0957–5820/04/$30.00+0.

00
# 2004 Institution of Chemical Engineers
www.ingentaselect.com=titles=09575820.htm Trans IChemE, Part B, March 2004
Process Safety and Environmental Protection, 82(B2): 105–127

REVIEW PAPER

FLIXBOROUGH: THE EXPLOSION AND ITS AFTERMATH

J. E. S. VENART*
Department of Mechanical Engineering, University of New Brunswick, Fredericton, New Brunswick, Canada

T
he Flixborough explosion was the largest-ever peacetime explosion in the UK. There
were 28 fatalities as well as the near complete destruction of the NYPRO plant in North
Lincolnshire by blast and then fire. An official Court of Inquiry was established and
charged with the responsibility of determining the cause(s) of the disaster and any lessons to be
learnt. This paper reviews the disaster, its investigation and inquiry, its attributed cause, and an
alternative possibility, and this redraws the lessons to be learnt for modern chemical plant
design and accident investigation. The reanalysis suggests that the failure was caused by a
complex two-step mechanism that resulted from the initial failure of only one bellows, a release
of about 10–15 tonnes of cyclohexane, and the detonation of the consequent vapour cloud with
an explosive effect of some 280 tonnes of TNT. This is compared with the single-step failure
process reluctantly acknowledged by the Court; the amount of release accepted, 40–60 tonnes
of cyclohexane, and its explosive effect, 6–16 tons of nitroglycerine. The results should be of
interest to industry, regulators, and loss prevention and risk assessment specialists.

Keywords: Flixborough; disaster; vapour cloud explosion; cyclohexane; loss prevention; risk
assessment.

BACKGROUND Office buildings. Adjacent, to the east, were Sections 7 and

On June 1, 1974, at about 15:53 UT, there was an explosion
and fire at the Nypro (UK) Ltd works near Flixborough. An
accidental release of cyclohexane at approximately
0.96 MPa and 155
C resulted in an vapour cloud explosion
(VCE) which caused 28 fatalities, the near-total destruction
of the 24 ha plant, severe injury to the public, significant
damage to many structures in the surrounding countryside
(Secretary of State for Employment, 1975), and scattered
debris as far as Anlaby on the outskirts of Hull 32 km away.
The plant was located on the east bank of the river Trent,
a tributary of the Humber, about 50 km west of the port of
Grimsby. On the west bank of the river, opposite the works,
was the village of Amcotts; the village of Flixborough was on
a small rise to the East; the town of Scunthorpe was about
1.5 km SE of the location: all suffered major distress though
no loss of life (Figure 1).
The catastrophe was initiated in section 25A of the
process stream during start-up while the mostly cyclohexane
feedstock, inerted with nitrogen, was under hot recycle
through the reactor train, R2521–R2526 (here called R1–
R6); only one-third of the 70 kg s1 full production flow
was circulating at the time. Section 25A was South of the
Laboratory and midway between it and the Project and Main
*Correspondence to: Dr J.E.S. Venart, Department of Mechanical Engi-
neering, University of New Brunswick, Fredericton, New Brunswick,
Canada, E3B 5A3.
E-mail: jvenart@unb.ca
27 (the Caprolactam Plant). The reactor train, which was the
source of release, was aligned primarily north–south and on
the side of the section.
Exactly 2 months before the explosion a temporary
508 mm outside diameter (o.d.) 304L stainless steel (SS)
pipe assembly was used to bridge the space between reactors
R4 and R6 as R5—which was leaking—had been removed.
This reactor was placed in waste ground in front of the
Project Office and alongside the Section. At the time of the
incident the wind was blowing from South-Southwest
4–7 m s1. The perceived centre of the explosion, as deter-
mined by several investigators, was identified as being at the
roadway just in front of the Main Office (Figure 1).
In the original oxidation sequence, all reactors were
connected by 711 mm inner diameter (i.d.) bellows and
711 mm o.d. stub-pipe assemblies (Figure 2). Each 12-
convolution 750 mm mean diameter bellows was made
from 316L SS sheet 1 mm thick. The convolutions were
37 mm high and at a 33.9 mm pitch; all were about 230 mm
long overall and welded to 316L 711 mm o.d., 6.75 mm
wall, loose flanged stub-pipes. The bellows and a backing
ring were fillet welded to each stub pipe (Foley, 1974a). In
the modification, however, the temporary 4.8 mm wall
thickness pipe-bridge, which used the two original bellows
and their 316L SS stub-pipe assemblies, had been installed
between R4 and R6 (Figure 3).
The 5 m high by 3.5 m diameter reactors were made from
13 mm mild steel with a 3 mm SS weld inlay liner. Each was
fitted with an interior overflow weir and inlet baffle plate to

105
106 VENART
Figure 1. NYPRO site plan showing Section 25A: the reactors; explosion origin, X; and CFD study area, cross hatched (modified from Figure 1, Secretary of
State for Employment, 1975, under the terms of Crown Copyright Policy Guidance issued by HMSO).
ensure level control and bottom feed as well as a gas sparger
for oxidation and a line for catalyst addition. The height of
flow over the weir was about 115–130 mm under full pro-
duction and 40 mm at the time of the incident (Artingstall,
1974). All but one reactor, R4, had an agitator (Artingstall,
1974). A 13 mm mild steel flanged skirt served as the reactor
base and this was affixed to a 460 mm thick reinforced
concrete slab atop stepped concrete plinths. Pressure equal-
Figure 2. Section 25A during construction (before vessels were insulated)
showing the six reactors and their connections (source J. I. Cox, 1976,
reproduced with permission from J. I. Cox).
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
ization between reactors was provided by individual off-
gas lines connected to a common header with pressure
relief.
The interconnecting pipes and their bellows had originally
permitted the hot pressurized feedstock to flow over the outlet
weir in the upstream reactors, through the bellows stub-pipe
connection into the downstream reactors with a fixed eleva-
tion difference of 178 mm at a velocity of about 0.3 m s1
through the nearly full pipes of full production flow. The
original freeboard of 230 mm also provided additional
pressure equalization capability, although, if pipes became
full and the off-gas vents blocked—say by a two phase
mixture—pressure balance could be compromised. The
‘dog-leg’ pipe adaptation, now, however, doubled the head
between R4 and R6 to 356 mm so there was a significant
weir ‘waterfall’ (velocities from 1.8 to 2.6 m s1 dependent
upon production—full=recycle) in addition to hydraulic
jumps in the very complex flow structure (Figure 4).
At full production the R4–B4 stub-pipe was nearly half
full at a velocity of about 0.5 m s1. Immediately after the
bellows, with transition into the 508 mm section, the flow
accelerated and became supercritical, ending in a significant
hydraulic jump near the lower mitre bend nearly filling the
pipe—due to a reduced freeboard of 120 mm—and then
slugging into the lower B6 bellows and thence into R6;
here the flow was quiescent (Teng-yang et al., 2000);
velocity at full production was again about 0.3 m s1.
 FLIXBOROUGH 107

Figure 3. Reconstruction of the R5 bypass assembly showing the two bellows and stub pipes, the 508 mm dogleg pipe and support scaffolding (source
Secretary of State for Employment, 1975, is reproduced under the terms of Crown Copyright Policy Guidance issued by HMSO).

Under the conditions of hot recycle the weir overflow
generated greater velocity and turbulence due to the
increased fall.
The modification made was thus much more than just a
geometric alteration. Significant changes were made to the
statics, the dynamics and the flow. First, due to the dog-leg
shape of the bridging pipe and the differences in pipe
diameters, there were a couple which latterly loaded
the pipe and the bellows in the vertical direction. Second,
the assembly was a mass–spring system with very little
support provided by the scaffolding, especially upon vertical
thermal expansion of the reactors. And, finally, there were
the significant discharge and flow modifications.
The process fluid in this portion of the plant, cyclohexane
(C6H12, MW 84), is a colourless liquid similar in some of its
properties to gasoline [flash point 20
C, auto-ignition
temperature (AIT) 245
C]. It has a normal boiling point
of 80
C and at 150–155
C saturation pressures, psat, of
0.55–0.60 MPa. When the liquid was saturated with nitro-
gen, or air—as when oxidation was occurring in the reactor
stream—the pressure was 0.96 MPa. Under these condi-
tions the liquid was 99.5–99.6% C6H12 saturated with

Figure 4. Reactor vessels R4 and R6 with the temporary bypass indicating liquid levels at full production (source Teng-yang et al., 2000, reproduced with
permission from IChemE).

