Académique Documents
Professionnel Documents
Culture Documents
FouhEthtion
Trevor Kletz
The Author
Knownthroughoutthe process
industriesas a gifted communicator
on safety matters, TrevorKletz has
wide knowledgeof both practice and
theory. Hejoined Imperial Chemical
Industries on graduating as a chemist
and spent eight years in research,
sixteen in production management
and the last fourteen as safety adviser
to the Petrochemicals Division.
On retiringfrom ICI he joined
Loughborough University of
Technology, at first full-time and
then from 1986 as a Visiting Fellow.
He has written nine booksand
more than a hundred papers on
loss prevention and process safety
and is a Fellow of the Royal
AcademyofEngineering, the
InstitutionofChemicalEngineers,
the Royal Society of Chemistry
and the American Institute of
ChemicalEngineers.
ftC
Trevor Kletz
IChem
Published by
Institution ofChemical Engineers,
Davis Building,
165189 Railway Terrace,
Rugby, WarwickshireCV2I 3HQ, UK
IChemEis a Registered Charity
Preface
The Institution
111
iv
References
1.
2.
Accreditation ofUniversityChemical EngineeringCoursesA GuideforUniversity Departments, November 1996, Appendix 1, paragraph 10.1 (Institution of
Chemical Engineers, Rugby, UK).
Anon, 1999, Interactive Training Package No. 034, Hazop and Hazan and
Multi-stage HazardStudy(Institution ofChemical Engineers, Rugby, UK).
Pagc blank
in original
Contents
Preface
Forethoughts
1
xi
1.1
Introduction
1.2
A note on nomenclature
1.3
Legal requirements
2.1
9
9
20
26
27
34
37
41
47
50
51
54
56
56
61
61
62
63
63
64
65
vii
67
67
71
72
75
77
3.1
Objective
77
3.2
78
3.3
3.4
3.5
105
130
3.8
3.9
143
3.6
3.7
80
83
120
133
3.10 Afinalnote
143
148
152
4.1
Introduction
152
4.2
153
158
4.5
Theunforeseen hazards
Theassumptions
4.6
Data
160
4.7
162
4.8
Human reliability
Therecommendations
4,9
164
4.3
4.4
154
159
163
165
168
5.1
5.2
169
5.3
Popular objections
5.4
Theregulator's view
to Hazan
VIII
168
181
187
of
193
195
195
6.2
196
6.3
Confidence limits
196
6.4
197
6.5
6.6
199
203
Hazop
Hazan
203
7.1
7.2
198
207
Conclusions
213
214
218
Index
224
ix
Pago blank
in original
Forethoughts
editedby
xi
Hazard identification
and assessment
'The great end ofltfe is notknowledge but action.'
T.H. Huxley(18251895)
1.1 Introduction
The techniquesfor identifying hazards for finding out what hazards are
and the techniques for assessingthosehazards
present in a plantor process
for decidinghow far we ought to go in removingthe hazardsor protecting
are often confused. Figure 1.1 may help to make the
people from them
differences clear.
The left-hand side shows someofthe methods used for identifying hazards
and problems that makeoperation difficult.
Some hazards and problems are obvious. For example, if we manufacture
ethylene oxide by mixing oxygen and ethylene close to the explosive limitwe
do not need a special technique to tell us that if we get the proportions wrong
there maybe a big bang.
The traditional method of identifying hazards in use from the dawn of
was to build the plant and see what
technology until the present day
happens
'every dog is allowedone bite'. Until it bitessomeone, we can say
that we did not know it would. This is not a bad method when the size of an
incidentis limited but is no longersatisfactory now that we keep dogswhich
Methods ofassessinghazards
Obvious
Experience
Codes ofpractice
Hazard analysis
(Hazan)
Hazan
identifies hazards
Assesses hazards
Preferred technique:
Selective technique:
use when othersfail
Quantitative
Done by a team
Alsocalled:
'What if'?'
Risk analysis
Risk assessment
Probabilistic riskassessment(PRA)
Quantitative risk assessment(QRA)
the Hazan before the Hazop is carried out.In a Hazop the operability part is as
important as the hazard part. In most studies more operating problems are
identifiedthan hazards.
Hazop and Hazan are often confused, and Hazop is sometimes used to
describe any technique for identifying hazards. Figure 1.1 and Table 1.1
should makethe difference clear. However, ifsomeone asksyou to carry out a
Hazop or Hazan on a design,flrst makesure that the questioner is clear on the
difference betweenthem and is usingthe terms correctly.
The techniques described in later chapters are sophisticated techniques
which enablecompanies to use their resources more effectively. They assume
that the general level ofmanagement is competent, that the plantwill be operated and maintained in the mannerassumedby the design team and in accordance with good management and engineering practice. In particular they
assume that protective systemswill be tested regularly and repaired promptly
when necessary.
Ifthese assumptions arenot true then Hazop and Hazan are a wasteoftime.
It is no use identifying hazardsor estimating theirprobability if no-onewants
to do anything about them; it is no use installing trips and alarms ifno-one is
going to use or maintain them. The time spent on Hazopand Hazan would be
better spent on bringing the safety consciousness of employees and management up to standard. The following is a summary of a paper by Atallah and
Guzman on doingthis in developing countries4 (and perhaps elsewhere):
Be patientwhenyou are waitingfor data, prompt whenaskedfor advice.
Include in your team someone who speaksthe local language.
Submit your report in draft for comment; justify your criticisms and
recommendations.
Photograph problemareas.
Visit the plant at night.
Wearall therequired protective clothing and followall the safety rules.
Expectto be askedabout subjects not coveredin theremit.
Provide theclient with copiesofreferences, codes, and so on, not just a list
ofthem.
damage.
umbrellaI estimate that I will get wet 20 times/year; if I go out today without
an umbrellaI estimate that the probability that I will get wet is 0.3 (30%). Risk
is thus a measure ofthe likelihood of specific consequences.
A hazard may be serious but the risk from it may be small. For example,
experience over many years shows that in the UK, on average, less than one
person per yearhas beenkilled by the transport of flammable chemicals. The
risk of being killedin this way is therefore small, less than I in 60 million per
person per year, though the hazard, the potential for damage and injury, is
large.
Figure 1.3 shows the definition of risk adoptedby the EuropeanCommunity for use in risk assessment10.
The consequences of a hazard may be immediate or long-term. Thus fires
and explosions and some toxicchemicals such as chlorine produceimmediate
injuries. Other chemicals such as asbestos produce ill effects only after many
years have passed. Ultraviolet radiation produces both immediate effects
(sunburn) and long-term effects(skincancer).
Some consequences are deterministic (that is, theyalwaysfollowexposure)
while others are probabilistic (that is, they may or may not follow). For
example. ifan objectis dropped from the top ofa structure it will alwaysfall to
the ground (deterministic) but the effects are probabilistic. It may kill
someone, may cause serious injury, may cause slight injury or may merely
Operation
Hazard analysis
tChemE
Thisbook
Risk assessment
IChemE
Identification of
hazards
Estimation of
how often
Estimation of
consequences
Comparison with a
criterion and a
decisiononaction
RISK
related
to the
is a
considered function
hazard
of
SEVERITY
ofthe
frequencyand durationofexposure
possible
harm that
and
canresult
fromthe
__________________________________________
considered
hazard
possibility of avoidingorlimitingtheharm
S
In theUK thelaw requires all employers to carry out a five-step risk assessment1 1,12:
(1)
(2)
(3)
(4)
(5)
Referencesin Chapter 1
2.
and Accidents Recur (Institution of Chemical Engineers, Rugby. UK, and Gulf
Publishing Company, Houston, Texas, USA).
7
Hazard and
operability studies
(Hazop)
'Since the destructionof/he Temple, the gift
ofprophecyhasbeendeniedtoprophetsand
bestoweduponscholars.'
