Failure Knowledge Database - 100 Selected Cases

Fire and Explosion of LPG Tanks at Feyzin, France

【January 4th, 1966 Feyzin, France】

Mitsuo Kobayashi (Graduate School of New Frontier Sciences, University of Tokyo)

Masamitsu Tamura (Graduate School of New Frontier Sciences, University of Tokyo)

This accident brought about an enormous disaster resulting from a small hu man
error. Th e d etails of th e accid ent sh ow well the d angers of handling LPG (Liquefied
Petroleum Gas) and force us to ca refully c onsider the loca tion of larg e tanks of LPG.
Furthermore, it is the f irst acciden t in which the phenom enon of BLEVE was m ade
clear.

In the early morning of January 4th, 1966, a large-scale explosion and fire
accident of LPG occurred in th e Feyzin r efinery of the F rance national petroleum

1
 Failure Knowledge Database - 100 Selected Cases

company at F eyzin in the subu rbs of L yon in the s outhern part of France. Three
operators op ened two 2-i nch valves, which we re mounted in series at th e botto m of a
1200m3 p ropane sp herical tank. Due to s ome problems, t he valves d id not close , and
LPG escaped. Because LPG is heavier than air, it spread along the ground as a cloud of
vapor to the highway at a distance of 60m an d the local road paral lel to the highw ay.
The traffic on the high way was st opped, bu t the traffic on the local road was not
stopped. The e scaped LPG caught fire by a car passing through th e local road and
exploded. T he fire r eached the tank from whic h the LPG first leaked , and exp losive
destruction, called BLE VE, of th e tank vapor p hase was ca used by th e tank fir e. T he
fire spread to five oth er LPG tank s in th e same tank ya rd and also damaged many
tanks of cru de oil, and other materials that we re near. It is said that eighteen p ersons
died and do zens of pe rsons we re injure d, but in ano ther estim ate, a maximum of 81
deaths and 130 injuries was gi ven. The monetary damage was said to be about 18
million US dollars at the time of the accident (70 million US dollars in 1990 dollars).
LPG is the most d angerous fu el tha t is made fr om p etroleum. In this a ccident,
the remark able characteristics of LPG a ppeared. Th e d angers of LP G includ e the
following. a) LPG can tr avel along t he g round for a long di stance, beca use th e vapor
specific gra vity is greater than air , allowing the formation of a “vapor cloud”. b) As a
result of the equilibrium relationship between the vapor pressure and the temperature,
an extre mely low te mperature res ults when t he p ressure d rops. c) A sp ecial for m of
explosive d estruction ca lled B LEVE is caused in the fire of a storage tank. Since t he
gas form of LPG is colorless and odorless, an explosion may occur though the leakage is
not noticed, although this was not the case in the accident of Feyzin. These dangerous
characteristics of LPG are the same for the general consumption as for industrial use.
BLEVE: The abbreviation of “Boiling Liquid Expanding V apor Explosion”, an
explosion p henomenon caused by the rap id p hase change (vaporization) of a
liquid. A l iquid substan ce stored in a pressurized vessel is in a vapor -liquid
equilibrium condition. When it is heated to a t emperature that is sufficiently
higher than the boiling point of the su bstance at atmospheric pressure, such
as when th e container is h eated in a fir e, t he p ressure also rises in the
container. When the container is dama ged under this condition and t he gas
escapes, the pr essure in the co ntainer drops t o atmospheric pressure in an
instant. The equilibrium state in the cont ainer is broken at this time, and an
explosion phenomenon results from the bumping of th e liquid that is rap idly
changed into the gas phase. This is called BLEVE.
The BLEVE phenomenon that occurs in the LPG tank was shown in Fig.1. First,

2
 Failure Knowledge Database - 100 Selected Cases

a fire was generat ed under the sph erical tank. The LPG in the tank w as heated, and
the tank internal press ure rose. The LPG vapor was discharged to the atmosphere
when the internal press ure rose above the s etting pressure of th e safety relief valve,
and this vapor caught fire. Th e surface of the wall above t he liquid l evel was strongly
heated, causing the strength of the wall to deteriorate. Finally, the wall ruptured under
the internal pressure. As a r esult of the rupture, the pressure inside the tank dropped
to atmo spheric pre ssure instantane ously, and the BLEVE phe nomenon happe ned.
The large vapor that resulted became an enormous fireball.