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
108
0.5–0.4% N2 or air; the vapour would have consisted of
about 61–68% C6H12 and 39–31% N2 or air (NIST, 1999).
The lower flammable (LFL) and upper flammable limits
(UFL) of a vapour–air mixture by weight, under atmo-
spheric conditions, are 3.5 and 20 wt% (1.3; 7.8 vol%),
respectively.
The scale of the disaster and damage can be appreciated
from photographs taken soon after the explosion and shown
in Figures 5 and 6.
In Figure 5 we are looking north over the rubble pile of
the Main Office building at Section 27. Notice the crushed
columns to the right (east) of the Section and the, amazingly,
still near-vertical lamp-post, all indicators that this location
was near the center of the explosion. Note also that the
columns and vessels on the west (left) side of the section are
uncrushed. Had the explosion occurred during the week
the three-storey Main Office would have been occupied.
In Figure 6 the car park to the southeast of the Main
Office building is shown. This spot is near where the main
entrance and Gate House would have been. To the right, the
Figure 5. Looking north over the rubble of the Main Office Building into
Section 27 (photograph reproduced with permission from the Scunthorpe
Evening Telegraph, www.thisisscunthorpe.co.uk=photosales).
Figure 6. The NYPRO car park to the east of the Main Office Building
(photograph reproduced with permission from the Scunthorpe Evening
Telegraph, www.thisisscunthorpe.co.uk=photosales).
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
VENART
east, are the remains of the warehouse and flaking plant
where there were four fatalities. Note that the damage to
vehicles in the car park suggests very large over-pressures—
out in this area there were two deaths, a chemical worker
found 55 m from the plant entrance and 12 m from the car
park perimeter and the Main Gate security guard found
120 m from the centre of the explosion and 30 m from the
car park perimeter. A tanker truck, in the centre background,
was completely crushed and its contract driver found nearby
at the plant weigh-bridge, some 120 m from the explosion
centre.
THE COURT OF INQUIRY
An Official Court of Inquiry was immediately appointed
by the Secretary of State for Employment, Michael Foot, ‘to
establish the causes and circumstances of the disaster and to
point out any lessons . . . to be learned therefrom’. The
Court comprised R. J. Parker, QC (Chairman), J. A. Pope
(Deputy Chairman), J. F. Davidson and W. J. Simpson and
heard from over 170 witnesses over a period of 70 sittings.
The essential problem faced by the Court was ‘to determine
what had caused the rupture of the by-pass.’ However, the
only possibilities considered in detail were: ‘(i) rupture of
the by-pass assembly through internal pressure; (ii) rupture
of the assembly in two stages; a small tear in the bellows
[B4] leading to an escape and a minor explosion causing
final rupture; and (iii) rupture of the ‘8 inch’ line at the
50-inch split leading also to a minor explosion causing
rupture of the by-pass assembly’.
Processes (ii) and (iii) were interpreted to require an
external minor explosion to trigger events; little evidence
for this could be determined, however. All required greater
than normal internal pressure. Possibility (ii) was ruled out
early in the investigations as extensive testing of six new
bellows showed that pressures in excess of those required to
activate the pressure relief system were necessary and also
that low cycle fatigue would not extend the very small
resulting tears (Batstone, 1975). So, despite a large internal
tear found in bellows B4 (Foley, 1974a), and compelling
eyewitness evidence, only three paragraphs in the final report
and a few pages in the Court’s record, no definitive studies
were further devoted to the two-stage rupture possibility.
Processes (i) and (iii) only were examined in depth and in
near equal measure in the Report of the Inquiry, the Records
of the Court and the supporting studies. Other possibilities
considered for an internal initiating event (e.g. peroxide
decomposition and the operation of a nitrogen purge) were
excluded, as was the likelihood of fatigue and a water–
cyclohexane superheated explosive boiling interaction (King,
1975, 1977, 1999). In this case recent experiments under-
taken for the UK Health and Safety Executive on behalf of
the Minister of the Environment (Snee, 2004) indicate that
water trapped in the bottom of R4 did not play a significant
role in the event; further experiments to examine the
influence of venting are planned.
On the basis of expert evidence by Professor D.E.
Newland (Newland, 1976), based on a dynamic analysis
and assessment of energy requirements for both bellows to
squirm and the by-pass pipe to buckle as one event, the
Court concluded, albeit with low probability, ‘that the
disaster resulted from a one stage failure of the 20-inch
assembly’—as a result ‘of conditions of pressure and
 FLIXBOROUGH
temperature more severe than any which had previously
prevailed but no higher than careful and conscientious plant
operators could be expected to permit.’ This failure process
was supported only partially by an extensive full-scale
experiment (Games and Waterhouse, 1974), which indicated
that bellows squirm and pipe-bridge buckling of the assem-
bly (as shown in Figure 3) would occur in about 100 ms.
Other experiments and bellows tests at the same and more
severe operating conditions were also conducted (Batstone,
1975). All these experiments were, however, unable to
reproduce fully the failure process accepted, the physical
evidence, and the experience of witnesses within their
observation time-frames. The Court did state, though, that
their conclusion ‘would be readily displaced if some [other
event of] greater probability to account for the rupture could
be found’ (Secretary of State for Employment, 1975).
The Court’s qualified conclusion as to cause has been the
subject of considerable controversy, especially regarding the
actual failure process (e.g. Ball, 1975, 1976; Butler, 1975;
Cox, 1976; Gugan, 1976; King, 1977; Warner, 1975;
Warner and Newland, 1975), the amount of cyclohexane
released, and whether the unconfined vapour cloud formed
in the release detonated (e.g. Gugan, 1979, 1980a,b; Ale and
Bruning, 1980a,b; Fu and Eyre, 1974; Phillips 1981). The
debate and study continue to this day (e.g. Gugan, 2000;
Høiset et al., 2000; King 2000; Kletz 2000; Swan 2000).
In view of the role this event has played in the chemical
and insurance industries and its pivotal function to loss
prevention and regulation, a further reexamination now
appears warranted. In this, greater attention will be focused
on possibility (ii)—rejected by the Court—as a two-stage
event was certainly experienced by most witnesses. In what
follows the bellows, the bridging pipe and its assembly into
the by-pass are modelled and a release scenario that consid-
ers the likely sequence of events initiated by the crack noted
in B4 is advanced. This is done because flow-induced
excitation and fatigue of bellows was not considered in
depth by the Court and this may have been responsible for
the initiation of events.
Each bellows had a first natural frequency of about 65–
71 Hz axially and between 325 and 416 Hz in the transverse
direction. The system (pipe, stub-pipes and bellows) reso-
nance frequency, as determined for a straight pipe, was
3.9 Hz axial, 15.7 Hz lateral (ends of the spool pipe in
phase), and 27.2 Hz for the rocking mode (EJMA, 1993).
The excitation of this very complex spring–mass system was
now such that resonance with portions of the flow was
possible (Teng-yang et al., 2000). Most of the lower
frequencies could have been excitable by the very broad
range of flow velocities present, particularly near the
hydraulic jump and regions of flow acceleration. The jetting
at the base of the weir outfall could also result in the axial
resonance of the bellows since these were about one pipe
diameter away.
With flow-induced vibration of bellows, no significant
response is observed up to some critical velocity and then,
as the velocity is increased further, the convolutions become
unstable. For uniform mean flows the critical Strouhal
number is St ¼ 0.45 (Weaver and Ainsworth 1989); for
non-uniform flows, as at the exits of elbows leading to
bellows and perhaps the weir outfall region, St ¼ 0.57
(Jakubauskas and Weaver, 1998a,b). The Strouhal number,
St ¼ ( fp)=V, where f is the bellows axial natural frequency
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
109
in Hz, p is the bellows pitch (m), and V is the mean flow
velocity (m s1). The flexibility of expansion joints makes
them susceptible to self-excited vibration, resulting in bend-
ing stresses that may produce fatigue failure (Weaver and
Ainsworth, 1989). Gigacycle fatigue damage to the thin wall
bellows may thus have occurred since between 20 million
and over 300 million cycles were possible within the range
of potential excitations present during the 2-month operating
period. In the fatigue testing of steels, trials are usually
terminated before 100 million cycles since a ‘fatigue limit’
is normally registered between 1 and 10 million cycles. How-
ever, tests conducted with high-frequency random vibration
(similar to those associated with the natural resonance of a
structure) show that the fatigue limit of steels can disappear
and that fatigue failures occur (Miller, private communica-
tion 1999; Au-Yang, 1999).
It is worthwhile to review the Court’s concerns on fatigue.
The only serious questioning on the subject was raised by
Dr Pope, Deputy Chair of the Court, and a mechanical
engineer, in his questions to Dr Foley (Foley, 1974d).
Drs Gill and Kitching (Gill, 1974; Kitching, 1974) as well
as Dr Ryder (Ryder, 1974) also raised concerns regarding
flow induced and other vibrations and fatigue. Dr Foley had
earlier stated that ‘detecting fatigue failure in thin sheets is
difficult’ (Foley, 1974e). In order to determine evidence for
fatigue in the bellows, over 26 m of failure surface would
have had to have been carefully examined. Dr Foley had
stated (Foley, 1974b,c) that his inspection was only made by
‘eye and a small lens’ and thus the possibility of fatigue,
especially at the interior tear in the upstream B4 bellows,
cannot be ruled out since a more careful examination would
have necessitated greater magnification. Dr Miller, too, was
questioned on this subject (Miller, 1974).
THE EVIDENCE
The Safety and Mines Research Establishment (now the
Health and Safety Laboratory of the Health and Safety
Executive) was responsible for the work of recovery, identi-
fication and examination of the wreckage and the conduct of
further testing. Creamer and Warner, Consulting Chemical
Engineers, were appointed technical advisors to the Court.
HM Coroner was A. A. Collins. The Home Office Patho-
logist was Dr A. Usher of the University of Sheffield (Usher,
1979).
Witness Statements
Humberside Police were in charge of the accident inves-
tigation (Humberside Police, 1974). Interviews were
conducted, and statements taken from nearly 300 indivi-
duals. These included 43 shift survivors, 54 off-site
witnesses, 78 employees on the June 1 7 a.m.–3 p.m. and
8 a.m.–4 p.m. shifts, as well as 114 other employees and
contractors. Dr J. I. Cox worked with many of the survivors
and witnesses to reconstruct events and confirm the timings
made in statements (Cox, 1976).
In 1974 Nypro had grown to employ 560 persons of
whom some 260 were shift workers (Westgate, 1975). June 1
was a Saturday and so only about 70 persons, employees
and contractors, were on site. Of these, Humberside police
were able to take statements from 43 individuals; nearly all
110 VENART
(42) indicated that the lead up to the explosion had consisted
of some noisy events in the vicinity of the southwest end
of section 25A—a ‘bump’, a release of high-pressure gas
(‘like a PRV vent’) for about 2–4 s, deep rumbling, more
noise—a ‘bang’—with debris being scattered upwards, a
‘roar like a jet engine’, massive upward and southern
releases concurrent with a westward developing ground
flow, and small fires about the base of the reactors, all
taking place over about 30 s prior to the explosion. In the
3–4 s initial period some observed something like a ‘tank
top’, or ‘sheet of metal’ 1 m diameter, being thrown up
and spiraling downwards, one on site, L. Harry, and three
off-site, J. A. R. Cooke, S. Hill, and H. Toyne. Many then
observed debris and dust being thrown upwards followed by
vertical and then horizontal jets of discharged material
(Humberside Police, 1974; Harry, 1974).
Of the eight survivors from the Laboratory, seven gave
evidence. Of these, five were specific in the description of
a slowly developing vapour cloud moving to the west over
the time period of their observation, 8–12 s. This cloud
was described as being about 1.5–2 m high and moving
at 5–6 m s1 along the ground across the clear space
between the Laboratory and the Project Office. Its origin
was described as either being near the top of R4 or at its
base, but at least between the two (R4 and R6) reactors.
None of the these witnesses described any northward flow,
i.e. any flow coming towards them. One described debris
being thrown up into the air very early into the event.
Other individuals, one about 90 m east and some 60 m
north of the Main Office building described a cloud of dust
and smoke coming round the southern end of Section 27
(Colquhoun, 1974). Another, some 500 m to the west
observed the development of the cloud from its initiation
as a ‘whisper of ‘‘steam’’’ to its enlargement into a ‘roaring’
jet going up-river (i.e. south) and ending up ‘like a fog’
across the car park (Ayre, 1974).
Various off-site witnesses, e.g. King (1974) about 500 m
away, reported seeing ‘dirty steam’ going up into the air,
followed by ‘what appeared to be an explosion that blew
vapour sideways into the works’ and then the major explo-
sion; Dickinson (1974), about 200 m away, saw a flash like a
large ‘creamy coloured torch’ coming towards him (i.e.
eastward) before the explosion. One described a ‘kettle
spout’ and another a ‘swan neck pipe’ in the vicinity of
Section 25A (Humberside Police, 1974).
A cine film taken immediately after the explosion indi-
cated what appeared to be a short duration moving and
bifurcated near-vertical torch-like fire from the southern end
of the section (Goodchild, 1975).
Physical Evidence
The enormity of the disaster can only be appreciated by
reference to the many photographs included in references
(Secretary of State for Employment, 1975; Sadee et al.,
1976=1977; Gugan, 1979; Roberts and Pritchard, 1982) and
others available in reports prepared for the investigation.
Photographs from the press, particularly those taken by
the Scunthorpe Evening Telegraph, reveal the extent of the
destruction, especially out into the car park south of the
offices where most automobiles were completely crushed
(Figure 6).
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
Blast
The Main and Project Office buildings (Figure 1) were
totally destroyed, likewise the Laboratory and Main Control
Room. W. Thomas, of Amcotts, saw what appeared to be the
interior explosion of the Project Office immediately prior to
the main explosion (Humberside Police, 1974). The 0.4 m
thick concrete second floor of the southeast corner of
Section 27 was displaced downwards (Roberts and Pritch-
ard, 1982), suggesting an aerial flammable cloud at this
location. A tanker truck (cf. Figure 6), located about 150 m
east of the southern end of Section 27 was completely
crushed by an overpressure estimated to be about 1 MPa
(Gugan, 1976). A manhole cover, located on the roadway, in
front of the Main Office building and midway between the
Gatehouse and Section 25A was broken by a similar over-
pressure (Gugan, 1976).
Physical damage to steel chemical plant components, for
example some of the reactors and strong pressure vessels
located in sections 25A, 7 and 27, and closest to the Main
Office building, exhibited crushing of their skirts or bases,
which indicated overpressures of the order of 1 MPa. R5,
located in the waste ground north of the Project Office,
however, only experienced crushing to its skirt on its eastern
perimeter, whereas its western base and the upper sides of
the weaker cylindrical section were undamaged [compare
Plate 7 in Secretary of State for Employment (1975) with
Figure 2.4 in Center for Chemical Process Safety (1994a)].
Crushing was also evident around the base of R6 but not the
upstream reactors (Plate 8 in Secretary of State for Employ-
ment, 1975): the fact that the base of R6 was crushed but not
its weaker barrel suggests that this was under pressure at the
time of the explosion. In the failure mode accepted by the
Court, both reactors R4 and R6 would have discharged
together. Since the upstream reactors had greater inventory
(there were five reactors; the non-return valve after R6
limited the downstream capacity to that of R6 only), their
discharge would have lasted longer than that from R6 and it
should have emptied earlier and not been under pressure.
Additionally the Goodchild cine film as well as Plate 5 of
the Court’s report, both taken after the explosion, show a
significant torch fire from the south end of the section.
Process columns to the southeast in Sections 25A, 7 and
27 exhibited considerable crushing at a variety of elevations
[e.g. Figure 6; Plate 8 in Secretary of State for Employment
(1975), Sections 25A and 27; and Figure 2 in Roberts and
Pritchard (1982)]. High pressure crushing damage was not
apparent, however, into the northern sections of the plant
(Gugan, 1976). This pattern is consistent with the explosion
of both a ground level (to the west out nearly to R5) and an
elevated flammable cloud that encompassed the Main Office
and Project buildings and the southern end of the plant. That
only the southern and SE outer perimeter of plant appeared to
have been subjected to such great pressures suggests that the
initial extent of the fuel air cloud had not penetrated exten-
sively into the northern sections of the plant such as would
have been necessary by the simultaneous dual bellows failure
scenario accepted by the Court. Furthermore, the determined
center of the explosion, just immediately north of the Main
Office building, and stated to have been at an elevation
of about 20–30 m (Sadee et al., 1976=1977), coincides
closely with the cloud development and its centre of mass
for the R4-only discharge, as will be shown later in the CFD
studies.
 FLIXBOROUGH
Pressures sufficient to crush the bases of the reactors can
possibly only be obtained with the detonations of fuel drop–
air vapour clouds where values of 1.9 MPa may occur
(Fishburn et al., 1981; Center for Chemical Process Safety,
1994b; Høiset et al., 2000). Pressures in closed vessel
deflagrations do not normally exceed 0.8 MPa. Detonations
of unconfined fuel–air clouds require significant ignition
energy (Fishburn et al., 1981; Center for Chemical Process
Safety, 1994b), thus requiring, perhaps, a prior gas explo-
sion (Harris, 1983) of, possibly, the Main Office and=or the
Project Office building to act as a ‘precursor’ or trigger for
the detonation. Ionospheric measurements made in the close
range suggested that there might have been a precursor, or
smaller disturbance, to the main explosion (Jones and
Spracklen, 1974) such as could have occurred with one or
both of these buildings exploding. Photographs of the rubble
and debris of both (Secretary of State for Employment,
1975; Sadee et al., 1976=1977) also suggest that an interior
explosion may have occurred there first followed by the
major blast. Alternately, jet-fire initiation is also possible
(Harris and Wickens, 1989).
Other evidence
Metallographic examination of the surfaces of the brid-
ging pipe, stub pipes and bellows indicated no unusual
features except on the surface and attached bellows
remnants of stub pipe R4. This showed a heavy grain-
boundary network of carbides which extended into the
structure (Foley, 1974a). Grain-boundary carbide networks
can be produced in austenitic steels by prolonged heating in
a carburizing atmosphere (i.e. cyclohexane vapours) within
the sensitizing range (400–800
C; Foley and Nicholson,
1974). Two-phase jetting from R4 should have cooled this
structure and thus prevented carbide formation. Upon cessa-
tion of a R4 discharge, however, the impact of a R6 jet fire
could have resulted in the carbide formation. Carbide
formation, along with zinc embrittlement, formed by high
temperature exposure and molten zinc contamination, was
found on portions of the ‘8-inch’ pipes. From this it was
possible to infer probable temperature distributions within
that piping structure (Foley and Nicholson, 1974).
Figure 7 and Plate 9 of the Court’s report show that
portions of the scaffold supports and the plinth crossing
walkway, in front of R6, have been impacted by something
and swept towards the base of R6. There was also an
unexplained impact dent in the front, i.e. facing R4, side
of R6 (Figure 7 and Artingstall, 1974). The inlet baffle plate
of R6 had been torn off and had impacted the agitator shaft,
bending it through an angle of 15
(Artingstall, 1974,
Gugan, 1979).
Experiments
Dog-leg pipe-bellows assembly
A test of the assembly was conducted at Flixborough
using a new pipe built according to the original design and
inserted in place of reactor 3 (R3) (Secretary of State for
Employment, 1975; Games and Waterhouse, 1974). The
bellows utilized were new with axial stiffness of 498 and
482 kN m1 for the B4 and B6 bellows, respectively. Scaf-
folding, similar to that in the original installation, could also
be used. Axial and lateral deflections were measured, both
cold and hot (155 and 160
C), with a chain used to simulate
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
111
Figure 7. Reactors R4 and R6 after the explosion showing reactors R4 and
R6, their stub pipes, the buckled dog-leg pipe lying on top of the R5 plinth,
the cross-walkway deformation, and damage to R6—dent in northwest side
and crushed support skirt (Flixborough file photograph, HSL Library,
Sheffield, reproduced under the terms of Crown Copyright Policy Guidance
issued by HMSO).
the dead load of the fluid inventory. Pressure was applied by
nitrogen gas. Strains in the pipe were also measured. There
were eight test runs, three at ambient temperature, three at
155
C, and two at 160
C. The maximum pressure utilized
was 0.964 MPa in all but the last two cases; for these the
bellows were squirmed at 1.06 MPa and then ‘burst’ at
1.53 MPa.
In the experiments at ambient temperature, internal pres-
sure caused the B4 end of pipe to move upward and the
scaffolds there were only lightly loaded. With increase in
pressure, the B4 end rose off the scaffold support as the B6
end moved downwards and the scaffolds there carried
greater load. However, under operating conditions, 150–
155
C, the assembly was mostly unsupported since the
reactors expand upwards about 6 mm.
Bridging pipe
Experiments on the 4.8 mm thick wall dog-leg pipe were
also carried out (Batstone, 1975). Two 75 mm pin joints
were located at the centre of the bellows at each end of the
original assembly where the bending moment was zero. An
axial load was applied horizontally at the lower end through
two pre-calibrated hydraulic rams each of 100 tonnes
capacity with 152 mm stroke. In order to support the vertical
force on the ram, two double row, self-aligning ball bearings
were located on the extended pin of the joints. These were
free to move on two horizontal supports mounted on the
floor as the rams extended but were prevented from twisting.
Two strengthened adapter pieces were welded to blank
flanges bolted to both ends of the pipe and pivoted on pin
joints fixed to vertical support columns. The pipe itself was
filled with water and internal pressure applied with a hand
pump. The axial load applied via the ram maintained the
internal pressure according to the relationship:
F ¼ pA (1)
where F was the horizontally applied axial load, p the
internal pressure, and A the effective cross-sectional area
of bellows (deff ¼ 747 mm). The internal pressure and
deflections at several positions were measured and recorded.
112 VENART
This pipe failed, i.e. buckled at its lower mitre joint at
1.89 MPa.
Bellows and stub-pipes
Batstone (1975) also conducted four individual water-
filled bellows tests with a facility designed to simulate half
the dog-leg pipe=double bellows assembly. As the distance
from the mid-point between the two mitre bends to the first
convolution of each bellows only differed by 17.5 mm, only
the lower, R6, bellows was evaluated. Axial and shear forces
on the bellows were measured as functions of internal
pressure along with vertical deflection measurements. The
pivot point of the pipe assembly was fixed 0.178 m above
the centreline of the bellows at zero internal pressure at a
distance of 2.4 m to the first convolution. The ends of the
bellows were welded to stub-pipes and these bolted to blank
flanges, the fixed end being bolted to the support frame and
the free end being welded to an I-beam which supported the
pivot point of the assembly.
The bellows all failed, i.e. squirmed, at an internal
pressure of about 1.2 MPa. In one of the tests, pressures
were cycled over selected ranges of up to 3456 times. In this
test, cracks did not occur until further pressurization to
1.5 MPa. These cracks were sealed and the test cycles
continued for 900 cycles (1.0–1.1 MPa) and a further 400
cycles (1.07–1.2 MPa); there was only a slight increase
(2 mm) in the three cracks which had formed at the ‘crimple’
points due to squirm. In a few experiments a 48 mm
diameter 1.82 m long scaffold support was provided with a
6.4 mm space between stub-pipe flange in order to evaluate
the influence these had on the behaviour. It was this
experiment that the Court used to rule out its possibility (ii).
A POSSIBLE INITIATING MECHANISM
A potential initiating mechanism consistent with the
possibility of fatigue and the evidence outlined above first
requires the initiating tear found near the top of the
upstream, B4, bellows (Secretary of State for Employment,
1975; Foley, 1974a; Foley and Nicholson, 1974). The
topmost convolutions of the B4 bellows nearest R4 are most
highly stressed (Venart and Tan, 2000) and the convolutions
susceptible to excitation by the flow.
The various steps in the failure process are portrayed in
Figure 8 and a tabulation of the sights and sounds that
would have been expected indicated in Table 1. The very
complex process can be summarized as follows.