RabbiEudemusof Haifa
'Thereis a way of gomg about one'swork in
chemicalengineeringmore certain and less
expensive than the time-honoured processof
trialanderror.'
George E. Davis34
21 What is a Hazop?
missed it is done in a systematic way, and each pipeline and each sort of
hazard is considered in turn. The study is carried out by a team so that the
members can stimulate each other and build uponeach other's ideas.
A pipeline for this purpose is one joining two main plant items for
example, we might start with the line leading from the feed tank through the
feed pumpto the first feed heater. A seriesofguide words are appliedto this
line in turn. The words are:
NONE
MORE OF
LESS OF
PART OF
MORE THAN (or AS WELLAS)
OTHERTHAN
Deviations
NONE
MORE OF
LESSOF
ie, no flow or
PART OF
MORE THAN
OTHER THAN
l0
11
When all the lines leading into a vessel havebeen studied, the guide word
OTHER THAN is applied to the vessel. It is not essential to apply the other
guide wordsto this item as any problems shouldcome to light when the inlet
and exit lines are studied. However, to reduce the chance that something is
missed,the guide words should be applied to any operation carriedout in the
vessel. For example, if settling takes place we ask if it is possible to have no
settling, reverse settling (that is, mixing), more settling or less settling, and
similarly for stirring, heating, cooling and any other operations (see Section
2.8.4,page 50).
Some team leadersuse 'Relief' as a backup guide word (see Section 2.11,
page 54).
Pay special attention to intermediate storage vessels. As arule,no change is
supposed to take placethereexcept emptying orfillingbutchangesintemperature orcomposition maytake place,particularly whenthe contentsare allowed
12
can be seen at a glance? If ball valves or cocks are used, can the handles be
fittedin the wrongposition?
Are spectacle plates installed whenever regular slip-plating (blinding)of a
joint (for maintenance or to prevent contamination) is foreseen?
Access is normally considered later in design, when a model of the plant
(real or on computer) is available, but the Hazopteam should note any points
that needspecialattention for example, valves that will haveto be operated
frequently or in an emergency, and shouldtherefore be easy to reach.
Ozog'7 describesa variation of the normal Hazop procedure in which the
guide wordsare appliedto equipment(including pumps) insteadoflines.
Start-up, shutdown and other abnormal conditions such as catalystregenerationshouldbe considered during Hazop as well as normal operation.
Table 2.2 (pages 1415) describes in detail the results of a Hazop on the
part of the design shown in Figure 2.2. More details are given in Section 2.5,
page 34. The procedure will become cleareras you go through each item in the
table in turn. To get the mostout ofTable 2.2, display Figure2.2 (pages 16-l7)
on a screen in front of the team, or give copies to each member, and ask
everyone to carry out a Hazop on it, with the discussion leaderacting as team
leader. The results can then be compared with those in Table 2.2.
However, do not considerTable2.2 to be the correctanswer. Those taking
part in the discussion mayfeel that the authors ofTable2.2 went toofar, or did
not go far enough, and they could be right.
Table2.2 was basedon a real study of an actualdesign. It is not a synthetic
exercise, but it is written upin more detail thanessential in a real life situation.
The use ofHazop is widespread and in the oil and chemical industries most
companies now say that all new designs are Hazoped or examined in a similar
way.However, becauseofthe work involved many old plantshavenever been
Hazoped. If they have beenextensively modified, as most have, then a Hazop
is well worth while.On an old refinery 17 Hazops over seven years resulted in
13
Table 2.2 Results ofHazopof proposed olefin/dimerization unit: line section from
intermediate storage to buffer/settling tank
(From Reference 5. Reproduced withpermission ofthe American Institute of
Chemical Engineersand Dr HG. Lawley. Copyright 1974 AIChE.
All rightsreserved.)
Guide word
Deviation
Pos,sible causes
NONE
No low
ii
MORE OF
More flow
More pressure
LESS OF
PART OF
More temperature
Less flow
Lesstemperature
(10) Winterconditions
Highwater
concentration in
Streaiii
High
concentration of
lower alkanes or
alkenesinstream
MORE THAN
Organic acids
present
(13) As for(12)
OTHER
Maintenance
14
Consequences
Actionrequired
conditions.
Asbr(I)
Coveredby (hi
storage operator
Asfor(
Coveredby (h)
Ic) Install kickbackon pumps
(d)CheckdesignofJI pumpstrainers
Asfor(I)
highway
Coveredby (b)
(e) Institute regularpatrolling and inspectionof
transfer line
Settlingtank overfills
it pumpoverheats
Hydrocarbon dischargedinto area adjacentto public
it
pressure
Line fractureorflangelead
if
temperature at intermediatestorage.
Ifnot,install.
Higher systcnlpressure
Increased rate
drain line
Line cannothe completelydrainedorpurged
ofconstruction
Jl transferpumps
Todrain
Hydrocarbonfrom
interiiediatestorage
Drain and N2purge
DON'T CHARGE A
CHARGEMORE A
CHARGELESS A
CHARGEAS WELLAS A
CHARGEPARTOF A (if A is a mixture)
CHARGEOTHER THAN A
16
From reactor
200CC
260psig
20C
300psig
I60C
290 psig
and reactor
To after-cooler
Drain andN2purge
REVERSE CHARGEA (that is, can flow occur from the reactor to the A
container?) This canbe the most serious deviation (seeSection A2.1, page61)
A IS ADDED EARLY
A IS ADDED LATE
A IS ADDED TOO QUICKLY
A IS ADDED TOO SLOWLY
Table 2.3 Some resultsofa Hazopofa batch process: making a cup of tea
The instructions studied (from apacketofone-cup tea bags) are givenopposite.
Note that someinstructions are implied for example, put water in kettle.
For more detailedinstructions see British Standard 6008and ISO 3103.
Step
Guide word
Deviation
NONE
No watercollected in kettle
All
NONE
No understanding
MORE OF
LESS OF
Temperature
TOO LATE
MORE OF
Pressure
18
'one-cup' tea bags and a few ofthe points that mightcomeout ofa Hazop. It is
easy for a team withoutpracticalexperience to cometo the conclusion that the
process is so hazardous and the result so uncertain that the task should not be
attempted.
2
3
4
5
6
Stir immediately
Leavefor 35 minutes depending on strength preferred
Pressthe bag against the side ofthe cup with a spoon and remove
Possiblecauses
Consequences
Action required
I No water supply
2 Tap fails closed
No tea
No tea or poortea
use in emergency
h Print instructions in
other languages
understand English
5 Wateris belowboiling
point
6 Distraction
Spillage
g Use timer
h Train operator to steady
cup withother hand
19
of
or
should
be
studied
in
example,conditioning equipment catalystchange
a similar way by listing the sequence of operations and applying the guide
Oncomputer-controlled plants the instructions to the computer (the applicationssoftware) should be studied as well as the line diagrams. For example,
if the computer is instructed to take a certainaction whena temperature rises,
the team considers the possible consequences of this action as well as the
consequences of the computer failing to take action. On a batch plant the
consequences may be different at each stage of the batch. On a continuous
plant the consequences may be different during start-up, shutdown, catalyst
regeneration, and so on.
The appendix to this chapter (see Section A2.6, page 65) describes a
dangerous incident that occurred because the design and operating teams
assumed that the computer would always take care of alarmsituations and did
not considerin detail the consequences ofeach actionat each stage.
Research chemist
Responsible
of alarms and
Independentteam leader
preventing tragedy'
Another featureofgoodteam members is a mental ragbag ofbits and pieces
of knowledge that they have built up over theyears. Such peoplemay be able
to recall that a situation similar to that under discussion caused an incident
elsewhere. They need not remember the details so long as they can alert the
team to possibilities that should be considered and perhaps investigated
further. For an example, see Section A2.7, page67.
Note that the team members, except for the team leader, are experts on the
process. They will,by this stage,havebeenimmersed in it for between one and
two years. Hazop is not a technique for bringing fresh minds to work on a
problem. It is a technique for allowing thoseexpertin the process to bring their
knowledge and experience to bear systematically, so that problems are less
likely to be missed.