1. Event
On J anuary 4th, 1966, at the Feyzin refinery in the suburbs of L yon in the
southern pa rt of F rance, th e valves c ould n ot close during work at a L PG spheri cal
tank, which contained 1200kl propane, and a fire and explosion occurred from the
leakage of LPG. The tank causing the accide nt was constructed in the tank yard where
eight spherical tanks (four 1200kl propane ta nks and four 2 000kl butan e tanks) were
located. There was another tank yard with four atmospheric pressure tanks containing
400kl of jet fuel (kerosene fraction) adjacent to the LPG tank yard.
Three operators opened the valves at the bott om of the L PG tank in order to
drain the free water that had accumulated at the botto m. They finish ed the wor k in
about 15 minutes, and they intend ed to close t he valves, b ut the valve s d id not close
perfectly. There were two valves that were placed in series, and the upstream valve did

3
 Failure Knowledge Database - 100 Selected Cases

not clos e. They left th e site withou t closin g th e downstrea m valve (or it could not be
closed). The alarm (a gas detector) indicating t he possibility of a gas leak rang in t he
control room, and a warning was issued in all of the Feyzin districts.

In the meantime, the l eaking LPG spread in t he d irection of th e high way that
was located 60m from the tank, whil e forming the vapor cloud. There was a small local
road running parallel to the highway . The traffic on the highway was stopped, but t he
traffic on the local road could not be stopped. One car drove past on the local road. The
driver recognized the gas flow and stopped the c ar several meters before the gas flow.
However, an enormous flame happened at th e moment, the flame flowed backward in
the direction of the tank, and the leaking tank was wreathed in flames. The fi re
extinguishing work was carried out on the circumference of the flame by sp raying
water and f ire extinguishing foam, and the adjacent sph erical tanks were c ooled by
water from the water -spraying noz zle of each tank. However, the safety valve of the
leaking tank op erated, and the gas sp outed out of the tan k. Th e first accident tank
exploded su ddenly, and then th e second and third tanks exp loded. In ad dition,
petroleum products, crude oil, and other materials caught fire, and it became a major
disaster.
As shown in this accident, if l eaked vapor diffuses to the atmosphere like a cl oud

4
 Failure Knowledge Database - 100 Selected Cases

without igni ting imm ediately then an exp losion may occur. This p henomenon ca lled
“Unconfined Vapor Cloud Explosion”, abbreviated as “UVCE”.

2. Course
At 6:30, in the early morning, three ope rators started the water draining
operation of the tank. T he d rainage flow rate is nor mally c ontrolled by ad justing the
downstream valve while the upstream valv e i s fully opened, but the actual meth od
used on that day was not found recorded anywhere.
It ap pears that the draining work was finished after about 15 minutes. First,
they tri ed to cl ose th e upstream v alve. Th e L PG leak sta rted becaus e that valve was
not closed completely. The closing of the se cond valve, th e downstream valve, was not
described. After a fe w minutes, they were wrapped in pro pane vapor, and the y walked
from the site unsteadily. The leaking LPG sp read in the di rection of the highway, with
partially vaporizing. The alarm rang in the control room at this time.
At around 07:05, the alarm rang through all of the Feyzin districts. However, “all
of the Feyzi n districts” may mean just the “wh ole Feyzin Refinery”, and it is n ot clear
in the literature whether the alarm rang throug h the whole town or through the whole
factory. Th e first fir e tr uck of th e f actory turne d out imm ediately. Th e fire brigad e of
Lyon arrived at 07:20. Apparently, at some time before the arrival of the fire brigad e of
Lyon, the traffic on the highway next to the refinery was stopped.
At a little before 07:30, a car came an d st opped on th e l ocal r oad w here th e
traffic had not ye t be en sto pped. The explosion appare ntly o ccurred just afte r the car
stopped. The fire increased in po wer, spread along the leaking LPG, and the tank wa s
wreathed with flames immediately.
At 08:40, the first tank exploded by the BLEVE phenomenon after spouting LPG
through the safety valv e. The s econd and th ird spherical tanks exploded five minutes
later. Two more spherical tanks and many oil tanks burned continuously.
In the next morning, the fire was extinguished at last.