The instability of the assembly and the near-resonance of
the upstream, B4, bellows with the forcing fluid flow
initiates fatigue damage which accumulates in the most
highly stressed, and least damped (i.e. the vapour space),
regions of the unit. Small cracks form.

Sufficient damage accumulates to cause the cracks to join
and initiate a tear which further opens, due to pressure,
and local vapour–gas depressurization occurs, Figure
8(a). There is a bang and a sound like a pressure relief
valve going off, Table 1. The vapour space of R4 and
B4 rapidly, 3–4 s, depressurizes down to the saturation
pressure of the cyclohexane, 0.55–0.60 MPa.

Pressure reduction to this level is nearly immediate in the
structure whereas the pressure response times are three to
15 times slower in the fluids.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127

As a result B6 rapidly expands and compresses B4 due to
this unequal pressure distribution, Figure 8(b), and the
now superheated and supersaturated cyclohexane starts to
boil with a ‘rumbling’ sound; the boiling becomes more
vigorous with time. The B4 bellows are hard compressed
from 228 down to 50 mm and all the convolutions
creased; B6 expands from 228 to 400 mm.

Owing to local boiling, a two-phase swell rapidly fills R4,
its top vent, and its R3 inlet as well as B4 and its stub pipe.
This restricts pressure equalization with R6. As nitrogen
comes out of solution and the fluid boils, the increase
in pressure, from the cyclohexane saturation pressure back
to operating conditions, generates a dynamic two-phase
head space impact and dynamic re=over-pressurization
(Gromles, 1984; Ramier and Venart, 2000). This immedi-
ately causes the compression of the downstream, B6,
bellows, the extension of B4, and the initiation of its
upwards ‘squirm’, Figure 8(c)–(e). Squirm is a result of
the extension, >170 mm, and due to fluid distribution in
the pipe. There is first the ‘thud’ as B4 is compressed and
then a minor ‘rumbling’ of the boiling process in the B4
region of the pipe-bridge, and then a further ‘thud’ as B6 is
now hard compressed by the repressurization occurring in
R4 and B4, Table 1.

The compression of B6, by this local overpressure, fixes
the constraint of the dogleg pipe at its R6 connection and
causes the process of B4 squirm to act over the full
356 mm eccentricity, Figure 8(d), with much less force
than needed by Newland’s 178 mm offset single stage
analysis (Newland, 1976).

This results in the collapse, by buckling, of the 0.5 m
pipe, at its lower mitre joint, as the B4 bellows continues
to squirm, then balloon, burst and is ripped off at its
normal operating pressure due to the ‘p–V’ work of the
expanding two-phase fluid, Figure 8(f), the unresisted
squirming force, and the massive reduction in the pipe’s
moment of inertia caused by buckling, Figure 8(g). It
should be noted that there is a reduction in moment of
inertia from a thin-walled cylinder to collapsed tube
greater than 4000 to 1.