The complexity of modern plantsmakeit difficult orimpossible to see what
mightgo wrongunless we go through the design systematically. Fewaccidents
occur becausethe design team members lack knowledge; most errors in design
occur because they fail to apply their knowledge. Hazop givesthem an opportunity to go through the design line by line, deviation by deviation, to see what
they havemissed.
The team should have the authority to agreemost changes there and then.
Progress is slow if every change has to be referred to someone who is not
present. The team members shouldtry toavoid sending deputies. They lack the
knowledge of previous meetings and might not havethe authority to approve
changes: as a result progress is held up. Somepeoplehave told me that this is
impracticable in their companies as all changeshave to be approved at a high
level. This does not matter so long as the team members feel confident that
most of their recommendations will he accepted withoutargument. However,
ifthediscussions in theHazop meetings haveto be gone through again, time is
wasted. In addition, the team mayhe temptedto add somefat so that the boss
23
has something to remove. But he may not know the fat from the meat.
I haveknown somepeople saythat thejob of the Flazop team is to identify
problems and that finding solutionsshould be left to the project team. If the
Hazop team is made up as I havesuggested, experience shows that it can find
solutions to most problems, withoutthe needforanothermeeting with many of
the same people present. However, some problems may have to left until
expertadvice has been obtained.
The team leaderoften acts as secretary as well as safety department representative. He writes up his notes after the meetingand circulates them before
the next meeting. As alreadystated,it is not necessary to write them up in the
degree of detail shown in Table 2.2 (pages 1415). Figure 2.3 shows a
suggested form for the first few actions agreed in Table 2.2. However, the
tendency today is to write up the notes in more detail than in the past, in the
style of Table 2.2 rather than that of Figure 2.3, so that the company can
demonstrate, if necessary, that it has done everything reasonably possible to
identify the hazards.
Some companies consider that all Hazops should be written up in great
detail. If the design is queried in the future, the Hazop records can be
consulted. There is some force in the argument but the extrawork is considerable and, in practice, most Hazopreports are rarely,if ever, consultedoncethe
plant is on line.
A numberofcomputerprograms are nowavailable forrecordingthe results
ofHazopstudiesas they arise. Copiesof the actionsagreed and thereasonsfor
them are available immediately after the meeting, without rewriting or
retyping. The display can be projected onto a largescreen, so that all the team
members can see it and can confirmthat they agree with the decisions. The
programs also remind the team of the deviations to be considered and their
usual causes. A survey in 1995 in the UK showed that about half the compaflies questioned were using computerized recording and the number is
growing. Table2.4 (page 26) shows some ofthe factors to be considered when
choosinga program. Turney32 says that these programs produce more effective meetings, more accurate action lists (and thus quicker action) and fewer
misunderstandings (seealso Section 2.6, page37).
A few weeks after the Hazop the team leadershouldcall the team together,
check on progress made and recirculate the report form (Figure 2.3) with the
'Follow-up'column completed.
Although Hazop is a valuable technique, no-one jumps out of bed on a
Monday morning shouting, 'Hooray! I've got a Hazop today!'. The need to
consider every deviation on every line can becometedious. Bewareof making
it more so by bureaucratic procedures such as insistingon excessiverecording
24
Study title:
Project No
Sheet 1 of
Line Diagram Nos
Date
No flow
Action Follow-up
review
by
comments
CM
IDE
DE
DE
Instituteregularpatrolling and
inspection oftransferline
CM
IDE
PE
CM
DE
or discussing everything twice (or three times) in the Hazop meeting and
afterwards with the boss or the projectteam.There is a net loss if in our eagerness to document everything and explain it to everybody we discover less
information worth documenting. If Hazop and similarsystemsare not acceptable to creativeminds, they will never succeed.
A Hazop usually takes 1.53 hours per main plant item (still, furnace,
reactor, heater, and so on). ifthe plantis similarto an existingone it will take
1 .5 hours per item but ifthe process is new it maytake 3 hours per item. inexperienced teams, of course,take longerthan experienced ones. References 40
and 41 describemore sophisticated methods ofestimating the time required.
Meetings are usually restricted to 3 hours, 2 or 3 days perweek,to givethe
team time to attend to their other dutiesand becausethe imagination tiresafter
3 hours at a stretch. If the members of the team have to be gathered from a
distance, longer periods of working, perhapsevery morning for a week, may
haveto be accepted. Resist any temptation to work 8 or more hours per day for
a week, as attention inevitably Ilags. It is the results of a Hazopthat are important, not the numberof hours spenton it.
The Hazop on a large project may take several months, even with two or
three teams working in parallel on different sections of the plant. It is thus
necessary to either:
(a) Holdup detaileddesign and construction until the Hazopis complete; or
(b) Allow detailed design and construction to go ahead and risk having to
modify the detailed design or evenalter the plantwhen the results of the Hazop
are known.
Ideally, the design should be plannedto allow time for(a) but ifcompletion
is urgent(b) may haveto be accepted.
Section 2.7 (page 41) suggests that a preliminary Hazop is carriedout on the
flowsheet beforedetaileddesign starts. This will take much less time than the
Hazop ofthe line diagrams.
Investigations of Hazop by a combined industry/university team showed
that time spenton explanation at the startofaHazopreducedthe time spenton
the Hazop itself. They also found that interesting or difficult cases can take
excessive time and that inexperienced teams tend to be too rigid in their
approach and that this causes delay. For example, teams usually discuss the
possible causes ofa deviation beforetheydiscuss the consequences, as ifthere
is no possible cause the consequences do not matter. However, experienced
teams are flexible and sometimes find it better to discuss the consequences
tirst
it is possible for a team to get carried away by enthusiasm and install expensiveequipmentto guard against unlikely hazards. The team leadercan counter
27
in Reference 3.
Title:
Reg. No.:
Operating methods
start-up
routineoperation
shutdown
preparation for maintenance
abnormal operation
emergency operation
layoutand positioning ofcontrols and
instruments
Engineeringmethods
trip and alarm testing
maintenance procedures
inspection
portableequipment
controllogic
Safety equipment
fire-fighting and detection systems
meansofescape
safetyequipment for personnel
Environmental conditions
liquid effluent
solid effluent
gaseouseffluent
noise
Engineeringhardwareand design
line diagram
wiringdiagram
plant layout
designpressure
designtemperature
materials ofconstruction
loads on, or strength of:
foundations, structures, vessels
pipework/supports/bellows
temporary orpermanent:
pipeworklsupports/bellows
valves, slip-plates
restriction plates, filters
instrumentation and controlsystems
tripsand alarms
staticelectricity
lightning protection
radioactivity
rate ofcorrosion
rate oferosion
isolation for maintenance
mechanical-electrical
fire protection ofcables
handrails
ladders
platforms
walkways
trippinghazard
access for:
operation, maintenance, vehicles,
plant, fire-fighting
underground/overhead:
services
equipment
(Continued overleaf)
31
Reliefand blowdown
(I) Introduceoralterany potential cause of
over/underpressuringthe systemorpart ofit?
(2) tntroduceor alter any potential cause of
higheror lower temperaturein the system or
part of it?
(3) Introduce a risk ofcreating a vacuum in the
systenior part of it?
(4) In any way affect equipmentalready
installed for the purpose ofpreventing or
mininli/ing over or under pressure?
Area classification
(5) Introduceor alter the locationofpotential
leaks of flammablematerial?
(6) Alter the chemical compositionor the
physicalpropertiesofthe process material?
i7) Introduce ness or alter existing electrical
equipment?
Safety equipment
(8) Require the provisionofadditionalsafety
equipment?
(9) Affect existing safety equipment?
Operationand design
(10) Introduce new oralterexisting hardware?
