Engineering Failure Analysis 13 (2006) 705–715 www.elsevier.com/locate/engfailanal

An acetylene cylinder explosion: A most probable cause analysis q John W.H. Price

*

Mechanical Engineering Department, Monash University, PO Box 197, Caulfield East, Vic. 3145, Australia Received 12 July 2004; accepted 4 April 2005 Available online 15 September 2005

Abstract This paper considers the explosion of an acetylene gas cylinder, which occurred in 1993 in Sydney. The failure caused severe fragmentation of the cylinder and resulted in a fatality and property damage. This failure is outstanding in two respects: the violence of the failure which occurred and the fact that the explosion occurred where there were no apparent ignition sources present and several days, possibly weeks after filling. There is no other recorded failure of this type in the modern history of acetylene cylinder traffic. The paper describes the failure and the circumstances surrounding it, examines the nature of the explosion which occurred and seeks an explanation of the events. Since there is no physical evidence remaining of how the failure occurred, the analysis concentrates on an evaluation of the most probable cause of the incident.
2005 Elsevier Ltd. All rights reserved. Keywords: Acetylene cylinder; Explosion; Fragmentation; Most probable cause analysis

1. Introduction 1.1. The acetylene gas cylinder explosion In 1993, an explosion occurred in a suburb of Sydney causing the death of a driver of a truck who was loading acetylene and oxygen cylinders at a transport yard. The force of the blast caused damage to buildings and nearby equipments. Fragments were found up to 200 m away [1].

q *

Paper presented at the First International Conference on Engineering Failure Analysis (Lisbon, Portugal, 12–14 July 2004). Tel.: +61 3 9903 2868; fax: +61 3 9903 2766. E-mail address: [email protected].

1350-6307/$ - see front matter
2005 Elsevier Ltd. All rights reserved. doi:10.1016/j.engfailanal.2005.04.014

706

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

During site investigations, it became evident that the explosion was related to the failure of a seamless carbon steel acetylene cylinder, which was being returned to a ship after delivery from the filling plant three days earlier. The cylinder had fragmented violently, the fragments caused rupture of other acetylene and oxygen cylinders and were the cause of impact damage in the surrounding area. Following the incident, there were fires and pressure wave damage to surrounding buildings and equipments. Nearly 200 fragments of the cylinder were collected, but accounted for only about half the mass of the original vessel. This paper discusses the phenomena involved in this kind of loading and proposes theoretical approaches designed to extract information about the nature of the incident, which occurred in the cylinder. 1.2. The cylinder The cylinder which exploded was one of a group of imported seamless steel cylinders, which are used on ships (see Fig. 1). The cylinders have a design with an outside diameter of 229 mm and a cylindrical length of 1.22 m with a specified wall thickness of 4.72 mm minimum in the cylindrical section. The steel for man^ y Þ of 235 MPa, UTS 411 MPa and elongation of 14%. Manufacture from ufacture has a specified yield ðU correct alloys was confirmed on retrieved fragments. The design filling pressure for the cylinder is 1.5 MPa and its internal volume is about 42 1. When full, the cylinder contains approximately 16 l of acetone, 6.2 kg of acetylene (depending on temperature) and 9.2 kg of porous mass. Total mass content is about 70 kg. The thin wall formula for a cylinder gives a minimum required thickness of 0.71 mm. The specified thickness of the cylinders is 4.72 and the actual thickness is 6.1 mm. A similar shaped cylinder is also in use for oxygen traffic, where the design pressure is typically 15 MPa. In oxygen service, the cylinder is fitted with a different valve system though the cylinder itself always has a right hand screw thread.

Fig. 1. A cylinder of the type which failed.