The B4 squirm, once past about 50 mm vertically, cannot
be resisted and the collapse of the pipe and destruction of
B4 continues. B4 is torn into three pieces.

The ripped off topmost section of B4 is projected
upwards and sails into the air and appears, to some
as a ‘blown off ‘‘tank top’’’; it was found only two
months later under extensive debris between the line
of reactors and the after-reactors (Foley and Nicholson,
1974); all other portions of both the B4 and B6
bellows were located immediately by investigators
(Foley, 1974a).

The shape of the buckled pipe is now somewhat like the
‘swan neck’, Figure 8(g)–(h), noted by some (Marshall,
1974; Gugan, 1974) with a significant jet discharge such
that it resembles a ‘tea kettle spout’. The B6 bellows
through all of this is, however, still intact, although
possibly severely damaged.

In the buckling process, Figure 8(g)–(h), the 250 kg liquid
contents of the lower end of the 0.5 m diameter pipe are
forcibly ejected into R6 and this impulse tears off its inlet
baffle and causes this to impact and bend the internal
154 mm diameter SS agitator shaft (Artingstall, 1974;
Gugan, 1979).
 FLIXBOROUGH
Figure 8. Successive stages in a crack-initiated two-stage bellows and pipe-bridge failure (reproduced from Venart, 1999, with permission from
INTERFLAM, www.intercomm.dial.pipex.com).

The above processes would have been very noisy as the
bellows burst and the dog-leg pipe buckled. Some of the
overhead scaffolding would have been projected upwards
along with other debris and discharged material. In the
process some pieces of scaffolding boards and insulation
material (insulation and its aluminum sheet cover) are
caught in the ‘nip’ of the collapsed pipe (Foley, 1974a;
Secretary of State for Employment, 1975). There would
have been also the development of a ‘roar’ due to the
massive two-phase discharge from R4 and the lesser two-
phase discharge from fluid trapped in the B4 end of the
pinched pipe-bridge additional to the ‘rumbling’ of the
boiling of the superheated cyclohexane in R4, Table 1.

The R6 internal pressure now expands B6 and frees its
constraint, Figure 8(h). The reaction of the ‘choked’ two-
phase discharge, which occurs through the B4 end of the
buckled pipe-bridge, bends and distorts the bellows
forcibly expanding and stretching its topmost elements
to their full extent. The pipe now appears as a ‘swan neck’
or a ‘tea kettle spout’.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
113

The major explosion occurs which strips off most of the
local pipe and vessel insulation.

The now swinging pipe is swept through the scaffolding
below by the force of the blast toward the base of the
R6 reactor (see Figure 7 and Plate 9 of Secretary of State
for Employment, 1975) and it deflects further and its
lower surface impacts the side of R6 denting it (Arting-
stall, 1974) as well as flattening the bottom of the pipe
after the lower miter-bend further (Foley, 1974a), Figure
8( j). This effectively cuts off all flow through the pinch.
Additionally, the topmost portion of B6 is stretched out
flat, without any convolutions being left, unlike the
companion piece of B4, where some convolution creases
were still intact (Foley and Nicholson, 1974). The posi-
tions of the ‘crimple’ points in B6 also differ significantly
from those in B4 (Foley and Nicholson, 1974), indicating
a different distortion process for this bellows.

In this dynamic process, the B6 bellows are torn off and
the rebound from the R6 impact and the impulse of the
two-phase release from the now open R6 end of the pipe
114
Table 1. Sights and sounds in the Flixbourgh by-pass pipe-bridge failure; Figure 8.
Time (s) Event (Figure 8) Sound
1–2 (a) ‘PRV’ vent
3 (b) ‘Thud’
3–4 (c)–(d) Bang and rumble
4þ (e) Greater rumble, larger jet
5þ (f)–(g) More severe rumbling
5–25 (h)–(i) Roar ‘like a jet engine’
25þ (j) First explosion—detonation
26 (j) Pipe bangs R6
27 (k–l) Crash and thud
27þ (l) New rumbling
spins the assembly clockwise downwards, Figure 8(k),
such that the flanges, after the lower mitre bend and its R6
stub-pipe, impact the top of the 0.46 m thick reinforced
concrete R5 plinth puncturing it, Figure 8(l) (Secretary
for State for Employment, 1975). In this process only the
R6 stub end of the pipe and its flanges are severely
distorted at the point of impact (Foley, 1974a). The
pipe, after impact, rebounds and falls over. In the mechan-
ism accepted by the Court the buckled end of the pipe
should have impacted the plinth slab.

These later sequences, Figure 8(i)–(l), occur after, the
major explosion.

Once the B6 bellows blows off, the resulting horizontal
northward developing torch flame is split by the R4
reactor and the eastern half envelopes the now distorted
bare ‘8-inch’ pipes causing the inferred temperature
distributions, observed creep swelling, zinc contamina-
tion and failures (Foley and Nicholson, 1974; Cottrell and
Swan, 1976). The bifurcated R6 jet flame explains the
cine film taken by a witness (Goodchild, 1975) and obser-
ved by others (Carter, 1974). High temperature carburiza-
tion is found only on the R4 stub-pipe and its attached
remnants of the B4 bellows (Foley, 1974a; Foley and
Nicholson, 1974). No carbide formation was observed
on the R6 stub-pipe nor its B6 bellows remnants as would
be required from the simultaneous jet fires resulting from
the Court’s accepted mechanism.
The major consequence of this sequence of events is that
a discharge through reactor R4 alone occurs prior to the
main explosion and is thus responsible for the blast. This is
only sufficient to provide a release of some 370 kg s1 of
cyclohexane for probably less than the 30–40 s determined
for initial events (Cox, 1976). This vent, of some 10–15
tonnes (Leclude and Venart, 1996, 1997), is very much
smaller than the 40–60 tonne estimates previously attributed
to the explosion (Sadee et al., 1976=1977; Roberts and
Pritchard, 1982; Bjerketvedt et al., 1997), since once the
flashed inventory of R4 is exhausted the two-phase choke
from R3 now limits its (R4’s) function to that of a ‘knock
out’ vessel.
Any estimate of the amount of cyclohexane released must
take into account the very complex interior geometry of the
reactor stream flow path in addition to the composition and
thermodynamic state of the flow. Figure 3 partially illus-
trates the interior of the reactors and the ‘by-pass’ pipe
assembly. Under the hot recycle flow conditions existing at
the time of the accident (0.04 m3 s1), the levels in the
reactors would have been 130–150 mm lower than under full
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
VENART
Sight Comment
Wisp to jet of ‘steam’ Many observers
Jets of ‘steam’
Two-phase jet
B4 bellows squirms
Upward debris, large kettle spout Bellows off, pipe buckles
Vertical venting ‘swan-neck’ pipe R4 jet discharge
Works office ‘white flash’
Vertical flames Dent in R6, B6 tears off
Bypass impacts R5 plinth R6 jet-fire on B4 remnants
R6 jet fire Eight-inch line failures
production flow (Artingstall, 1974). The inlet–outlet pipes
of all reactors would thus have been nearly full aside from
the outlet pipe leaving R4. Here the level at the foot of the
overflow weir would have been much lower and under full
production flow, 0.125 m3 s 1, the 0.71 m discharge pipe
would have been only about half full.
With the release mechanism outlined earlier (Figure 8,
Table 1), the discharge would have initially been a vapour jet
or jets through the fatigue cracks, which then would have
extended by pressure to the extent noted by Foley (1974a) in
the B4 bellows remnant. The volume of vapour discharged
would essentially be that amount trapped in the top of R4
and in the upper reach of the B4 end of the dog-leg,
approximately 12.5 m3. A determination of the quantity
and an estimate of the time can be made based upon the
assumption of sonic vapour-only flow through a crack the
length noted by Foley (1974a). Then, dependent upon
the assumed crack width (2 mm) and its discharge coeffi-
cient (1.0), a discharge of about 0.1 tonnes could occur in
about 10 s. This assumes that the discharge process is
choked and takes place down to a condition where the
liquid will just commence boiling, i.e. 0.55 MPa, the satura-
tion pressure of pure cyclohexane.
Next it is assumed there is a point before all the vapour is
exhausted where there will be sufficient superheat in the
liquid for it to commence boiling and swell up and fill the
R4 vessel. At the same time the pressure differences
between the reactors will cause each connection between
the reactors to fill and essentially two-phase ‘choke’ such
that the discharge process after the vapour release will
consist of the sequential two-phase choked discharge from
each reactor in turn starting from R4. Further discharge from
this point in time will have R4 essentially acting as a
separator for the subsequent discharges from the upstream
reactors R3 to R1. The inlet baffles and the outlet weirs of
each reactor will essentially aid in this process as they will
additionally restrict flow beyond that provided by the two-
phase chokes of the reactor inlet–outlet nozzles.
The energy for boiling comes at the expense of a loss of
internal energy of the liquid and a determination of the
amount of material that must be vaporized to fill the R4
vessel can be made using mass and energy conservation as
well as Clausius Clapeyron considerations in order to
deduce pressure. Since the quantity of nitrogen dissolved
in the cyclohexane is small (0.2%), the vapour phase
depressurization is assumed complete when the pressure
has fallen to 0.54 MPa (i.e. just sufficient superheat to
evaporate enough liquid to fill the reactor). At this point
the vapour discharge is complete and the bellows squirms
 FLIXBOROUGH
and blows off. A two-phase discharge now commences from
the full 0.71 m R4 outlet.
The average void fraction in R4 is based upon the original
vapour volume, i.e. 0.2. To estimate the initial discharge
void fraction the non-homogeneous bubble distribution in
the vessel must be considered, (Gromles, 1984), where a
distribution factor (Kataoka and Ishii, 1987) is taken. From
this we can now deduce all the necessary properties for the
fluid at the entrance to the nozzle.
The level within the R4 reactor must now be tracked and
here the rate of vapour production is compared with the rate
that volume can separate via the pool free surface (Wehmeier
et al., 1994), where the velocity of separation (Kataoka and
Ishii, 1987) can be utilized. When the separated volume
exceeds that being produced, the volume of the two-phase
swell diminishes and the preceding reactor (R3) will com-
mence discharging into R4. Using a variety of two-phase
flow models (Leclude and Venart, 1996) with discharge
coefficients of 0.5 and over the pressure range 0.55–
0.45 Mpa, it is estimated that about 10 tonnes flash from
R4 and are discharged in between 10 and 18 s, dependent
upon the flow model chosen.
Both reactor R3 and R4 must now be considered since R4
is now essentially full of vapour at a pressure of 0.43 MPa.
R3 now starts to discharge into R4. Finally only about
10–16 tonnes of cyclohexane are discharged and not the
40–60 tonnes previously thought.
Thus far, the postulated complex behaviour of the
bellows–pipe assembly has been described, such as might
has been initiated by a fatigue-initiated tear in the B4
bellows. Finite element analyses of the full-scale experi-
ments undertaken for the Court are described next and then
the remainder of the paper examines the discharge processes
for the case postulated by the Court and that which would
have resulted under the circumstances outlined above; it will
thus compare the developments of the vapour cloud noted
by witnesses and the consequences of its ignition.
FINITE ELEMENT ANALYSES
ANSYS 5.3–5.5 (Swanson Analysis, 1994) non-linear
elastic-plastic large deformation finite element (FE)
models were made of the individual bellows and the
assembly of bellows, stub pipes and dog-leg illustrated in
Figure 3. Additional simulations of the experiments carried
out for the Court were also conducted. The required
mechanical properties of 304L and 316L SS were deter-
mined either from test (Venart and Tan, 2000) or from the
report of Foley (1974a) and the literature.
The elastic lateral stiffness of a bellows decreases linearly
with increasing pressure (EJMA, 1993; Newland, 1964;
Snedden, 1981, 1985) and, upon sufficient increase, the
bellows can become unstable, usually resulting in rapid
gross lateral deflection or ‘squirm’. The lateral stiffness
and response of a universal expansion joint, i.e. two bellows
interconnected by a straight length of pipe or ‘spool’ piece,
has been considered by Newland (1964). The lateral stiff-
ness for two bellows connected by an offset spool piece,
such as here, was undertaken by Newland (1976); however,
no determination appears to exist for the determination of
natural frequency for this case.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
115
The Bellows
For small deflections (i.e. elastic behaviour), bellows can be
simplified by only half a convolution. The axial and transverse
stiffness from such an analysis can be regarded as a reference
base for the verification of any later more detailed, and
therefore necessarily less refined, FE model. Although the
results are somewhat dependent upon the physical properties,
Venart and Tan (2000) made an analysis using 2688 eight-
node isoparametric two-dimensional symmetric elements
with 8753 nodes of a model of the Flixborough bellows that
resulted in a determined cold axial stiffness of 403.8 kN m1
and transverse stiffness of 5558 kN m1.
With large transverse deflections, contacts between adja-
cent convolutions occur and half a whole bellows must be
modelled. The shape of the bellows is asymmetric, i.e. the
overall shape of the deformed structure possesses 180

