall of the tanker, due to overheating caused by the flame generated from the pool of propane and jet fire, lapping

4 Investigation of an LPG accident with different mathematical applications 343 on the surface of the tank. The typical pulsating effect of this kind of phenomenon, is not considered by most models (normally referring to a steady combustion with an average heat emission divided among conduction, convection and radiation). Considering that 30% of the developed heat is propagated by radiation [1], we can consider 70% of the combustion heat of LPG to evaluate the overheating. Relevant to the failure of the tanker construction material, it must first be noted that only a third of the total surface of the tanker was greatly affected by overheating by the flame action (as verified by a second trial). Of the remaining parts, as far as radiation is concerned, only superficial effects have been found. This calculation must take into account the irregular shape of the rupture area with dimensions of 40 x 60cm, located in the connection between the cylinder and the hemyspheric bottom of the tanker. The size of the opening is important for analysis of the release of the burning gas: according to the above description, it cannot be classified as a classic fireball, as related to BLEVE, which is normally generated by a very large opening or by the collapse of the vessel. 2.5 Consequences of the fire in the second tank: BLEVE The flash fire and the subsequent fire around the first tanker, damaged the loading arm connected to a nearby second tanker, causing the failure of gaskets and other components. LPG flowing from the damaged connective device caught flame impingement from the second tanker and brought about a final phenomenon that can be classified as a BLEVE. 3 Analysis and reconstruction of the events 3.1 Study of the crack phenomenon in the first tank Even if several models are available to study similar events, the peculiarity of the accident described above, which took place with a partial rupture of the vessel instead of complete destruction as in the classic BLEVE phenomenon, required some adaptation of a mathematical model, particularly in carrying out a thermomechanical analysis relevant to mobile vessel fire damage and containing a two-phase flammable fluid. This was carried out according to the calculation code ANSYS 5.3, allowing the study of structures under mechanical and thermal stress, also variable during the time. Due to the thickness of the vessel wall (about 1cm), the theory of thick-walled pressure vessels was used. The model of the structure is explained in Figure 1. The fire around the rear of the tanker was simulated with a variable thermic flux, whose intensity was calculated by considering complete combustion of the gas flowing out of the tank and if there was a likelihood of an unburnt fraction of LPG. The thermal load, computed according to the above criteria, was related to the part of the tanker wall damage caused by the flames of the pool-fire and its variation in dependence of time corresponds to LPG FIRE CURVE reported in UNI.ENV standard (see Figure 2).

5 344 F. Zenier, F. Antonello, F. Dattilo and L. Rosa Figure 1 Model of the structure Figure 2 Temporal temperature increase

6 Investigation of an LPG accident with different mathematical applications 345 Also in the hypothesis, in computing the simplification of isotropic behaviour of the construction material, two aspects related to the construction criteria of this kind of vessel must be underlined. Taking into account that a spherical shape has a higher resistance than a cylindrical one, the thickness of hemispheric bases is normally thinner than the thickness of the cylindrical walls. In a stress situation created by fire, the strains in the hemispheric part are different from the ones in the cylindrical part; one can say, to simplify, that in the cylindrical part there are three components (radial, longitudinal, tangential) whilst in the hemispheric part there are only two components (radial and tangential). This causes a state of tension located at the interface of thecylinder-sphere because the cylinder tends to open more than the sphere but this deformation is blocked by the sphere itself. The spherical part transmitted a force moment and a shearing stress to the cylinder and vice versa and this new tension condition, caused by fire, must be added to the normal stress condition related to a vessel working under internal pressure. The simplified result is reported in Figure 3. Figure 3 Tension state located at the interface cylinder-sphere In spite of the simplified hypotheses adopted, the above considerations take into account the most important aspects related to the weakening and the rupture of the vessel. The application of the program allowed the determination of the temperature distribution at the moment of rupture, obtaining the stress configuration of the structure, relevant to the thermic and pressure loads, the value of the plastic deformation and the location of the rupture point (Figure 4). The obtained results are in accordance with the reports of the people attending the accident.

7 346 F. Zenier, F. Antonello, F. Dattilo and L. Rosa Figure 4 Location of the rupture point 3.2 Analysis and reconstruction of the events by simulation models The amount of gas still present in the first tanker after the explosion and the vapour release has been calculated on the basis of the combustion time and the vaporization rate of propane, provided that the controlled combustion of the contents lasted eight hours. Starting from this evidence we have reconstructed some of the events to verify if the calculations performed would produce the same conclusion. The models listed below were utilized to simulate the events: ARCHIE (Automated Resource for Chemical Hazard Incident Evaluation) [5]. DEGADIS 2.1 (DEnse GAs DISpersion US Coast Guard, GRI and API US EPA). HGSYSTEM 3.0 developed by Shell and distributed by API (Ed ). RMP*Comp (only for flow rate calculation in the alternative mode) US EPA [6]. SIGEM-SIMMA developed by TEMA SpA. and used by the Italian National Fire Brigade. STAR (Safety Techniques for Assessment of Risk rel 3) developed by ARTES Srl. The reference points for the study of the different phenomena, obtained by information supplied by those in attendance, can be summarized as follows:

8 Investigation of an LPG accident with different mathematical applications The initial release of LPG, without fire, lasted about 50 minutes. 2 After the flash fire there was a mixed fire (pool and jet fire) with flames engulfing the rear of the tank, causing the partial rupture of the shell of the tank with an intermediate phenomenon between fireball and jet fire (transient jet fire) [7]. 3 Due to the damage caused by flash fire on the loading arm there was a fire involving a second tanker which, after some ten minutes, collapsed, producing a BLEVE. The flow rate has been determined by considering the presence of a deep tube, 1 metre long and an orifice with an equivalent diameter of 25 mm. LPG temperature was 278 K and equilibrium pressure about 5.4 bar (abs). The results of the calculations are reported in Table 2. Table 2 Results of calculations Model Flow rate (kg/s) Flow condition ARCHIE 11.6 (peak) two-phase RMP*Comp (1) 8.87 SigemSimma 0.46 gas STAR 2.25 two-phase (1) The calculation was performed with an alternative scenario, vapour cloud fire option and hole in liquid space; with the pipe release option the flow rate result is 11.7 kg/s Since the flow, even if not in a steady condition for the probable presence of ice in the orifice, lasted about 50 minutes and a third of the content was emptied, the only consistent result is the one referring to a flow rate of 2.25 kg/s. Taking into account the above time (50 minutes), a total amount of 6.75 tons of LPG was ejected, that is about one third of the initial content (21.4 tons), as testified by people present at the site. The quantity of LPG evaporated during the flash can be obtained by adding the amount of LPG evaporated inside the tube and the amount of LPG evaporated during the isenthalpic flash, that is about 23% of the total flow. To evaluate the concentration in the air, several models have been utilized, each one associated with the relevant flow rate; it must be underlined that the atmospheric conditions (stability class F and wind velocity 0.51 m/s) are of limited acceptability for most models. The results are reported in Table 3 (reference concentrations: Lower Flammability Limit LFL and 50% of LFL): Table 3 Results of computations Model Flow rate LFL 50% LFL Flammable mass ARCHIE peak 11.6 kg/s 140 m 203 m 1800 kg Degadis 2.25 kg/s 160 m 205 m 1030 kg HGSystem 2.25 kg/s 33 m 55 m n.c. RMP*Comp 8.87 kg/s <160 m n.c. n.c. SigemSimma 0.46 kg/s (gas) n.r. n.r. n.r. STAR 2.25 kg/s 38 m 55 m 157 kg n.c. = not calculated n.r. = not reached

9 348 F. Zenier, F. Antonello, F. Dattilo and L. Rosa Regarding the ARCHIE model, it is clear that the vapour flow rate adopted for dispersion is lower than the peak value indicated in the results, but this datum is not supplied. The results of the DEGADIS model, in accordance with the ARCHIE model, seems affected by the limit related to the wind velocity (minimum value suggested 1 m/s). A similar limit (1.5 m/s) is required by the HGSystem; however, the final result is considerably different in comparison with the result of the ARCHIE model. The results obtained with RMP*Comp are interesting from the viewpoint that a relevant flow rate produces the same LFL distance of DEGADIS, which uses a lower flow rate. The SigemSimma model lacks a calculation routine for dispersion of heavy vapours and aerosols and the evaluation of the flow rate refers to the gas phase only. The STAR model (box model developed from [8]), which considers the presence of obstacles or buildings in the propagation of the gas, offers an indication of gas accumulated in the area in front of the building, without, in substance, modifying the distances of the flammability limits. Disregarding the simulation of the initial flash fire, the radiation caused by the burning gas released from the first tanker and the fireball and BLEVE related to the second tanker, will be analysed. The phenomenon related to the first tanker in concomitance with the burst or structural rupture can be regarded as a transient jet release, which occurs when the substance in the vessel is not overheated [9]. According to relations obtained from different sources [7,10], the required time for the collapse of a vessel is a function of the geometry and characteristics of the flame engulfing the vessel; with the model of turbulent flame on the top of the vessel, a time of seven minutes can be estimated, while for pool flames a time of 15 minutes can be estimated. On the basis of the survey, the opening of the first tanker corresponds to a hole 50cm in equivalent diameter; according to this indication the following criteria were adopted to evaluate the effects of the phenomenon: simulation of a jet fire with flow rate equal to the initial flow rate, with pressure 17 bar and temperature 323 K, evaluation of the jet fire duration by calculation of the amount of the released gas for the time necessary to reach the atmospheric pressure inside the tanker, evaluation of the radiated energy. The initial flow rate was calculated by two models, with very similar results (503 kg/s by SigemSimma; 523 kg/s by STAR). The duration of the release was calculated according to the STAR model (about three seconds) and by evaluating the amount to be released to reach the atmospheric pressure in the tank starting from the initial pressure 17 bar. Considering the quantity of LPG released before the burst (about 13m 3 ) and the volume of the tanker (52m 3 ) filled by 80%, the volume occupied by the gas, before the burst, was 28m 3 with a pressure of 17 bar, equivalent to 456m 3 of gas, that is kg. With a flow rate equal to 520 kg/sec, about 24 seconds are necessary to decrease the internal pressure to the atmospheric value. The indications supplied by jet fire models were considered to take into account the limited duration of the phenomenon by transforming the radiation into energy, as indicated in Table 4.