32
Date
Plant Manager
Checked by
ngineer
a time when companies are reducing manning and the over-tfties are looked
upon as expenses to be eliminated rather than assets in which thirty years'
salary has been invested. Seniormanagers should systematically assess, from
time to time, the levels of knowledge and experience needed and ensure that
they are maintained. This is an area where systematic methods have not been
applied as thoroughly as elsewhere. In the UK the Health and Safety Executive
has recently instructed a major company to set up a formal systemfor controlling changesto its organization.
2.4.5 'Do it for us'
to say to a design contractor, 'We are understaffed and you are the experts, so why don't you do the Hazop for us?'23.
The client should be involved as well as the contractor becausethe client
will have to operate the plant. The Hazop gives the client's staff an understanding of the reasons for various design featuresand helps them write the
operating instructions. Even if the client's staff know little to start with about
the problems specific to the particular process, they will be able to apply
general chemical engineering and scientific knowledge as well as common
Companies have been known
design. 'Don't bother me now. We'll be having a Hazop later on. Let's talk
about it then'.
33
1-IAZOP
AN!) HAZAN
2.4.9 Relevance
Thoughthe members ofa Hazop team havethe necessary knowledge they may
fail to see its relevance. Thus, they may not realize that an open vent on a
vessel is, in effect, a relief valveand should be treated with the same respect.
Its size should not be altered unless we have gone though the same procedure
as we would gothrough before changing the sizeofa reliefvalve, and it should
be registered for regular inspection. Another example: chimneys are commonplace, we all know how they work,but we sometimes fail to recognize that an
open drain and an open venton the same unit may producean upward flow of
air, in effecta chimney48.
Leathley and Nicholls suggest that presenting case studies(sometimes from
unrelated industries) beforea Hazopcan widenthe team's view ofwhat might
happen and encourage widerthinking49.
Table 2.2 (pages1415) gives the results of a Hazop on the plant shown in
Figure 2.2 (pages 1617). It shows the feed section of a proposed olefin
dimerization unit and details are as follows.
34
(I) Right at the start we see that the first two actions required are a software
one and a hardwareone, thus emphasizing that Hazop is not just concerned
with the hardware. This flrstitem brought thecommissioning manager'sattention to the fact that his raw material came from a storage area 1 km away
controlled by a different manager and operators who did not haveto cope with
the results of a loss offeed. Whosejob was it to monitorthe stock and see that
it did not run out?Although the storage operatorwas Ofl the job, the plantoperators had more incentive as theyhad to deal with the consequences ifthe stock
runs out.
Note that a deviation in one line may produce consequences elsewhere in
the plant. Thus 'no flow' in the line we are studying in this example may have
effectsfurtheron in the plant,in the line leadingto the reactor, where'no flow'
may result in higher temperatures and the formation of polymer. In a batch
process a deviation at one stage may have consequences at a later stage (see
Section A2.9. page 71).
(I )(b) A low flow alarmmight be installed instead ofa low level alarmbut it is
better to measure directly what we want to know, and the low level alarm is
cheaper.
(3)(c) Note that a kick-back line is shown after pumpJ2 on the next line to be
studied. A kick-back is cheaperthan a high-temperature trip and requires less
maintenance. Students shouldbe reminded that the lifetimecost of an instrument is about twice the capital cost (afterdiscounting) if testing and maintenance are included. Instruments (and computers) cost twice what you think
35
to passgas or liquid.
(5)(g) Locking-off the bypass makes it harderto open it quickly if the control
valve fails shut. Do we need a bypass?How often will the control valve fail
shut?
(5)(h) The team members might havedecided that they wished to increase the
sizeofthe buffer/settling tank, originally sufficient for 20 minutes settling time
but reducedby the action proposed. If so, theymighthavefound that it was too
late to do so asthe vessel was on the critical pathandhad already beenordered.
Section 2.7 (page 41) recommends a preliminary Hazop on the flowsheet at a
time whensuch changes can be made.
(6) This item introduces students
met before.
Note that we often have more than one chance to pick up a hazard. When
discussing no flow' item (3)] the team members realizedthat line blockage
would cause a rise in pressure but they decided to leave discussion of the
consequences until they cameto the deviation 'more pressure'. If theyhad not
realized, when discussing item (3), that line blockage could cause a rise in
pressure, then they had another opportunity to do so later. Sections 2.8.4and
A2.8 (pages 50 and 67) describe other examples.
(9) Somedrainsin Figure 2.2 are shown blanked, others not. All drainsshould
be blanked unless used regularly by the process team.
man in another plant who may not realizeits importance and does not haveto
handlethe consequences if the water goes forward.
An automatic controllerto remove water, operatedby the interface level
indicator, is not recommended as ifit failsoil will flow to drain and maynot be
detected.
(1 2)Havethe distillation columns
available for recordingthe results ofstudies(seeSection 2.2. page 24), and the
programs can also remind teams of the possible causes of various deviations
and possible remedies so that they are less likely to overlook them.Thus ifthe
team is considering no flow' in a pipeline, the computercan remind them that
possible causes are an empty suction vessel, a pump failure (which in turn
could he due to failure of the power supply, the motor, the coupling or the
pump itself), a blockage, a closed valve, a slip-plate, a broken pipe or high
37
pressure in the delivery vessel. Pitt et a150 have devised a procedure for calcu-
lating the effects of deviations. However, these programs are not what people
mean when they ask the question about computers and a Hazop. They are
asking if the computer could examinethe line diagram,say what deviations
can occur,and why, and suggest changes to the design or method ofoperation,
perhaps using an expert system. Before answering this question, two points
shouldbe considered.
The first is that Hazop is a creativeexercise and those who are best at it are
people who can let their minds go free and think of all the possible ways in
whichdeviations might occur and possible methods of prevention and control
(seeSection 2.2, page 20). To quote from a bookon artificial intelligence25:
these sort of techniques ... ,nay eventual/vproduce machines with a
capacityfor manipulating logical rules that will match, or even exceed, our
own. But logic is just one aspect ofhuman intelligence, andone whose importance can easily be overrated. For ...ftictorssuch as intuition andflairpay a
very large part in our thinking, even in areas like science wherelogic ostensihls'reigns supreme. For example, most ofthe scientists whohave recounted
how they came to make an important discovery or to achieve a significant
breakthrough have stressed that when they fbund the answer to the crucial
problem they intuitively recognised it to he right and on/v subsequent/v went
back and worked out why it was right.'
The secondpoint is that the knowledge used in a Hazopis 'broad and deep'
while expert systems are suitableonly for 'narrow and deep' knowledge26.
The knowledge used in a Hazop canbe divided into fourtypes26 (seeFigure
2.5). The following examples of each type are taken from the Hazop of the
dimerization plant described in Section 2.5:
Plant-specific knowledge
Plant specific
Generalprocess engineering
Generalscientific
For example:if a line is broken, the contents will leakout; the men whohaveto
cope with the effects of plant upsets are more likely than other men to take
actionto prevent them;a mancannot hearthe telephone ifhe is out ofearshot.
The difficulties here are greater still and may be beyond the power of any
expert system. To quote from Reference 25 again:
39
interactions between people. So hazardteams are unlikely to become redundant in the foreseeable future.
So far there has been little industrial experience of these techniques, but
industry has been involved in their development.
Duringa Hazop study, particularly whenthe technology is new to the team,
someone often half-remembers a hazard. It would be useful to be able to call
up detailsof hazards, of accidents which they have caused and of the actions
recommended to prevent a recurrence. Although computerized databases are
available they suffer from a common weakness: they are eithergo or no-go
that is, they find a precise match with the chosen keywords or they do not. To
overcome this Chung et al are developing a fuzzy search tool which uses
case-based reasoning. The key words are arranged in hierarchies resembling
family trees. If the program cannot find a precise match it looks for matches
with the parents or siblings of the keywords and, ifthat is unsuccessful, with
more distant relatives. For example, suppose we wish to find information on
the road transportof sulphuric acid. If no match can befound,the program will
look for matches with the rail transport of sulphuric acid, with its transportby
any means, or with the road transport(or just transport) of other acids, or for
their storage. If these searches fail it mightlook for the transportofcorrosive
chemicals or their storage.