rotational symmetry about the mid-point of the neutral
axis. Three dimensional (3D) shell and surface contact ele-
ments are thus necessary. An element size of 12  12 mm
yields 11,760 elements and 11,979 nodes to mesh half
a bellows. The computational requirements for this are
extremely large.
An extension of this work by Li et al. (to be published)
increased the total number of elements to nearly 3000 and
considered in detail elastic and plastic behaviour of bellows
as functions of pressure as well as large-scale extension and
compression. Figure 9(a) illustrates the model used to
consider axial loading with large-scale compressive and
tensile loading. Axial elastic spring rates under ‘zero’
pressure for compression and extension are 516 and
527 kN m1, respectively (Figure 9b); in these simulations
the axial elastic spring rate increases slightly (1%) with
pressure increase to 1.1 MPa. Experimental values, obtained
from the manufacturer, and presumably not under pressure
(Batstone, 1975), were between 485 and 578 kN m1. Large
deformation behaviour (Figure 9c) indicates that as the
bellows are compressed the convolution crests and roots
start to plastically deform, and that at about 50 mm and
22 kN, with very small increase in load, the bellows
shorten to 110 mm where greater resistance sets in due
to convolution compaction—with further compression, to
160 mm, the joint is now effectively rigid and the load
rises to 170 kN. In extension, plastic deformation com-
mences again at about 50 mm and a load of 21 kN. The
roots and crests plastically deform as the sheet straightens
out to about to 230 mm at 31 kN—further extension now
requires greater load and the bellows extend finally to
733 mm and start to tear at a load of 900 kN with the
critical elements experiencing a 37% reduction in area.
These results indicate that, if a pressure difference of only
0.1 MPa developed between R4 and R6, the respective
bellows sets, if axially constrained, would be compressed
and extended 160 mm due to the difference between the
bellows mean diameter and the 508 mm pipe; this could
occur if the R4 vent and the lower section of the bridging
pipe became full of two-phase fluid and ‘choked’. Under
operational conditions with little or no vertical support
or upwards restriction, incipient squirm may have been
possible.
The ‘zero’ pressure transverse, or lateral, elastic spring
rate of the model was 7474 kN m1 (the lateral elastic spring
rate decreases 25% with pressure increase to 1.1 MPa).
116 VENART
Experimental values, obtained from the manufacturer, and
again presumably not under pressure (Batstone, 1975), were
between 7567 and 9015 kN m1. The differences in these
results could be due to non-isotropy due to the forming
process and the choice of a uniform wall thickness in the
model formed bellows that were thinner at their crests and
roots. Figure 10(a) shows the deformed shape of the bellows
joint at an internal pressure of 0.96 MPa and a transverse
displacement of 42 mm. Figure 10(b) illustrates the trans-
verse loads required to effect the lateral displacement at
0.96 MPa as a function of extension. Four main stages can
be observed. The first stage is a linear elastic range; here the
load–deflection curve is linear and the stress in the bellows
is below the yield strength. In the second stage primary
yielding occurs localized at the tops or bottoms of the
convolutions where stress levels are high. With increased
loading, the bellows go into a third stage where contact
between convolutions occurs. This stiffens the bellows in the
lateral direction. The final stage is one of general plastic
deformation where the whole bellows yields and the slope of
the load–deflection curve decreases until the bellows
becomes unstable and final rupture may occur if bellows
movement is not resisted.
Snedden (1981) experimentally obtained similar load–
deflection curves up to the point of instability. He developed
simplified analyses to determine the collapse pressure and
evaluate the deformations observed (Snedden, 1985). He
considered the lateral stiffness, once convolution contact
occurred, to be infinite and thus any constraint to be ‘rigid’
though, as can be seen from his results, this is far from the
case (Figure 7 of Snedden, 1985), and a more gradual
approach to failure develops in reality and the FE model.
The limiting internal design pressures, based on column
instability (both ends rigidly supported) and no axial exten-
sion, were estimated by EJMA (1993) methods to be 2.1 and
1.8 MPa for the cold and hot bellows, respectively.
Bridging Pipe
A model was constructed to simulate the experiment on
the bridging pipe alone (Batstone, 1975) described earlier
(Li et al., to be published). Internal pressure in this model
was increased gradually and the axial loads at the pivot as
well as the vertical deflections computed. Two cases were
considered; one that for a ‘new’ pipe, i.e. as constructed, the
other plastically deformed to give a pre-test axial set of
19 mm. A comparison between measured and simulated
loads is shown in Figure 11. This indicates that failure of the
pre-plastically deformed pipe occurs at a similar pressure.
Although there are some differences in the results, agree-
ment is reasonable, with the FE model generally more
compliant. The final measured and simulated squirming
pressures are, however, very nearly the same (1.89 MPa).
Bridging Pipe and Stub Pipes
The FE simulation of this modification of the experiment
conducted by Batstone (1975), was simpler than that for the
bellows assembly (Venart and Tan, 2000). The half pipe and
its stub pipes were represented with over 2200 shell
elements with strengthened adapter pieces, which connected
the blank flanges in turn connected to pin joints, as shown in
Figure 12(a), which additionally illustrates the deformation.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
The weights of the heavy flanges, with its bolts, nuts and
washers, the adapters and the fluid-filled pipe were
converted into equivalent density variations of the shell
and beam elements similar to that used in the bellows
simulation. Internal pressure and axial force were applied
to the pipe in a step-wise fashion until failure. Both room
temperature and hot evaluations were conducted. Large-
scale axial loading of the structure was also carried out at
an internal pressure of 0.96 MPa.
The room temperature simulation failed by bucking at the
lower mitre joint at 2.16 MPa; the hot simulation failed at
1.87 MPa; Batstone (1975) recorded a similar cold failure
pressure, i.e. 1.89 MPa. The large-scale deformation simula-
tion at 0.96 MPa required over 800 kN to initiate failure
(Figure 12b).
In order to evaluate the influence of end constraint on
pipe behaviour a further FE study was conducted using this
model (Figure 13a and b). However, this employed a ‘hard’
or fixed constraint at the B6 end that restricted rotation of
the pipe and thus simulated the constraint of a compressed
B6 bellows. This model indicated that an internal pressure
of only 1.09 MPa could initiate buckling failure of the pipe
at its lower mitre bend. Further, at an internal pressure of
0.96 MPa, an axial force of only about 400 kN is required to
initiate buckling of the pipe, Figure 13(c). Relaxing the B6
constraint would obviously increase the required pressure
and load. However, such a result indicates that, had a
differential pressure occurred across the two reactors, such
as could have happened by only one bellows suddenly
tearing and depressurizing, the subsequent repressurizing
of R4 coupled with the squirm of B4 could lead to buckling
of the pipe assembly—all now possible at near-operating
pressure.
The Bridging Pipe–Bellows Assembly
Both static and dynamic FE analyses for an assembly
made of the dog-leg pipe, its two stub pipes and the bellows
joined to the reactor stub pipes were made based on the
previous models (Venart and Tan, 2000). The model setup,
loads and weight conversions were similar to those used
earlier. The main difference was that the pipe was not fully
filled with cyclohexane (Teng-yang et al., 2000), and that a
scaffold system could be set to support the pipe assembly
(Figures 3 and 14).
Four load steps were used in the FE simulation of the
complete assembly: (1) initially the by-pass pipe rested on
the scaffolding without internal pressure; (2) the system was
next pressurized; then (3) the assembly was set free due to
the upward thermal expansion of the reactors (6.4 mm); and
(4) finally the pipe was allowed to thermally expand axially
(approximately 10 mm) along with the radial expansion of
each reactor and its stub-pipe (about 2.5 mm). At the lower
or R6 end contact and deformation of the scaffolding was
permitted if the downward assembly deflection exceeded
6.4 mm.
In the model it was assumed that the upper pipe was half
full and the lower pipe full of liquid cyclohexane. It was also
assumed that the scaffolds were made from normal structural
steel; these were simulated by beam elements with contact
elements to prevent any loading when the pipe contact was
lost. Elastic–plastic load–deflection relationships were used to
simulate the pipe, the bellows and the scaffolding.
 FLIXBOROUGH 117

Figure 9. Axial load–extension of the finite element model of the Flixborough bellows; (a) the bellows model, (b) small extension–load behaviour, (c) large
extension–load behaviour (Li et al., to be published).