3. Ca use
3.1. Cause of the leakage
It seems th at the leakage was caused by the freezing of th e valve by the d rain
work. However, the freezing is a result and not a cause in fact. For LPG, in this case
propane, th e t emperature d rops t o -40℃ if th e pressu re is low ered to atmospheric
pressure. At this temp erature, not only is the moisture i n the air frozen, but the
moisture also reacts with LPG , forming a so lid hydrate. The hydrate formation does

5
Failure Knowledge Database - 100 Selected Cases

not require an extremely low te mperature. Apparently, either the valve handle was
stuck by the frozen moisture or the valve could not be closed tightly as a result of the
hydrate formation.
Considering the possib ility of ice form ation, usually t wo drain valves and
sampling valves are m ounted on th e botto m of LPG tanks in series . A n examp le of
this situation is shown in Fig.3. The usua l operation for opening the valves at the
beginning of use is as f ollows. First, the upstream valve is opened fully, and the inlet
side of the downstream valve is pressurized. Next, the downstream valve is gradually
opened starting fro m the slight open, ad justed to cr eate the n ecessary flow rate .
Because the downstrea m valve inlet is pr essurized, the t emperature at upstrea m
valve d oes not d rop. T he outl et sid e of the d ownstream valve is coo led und er th e
operation, and it is also important t o ensure that there is s ufficient distance so tha t
the upstream valve does not drop to a low temperature. This accident was believed to
have been caused as follows: th e upstream v alve was not op ened fully and the
pressure at the outlet o f the first v alve was ne ar the atmo spheric p ressure, so the
upstream v alve was c ooled by the pressure d ifference. It i s absolutely forbid den t o
operate valves lik e this. Therefore, the cause of the accident is either the operator’s
human error or a mista ke on the management side in not t o explaining the meaning
and proper execution of the operation.
Moreover, another one of th e causes is that t he operator d id not cl ose the
downstream valve. T here is n ot a cl ear description of why t his valve w as not cl osed.
It is p resumed that either i ce was gen erated in th e d ownstream valve lik e in t he
upstream v alve or a ra pid large LPG leak occurred at th e end of d rain work. If a
large LPG leak occurred, the operators may have panicked when they could not close
the upstream valve and therefore may have not remembered to close the downstream
valve. Basically they ha d to d ecrease the downstream valve opening when the LPG
began to appear in the drain water. Then the valve could have been closed quickly at
the end of drain work. Either they forgot to carry out this procedure or they could not
close the second valve because it had also become stuck by freezing.

6
Failure Knowledge Database - 100 Selected Cases

3.2. Cause of the ignition … (Refer to Fig.2)

It is presumed that the leaked LPG spread along the ground and was ignited by
a car that w as driving down the local road . The propane va por was thought to have
been ignited by a cigarette of the d river or the high temperature of the engine of the
car that had just stopped.
It is not known why the traffic on the local road could not be stopped, though the
traffic on the highway c ould be stop ped. If there was enou gh time, an d if a p erson
who knew the roads in the area well had led the traffic r estriction, it se ems that the
traffic on the local road would also have been stopped early.
Although th e d irect cau se of th e ig nition is th at the stop ping of the tr affic was
delayed, the fund amental p roblem see ms t o be the l ayout of t he area. The
construction of the high way had been permitted with only a distance of slightly over
50m fro m t he g roup of en ormous LPG tanks. Mor eover, from the p hotographs, i t

7
Failure Knowledge Database - 100 Selected Cases

appears that there was no dike around the LPG tanks. In this situation, when a large
amount of LPG leak ed, the LPG v apor flows a long the gr ound to th e highway. Th e
main cause s of the ign ition are th at the d istance r egulation was insufficient and
there was no dike around the tanks.
Another possible cause of the ignition is the the ory that a static elect ricity spark
occurred w hen the L PG began spouting from the safety valve. Ho wever, if the
ignition occurred at the initial stage of the leakage, a fire should have occurred, but
an exp losion should n ot have hap pened be cause en ough L PG was not accumulate d.
Furthermore, after a la rge amount of leakage, the LPG might not ignite because the
LPG con centration mi ght be higher than t he exp losion limit. Th erefore, it is
considered that the possibility of the static electricity theory is low.