The program is not intended merely, or even primarily, for use by Hazop
teams.It could be used by designers, by anyone lookingfor information and, in
a somewhat different form, by process operators. In this case information
would be displayed automatically when hazardous conditions are
approached5355.
operating problems to light at a time whenthey can be put right with an indiarubberrather than a welding set, but at a time when it is too late to makefundamental changes in design.
For example, referring to Section 2.5, note (12) (page 37), the Hazop might
bring to light the fact that the concentration of light ends mightvary markedly
from design and that the still shouldbe redesigned to allow for this. It is probably too late to do this; the still may havealready been ordered. Section 2.5,
note (5)(h) (page 36), contains another example: by the time of the Hazop it
mayhavebeen too late to increase the size of the settling tank.
Such problemscan be picked up earlier if a preliminary or coarse-scale'
Hazop is carried out on the Ilowsheet before it is passed to the engineering
department for detailed design, a year or more before the line diagrams are
available. Like a normal Hazop it can be applied to continuous and batch
plants.
These arejust a few of the 66 points that came up during three three-hour
meetings. Many of the points would have come up in any case but withouta
Hazopmany might have been missedor might not have come up until it was
too late to change the design.
While the results of several line diagram Hazops have been described in
detail (see the list at end of Section 2.5, page 37), very few fiowsheet Hazops
havebeendescribed in the same way.Table2.6 (pages 44-45) lists someofthe
42
Deviation
Consequences
Higherreactortemperature
Lowerreactortemperature
Pooror no reaction;
poor quality product
*,
LOWER
No flow
LESS (ethylene)
Less flow
Levelbuild-upin reactor
System upset;
productquality affected;
system shutdown
MORE (initiator)
More flow
More polymerization;
possibility of runaway
conditions; productquality
offspecification
LESS(initiator)
Less flow
Less polymerization;
reactortemperature
imbalance affects
downstream equipment such
as heat exchangers
Causes
Recommendedactions
Provide temperaturecontrol
Provide high temperature sensor/alarm
Coolanttemperature low
Meltpump I fails
Make-uporrecyclecompressor
failure
Initiator pumpmalfunction
AVOID (the
need). Table 2.7 (from Reference II) is an extractfrom an early
criticalexamination of a flowsheet.
Even a coarse-scale Hazop is too late for some major changes in plant
design. A similartypeofstudyis needed at the conceptual or business analysis
stage when we decide which product to make, by what route and where to
locate the plant. For example, at Bhopal in 1984 an intermediate, methyl
isocyanate (MIC), leakedout ofa large continuous plant and killedover 2000
people. Ifthe same raw materials are allowed to react in a different order, no
MICis produced. It is too late to suggest at the flowsheet stage that the orderof
reaction, on a continuous plant, should be changed. That decision has to be
made rightat the beginning of the design process (seealso Section A2.2. page
62).
Alternatively, ifwe use the MICroute we can reduceor eliminate the intermediatestock and use the MIC as soon as it is formed. The decision to do so
can he made at any time, even when the plant is on line, but money will be
saved ifthe decision is made early in design.
Table 2.7 An extractfrom the critical examination ofa flowsheet showing the
generation ofalternatives by successive questioning
(From Reference II)
Statement: Designa distillationcolumn
Successive questionsand answers
Alternativeideas generated
Why?
To separate A Irom B.
Why?
Because the recycle reactor won't crack
A mixed with B.
make B.
Why?
Because the furnace temperature isn't
highenough.
Why?
Because tube materials won'tstand a
highertemperature.
46
A clever man has been described as one who finds ways out of an
unpleasant situation into which a wise man would never have got himself.
Wise men cany out safety studiesearly in design.
Of course, every company carriesout many studies beforeembarking on a
design. What is lacking in most companies at the conceptual and flowsheet
stages of projects, however, is the systematic. formal, structured examination
which is characteristic of a Hazop. The normal Hazop questions are not
suitable at the conceptual stage but Chapter 10 of Reference 15 suggests some
alternatives. It also gives many examples of hazards that have been or could be
reduced or avoided by Hazop type studiesat the conceptualor tlowsheet stages.
A nuisance during a conventional Hazop is the person who asksif the right
product is being made in the right way at the rightplace. It is by then far too
late to ask such questions.If the person asks them then, perhaps there was no
opportunity to ask themearlier.
or repeat units
where people feel that the full treatment is unnecessary. 'It is only a storage
project and we have done many of these before!' 'It is only a pipeline and a
coupleofpumps.' 'It is only a service system.'
47
Restriction
Ifdesigners talk like this, suggest they try a Hazopand see what comes out
of it. Afterthefirstmeetingor two they usually want to continue.
Figure 2.7 shows part of a line diagram on which the design team was
Figure 2.8 When the automatic valve closed, the pump was overpressured
A Hazop had been carried out on the plant,but this section was not studied
as it was 'only an off-plot' a tank,apumpanda few valves too simplefor
any hazardsto pass unnoticed, or so it was thought. Consideration of 'reverse
flow' through the kick-back line (or 'more of pressure' in the filling line)
would have disclosed the hazard.
Afterthe incident the kick-back line was rerouted backto the tank.
2.8.3 Servicesystems
All service lines (including steam, water, compressed air, nitrogenand drain
lines) should be 'Hazoped'as well as process lines (seeSections A2.3 and A2.5,
pages 63 and 64). Pearson16 lists some of the questions which arise during
49
1-IAZOP
AND HAZAN
Powersupply
LZ Highlevel trip
LC Level controller
design engineer, responsible for controlling the cost, was opposed: this, he
said, would be gold-plating. A simple calculation (see Section 3.5 on page 105
for an explanation ofthe terms used)helpedto resolve the conflict.
The trip will have a fail-danger rate of about once in two years. With
monthly testing the fractional deadtime will be 0.02.
The demand rate results from the failureofthe level controller. Experience
shows that a typical figure is onceevery two years or 0.5/year. A hazard will
therefore occur once in 100 yearsor, more precisely, thereis a I in 100 chance
that it will occur in any one yearor a I in 10 chancethat it will occurduringthe
10-year life ofthe plant. Everyone agreed that this was too high.
They also saw that therewas more thanone way of reducing the hazard rate.
They could improvethe control system and reduce the demand rate, or they
could improve the trip system and reducethe fractional dead time. It may not
he necessary to duplicate all the trip system; it may be sufficient to duplicate
the trip initiator.
If thehazardunder discussion is a runaway reaction, then quantihcation is
more difficult. A key question to ask, according to Stoessel57,is, 'If cooling is
lost, howlong do we havebeforea runaway occurs?' Iflthe time is less than20
minutes, automatic protection is probably necessary. It maybe necessary for a
longertimescale if the operatorcoversmany units.Another key question is, 'If
a reaction mixture is left standing, and the cooling cannotpreventa runaway,
howlong do we have beforea runaway occurs?' Ifthe time is less thana day an
alarm or automatic protection maybe necessary.
Hazop was pioneered in the chemical industry (see Chapter7) and soonspread
identifying ways in whichcontamination could occur rather than other operatingand safetyproblems. This section discussessomeother applications.
In considering whether or not Hazop could be applied in a new context,
remember that Hazop grew out of critical examination (see Sections 2.7 and
7.1, pages 45 and 203) and that the original form ofthe techniquemay he more
suitable than the modification (Hazop) developed to meet the process industry's needs.
Hazop has beenappliedto laboratory design1 and to laboratory operations.
One study of a new operation disclosedthe fact that the chemists intended to
convey cylinders of hydrogen cyanide to the top floor in the lift!