10 Investigation of an LPG accident with different mathematical applications 349 Table 4 Results of computations Model 350 kj / m kj /m kj / m 2 ARCHIE 49m - 106m SigemSimma n.r. n.r. n.r. STAR n.r. n.r. 110m The results seem to be overestimated because people present at less than 100m were not injured. We must underline that the evaluation of the phenomenon is affected by simplification, mainly because the decrease of the flow rate and gas density was not related to the decrease of pressure. The phenomenon regarding the second tanker approaches the BLEVE theory, with a total collapse of the tanker and fragments and metal splinters found 500m from their origin. By application of the simulation models and considering the collapse conditions similar to the conditions of the first tanker (p = 17 bar; T = 323 K; M = 800 kg LPG), the results plotted in Figure 5 were obtained. Figure 5 Simulation results In comparison with the previous calculation, the results seem to be more homogeneous, but, also in this case, over estimated. It must be considered that for small quantities the models supply conservative estimates which are often not in accordance with the practical results; the results are more accurate for phenomena of great magnitude. With regard to the projectiles, two of them, of considerable size, were found at a distance of about 500m; the first was a part of the tanker shell with dimensions of 1.2m

11 350 F. Zenier, F. Antonello, F. Dattilo and L. Rosa by 1.2m and thickness of 10mm, the second was part of a pipe (1m long), probably part of the deep tube. The simulation of the fragmentation and projectiles is affected by great uncertainty related to the original shape of the fragments and to the angle of attack. With the exception of the STAR model, based on NASA studies and publications [11,12], it is not possible to consider correctly the shape and size of the fragments; for this reason the results obtained by using models other than STAR look quite unrefined. In fact, the model SigemSimma gives a range of m, while the results of the STAR model, considering an angle of attack of 35 on the basis of the reported evidence, are: 670m for the first piece and 520m for the second. The Birk simplified model [13] giving a distance of 610m referred to a fragment of significant size from a 15m 3 tanker BLEVE. 4 Conclusions With reference to the historically registered accidents, the one described in this study, related to a particular sequence of a great number of events and circumstances, cannot be considered as a single case. Most accidents are constituted by different events generating complex phenomena: one major problem with the description of these phenomena by simulation models, is that their main purpose is to give an idealization of reality through calculation algorithms, not to describe it in all of its complexity. So, the use of these models appears unsuitable for the detailed reconstruction of complicated cases as the one under consideration and in particular for the simulation of the diffusion of heavy gases in the presence of obstacles; in this case the use of 3D models is suggested, particularly in the case of the absence of wind. Nevertheless, with reference to the content of this study, it is clear that the answer supplied by the models adopted for risk analysis appears prudent, but, however, useful for forecasts and predictions to use in safety reports. Finally, it must be emphasized that the indications which come from simulations models should always be analysed in relation to experience, comparing the output result with reference situations (for example, historical cases). References 1 A Guidance Manual for Modeling Hypothetical Accidental Releases to the Atmosphere (1996) API publication No Methods for the calculation of the physical effects of the escape of dangerous material ( Yellow Book ) (1979) Report of the Committee for the Prevention of Disasters, The Directorate General of Labour Ministry of Social Affairs, The Netherlands. 3 Lees, F.P. (1996) Pasquill s stability categories (Pasquill, 1961), Loss Prevention in the Process Industries - Tab / pp.89 94, 2nd edition. 4 Lees, F.P. (1996) Relation between Pasquill stability categories and parameters of some typing schemes (after Sedefian and Bennet, 1980), Loss Prevention in the Process Industries Tab / pp.94, 2nd edition. 5 ARCHIE Federal Emergency Management Agency, US DOT, US EPA. 6 RMP*Comp Ver 1.06 EPA Risk Management Planning RCRA Superfund and EPCRA US.

12 Investigation of an LPG accident with different mathematical applications Birk, A.M. (1997) BLEVE research, Queen s University of Kingston, Ontario, Canada, ( 8 CRUNCH a dispersion model for continuous release of a denser than air vapour into the atmosphere (1983), SRD R229, UKAEA. 9 Le esplosioni BLEVE: rischi e misure preventive (1987), Rivista Antincendio, Agosto. 10 Guide for pressure relieving and depressuring system (1997), API RP Workbook for estimating effects of accidental explosions in propellant ground handling and transport system (1978), NASA Report Workbook for predicting pressure wave and fragment effects of exploding propellant tanks and gas storage vessels (1977), NASA Report Birk, A.M. (1996) Hazard from propane BLEVE: an update and proposal for emergency responders, Journal of Loss Prevention in the Process Industries, Vol. 9, No. 2.