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
118 VENART

Figure 10. Transverse load–deflection of a finite element model of the Flixborough bellows (Li et al. to be published); (a) three-dimensional FE model—
lateral loading: equivalent Von Mises strain at 0.96 MPa at 0 m axial extension. The FE results are based on surface-to-surface contact with a 0.4 friction
coefficient; the lateral deflection is 41.75 mm. (b) three-dimensional FE model—lateral loading as a function of extension at 0.96 MPa. FEM simulations
based on point-to-point contact elements.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
 FLIXBOROUGH
Figure 11. Comparison of Cremer and Warner dog-leg pipe experiments
with the FE simulation of the dog-leg pipe (Batstone, 1975), where: (s) FE
simulation0 is based on a new pipe; (m) FE simulation was based on the pipe
with 0.75 inch (19 mm) axial plastic deformation (Li et al. to be published);
and (----) experiment; failure at 1.89 MPa.
The vertical deflections at the two ends of the pipe,
obtained for an internal pressure or 0.96 MPa, are listed in
Table 2. Compared with the experimental measurements, the
trends are similar, however, there are some differences in the
values, particularly for the upper bellows where the experi-
mental deflections exceed the FE results by some 35%.
The maximum stresses occur along the tops and bottoms
of the convolutions, particularly at the second to last
convolutions of both ends at normal operating conditions.
Some parts of these regions appear to be coincident with
portions of the internal tear noted by the Court (Secretary of
State for Employment, 1975; Foley, 1974a). Stress levels in
the lower bellows are similar to those in the upper bellows;
however, damping is greater due to the fluid inventory. The
stresses in the experimental bellows at B4 would, however,
have exceeded these values since deflections were larger.
At temperature and pressure, the upper (R4) end of the
pipe moves towards R4 by 0.67 mm; the absolute axial
displacement of the other end of the bellows is away from
R4 by 2.89 mm so that the axial compression of the B4
bellows is 3.55 mm. At the lower (R6) end of the pipe the
assembly moves away from R6 by 0.037 mm with the
absolute axial displacement of other end of the bellows
moving away from R6 by 3.08 mm; the axial compression of
B6 is thus 3.05 mm. The pipe, between bellows, shortens
under these conditions by 3.51 mm compared with a mea-
sured value of between 5 and 7 mm (Games and Water-
house, 1974), Table 2.
A FE modal analysis of this model was also carried out in
order to evaluate the first three natural frequencies and
vibration modes of the cranked pipe assembly. In this
instance the mass of fluid was converted into a density
increase as a function of location. The total weight of the
pipe, flanges, bolts, bellows and insulation was about 1.13
tonnes and its liquid inventory some 600 kg, of which about
375 kg were distributed in the lower half of the pipe under
full production conditions. The first three natural frequen-
cies and vibration modes were found to be: first natural axial
frequency, f1 ¼ 3.98 Hz; second natural, and transverse
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
119
(vertical) frequency (coupled with bending of the pipe),
f2 ¼ 12.35 Hz; and third natural frequency(rocking mode)
f3 ¼ 30.16 Hz. The equivalent EJMA (1993) determinations
are 3.93, 15.7, and 27.2 Hz respectively. The axial mode
agrees well but the inclusion of the offset pipe and its
bending appears to reduce the transverse and increase the
rocking modes.
Three main harmonic forces are induced with the flow
system (Teng-yang et al., 2000). One is the axial load that
results from the flow fluctuations at the diameter change in
the upstream pipe. Another is the transverse loading due to
the change in fluid mass at the lower hydraulic jump. Both
these forces are in phase and at the same frequency as the
hydraulic-head change. Owing to the hydraulic jump after
the upper mitre bend the mass in the lower half of the pipe
would have oscillated by about 60 kg (3.5% of the total
system mass) at a frequency of between 1 and 2.5 Hz. A
third force is the level fluctuation in the weir waterfall pools;
levels in similar sized outfalls vary with a frequency of
about 7 Hz (Rajaratnam and Chamani, 1995). Further, the
plunge pool jet velocity, especially at the lowest flow, is
sufficient to excite axial resonance of the bellows.
The comparison between the finite element models and
the experiments undertaken above indicate that the models
reasonably represent the behaviour of the bellows, bellows
and stub-pipes as well as the bridging pipe.
THE CFD ANALYSES
A transient three dimensional CFD model was developed
for a 130 (x) by 130(z) by 30 (y, elevation) m volume of the
region around Section 25A (Venart, 1999). This area
encompassed the reactors and after-reactor vessels and
their plumbing (especially the ‘8-inch’ lines) in addition to
Section 7 (and a portion of Section 27), the Laboratory, the
Control Room, as well as the Main and Project Office
buildings (Figure 1). The computations were carried out
using the commercial code FloSYS (Flomerics, 1994),
which is a CFD solver operating in Cartesian coordinates.
Solution were performed throughout the rectangular
cuboid domain for the velocities u, v, w in the x, y, and z
directions respectively, fluid static pressure, p, temperature, T
in the gas and solid regions, and concentration of the secon-
dary medium (cyclohexane), c. Several tests were evaluated
for solution convergence, and as a compromise between
accuracy, computational requirements, and resources, a grid
system of 98  41  68 nodal points in the x, y and z
directions, respectively, was selected. The smallest cell size,
commensurate with the smallest scale of detail, the ‘8-inch’
pipe, was 0.1 m in the x and z directions and 0.05 m in the y
direction; the largest cell sizes were on the upper domain
corners; these were typically 3  2.3  3.3 m in the x, y and z
directions, respectively. The grid, where required, was set
automatically to reconcile with internal solid surfaces, parti-
cularly in the reactor near field.
The equations solved formulate the principles of conser-
vation of mass, momentum in all directions, thermal energy
and mass of secondary medium (cyclohexane). Turbulence
modeling is performed by using a ‘standard’ two-equation
(k–e) turbulence model with near-wall grid node ‘wall
functions’ used to evaluate surface friction and surface
heat transfer. The effects of turbulence are represented via
120 VENART

Figure 12. (a) Flixborough dog-leg pipe FE model, B6 ‘pin–pin’ constraint model, equivalent Von Mises stress contours for the pipe when Ux ¼ 0.365 m and
Uy ¼ 1.0675 m at the left pin as predicted by the FE model. (b) Nodal force load Fx (kN) vs deflection, Ux (axial) at the left pin predicted by FE simulation,
pin-pin model.

a turbulent viscosity and thermal diffusivity where these
quantities are linked by the turbulent Prandtl number, which
takes a constant value of 0.9.
Buoyancy effects due to thermal expansion and the heavy
vapour dispersion of the evaporated cyclohexane are
included in the y direction as a gravitational source term.
The variation of properties, in this particular case, density, is
assumed to be governed by the ideal gas laws. The mole-
cular viscosity, thermal conductivity and diffusivity were
taken to be those for air and constant.
Plant equipment, pipe racks and supporting structures,
other than the five reactors and the associated after vessels
and plumbing, were treated as flow obstacles with a uniform
resistance and porosity of 0.5. Buildings and other large
structures (the Main Office, the Project Office, the Labora-
tory and the Control Room) were assumed to be impervious
to the flow and adiabatic. The prevailing wind was assigned
by steady vector components as boundary conditions on the
free boundaries of the domain. The surface of the ground
was set to a uniform roughness of 0.01 m.

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
 FLIXBOROUGH 121

Figure 13. (a) Flixborough dog-leg pipe FE model, B6 ‘hard’ constraint, Fx ¼ pA, failure initiated at 1.08 MPa. (b) Flixborough dog-leg pipe FE model, B6
‘hard’ constraint, Fx ¼ pA, failure at 1.08 MPa, axial (x) and vertical (y) deflections as a function of ram load, Fx. (c) Equivalent Von Mises strain contours for
the pipe when Ux ¼ 0.365 m and Uy ¼ 1.0675 m at the left pin as predicted by the FE model. (d) Nodal force load Fx (kN) vs deflection, Ux (axial), Uy
(vertical) at the left pin predicted by FE simulation, pin–fixed model.

The mass source terms, representing the flashing two-
phase discharges from the reactors, were evaluated from
prior choked flow determinations and the assumption of
sonic, vapour only, flow at the reactor, R4 and R6, outlet
nozzles (Leclude and Venart, 1997). The properties of
cyclohexane and its dissolved nitrogen were obtained from
the NIST (1999) mixture property program, STRAPP.
A near-field isometric view of the reactor train as may
have been seen from the Main Office is shown in Figure 15
as represented in the Flomerics code. In the figure, the

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
122 VENART

Figure 13. Continued.

walkways along the reactor train are represented by porous
resistance areas, and the ‘8-inch’ pipes connecting the two
after vessels together are indicated. The cuboid representa-
tions of the vessels are also shown.
The FloSYS ASCI output files were used as input files to
be processed by Tecplot (Amtec Engineering Inc., 1996) to
extract the iso-contours for the lower (LFL) and upper
(UFL) flammable limit (0.037 and 0.198 wt.%, respectively)
and provide interactive visualization through time-
sequenced plots of the iso-surfaces.
Simulations were preformed for two cases: (i) the simul-
taneous discharge from both R4 and R6, as the base case
representing that process accepted by the Court, and (ii) a
discharge from only R4 prior to the explosion. In both

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
 FLIXBOROUGH 123

Figure 14. FE model of the Flixborough by-pass assembly and its supports (reproduced from Venart and Tan, 2000, with permission from ICPVT).

Table 2. FE analysis of the assembly and comparison with test data (Games and Waterhouse, 1974), vertical deflections (Venart and Tan, 2000).

Deflection at upper, B4, Deflection at lower, B6,

Case Condition end of pipe, mm end of pipe, mm
FE Initial (a) Full scaffold support no internal pressure 0.72a 0.77
FE @ 20 C (b) Partial scaffold support, 0.964 MPa 3.41 3.77
FE @ 150 C (c) Partial scaffold support, 0.964 MPa 6.93 4.49
(a)–(b) FE cold Scaffold support, 0.964 MPa, cold 4.13 [10.66]b 3.00 [3.26]
(a)–(c) FE hot Partial scaffold support, 0.964 MPa, hot 7.65 [10.27] 3.72 [3.26]
a
(þ ) Upwards, () downwards displacement. bAverage experimental values.

instances, the discharges were taken as choked sonic cyclo-
hexane vapour at 370 kg s1 and 150
C. The results are
available in the form of ‘movies’ or animations both for the
‘far field’ and ‘close in’ regions of the computational
domain.
In the figures discussed here the view is as from near the
top of the Hydrogen Plant looking towards the Main Office
building (Figure 1) at about 30 s into the discharge process.
The UFL is represented as the panelled interior, and the
LFL, the wire-frame exterior, surfaces.
The transient development of the cyclohexane vapour
cloud at its LFL and UFL over a period of 30 s is now
briefly described. Figures 16 and 17 illustrate the LFL=UFL
cloud distribution at 12 s as it would have been seen from
near the top of the Hydrogen Plant.
Simultaneous Discharge from both R4 and
R6 (Figure 16, ‘Movie’ 1)
Even at 1 s into the event the simultaneous discharge
process accepted by the Court results in an aerial cloud from
the discharge of R4 riding up and over both the jet discharge
from R6 and the top of this vessel and reaching a height of

Figure 15. A near-field view from southeast of the reactor train and the Figure 16. Southeast view from the top of the Hydrogen Plant looking
after-reactors as might have been seen from the roof of the Main Office towards the Main Office Building: simultaneous discharge of R6 and R4 at
building (reproduced from Venart, 1999, with permission from INTER- 12 s, LFL and UFL envelopes (reproduced from Venart, 1999, with
FLAM, www.intercomm.dial.pipex.com). permission from INTERFLAM, www.intercomm.dial.pipex.com).

Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
124 VENART
Figure 17. South-east view from the top of the Hydrogen Plant looking
towards the Main Office building: R4 discharge only at 12 s, LFL and UFL
envelopes (reproduced from Venart, 1999, with permission from INTER-
FLAM, www.intercomm.dial.pipex.com).
about 20 m. This cloud proceeds southerly into the prevail-
ing wind and towards the opening between the two office
blocks. At the same time, the R6 discharge is split by R4
into two jets and channeled, by the R4 jet above, the reactor
train walkways and the after-reactor and reactor vessels to
form a dense gas cloud moving predominately at ground
level along the reactor train in a NNW direction. After 2 s,
the LFL of this cloud has reached the northern end of the
reactor train, i.e. R1; there is, however, relatively little
westward development of this cloud due to the intense
momentum of the R6 discharge split by the R4 reactor.
At 8 s, the R4 cloud is still airborne (24 m elevation at
the LFL) and much larger and starts to slump towards the
ground in a southerly direction. The LFL makes a footprint
around the northwest corner of the Main Office building and
commences exiting from the domain in a southerly direction
between the two buildings (i.e. into the car park); the cloud
does not, however, contact the Project Office. The R6 cloud,
though, now has developed its LFL envelope up to 16 m
elevation, and is, at its north-northwest ground boundary,
approaching the Laboratory with little westerly develop-
ment. The height of this cloud would no doubt have been
restricted by the floors in the rest of the open plant structure
of section 25A and so the 16 m vertical extent of the LFL is
perhaps an artefact induced by the use of uniform porosity
and resistance.
By 12 s (Figure 16), part of the R4 UFL envelope has
touched down between the office buildings and the LFL cloud
has proceeded out into the car park at a mean velocity of some
4 m s1. The other portion of the R4 LFL cloud has exited
to the east from the space between the north face of the
three-storey Main Office building and the southern end of
sections 25 and 7. The R6 LFL envelope by this time has
started to surmount the two-storey Laboratory building and is
commencing contact with the Control Room; the westerly
spread of the LFL has reached the westernmost end of the
Laboratory.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
After 18 s, the R6 UFL envelope has reached the Control
Room and has topped the two-storey Laboratory building.
Both the LFL and UFL envelopes have exited from the
western boundary of the domain towards the hydrogen plant.
The R4 discharge LFL cloud continues to exit between the
Main Office block and the southerly ends of sections 25 and
7 as well as southerly between the space with the Project
Office; the UFL of this envelope has now mounted the
northwest corner of the office block.
By 22 s, the R6 LFL and UFL envelopes have both
surmounted the Laboratory and the Control Room and are
proceeding across the western domain boundary towards the
hydrogen plant. The centre of gravity of the main body of
this cloud is situated just north of R1. Both the LFL and
UFL R4 envelopes are now exiting from the southern
domain boundary between the two offices and into the car
park; the eastward LFL cloud still exits from the eastern
boundary in the space between the Main Office and the
southern end of section 7.
At 30 s, the R6 discharge clouds have proceeded northerly
past the Control Room and Laboratory. The R4 discharge
cloud is still split by the Main Office block into two sections;
one exiting into the car park, the other to the east between
the office and the southern ends of sections 25A and 7.
Single Discharge from R4 (Figure 17, ‘Movie’ 2)
The transient development of this vapour cloud differs
significantly from that above. At the commencement of the
event, a series of choked vapour discharges as from a slit, set
to be approximately similar in area to that noted in the piece
of the B4 bellows by Foley (1974a), is portrayed for about
4 s. This set appears as a vertical jet that increases and then
diminishes in flowrate.
The first portion the discharge from the R4 is a westward
proceeding jet developing over a time span of 1 s. The next
sequences show the interaction of the jet impact from R4 as
it hits R6 when the B4 bellows is blown off and the jet
formed is deflected upwards. By a further 4 s, the shape of
both the LFL and UFL envelopes are roughly cylindrical;
the height of the LFL cloud is about 25 m high with a
cylindrical ground footprint of some 30 m diameter for the
LFL and 15 m for the UFL.
At 10 s, the prevailing wind is distorting the LFL envel-
ope; both the UFL and LFL footprints commence a west-
ward progression across the open space in front of the
Project Office and northward to the base of the R2 reactor
(LFL). The LFL southern envelope approaches the north-
west corner of the Main Office building.
In 12 s (Figure 17), the LFL envelope’s southeast extre-
mity has impacted the northwest corner of the Main Office
building and has commenced exiting from the eastern
domain boundary between the southern ends of section
25A and 7 and the northern wall of the office building.
The base of the LFL footprint has progressed northward to
the base of R1 with the northern boundary of the UFL just
past the base of R4.
At 16 s, the momentum of the jet has pushed the UFL
envelope to the northwest corner of the Office building. The
LFL envelope in its eastern branch is now greater in
elevation than the office building with a southern branch
pushing past the west wall of the Office building and
commencing to exit from the southern domain boundary
 FLIXBOROUGH
into the parking lot. The northern extent of the LFL
footprint has developed just midway of the last after-reactor.
Westward development of the cloud has just reached the
northeast corner of the Project Building with the edges of
both the LFL and UFL in line with the east–west axis of R6
pushing some 20 m past R6 at somewhat less than 1 m s1.
Within 20 s, due to the continuing momentum of the R4
jet, the UFL envelope detaches from the eastern wall of the
office block to also progress in the direction of the parking
lot. The LFL southern branch fills the space between the two
offices and further pushes out into the parking lot at over
8 m s1; the eastern LFL development has topped the Main
Office block and is exiting eastward at some 4–5 m s1.
Westward development of both envelopes has now hit the
northeast corner of the Project Office and progressed
westward some 8 m along its northern wall in the case of
the LFL.
By 21 s, the LFL envelope at ground level has progressed
westward some 20 m along the Project Office northern wall
with the UFL cloud only 8 m behind. The northern devel-
opment of the LFL envelope has only just reached within
22 m of the Laboratory as a rather flattened pancake-shaped
cloud. Both the LFL and UFL envelopes have pushed
southward out into the parking lot; the UFL cloud has
become completely detached from the Main Office block
and slides along the eastern wall of the Project Office. The
LFL envelope continues its eastern progression over the top
of the Main Office block and around behind sections 25A
and 7, exiting eastward at some 4–6 m s1.
In the sequence of frames following 21 s, a pseudo-steady
state has developed for the southern and easternmost exiting
domain discharges; there is a strong UFL and LFL ground
jet into the parking lot along with a less strong, but
volumetrically greater, one at the LFL to the east channelled
by the north wall of the Main Office building. Aside from
these two discharges to the west, both the LFL and UFL
continue to enlarge with little northern development. The
western ground cloud has progressed towards the eastern
edge of R5.
Summary of the Results of the CFD Simulations
There were 28 fatalities in the explosion; of these 18
individuals were in the Control Room area as well as three
each in the Gate House and Warehouse=Flaking Plant
(Humberside Police, 1974). In the Laboratory (separated
by less than 12 m from the Control Room, Figure 1) the
eight survivors were able to observe the release over a period
estimated to be 8–12 s. They then appreciated the threat
and ran from the building. Other witnesses, further away,
likewise directly observed the developing vapour cloud, the
resultant fires and the explosion. These observations, when
Table 3. Analysis of the Flixborough ionospheric records of Jones and Spracklen (1974) and Krasnov et al. (2003).
Radio trace Y (degree) j (degree)
T1 7 R1 10.0 206
T2 7 R1 15.8 176
T3 7 R1 15.2 214
T1 7 R2 18.2 205
T2 7 R2 22.7 192
T3 7 R2 22.6 212.3
Mean — —
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
125
compared with the results of the CFD simulations of events
show remarkable agreement with the cloud development
portrayed in the two-stage event.
The two CFD simulations of the possible accidental
cyclohexane releases in the Flixborough disaster permit
the following conclusions.

In both cases, the R4 discharge results in ground level
vapour clouds that exit to the east, from around the
southern end of sections 25A and 7, and to the south as
a jet channelled between the works and Main Office
buildings out into the parking lot.

As well, both instances exhibit conspicuous vertical
cloud-like accumulations near the southern ends of
Sections 25A, 7 and 27 and located on the determined
focus of the blast.

Only in the case of the single discharge from R4 is there
the significant westward development of a ground-level
cloud that corresponds directly to witness recollections.

In the situation with the two simultaneous discharges, a
strong split jet (from R6) is formed that develops into a
rapidly northward moving ground-level cloud, with little
westward progression. This cloud reaches the Laboratory
and Control Room areas in less than 12 s.

In the case of the R6 discharge the LFL=UFL distribution
around the ‘8-inch’ pipes, if ignited as a jet fire, can be
used to explain the inferred temperature distribution to
these pipes.
BLAST MEASUREMENTS
Grover (1974) made seismic recordings of the Flixbor-
ough explosion and determined that the time of the explo-
sion was 15:52:15.5  2.1 UT. Analyses of infrasound
measurements were also made. Both were compared with
a large blast at sea recorded by the same set of instruments;
this suggested that the bast was equivalent to between 6
and 16 tons of nitroglycerine, or 8.6–23 tonnes of TNT
(Meyer, 1987).
Jones and Spracklen (1974) made six Doppler ionosphere
records of the Flixborough explosion using two transmitters.
No estimate of blast was given but its occurrence was
estimated to be at about 15:50:35 UT.
In a reanalysis of the ionospheric records, Table 3,
Krasnov et al. (2003), determined that the explosive yield
was equivalent to 283  38 tonnes of TNT and that the blast
took place at 15:52:08  6 UT. A comparison of this yield
with that resulting from the detonation of fuel-drop–air
clouds (Fishburn et al., 1981) suggests that the Flixborough
explosion was the result of the detonation of only 14  2
tonnes of cyclohexane.
D (km) hr (km) te (UT) Q (tonne TNT)
65.8 228 15:52:16 210  35
101.2 228 15:52:27 290  35
104.8 228 15:52:18 390  55
118.5 218 15:52:06 405  50
167.7 222 15:51:54 230  40
173.7 224 15:51:47 175  40
— — 15:52:08  6 283  39
126 VENART
DISCUSSION AND CONCLUSION
The evidence and studies outlined above support the case
that the Flixborough explosion was caused by a two-stage
event that resulted in a release from only one reactor, R4,
that was limited to about 10–15 tonnes of cyclohexane. The
initiation of this failure may have been fatigue of the B4
bellows. A subsequent release from R6 was responsible for
the exposure to the ‘8-inch pipes’ and the carbide formation
on the R4 stub pipe and its bellows remnants. A blast in the
Main Office building, the Project Office building, or both
together, and=or the R4 jet-fire may have ‘triggered’ the
detonation of the cloud formed; if true, this is one of a very
few genuine vapour cloud detonations.
The consequences of these results should:

prompt industry, insurers, and regulators to initiate
reviews of ‘reportable’ inventories in chemical plant and
in transportation,

initiate further, more detailed, studies of chemical plant
safety, security and placement by industry, government
and universities; and