3.3. Cause of the first explosion

Before the accident it was assume d that the destruction and exp losion of th e
LPG tank would not occur ev en if the ta nk was wreath ed with fir e, be cause t he
pressure relief valve (safety valve) woul d op erate and d ecrease th e tank p ressure
when the tank pressure rose by vaporization of the liquid LPG remaining in the tank
due to tank heating. However, in fact th e tank exploded, scattering ma ny fragments
of various sizes.
The e xplosion o f the ta nk is e xplained by the BLEVE phenomenon, which has
already been introduced (Refer to Fig.1). The accident at Feyzin is the first one that
has been explained by the BLEVE phenomenon.
The water s howering devices havin g at least a certain prescribed capacity are
required f or spherical and cylindrical LPG ta nks in Japan at present. Ho wever,
water show ering f or cooling fro m t he top of t he tank ap peared not to have bee n
executed at the tank, therefore, b y heating t he tank vap or p hase, the B LEVE
phenomenon happened and the explosion occurred.

3.4. Cause of the following explosion and fire

The nearby tank exploded as a result o f the BLEVE phe nomenon, a nd the
fragments of the tank w ere scattered. Th e scattered fragments dam aged the piping.
However, only the piping damage might not have been enough to induce the
following explosions of a djacent tanks. Some causes of explosions of the other tanks
were being mentioned, and all of t hem wer e not c onfirmed, but they seem to hav e
somewhat reasonable explanation in it.
First, the ot her tanks w ere cool ed, but the c ooling capacity for each ta nk may

8
 Failure Knowledge Database - 100 Selected Cases

have be en insufficie nt. The BLEVE phe nomenon is a natural re sult if the cooling
capacity is insufficient. Second, the legs of th e spherical t ank wer e n ot made of a
fireproof structure. The strength of the legs d ecreases rapidly when the y are heate d
because they are m ade of iron. Eventually, the spherical t ank falls down (Refer t o
Fig.4.). Third, the distance between tank s was so short t hat each tank could be
affected by the adjacent tank. The distance between the centers of the tanks was only
27m according to the plans for the facility (Refer to Fig.2.). As the tank diameter was
about 14m f or the propane ta nks and about 16m for the bu tane tanks, the distance
between two propane tanks was 13m and between a propane tank and a butane ta nk
was 12m. The minimu m distance between LPG spherical tanks regul ated at present
in Japan is the same as the diameter of the la rger tank, in this case, 16m, which
seems insufficient. In a ny case, th e distance between th e t anks was even smaller
than this Jap anese reg ulation value. C onsidering the heig ht of th e fla re, first 75m
and finally 30m, the distance between the tanks might not have been enough.

4. Process of cause elucidation

As there are not any first hand report s of the accid ent or material s that show
results of investigation into the actual conditions, it is no t clear what kinds of method
were used f or the cause elucidation. Generally, the facts co llected from i nterviews and
the reports remained in the control room were lined up in order of time, and then they
were examined with rational consi deration. F or this accident, th e cau se would have
been elucidated using this general method.

5. Immediate action
The sounding of the alarm in the control room was the first notice of the accident,
as shown in the “C ourse” se ction above. Although there is no cle ar descriptio n in the
reports and other available materials, it is presum ed that the gas leak d etector
operated. According to lecture materials that quoted from the Paris Match Magazine, a
warning had been issu ed by the o peration of the gas d etector to all of th e F eyzin
districts. Th e moment t he alarm rang, the first fire engine owned by the comp any
turned out, and the company guards stopped the traffic on the highway. In other words,
the followin g immed iate actions w ere carried out: alarm, urgent fire brigade, and
traffic regulation.