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

707

1.3. The nature of explosions and definition of terms As described earlier, the fragmentation of the vessel means that this was a detonation explosion with the speed of the explosion front moving at or above the speed of sound in the steel. Some useful relevant velocities are given in Table 1. The development of detonation explosions is normally thought of as having three stages which follow one after another:
initiation: which is some minor energetic event such as high temperatures, flames, sparks caused by electricity, lightning, or electrostatic charge, impact, and some chemical effects,
deflagration: which is effectively a flame front or reaction wave propagating at subsonic speed, and
detonation: which occurs in situations where the reactants and physical geometry cause the reaction front to become supersonic. The term shock wave is used for a pressure transient in fluid which travels faster than the speed of sound. Acetylene can explode without the presence of oxygen or air, since it is an unstable substance which can decompose. In the presence of oxygen, the combustion of acetylene has an even greater energy release (heat of combustion at 1 atm, 20 C) C2 H2 ! 2C þ H2 þ 54:19 kcal=mol ¼ 8:7 kJ=gðC2 H2 Þ C2 H2 þ 2:5O2 ! 2CO2 ðgÞ þ H2 Oð1Þ þ 312 kcal=mol ¼ 50 kJ=gðC2 H2 Þ

ð1aÞ ð1bÞ

Both oxidation and decomposition can proceed by deflagration or detonation. Deflagration in acetylene tends to accelerate and Miller [2] and Sutherland and Wegert [5] give examples of deflagration to detonation transitions. These tests tend to cover systems with gaseous acetylene at close to atmospheric pressure, which is the typical condition used for distribution through piping systems. Very little of the reported data covers pressures as high as 1.5 MPa, the filling pressure of acetylene cylinders though high pressures appear to favour detonation. No information has been reported about explosions involving acetone or mixtures of acetylene in acetone in the presence of pure oxygen. 1.4. Protection systems against explosions in acetylene cylinders Deflagrations can occur in the supply tube, so the cylinders must have systems for quenching the flame. Acetylene cylinders contain two protection systems for quenching deflagrations in the free gas.

Table 1 Some relevant velocities for consideration of explosive loadings Physical phenomenon

Approximate velocity, m/sa

Sound in air at 1 atm, 20 C Sound in acetylene Sound in water Sound in steel Deflagration wave in acetylene/air mix at 1 atm Detonation wave in pure acetylene [2,3] TNT [4]

344 490 1500 4700–5000 Wide range up to sonic velocity 1800–2000 3190 (19 mm diameter) 4815 (152 mm diameter)

a Velocity data for shock waves are only indicative because there is significant dependence on factors such as confinement, shape and the mass of explosive involved.

708

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

The first protection is that the acetylene is mainly dissolved in liquid acetone inside the cylinder. In the early history of the use of acetylene, the bottling of high pressure acetylene or liquid acetylene proved to be highly dangerous. Dissolving the acetylene is used as a way of putting a reasonable mass of acetylene into the cylinder at moderate pressures. Acetone has a remarkable affinity for acetylene gas in normal operating temperatures and at 15 C and 1.5 MPa can absorb 554 g of acetylene per kg of acetone. Dissolving the acetylene in acetone is quite a complicated process that delays filling operations and increases the number of processes required for filling. The role of acetone in the suppression of an explosion is not entirely clear, since acetone could participate itself in a reaction with oxygen. Acetone appears to have the effect of delaying the transfer of the effects of deflagration in the gas space to the dissolved acetylene [6]. The delay involved is in the order of tens of seconds, which is adequate to quench the reaction. The second protection system is that the volume of the cylinder is filled with a monolithic porous mass of about 90% void. The porous mass is made of a specially formulated fibrous concrete type substance that is poured into the cylinder during manufacture and baked solid. The porous mass used to be made of granules or loose fibre and contained significant amounts of asbestos. In recent years, monolithic masses were composed of proprietary formulations involving calcium silicate. The porous mass must not contract after filling and must be demonstrated to be stable through a series of drop tests. The exact way in which the porous mass can prevent detonation is not entirely clear from the literature. The mass involved is quite small, since the porous material is 90% void. The important aspect may be that the porous mass prevents any free fluid being present. Free fluid could swill around and develop bubbles. Such bubbles have been shown to sensitise some solutions to initiation from impact, though there is no published literature relating to sensitisation of acetylene solutions in acetone by this technique [6,7]. Acceptance of the design of acetylene cylinders is on the basis of tests in which deflagrations are initiated in the stem area above the cylinder. The deflagration must be shown to be quenched by the contents of the cylinder without fragmenting rupture. There are visual inspections of the cylinders through the opening every ten years and the mass of acetone is checked during every filling. What is not clear in the literature is what happens when the acetone solution is hit by a fully developed detonation. In fact, the solution may be able to participate in the detonation if oxygen is present. There is some evidence dated 1897 [8], which suggests that above a pressure of 1 MPa a detonation could occur in the solution if adequately initiated. 1.5. Initiation in acetylene For an explosion to occur in the cylinder, it is necessary for it to have been initiated. In the incident in Sydney, the only possible cause of initiation seems to be impacts on the cylinder caused by handling the cylinder on a truck. However, the energy available from such an initiator is very low in the region of a small fraction of a Joule. In acetylene gas, there appears to be the possibility of a direct initiation, where the deflagration step is either very small or is omitted [3]. The presence of oxygen makes detonation possible at lower levels of initiator energy (which can be termed as being more ‘‘sensitive’’). There is some information about direct initiation using sparks and other initiators in acetylene–oxygen gas mixtures up to 1 atm in [9]. 1.6. Literature concerning violent fragmentation Published information on explosions in similar gas cylinders is rare, but reports were available about two relevant explosions. One relates to an acetylene cylinder where oxygen was present in Munich 1992 [10] and another in an oxygen cylinder (which must have received some fuel) in Corpus Christi, Texas in 1963 [11].