51
Hazop has alsobeen applied to the manufacture of a product usinggenetically modified organisms (GMOs)28. A modification of Hazop known as
GENHAZ has been proposed for identifying ways in which GMOs might
affect the environment29.Table2.8 isan extract from a hypothetical GENHAZ
study: the proposed experimental insertion into potatoesof an imaginary gene
(TP) that is toxic to a specific caterpillar. The studyraises questions for investigation; theycannothe answered on the spot.
2.10.1 Mechanicalhazards
Knowlton2 has describedthe application of Hazopto some mechanical problems. For example, a sterilization autoclave had to be loaded with a stack of
trays using a fork-lifttruck. Application of the deviation 'more of' disclosed
that if the driver moved the load too far forwardit could damage the rear wall
of theautoclave. Application of thedeviation 'as well as' disclosed that if the
driver raised the load it could damage an instrument that measured the
humidity and perhaps also damage the roof.
Similarly, too rapid operation could cause spillage and led the team to ask
how spillages would be handled.
2.10.2 Nuclear power
The nuclear power industry was slow to adopt Hazop, preferring instead a
technique known as failure mode and effect analysis (FMEA) (see Section
2.11. page 54).
In Hazop we startwith adeviation and ask howit might occur.For example,
'more offlow' in a pipeline mightbe caused by the failure of aflow controller.
There will probably be other possible causes as well (see Table 2.2, pages
1415).In FMEAwestart with a component andwork out the consequences of
failure. If we start with the flow controller, one of the consequences of its
failure may be too high a flow in a pipeline. There will probably be other
consequences as well.
in the line diagram sense, the essentials of a nuclearreactor arerelatively
simple: a hot coreheats water.In this senseit is much simpler than the average
chemical plant. On the other hand, the nuclear reactor contains far more
protective equipment to prevent it getting out of control and to commission
emergency cooling systems, and so on. The obvious first approach of the
nuclearengineers was therefore to ask, 'What will happen if a component of
the protective systems fails'?' and then examine each component in turn.
However, the cooling systems(normal and stand-by) and service lines on
nuclear power stations would benefit from Hazopand this is now recognized.
52
WHEREELSE
Deviation:
Consequences:
Causes:
Actions:
Deviation:
Consequences:
Causes:
Actions:
53
water,pumps,andso on. Some 'What if' analyses are more detailed. They ask,
for example,for each pipeline, what will be the result of more or less flow,
temperature, pressure, and so on. If we also ask, as we obviously should, if
these deviations are possible, then we havegot a Hazop.
Fault trees(Section 3.5.9, page 113), mainlyused as a method ofestimating
theprobability ofan event, have sometimes beenrecommended for identifying
hazards. A fault tree is the reverse of an FMEA. In FMEA we start with a
component failure and deduce possible results. In a fault tree we start with a
top event' such as a fire or explosion and work back to find the errors and
component failures that could lead to it. Its weakness as a method of identification is that we may not realize that certain top eventscan occur and therefore
not look for the routes to them.Fault treestell us howtop eventsoccurbut not
what top eventscanoccur.
Auditsand inspectionsare a necessary complement to Hazops because they
can tell us whether or not the plant is built, operated and maintained in accordance with the design assumptions. They are particularly necessary during and
after construction as the failure of construction teams to follow the design in
detail or to follow good engineering practicewhen details are left to them is a
major cause of incidents65. A weakness of many auditsis that they check that
methods of working are sound and are followed but do not check that all the
hazards havebeen identified. Turney and Roff havedescribed a 'process hazards
review (PHR), a mixture of What if' and check-lists, which is designed to
overcome this. Many past incidents were studied to identify possible hazards.
Unlike many ofthe techniquesdescribed in the literature, over a hundred studies
had been carried out by the time their paper was published66.
Auditors are not policemen. Theirjob is to spot the hazards, physical and
procedural, that the plantstaff have missed through lack of specialized knowledge, shortage oftime or overfamiliarity.
STOPHAZ is a group of computer programs designed to bring hazards to
theattention ofdesigners at an early stage and thus reduce thenumberofproblenis that are not discovered until a Hazop is carried out late in design67. It
includesAuto-HAZID, describedin Section 2.6, page 37.
Several attempts have been made to comparethe effectiveness of various
identification techniques. According to Turney and Pitblado, a study of past
incidentsshowed that Hazop could haveprevented 29% ofthe design incidents
and 6% of the operational incidents, a higherproportion than any other technique.Reviewsof human factors could haveprevented 24% ofthe operational
incidents68.
one continTaylor69 has describedan experiment in whichtwo designs
one
batch
were
each
studied
in
various
found
80% ofthe
uous,
ways. Hazop
55
faults on the continuous plant but only 22% of those on the batch plant.
However, his batch Hazop did not include consideration of the deviations
listed in Section 2.1.1 (page 16) which were considered under action error
analysis' rather than Hazop. Most of the other faults were detected during
commissioning and were not spotted during the Hazop becausethe team did
not havethe necessary knowledge. As stated in Section 2.4.4(page 30), Hazop
is no substitutefor knowledge and experience and its effectiveness depends on
theknowledge and experience ofthe team.Accordingto Skelton, even inexperienced teams, such as students on a Hazop course, find about 80% of the
hazardsand those missed are mainly minor72.
As the use of any technique becomes more widespread its quality is liable to
decrease. There is therefore aneedto beable to auditthe qualityof a Hazop. At
a workshop on Hazop held in 1995 theauditing of Hazop was selectedas the
most pressing current topic70. The best method of auditing is to sit in on a
because the Hazop is complete
Hazop. Ifthat is not practicable forexample,
Rushton71 has described an audit scheme. The auditor samples the documentation produced by theHazopand looks for evidencethat various modes of
operation such as start-up, shutdown and maintenance havebeenconsidered in addition to normal operation, that the knowledge and experience ofthe
team members were adequate, that the same people attended throughout and
did so regularly, that the recommendations made were carriedout,andthat any
late changes in design were studied. If the plant has already been commissioned the auditorshould examine the problems that have arisen and see ifthey
could reasonably havebeen spotted duringthe Hazop. Altogetherthere are six
pages of suggested questions. The auditorshouldtalk to the team members to
gaintheir impressions and assesstheir knowledge and experience.
2.13 Conclusion
Carling30 has described the effects of using Hazop in his company. The bene-
fits went far beyond a simple list of recommendations for a safer plant. The
interaction between team members broughtabout a profound change in individual and departmental attitudes. Staff began to seek one another out to
discuss possible consequences ofproposed changes, problems were discussed
more openly, departmental rivalries and barriers receded. The dangers of
working in isolation and the consequences of ill-judged and hasty actions
became better appreciated. Knowledge, ideas and experience became shared
more fully to the benefit of the individual and the company.
56
Carting's companyadoptedHazop after experiencing several serious incidents. Buzzelli writes31, 'For an industry so proud of its technical sal'ety
achievement it is humbling to have to admitthat most of our significant safety
improvements were developed in response to plantaccidents'.
It does not have to be so. Hazop provides us with a lantern on the bow
(Chapter 1). a way of seeing hazardsbeforethey wreckour ship.
References in Chapter
I.
2.
3.
4.
5.
6.
7.
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
18.
19.
20.
21.
Ii
57
22. Pegram, N., 1990, The Chemical Engineer, No. 482: 37.
23. McKelvey, T.C. and Zerafa, M.J., 1990, Vital 1-lazop leadership skills and techniques, American Institute oJ Chemwal EngineersSummer NaiionalMeeting, San
Diego, California, 1922August.
24. Rushton, AG.. 1989. Computerintegrated process engineering, Symposium Series
No. /14,27 (Institution ofChemical Engineers, Rugby.UK).
25. Aleksander, 1. and Burnett, P., 1987, Thinking Machines, 107. 196 (Knopf, New
York. USA).
26. Ferguson, G. and Andow, P.K.. 1986, Process plant safety and artificial intelligence, World Congress of Chemical Engineering, Tokyo,Paper 14153. Volume
II, 1092.