6. Countermeasure
There are many aspects that need to be considered in making countermeasures:

9
 Failure Knowledge Database - 100 Selected Cases

the cause of the leak age that triggered the accid ent, t he cause of ignition, t he
occurrence of the BLEVE phenomenon, the spreading of the fire to the other tanks, and
so on. The countermeasures made by the company itself and the countermeasures that
made in relation to the local population are described below.

6.1. Countermeasure against freezing of valves

a) Generally, redundant sampling valves and drain valves are installed in series. In
the tank involved in this accident, the double valve was already installed. In the LPG
tank, d oubling of the d rain valves is a minimum requir ement. Th e re ason why the
double valve is a necessary choi ce is that in t he cas e of a single valv e, th ere is a
strong possi bility that the valve main bo dy will be cool ed, hur ting the closeout
function of t he valve by icing of the water in the air, generation of the h ydrates, and
other facto rs. More over, the d istance between t he two valves should be sufficiently
long so that the low temperature of the se cond valve will not affect t he first valve.
The size of the second valve should be 1/2 inches or less for sampling and as small as
possible for draining work. Fig.4 shows the composition of the draining and sampling
valves of the tank where the accident occurred. It is desirable that a distance of 1m or
more should be kep t between valve A and valve B and between valve A and valve C.
Also, the size of valve B should be 3/4 inches or less.

10
 Failure Knowledge Database - 100 Selected Cases

b) Adequate education and training of the u se of d ouble valves is n ecessary. Th e

operation described below has to be d one completely: the firs t valve mus t be op ened
fully and flow adjustment must b e done only with the second valve, because the
temperature of the first valve must not to d rop. The reason is shown in the “Cause”
section. Also, a thorough monitoring of the draining work should be attempted.

6.2. Countermeasure for other facilities

In Japan, the following minimum standards are required by law.
a) Installation of a d ike: LPG d oes n ot vap orize immedi ately, and even if it is
vaporized, LPG spreads along the ground, because it is heavier than the air. A dike is
effective for preventing the spread of LPG.
b) Facilities for water showering f rom the tank top: as a countermeasure against
the BLEVE phenomenon, the tank wall should be cooled.
c) The fireproof structure of th e legs: if th e legs of th e tank are made of iron, they
will be damaged by heat.
d) Keeping the distance between tanks to p revent fires from spreading to and from
adjacent tanks. In Jap an, the d iameter of t he larg er t ank is required as th e
minimum d istance bet ween high pressure ga s tanks at present. If i t is possible,
greater distance between tanks is desirable.
e) Installation of a gas detector for the detection of leakages.
f) The outlet of the sa fety valve: although Jap anese regu lations sp ecify that the
outlet of the safety valve should be "mounted at a safe position", in fact the outlet has
often been mounted directly on the tank. This position cannot always be “safe”.
g) The aut omatic cutof f for the excess flow rate (an excess flow valve): although
there is no direct relationship to this accident and there is no regulation in Japan, it
is desirable that a device should be installed that can automatically cut off the flow
when the f low rate increases over the set rate from the tank bottom. This
countermeasure has already been adopted in some countries, including Taiwan.
6.3. Countermeasure considered in relation to the society
The d amage by the explosion and fire is enormous when a large amou nt of LPG
leaks and spreads. The corpor ations that store and handle large amounts of LPG must
consult with the local government for desi gning an adequate commu nication system
that includes communication with the local government, notice to the local inhabitants,
and instructions for how to restrict traffic in the case of an LPG leak.
The location of a large LPG storage f acility must be studied carefully in relation to
the environment and facilities surrounding th e corporation because of the potent ial

11
 Failure Knowledge Database - 100 Selected Cases

damage resulting from the l eakage of a larg e a mount of LPG. In Japan, the distance
regulation between high pressure gas treating facilities and the buildings for public use
is carried out.