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

709

The work investigating the Corpus Christi incident included a series of tests involving explosions of cylinders containing various fuels and oxygen or air. These tests are perhaps the most useful practical information concerning fragmentation of cylinders currently available. From the pictures that were published, it is clear that the result most closely replicating the sort of fragments seen in Sydney was obtained with a ‘‘2 parts hydrogen 1 part oxygen mixture’’ at 12.06 MPa. This explosion, it must be pointed out, involves entirely gaseous reactants, demonstrating that at high pressures, gases can cause fragmentation giving the appearance of the effects of condensed phase explosion as discussed earlier. The rarity of the event is attested to by the fact that there is no known precursor event where a cylinder of acetylene is known to have fragmented violently in the history of cylinders containing porous mass. One of the international experts in the legal proceedings estimated that there have been 9 billion fillings of such cylinders [12]. 2. Investigation A large number of fragments were provided to Monash University including parts of the neck, bottom, cylindrical walls and fittings (see Fig. 2). These fragments represent less than half the mass of the cylinder and the collection shown in Fig. 2 does not include the smaller fragments which were also collected. The detail of the failure can be seen on a typical part, Fragment 2, as shown in Fig. 3. This fragment is approximately 360 mm long, 40 mm wide and approximately 4.1 mm thick (about 33% thinner than the presumed original thickness) and fairly flat. This specimen is relatively undamaged by subsequent contacts and appears to come from the plain cylindrical part of the vessel. It is surrounded on all sides by about 80 separate plastic failures some of which are shown in Fig. 3. The failure surfaces are characterised by being at approximately 45 to the original surface of the cylinder. Visual microscopy of specimens at 1000· magnifications reveals similar metallographic structure in the exploded cylinder as in an undamaged cylinder. The main difference is that in the exploded cylinder, the grains had been significantly distended in the circumferential direction but less so in the longitudinal direction of the cylinder. There was no evidence of micro cracking or cracking parallel to the surface of the

Fig. 2. Some of the larger fragments provided to Monash.

710

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

Fig. 3. Fracture surfaces on fragment (2· magnification). Fracture plane reversals are indicated by arrows.

cylinder. Some of the more severe effects of explosive interaction, such as twinning and shock induced martensite [13], were not observed on the several samples inspected microscopically. Other investigations were conducted including examining all potentially relevant filling records. Anomalies were discovered in some of these and the evidence is that the practices at the filling plant did not always follow the companyÕs written procedures. The filling procedure has some moderately complicated steps involving weighing and measuring the pressure and pressure of the cylinders, because the quantity of two fluids (acetone and acetylene) has to be checked both before and after the filling. While the violations of the filling procedures could lead to problems, no direct link was established to the explosion.