27. A 4th century theologian quoted by N. MacGregor, 1991. RoyalSociety ofArts
journal, 139 (5415): 191.
28. Gustafson, R.M., Stahr. J.J. and Burke, D.H., 1987, The use of safety and risk
assessment procedures in the analysis of biological process systems: a ease study
ofthe VeraxSystem 2000, ASME /05th WinterAnnualMeeting, 13/S December.
29. Royal Commission on Ensironmental Pollution, 1991, Fourteenth Report:
GENHAZ
A System for the Critical Appraisal of Proposals /0 Release
Genetically Moditted Organismsinto the Environment(1-IMSO, London. UK).
30. Carling. N., 1986. Hazop study ofBAPCO'sFCCUconiplex,American Petroleum
institute Committee on Satciv and Fire ProjectionSpringMeeting. Denver, Colo-
rado, 8/ April.
31. Buzzelli, D.T., 1990, Plant/OperationsProgress,
9(3): 145.
32. Turney. RD., 1991, The application of Total Quality Management to hazard
studies and their recording, SymposiumSerie,sNo. /24, 299 (Institution ofChem-
58
Hazan
and
48. Kletz, T.A.. 1998, What Went Wrong Case HistoriesofProcessPlant Disasters.
4th edition, Section 17.13 (Gulf Publishing Company.Houston, Texas,USA).
49. Leathley, B. and Nicholls, D., 1998. LossPreventionBulletin, No 139:8.
50. Pitt, M.J., Flower, J.R. and Ben-Emhmmed, M.K., 1995. Computer simulation in
SI.
52.
53.
54.
55.
gence, 9: 129.
56. Kavianian, H.R.. Rao, J.K. and Brown, G.V.. 1992, Application ofHazardEvaluation Techniques to the Design of Potentially Hazardous industrial Chemical
Processe,r (US Department of Health and Human Resources, Cincinnati. Ohio,
USA).
59
1-IAZOP
ANI) IIAZAN
65. Kletz, TA., 1994, Learning trom Accidents, 2nd edition, Chapter l6
(Butterworth-Heinernann, Oxford, UK).
66. Turney, R.D and Roff.M.F., 1995, in Mewis. J.J. etal, LossPreventionand Safi-tv
Promotion in the Proces.r Jndustrie.r Proceedingsof the 8th International
Symposium.93 (Elsevier, Amsterdam, The Netherlands).
67. Preston, M.L. and Richards. D.C., 1995, STOPHAZ: A tool supporting safer
process design, Symposium Series No. 139, 5 17 (Institution of Chemical Engineers, Rugby, UK).
68. Turney, R. and Pitbiado,R., 1996, Risk Assessment in the Process industries, 14
(institution ofChemical Engineers, Rugby, UK).
69. Taylor, JR., 1982, Evaluation ofcosts, completeness and benefits for risk analysis
procedures, international Symposium on Risk and Safety Analysis, Bonn,
Germany, 68July.
70. Turner,S., 1996. The Chemical Engineer, No.606: 13.
71. Rushton, AG., 1996, Quality Assurance of Hazop, Report No. OTO 96 002
(Health and Safety Executive. Sheffield, UK).
72. Skelton, R.L., 1998, Loss PreventionBulletin, No. 142: 12.
The European Process Safety Centre, the Chemical Industries Association and the
Institution of Chemical Engineers arejointly revising Reference I above,for publication in late 1999.
The International Electrotechnical Commission has prepared a draft standard(IEC
61882),defining Hazop. It may be issuedin final form in 2000and copied as a British
Standard.
60
Appendix to Chapter 2
Some accidents that could have
been prevented by Hazops
A2.1
Reverse flow
Many accidentshave occurred because process materials flowed in the opposite direction to that expected and the fact that this could occur was not foreseen. For example, ethylene oxide and ammonia were reacted to make
ethanolamine. Someammonia flowed from the reactor, in the wrong direction,
along the ethylene oxide transferline into the ethylene oxide tank,past several
non-return valvesand a positivepump. It got past the pumpthrough the relief
valve whichdischarged into the pump suction line. The ammoniareacted with
30 m3 of ethylene oxide in the tank which ruptured violently. The released
ethyleneoxide vapour exploded causing damage and destruction over a wide
area1.
A hazard and operability study would have disclosedthe fact that reverse
flow could occur. Reference 7 of Chapter2 describes in detail a Hazop of a
similarinstallation.
On another occasion some paraffinpassed from a reactor up a chlorine
transferline and reacted with liquid chlorine in a catchpot. Bitsof the catchpot
were found 30 m away2.
On many occasions process materials have entered service lines, either
becausethe service pressure was lower thanusual or the process pressure was
higher than usual. The contamination has then spread via the service lines
(steam, air, nitrogen, water) to other parts of the plant. On one occasion
ethylene entered a steam main through a leaking heat exchanger. Another
branch ofthe steam main supplied a space heaterin the basement ofthe control
room and the condensate was discharged to an open drain inside the building.
Ethyleneaccumulated in the basement, and was ignited (probably by the electric equipment, which was not protected), destroying the building. Again, a
Hazop would have disclosed the route takenby the ethylene.
Forother examples of accidents due to reverse flow that could be prevented
by Hazop, see Reference 3.
61
A2.2 Bhopal
The sumpshownin Figure 2.10 contained water with a layerof light oil on top.
Welding had to take placenearby so the sumpwas emptied completely with an
ejector and filled with clean water to the level of the overflow pipe. When a
spark fell into the sump, there was an explosion and fire. The U-bend had not
been emptied and there was a layerof oil in the bend on top ofthe water.
A Hazop would havedisclosedthe hazard if the preparation of the equipment for maintenance had been considered. The equipment got little consideration during design as it was not part of the main plant, only a system for
collecting a wastewater stream (seeSection 2.8, page47).
62
APPENDIX TO CHAPTER
Overflow to drain
(12 inch diameter)
Figure 2.10 The sump was emptiedand filledwithclean waterhut oil was
left in the U-bend
Vent
Reactor
1-IAZOP
ANI) HAZAN
would flow by gravity into the reactorand coolthe contents. Unfortunately the
designers overlooked the fact that when the reaction started to run away the
pressure in the reactorwould rise. Whenthe valve was opened the water was
blown out ofthe vent! The reactorexploded and the subsequent fire destroyed
theunit9.
Rupture
Hot furnace
Steam
(start-up power supply)
To waste
heat boilers
Condensate make-up
Figure 2.12 When the steam valve was opened, condensate enteredthe hot line
from the furnace
64
APPENDIX TO CHAPTER 2
The plant instrumentation had originally been very well organized but, as
instruments were removed and others added, it became difficult to tell which
instruments were connected to which power supply. All modifications,
including modifications to instrument and electrical systems, should be
reviewed by Hazop or. if they are minor, by a similartechnique (see Section
2.4.3, page 28).
After the incident the steam drum was made larger so that it contained
enough condensate to remove residualheat from the process withoutmake-up,
an inherently safer design It)
Vent
Vapour
Cooling
waler
Computer
H
Figure 2.13 Computer-controlled batch reactor
65
computer had just started to increase the cooling water flow to the reflux
condenser. The computer kept the flow at a low value. The reactoroverheated,
the relief valve lifted, and the contents of the reactor were discharged to
atmosphere.
The operators responded to the alarm by looking for the cause ofthe low oil
level. They established that the level was normal and that the low-level signal
was false, hut by this time the reactorhad overheated. A Hazop had beendone
on the plant but those concerned did not understand what went on inside the
computer and treated it as a black box'
something that will do what we
want it to do withoutthe need to understand what goes on inside it. They did
not Hazop the instructions to the computer.
Whatthey should havedone is:
(I) Ask precisely what action the computer will take for all possible deviations (reverse how, more flow, loss of power, loss of input or output signal,
and SO on).
(2) Ask what the consequences will be.