7. Knowledge
a) The handling of LPG may cause proble ms with freezing due t o formation of ice
from the moisture of the surrounding air and /or formation of hydrates when the L PG
pressure is reduced to atmospheric pressure. Devices and meth ods that prevent t he
occurrence of low t emperature haza rds are i mportant for operations such as draining
work or sampling when the pressure may be reduced to a tmospheric pressure. This is
an imp ortant issue fo r all p ersons who ar e related to LPG p rocessing and LPG
handling, as well as to petroleum refining and petrochemicals.
b) Leaked LPG is a ve ry d angerous mate rial because the vapor densit y is h eavier
than air , the ignition p oint is low and it evap orates to fo rm a c ombustible gas-air
mixture ea sily, and th e vap or is colorless and od orless. Sufficient attention a nd
communication measures are necessary for storage and handling.
c) A sp ecial form of th e exp losion can occur in LPG tanks because of the B LEVE
phenomenon. Th e d anger of th e explosion was increased i n this accid ent by th e f act
that the stress that can be supported by th e tank decreased when the part of th e wall
without a liquid was heated, because the tank was d esigned based on calculations of
the str ess t hat can be supported by the i ron wall at ambient te mperature. A sh ower
system with sufficient capacity for spraying cooling water on th e tank fire from t he
tank top is a necessary measure. Although the tank internal pressure is given by the
equilibrium pressure of the liquid at the te mperature in the tank , the temperature in
the vapo r p hase m ay be inde pendently de termined. It is an e xample o f lo sing an
equilibrium relation in the case of localized heating.

8. Influence of failure
Although th ere are various d ifferent rep orts, it appe ars th at 18 to 81 persons
died and about 80 to 130 persons were in jured, including the members of the
fire-fighting brigade. The physical damage included five LPG tanks, many atmospheric
tanks conta ining crude oil and jet fuel, and so on . Th e Feyzin Ref inery suffered
extensive d amage. R eports that m entioned th e d amage to the n eighborhood area are
not found. One report estimates the total amount of damage to be about 18 million US
dollars in dollar values at the time of th e acci dent, or 70 million US dollars in 1990
dollar values. However, considering the actual amount of damage to the neighborhood,

12
 Failure Knowledge Database - 100 Selected Cases

the c ost of t he r econstruction of th e factor y, th e non- operation losses , and so on, the
monetary damage must be much larger.

9. On the side
There are many examples of accid ents involving LPG. The exp losion of the Pemex
Co. LPG terminal in Mexico City, which occurred in November 1984, destroyed six
large spheri cal tanks, 48 small hori zontal cy lindrical tanks, and much of the nearby
residential area, and over 500 persons died , over 4,000 persons were injured, a nd
several tens of thousands of houses were severely damaged. The cause o f that accident
was not des cribed clearly. The site limits might not be clear and apparently LPG was
often stolen from the facility , and thus a large LPG leakage seemed to have occurred.
The L PG sp read along the gr ound in the sa me way as in the Feyzin accident and
ignited somewhere else, causing the t errible disaster. In J apan, there also are many
reports of awful disaste rs due to neglect of danger at LPG filling stations and other
LPG handling facilities.
A great nu mber of reports and explanations of this accid ent exist, and contents of
each r eport are differen t one oth ers, so the facts are n ot s o clear. In th is report , it is
assumed that propane leaked from an un-closed valve due to hydrate formation during
draining work, and the vapor was ignite d by a car traveling on the local road near the
refinery.

13
 Failure Knowledge Database - 100 Selected Cases

References
*Frank P . Lees, “Loss Prev ention in the Process Industries”, p8 90, 898-8 99,
Butterworth-Heinemann (1980)
*Tetsuzo Ki tagawa, “Leakage of LP gas by i mpossibility of closing of t he sampling
valve”, “Analysis of Explosion hazard”, p161-164, Nikkan Kogyo (1980)
*Toru Miura, “In the LPG pressu re storag e t ank acciden t”, J.High Pressure G as
Safety Institute of Japan, 237, 29-33 (1990)
*Chemical I ndustries Association, “ The acci dent disaster case. LPG spherical tan k
Accident”, “Safety countermeasure technology of the chemical plant 4 – Disaster case
and countermeasure”, p 230-231, Maruzen (1979)

14