3. Discussion 3.1. Explanation of mode of failure When analysing very high speed incidents, such as the Sydney acetylene cylinder explosion, there is a need to understand the very different dynamics involved when compared to the normal experience of pressure vessel failure. The author has published a paper on the basics of this [14]. To obtain very small fragments, the pressure transient in the fluid must travel faster than the speed of sound in the metal. The transient pressure can be extremely high (perhaps tens of GPa) and applied for only a very short period of time. In this case, there are profound implications on the resulting interaction with the enclosing structure such as a pressure vessel. A hypothesis of how the fragments were produced in the Sydney acetylene cylinder explosions is as shown in Fig. 4. The loading occurs in a small band (marked ‘‘reaction region’’) which is travelling faster than the speed of sound in the fluid. It is followed at a later stage by a pressure relief wave associated with the expulsion of gas from the breaks in the pressure envelope. In Fig. 4, the deformation of the cylinder occurs in three zones following the passage of the detonation wave. In Zone A, the cylinder expands without failing, in Zone B the cylinder starts to rupture longitudinally as the ductility of the material is exhausted in the circumferential direction. In Zone C, the strips formed are hypothesised to break further as the kinetic energy of the material continues to stress the material beyond its failure point.

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

711

Fig. 4. The hypothesised process of fragmentation in a cylindrical shell loaded by a detonation.

The kinetic energy imparted to the shell by the pressure in the reaction zone is adequate to cause at first bulging (Zone A), then linear tearing (Zone B) and finally fragmentation (Zone C). The pressure in the vessel is relieved by a front that moves at the speed of sound. Vf is the velocity of the shock wave associated with the detonation and Vo is the radial velocity imparted to the cylindrical shell by the passage of the high pressure. For the speed of the reaction front to approach the speed of sound in the metal (see Table 1), it is necessary for the reacting medium to be a condensed phase. This implies that the contents of the vessel were at the time of the explosion a solid, liquid or possibly highly pressurised gas. 3.2. Cause of the incident The focus of the investigations as to cause is related to three main issues:
Why were the contents of the cylinder sensitive to this initiator?
Why was this error not detected?
If the error occurred during filling, how can explosion occur at a period of days to weeks after filling? In the case of explosions, there is often the situation where there is almost complete destruction of any detailed physical evidence about the prior condition of the material that exploded or the initiator. This was the situation for this incident. The investigators were thus faced with trying to sort through a series of suggestions looking for the most probable cause.

712

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

Given that there was no evidence for two important possibilities for initiation, deliberate sabotage and preceding fire, the only remaining sources of initiation are minor events such as dropping the cylinder. The only way that such a low energy initiator could result in detonation is if the cylinder was filled with some sensitive and energetic material. Several suggestions as to the sequence of events which led to the explosion were made by experts in the legal proceedings [15]. These are listed in Table 2. In the comments column, the author has rejected two of the proposals and combined the remaining into three separate sequences. The three remaining sequences have been drawn in the form of a fault tree in Fig. 5. 3.3. Quantification In common law countries such as Australia in civil cases, the standard of proof is known as ‘‘on the balance of probabilities’’. Since a fault tree has been drawn for this incident, a quantitative analysis can be attempted which may be helpful for this standard of proof. Below, this is attempted using the techniques of probabilistic analysis [16].

Table 2 Event sequence considered by various experts Suggested sequence: number

Suggested initiating event and its location

Normal main line of protection (This must be violated for the explosion to occur)

Gestation period

1

Sensitive contents added or liquid full. Location: filling plant Fill acetylene cylinder with oxygen on the oxygen line. Filling plant Fill acetylene cylinder with oxygen. At third party (i.e., not at filling plant) Fill acetylene cylinder with oxygen. At another port of call or on-board ship

Checks at filling plant, acetylene line Checks at filling plant, oxygen line

Days to weeks

Checks at filling plant

Days to months

Transport yard should send all incoming cylinders to contracted filling plant before reuse Transport yard does not fill cylinders Checks at original manufacturer and filling plant