revise the accepted cause and consequences of the Flix-
borough explosion.
In addition finite element and computational fluid dyna-
mics analyses of accidents, such as used in this study, can be
useful tools in incident investigations. Though these tools
were not widely available at the time of the incident their use
today should be mandatory.
As a further point of concern in these times of heightened
security it should be noted that many local delivery services
routinely transport 10–15 tonnes of product, e.g. LPG,
which if released to form a vapour cloud and then deliber-
ately detonated could cause similar blast damage and even
greater loss of life than in this case; a possibility that should
be of special concern to security services.
REFERENCES
Ale, B.J.M. and Bruning, F., 1980a, Unconfined vapour cloud explosions,
Letter, Chem Eng, 352: 47–49.
Ale, B.J.M. and Bruning, F., 1980b, Unconfined vapour cloud explosions,
Letter of reply, Chem Eng, 358: 494.
Amtec Engineering Inc., 1996, Tecplot User’s Manual, version 7
(Bellevue, WA).
Artingstall, G., 1974, Appraisal of the damage in the reactor vessels in
Section 25A, Report no. 4, Part I, Safety in Mines Research Establishment
(SMRE), September 11.
Au-Yang, M.K., 1999, Common mistakes and misconceptions in flow-
induced vibration and coupled fluid-structural dynamics analysis, ASME
PVPD, 29: 6.1–6.17.
Ayre, E., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty-fourth
Day, November 15.
Ball, J.G., 1975, After the Flixborough Report: do we know the real truth?,
Proc Eng, 56(12): 39–46.
Ball, J.G., 1976, Was this what caused the Flixborough explosion?, Proc
Eng, 57(1): 53–60.
Batstone, R.J., 1975, Description and results of bellows test program and 20
dog leg pipe tests, Cremer and Warner Report no. 7, January 7.
Bjerketvedt, D., Bakke, J.R. and van Wingerden, K., 1997, Gas explosion
handbook, J Hazard Mater, 52: 1–150.
Butler, P., 1975, The lessons to be learned from the Flixborough inquiry,
Engineer (GB), 241(6247): 20–24.
Carter, B., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Fourth Day,
September 12.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
Center for Chemical Process Safety (CPPS), 1994a, Guidelines for evaluat-
ing the characteristics of vapor-cloud explosions, flash fires, and BLEVEs
(American Institute of Chemical Engineers, New York, USA), Section
2.4, Fig. 2.4, p 12.
Center for Chemical Process Safety (CPPS), 1994b, Guidelines for evaluat-
ing the characteristics of vapor-cloud explosions, flash fires, and BLEVEs
(American Institute of Chemical Engineers, New York), Appendix F,
pp 363–382.
Colquhoun, D.C., 1974, Minutes of Proceedings at the Court of Inquiry; into
the disaster which occurred at Nypro (UK) Ltd, Flixborough, Third Day,
September 11.
Cottrell, A.H. and Swan, P.R., 1976, A metallurgical examination of the
eight-inch line, Chem Eng, April: 266–274.
Cox, J.I., 1976, Flixborough—some additional lessons, Chem Eng,
May: 353–358, 387.
Dickinson, J., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Second Day,
September 10.
EJMA, 1993, Standards for the Expansion Joint Manufacturers Association,
Inc., 6th edition (Tarrytown, NY, USA).
Fishburn, B., Slagg, N. and Lu, P., 1981, Blast effect from a pancake shaped
fuel-drop–air cloud detonation (theory and experiment), J Hazard Mater,
5: 65–75.
Flomerics, 1994, FloVent Reference Manual, Document no.
FloVent=RM=0994=1=0 (Flomerics Ltd, Surrey).
Foley, J.H., 1974a, Metallurgical examination of the damaged pipes from
section 25A, Report no. 1, Part I, Safety in Mines Research Establishment
(SMRE), September 10.
Foley, J.H., 1974b, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty-third
Day, November 14, pp 34–35.
Foley, J.H., 1974c, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty-fourth
Day, November 15.
Foley, J.H., 1974d, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty-seventh
Day, p 26–27.
Foley, J.H., 1974e, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty-seventh
Day, p 21, para 5.
Foley, J.H. and Nicholson, C.E., 1974, Metallurgical examination of the
damaged pipes from Section 25A, Report no. 1, Part III, Safety in Mines
Research Establishment (SMRE), November 8.
Fu, W.B. and Eyre, J.A., 1974, Unconfined vapour cloud explosions, Letter,
Chem Eng, 359=360: 368.
Games, G.A.C. and Waterhouse, D., 1974, Construction of and tests on a
reconstructed bridging pipe assembly, Report no. 2, Part II, Safety in
Mines Research Establishment (SMRE), November 10.
Gill, S.S., 1974, Minutes of Proceedings at the Court of Inquiry; into
the disaster which occurred at Nypro (UK) Ltd, Thirty seventh Day,
pp 51–52.
Goodchild, M.B., 1975, Minutes of Proceedings at the Court of Inquiry; into
the disaster which occurred at Nypro (UK) Ltd, Flixborough, Fifty-fourth
Day, January 24, pp 24–26.
Gromles, M.A., 1984, A simple approach to transient two-phase level swell,
Multi-phase Flow and Heat Transfer III, Vol 1 (Miami Beach, FL, USA),
pp 527–538.
Grover, F.H., 1974, Infrasonic and seismic wave records from the Flix-
borough and St Bridget explosions, AWRE Report no 046=74, p. 27.
Gugan, K., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty-ninth
Day, p. 17, November 22.
Gugan, K., 1976, Flixborough—A combustion specialists view, Chem Eng,
341–352, 387.
Gugan, K., 1979, Unconfined Vapour Cloud Explosions (IChemE, Rugby,
UK).
Gugan, K., 1980a, Unconfined vapour cloud explosions, Letter of reply,
0
Chem Eng, 358: 491–493.
Gugan, K., 1980b, Unconfined vapour cloud explosions, Letter of reply,
Chem Eng, 359=360: 567.
Gugan, K., 2000, Flixborough kicking, Letter, Chem Eng, 25: 30.
Harris, R.J., 1983, The Investigation and Control of Gas Explosions in
Buildings and Heating Plant (E&FN Spon, London, in association with
British Gas, UK).
Harris, R.J. and Wickens, M.J., 1989, Understanding vapour cloud
explosions—an experimental study, 55th Autumn Meeting of the Institute
of Gas Engineers, Kensington, London, UK.
 FLIXBOROUGH
Harry, L., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Third Day,
September 11.
Høiset, S., Hjertager, B.H., Solberg, T. and Malo, K.A., 2000, Flixborough
revisited—an explosion simulation approach, J Hazard Mater,
77(1–3): 1–9.
Humberside Police, 1974, Major Incident files, Explosion, Nypro (UK) Ltd,
June 1, 1974. Books 1–4 plus photograph albums 1–6.
Jakubauskas, V.F. and Weaver, D.S., 1998a, Transverse vibrations of bellows
expansion joints. Part I: fluid added mass, J Fluids Struct, 12:
445–456.
Jakubauskas, V.F. and Weaver, D.S., 1998b, Transverse vibrations of bellows
expansion joints. Part II: beam model development and experimental
verification, J Fluids Struct, 12: 457–473.
Jones, T.B. and Spracklen, C.T., 1974, Ionospheric effects of the Flix-
borough explosion, Nature, 250: 719–721.
Kataoka, I. and Ishii, M., 1987, Drift flux model for large diameter pipe and
new correlations for pool void fraction, Int J Heat Mass Transfer, 30(9):
1927–1939.
King, E., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Second Day,
September 10.
King, R., 1975, Flixborough—role of water re-examined, Proc Eng,
September, 69–72.
King, R., 1977, Latent superheat—a hazard of two-phase liquid systems,
Chem Proc Haz IChemE Symp Ser, 49: 139–145.
King, R., 1999, Flixborough, 25 years on, Process Eng, June: 27–28.
King, R., 2000, Flixborough: re-open public inquiry, Letter, Chem Eng,
May 25: 30.
Kitching, R., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Thirty seventh
Day, 51–52.
Kletz, T., 2000, Flixborough, no whitewash, Letter, Chem Eng, April 20: 18.
Krasnov, V.M., Drobzheva, Ya.V., Venart, J.E.S. and Lastovicka, J., 2003, A
re-analysis of the atmospheric and ionospheric effects of the Flixborough
explosion, J Atmos Solar-Terrest Phys, 65: 1205–1212.
Leclude, J. and Venart, J.E.S., 1996, Single and two-phase discharge from a
pressurized vessel, Rev Gen Therm, 35: 503–516.
Leclude, J. and Venart, J.E.S., 1997, Determination of the amounts
discharged in the Flixborough disaster, unpublished work, University of
New Brunswick, Canada.
Marshall, H., 1974, Official Statement to Humberside Police, H Div.,
Scunthorpe, June 9, 1974, Humberside Police Book 1, pp 167–169.
Meyer, R., 1987, Explosives, 3rd edition (revised) (VCH, New York, USA)
pp 239 and 367.
Miller, K.J., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd, Flixborough, Fifty-first Day
and Fifty-second Day, pp 16 and 27.
Newland, D.E., 1964, Buckling of double bellows expansion joints under
internal pressure, J Mech Eng Sci, 6: 270–277.
Newland, D.E., 1976, Buckling and rupture of the double bellows
expansion joint assembly at Flixborough, Proc R Soc Lond, A351:
525–549.
NIST, 1999, Standard Reference Database 4; Thermophysical Properties of
Hydrocarbon Mixtures, Supertrapp, Version 3.00 (National Institute
Standards and Technology, Gaithersburg, MD, USA).
Phillips, H., 1981, Unconfined vapour cloud explosions: a new look at
Gugan’s book, Chem Eng, 369: 286–287.
Rajaratnam, N. and Chamani, M.R., 1995, Energy loss at drops, J Hydraul
Res, 33: 373–384.
Ramier, S. and Venart, J.E.S., 2000, Boiling liquid expanding vapour
explosions: dynamic re-pressurization and two-phase discharge, Hazards
XV, IChemE Symp Ser 147: 527–537.
Roberts, A.F. and Pritchard, D K., 1982, Blast effect from unconfined vapour
cloud explosions, J. Occup Accidents, 3: 231–247.
Ryder, D.A., 1974, Minutes of Proceedings at the Court of Inquiry; into the
disaster which occurred at Nypro (UK) Ltd., Flixborough, Thirty seventh
Day, pp 57–58.
Sadee, C., Samuels, D.E. and O’Brien, T.P., 1976=1977, The characteristics
of the explosion at the Nypro (UK) Flixborough Plant on 1st June 1974,
J Occup Accidents, 1: 203–235.
Trans IChemE, Part B, Process Safety and Environmental Protection, 2004, 82(B2): 105–127
127
Secretary of State for Employment, 1975, The Flixborough Disaster
(HMSO, London).
Snedden, N.W., 1981, The strength and stability of corrugated bellows
expansion joints, PhD Thesis, Churchill College, Cambridge, UK.
Snedden, N.W., 1985, Analysis and design guidance for the lateral stiffness
of bellows expansion joints, Thin-Walled Structures, 3: 145–162.
Snee, T., 2004, Flixborough: water hypothesis laid to rest?, TCE, February,
pp 28–29.
Swan, R., 2000, This is beginning to look like a cover-up, Editorial, Chem
Eng, April 6: 2.
Teng-yang, R., Sollows, K.F. and Venart, J.E.S., 2000, Froude modeling of
the Flixborough ‘By-pass’ pipe, Hazards XV, IChemE Symp Ser, 147:
139–154.
Usher, A., 1979, The Flixborough disaster, Med Leg J, 43(3): 84–102.
Venart, J.E.S., 1999, Flixborough; a final resolution?, INTERFLAM’99, Vol
1, Grayson, S. (ed.), pp 257–272.
Venart, J.E.S. and Tan, D.M., 2000, Flixborough; twenty-five years after, the
final resolution?, Proceedings of ICPVT-9, Sydney, April 9–14, Vol 2, pp
525–539.
Warner, F., 1975, The Flixborough, Chem Eng Prog, 71(9): 77–84.
Warner, F. and Newland, D.E., 1975, Flixborough Explosion—Mechanical
engineering provides the key, Chart. Mech. Eng. (GB), June: 76–81.
Weaver, D.S. and Ainsworth, P., 1989, Flow induced vibration in bellows,
ASME J Press Vessel Technol, 111: 402–406.
Wehmeier, G., Westphal, F. and Friedel, L., 1994, Pressure relief system
design for vapour or two-phase flow?, Trans IChemE, Part B, Proc Safe
Env Prot, 72: 142–148.
Westgate, K., 1976, Flixborough—the human response, University of
Bradford, Disaster Research Unit, Occasional Paper no. 7, January 1975
(reprinted September 1976).
ACKNOWLEDGEMENTS
Many individuals have contributed to this work. Specifically, the staff at
the Health and Safety Laboratory Library in Sheffield, and in particular
Linda Heritage, are thanked for their help in making the Minutes of
Proceedings of the Court of Inquiry and many other documents and reports
dealing with the disaster available over an extended period of time; C.E.
Nicholson gave permission to reference many of the SMRE and other
reports available. The Scunthorpe Evening Telegraph, and in particular Ms
Jackie Cole, provided access to many photographs and material from their
archives dealing with their staff’s coverage of the accident. Professors
T.B. Jones and Terry Robinson of the University of Leicester gave freely of
their time to explain and elaborate on the ionospheric effects of the
Flixborough explosion; V. Krasnov and Ya Drobzheva of the Institute of
Ionosphere, Ministry of Science and Education, Kazakhastan helped
analyse these records. Dr John I. Cox provided details of his precis on
the statements and re-enactments of eyewitness made for the Inquiry and
other material. Ralph King and his colleague, Ronald Hirst, provided details
regarding what influence water in the reactors may have had on the accident.
The personnel at the BBC Film Archive Unit in London assisted in the
review of the background video and cine film records for the event. BBC
North, and especially Christine Hamill and Close-up North series producer
Ian Cundall made arrangements for an interview with one of the Laboratory
eyewitnesses, Brian Pettitt, and prepared a programme historically reviewing
the event and its consequences. The northeast Lincolnshire Archivists,
J. Wilson and J. Veltham, allowed access to their comprehensive files on the
incident at the as well as details of the coroner’s report. My colleagues at the
University of New Brunswick and many graduate students have all
substantially contributed through discussion, critiques, and other studies.
Drs D.M. Tan and G. Li conducted the ANSYS finite element (FEM)
analyses and assisted in the preparation of the Tecplot ‘movies’ of the
Flomerics CFD and the FEM results. Any errors and omissions are the
author’s alone.
The manuscript was received 14 July 2003 and accepted for publication
7 January 2004.