Days to months

Subset of Sequence 3

Nil to weeks

No equipment to do this: rejected Treated as a subset of Sequence 2

Faults are undetectable

Years to nil

2

3

4

5 6

7

Fill acetylene cylinder with oxygen. Transport yard Defective porous mass. (Original manufacturer.) ‘‘Overfilled so that a super saturated acetylene in acetone liquid formed.’’ Filling plant Bubbles in liquid. Contents sensitised by impurities

Comments on how this sequence will be considered in the probability analysis

Days to weeks

Years and days to weeks

Faults proposed are due to minor events which are undetectable. If this sequence can occur, then it should be a common event. Explosions are not a common event, so this suggestion can be rejected as being speculative

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

713

Fig. 5. Fault tree related to Table 2.

3.3.1. Initiating sequences A rough ranking of initiating sequences is possible as follows: Sequence 2 is chosen as the base example, since some errors of filling had been detected at the filling plant in question. This is given a probability of a. Sequence 3. There is some evidence that this sequence has occurred given by another expert (though details were not given). This will be assumed to be the same probability per bottle as Sequence 2, i.e., a. However, the frequency of bottles arriving at the transport yard also has to be considered. There is supposed to be no filled bottles coming from third parties, that is, off ships or from other users. A filled bottle is required for there to be a condensed phase explosion. As a result, the total probability for this sequence will be at least 2 or 3 orders of magnitudes lower than sequence 2. Let this have a probability of a/100. Sequence 1 is speculative, since no known cases have been reported. There was some debate between the experts as to whether this sequence was possible since the excess mass in the filling line would have been readily detected, though against this was evidence of improperly weighed cylinders. The probability for this case was put at 2 or 3 orders of magnitude lower than Sequence 2, that is, a/100. 3.3.2. Lines of protection The lines of protection in each of the three sequences depend on the vigilance and systematic behaviour of people at the cylinder filling company. In Sequence 3, it also depends on the transport company doing what they claim they do, that is send all received cylinders to the filler. For this reason, the lines of protection are of the same order in their success rate for the three scenarios, say, b. A lower rate for Sequence 3 is required, since the handling error has to occur at both the filler and the transport yard. Thus, the probability of this will be b2 but will be conservatively for our purpose set to b/2 for simplicity.

714

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

3.3.3. Gestation period Gestation period is only a few days or weeks for faults emerging from the filling plant. Some suggestions have been made for how such a period is possible, including cooling processes and mixing processes. Given the ability to survive to the correct gestation period a probability of c. The gestation period for faulty cylinders emerging from overseas locations is much more problematic, since this would be longer and more handlings would occur in the period. Thus, the probability of Sequence 3 should be reduced by an order of magnitude from this sequence, that is, c/10. 3.3.4. Ratios of total probabilities The various probabilities in each sequence are multiplied together to get the comparative probability of each sequence as follows: PrðSequence 1Þ ¼ a=100  b  c ¼ abc=100; PrðSequence 2Þ ¼ a  b  c ¼ abc; PrðSequence 3Þ ¼ a=100  b=2  c=10 ¼ abc=2000. The values of a, b and c are not known. However, because all of the sequences contain these factors, the probabilities of the sequences can be compared. Sequence 2 is 100 times more likely as that of Sequence 1 and 2000 times more likely than Sequence 3. This means that Sequence 1, filling of an acetylene cylinder with oxygen at the filling plant, has near to 99% probability of being the cause of the accident. Even if the other sequences (having about 1% probability) are considered, quality checking failures at the filling plant are still likely to have been involved in the explosion.

4. Conclusion The paper describes the fragments resulting from an extremely violent and unusual acetylene gas cylinder accident, which occurred in Sydney in 1993. The failure is significant because it appears that the cylinder was being handled in a fairly normal way at the time and there was no readily apparent fault with the cylinder. Millions of such cylinders are in use worldwide with extremely few incidents. No identical incident has been found in the literature or identified through enquires with other parties. In previously published work [14], it was established that the fragmentation observed could only be caused by an incident in which a detonation occurred with a velocity at or above the velocity of sound in the steel of the cylinder. Such detonation velocities are only possible in the condensed phase, that is in a liquid or solid phase or possibly in a highly compressed gas. Information from a previous investigation [11] suggested that highly compressed oxygen (over 10 MPa) with a fuel could cause such fragmentation even in the gaseous phase. Given that such a destructive detonation occurred, the focus of the investigations as to cause related to three main issues:
Why were the contents of the cylinder sensitive to this initiator given that the only initiator which seems to be available for the explosion is a low energy impact?
Why was this error not detected at the filling plant?
If the error occurred during filling, how can explosion occur at a period of days to weeks after filling? Though no definite answer to these questions can be given, a most probable cause analysis using fault tree mathematics suggests that filling the acetylene cylinder with high pressure oxygen was more likely than

J.W.H. Price / Engineering Failure Analysis 13 (2006) 705–715

715

other possibilities. This eventuality is supposed to be impossible because of the colouring of the bottle and the fact that the fittings have different threads. However, both of these can be violated.
In the specific case of this type of bottle, the steel shell is also in use as an oxygen container and both types of cylinder have a right hand thread at the neck to valve connection.
The colouring of the cylinders actually varies through life due to corrosion (which is quite like the acetylene maroon), oil or other contaminants (which is like the oxygen black) and repainting at inspection stations (where the valve is also removed for internal safety inspections).
The delay to the explosion is explained by the time it took for the oxygen and the acetone to mix adequately. It seems possible at the very low frequencies suggested for this type of incident (1 in 9 billion fills) that these events may have occurred in the Sydney case, and that this was the most probable cause of the incident. It is to be noted that due to out of court settlements, no court judgement on this issue has been recorded.

Acknowledgements The incident was initially investigated and reported by the Workcover Authority of New South Wales and the author is indebted to the assistance of Lucian Kent and Gary Mason of the Authority.

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15]

[16]

Fitzpatrick E. Man dies in gas tank blast. Sydney Morning Herald 1993;11(October):2. Miller SA. Acetylene, its properties, manufacture and uses. London: Ernst Benn; 1965. Baker WE, Tang J. Gas, dust and hybrid explosions. Amsterdam: Elsevier; 1991. Fordham S. High explosives and propellants. Oxford: Pergamon Press; 1980. Sutherland ME, Wegert G. An acetylene decomposition incident. Chem Eng Prog 1973;69(4):48–51. Miller SA. Acetylene, its properties, manufacture and uses. London: Ernst Benn; 1965. p. 502. Lee HS. Comments on the conclusions of an investigation of the explosion of an acetylene gas cylinder explosion. 36 Doody St., Alexandria, NSW; 10th October 1993. McGill University, Canada; 10 December 1996. Miller SA. Acetylene, its properties, manufacture and uses. London: Ernst Benn; 1965. p. 501. Lee JHS, Matsui H. A comparison of the critical energies for direct initiation of spherical detonations in acetylene–oxygen mixtures. Combust Flame 1977;28:61–6. Kuegel P. Acetylene cylinder explosion. Munchen: TUV, Bayern Saschen; 26 October, 1993 [private communication]. Mathews LG. Investigation of an oxygen cylinder explosion. In: Compressed gas association meeting, US; 1964. Tribolet RO. Report for the purposes of workcover prosecution; August 1997. Walsh B. The influence of geometry on the natural fragmentation of steel cylinders. Report 533. Department of Supply, Defence Standards Laboratories, Melbourne; January 1973. Price JWH. An acetylene gas cylinder explosion. ASME J Press Vess Technol 1998(February):62–8. The suggestions were made in the course of court proceedings by various parties including the current author, Lee HS. Comments on the conclusions of an investigation of the explosion of an acetylene gas cylinder explosion, 36 Doody St., Alexandria, NSW; 10th October 1993. McGill University, Canada; 10 December 1996.Tribolet RO. Report for the purposes of workcover prosecution; August 1997.Green AR. Report on the acetylene explosion, 36 Doody St., Alexandria; 10/10/93. Department of Safety Science, University of New South Wales; 12 June 1997. Kumamoto H, Henley EJ. Probabilistic risk assessment and management for engineers and scientists. New York: IEEE Press; 1996.