(3) Ifthe consequences are hazardous or prevent efficientoperation, consider
what alternative instructions might he given to the computer or what independent backup system might he required.
The incident provides a goodexample of the results of blanketinstructions
(to computers or people) such as, 'When a fault develops, do this'. All faults
shouldbe considered separately duringa Hazop,for all operating modes. The
action to be takenduring start-up maybe differentfrom that to be takenduring
normal running or later in a hatch. This is a lot of work, but is unavoidable if
accidents are to be prevented.
As technologists we like to know how machines work and like to take them
to bits. We should extend this curiosity to computer programs and not treat
them as 'black boxes'. It is not necessary to understand all the details of the
electronics, but it is necessary to understand the details of the logic
to know
precisely what instructions havebeen given to the computer.
There may have been a misunderstanding between the operating manager
and the softwareengineer. Whenthe manager asked forall controlled variables
to be left as they are whenan alarm sounds, did he mean that the cooling water
flow should remain steady or that the temperature should remain steady?As
stated in Section 2.2 (page 21). when a computer-controlled plant is 'Hazoped'
the software engineershould be a member ofthe team.
An amusing example of a failure to consider all eventualities occurred
during the night when summer time ended. An operator put the clock on a
computer back one hour. The computer then shut the plant down for an hour
until the clockcaught up with the program'7.
66
APPENDIX TO CHAPTER
computer-controlled systems.
A2.7 Abbeystead
an
explosion in a water
pumping station
50 mm (2 inch)
return line to plant
From plant
line to sea
Figure 2.14 Simplified line diagram ofthe waste disposal system at Sellafield
As a result of the human error some material which was not suitable for
discharge to sea was moved to the sea tanks (seeFigure 2.14). This should not
havemattered as BNFI. thought it had 'secondchance' design the ability to
pump material back from the sea tanks to the plant. Unfortunately the return
route used part ofthe discharge line to sea. The return line was 2 inches diameter, the sea line was 10 inches diameter, so solids settled out in the section of
the sea line where the linearflow rate was low and were later washed out to
sea. The design looks as if it might have been the result of a modification.
Whetherit was or not,it is the sort of design errorthat would be pickedup by a
1-lazop.
APPENDIX TO CI-]APTER
Team leader 2
'Onecan never be absolutely certain that all possible situations are considered
during a Hazop, but I feel reasonably certain that this operability problem
would have been discussed in some detail (providing the technique was
applied by experienced people) underone or more ofthe following headings:
(a) NO FLOW: One reason for 'No flow' in the 2 inch line could be wrong
routing for example, all the off-spec material entering the seadue to leaking
valves, incorrect valve operation, etc. How would we know that we were
putting off-spec material into the sea?
(h) LESS FLOW: Again, leaking valves would allow off-spec material into
the sea, and a reduced flow to the plant, etc. Also, possible restriction or
blockage due to settlement ofsolids would certainly be discussed.
(c) MORE FLOW: The team would have checked design flow rates and
commented on the differentvelocities in the 10 inch and 2 inch line sections
and possible consequences.
(d) COMPOSITION CHANGE/CONTAMINATION: The team would have
questioned methods of analysis, where samples were taken, and how we
ensured that the contents ofboth the sea tank and the 10 inch line section were
suitable to dump into the sea. Indeed, when the 10 inch route to the sea was
studied the problem ofcontamination would againbe discussed.
69
Team leader 3
'I believe that theline ofquestioning would be as follows:
(a) NO FLOW: Misrouting opening of the 10 inch sea line in error when
material should be returned to the plant for reprocessing; this would raise
furtherpoints of sampling, valvelocations and the need for interlocks.
(h) REVERSE FLOW: Direct connection between plant and sea via the
common manifold whatpreventshackflowand howreliable isthe system?
(c) LESS FLOW: Contamination implications of incomplete purging of
the systembetween batchdischarges. How will the operatorsknow that the sea
tankand dischargeline havebeen emptied and purged following a discharge?
What are the consequences of contamination due to accumulation of material
in dead spaces in the common dischargesystem? A team with knowledge of
slurry-handling plants would be aware ofthe problems of deposition resulting
from reduced flow velocities. For example, it is common practice to provide
recirculating ring mains on centrifuge feed systems to avoid deposition and
blockage.
would he raised.
APPENDIX TO CFIAPTER
Circulation line
Valves closedbut
Distillation
feedvessel
To distillation
column
water can so easily turn up as the result of corrosion, leaking valves, failure to
disconnect a hose or accumulation in a dead-end or becauseit has been left
behind after a wash-out.
Can the presence of water (or anything else) cause formation of a separate
layerand, if so, what will he the consequence?
Forany deviation, look for consequences in other parts of the plant and at
later times, notjust for local and immediate ones (seeSection 2.5(1), page35).
Unexpected formation of a separate layer was the cause of one of the few
serious criticality incidents that have occurred on nuclear processing plants. In
1958. at Los Alamos, USA, the liquid in four tanks had to he washed with
solvent to recover some plutonium. Each tank should have been treated separately hut instead their contents were combined in a single tank, together with
plutonium residues that had accumulated in the tanks over a period of seven
years. The acid present in one ofthe streams caused an emulsion to break and the
plutonium concentrated in the upper layer. This layerwas too thin to be critical
but when the stirrerwas started up the layerbecame thickernearthe axisofthe
stirrer and criticality occurred. One man was killed. Afterwards unnecessary
transfer lines were blocked to reduce opportunities for incorrect movements.
A review of criticality incidents shows that many could have been
prevented by Hazop as they were due to reliance on valves which leaked,
excessivecomplication, unforeseen flows through temporary lines, inadvertent
siphoning and entrainment.
APPFND1X TO
CHAPTtR 2
To waterway
valve
Figure 2.16 Shouldwe assume that the hose mightleak and the two valves mightbe
left open all at the same time?
(Reprinted by permission of Hydrocarbon Processing, April 1992, copyright 1992 by
Gulf PublishingCo. all rights reserved)
Accidents are sometimes said to be due to an unlikely coincidence that
could not havebeenforeseen, but theyare usually not true coincidences. As in
this case, two (or more) failures are latent or ongoing faults that exist for
significant periods of time. When a third failure occurs, an incident is
inevitable.
A2.10.2 The need for specialized knowledge
A vessel containedliquid sulphur (melting point 120C). A Hazop was carried
out on the t1owsheet the team considered 'more of pressure' and decided that
the precautions taken to prevent choking of the vent, which included a lute,
were adequate. At a later Hazop ofthe line diagram, when considering 'more
oftemperature'. someonepointed out that the viscosity ofsulphur rises sharply
aboveabout 200C. This temperature could not be reached in normal operation
hut could he reached ifthe vessel was exposed to fire. The sulphur in the lute
could then become so viscous that it would prevent relief of the vessel. The
relief systemhad to be redesigned'9.
A solvent tank was ventedthrough a seal pot.An electricheaterwas added
later. The reason is not stated in the report, but was presumably to prevent
freezing in cold weather. The modification was Hazoped hut all the members
of the team were chemicalengineers no electrical engineeror representative
of the supplierwas present. None of the chemical engineersrealizedthat the
temperature ofthe heatercould rise abovethe auto-ignition temperature of the
solvent ii' the liquid level in the seal pot was lost20.
73
APPENDIX TO CHAPtER 2
Alil
a plant for a
for irradiating cancer patients. They can he irradiated directly or with X-rays
generated by the electronbeam hitting a target. Much higherenergybeams arc
used to produce X-rays than for direct irradiation. As the result of a software
error a number ofpatients were directly irradiated with high energy beams. A
systematic hazard identification procedure would have shown that absenceof
the target was potentially dangerous and that it should be physically impossible to operate in high energy mode unless the target was in place. The fatal
error was relying on software interlocks23'24.
Acknowledgements
Thanks are due to Messrs. H.G. Lawley, FR. Mitchell and R. Parvin for assistance with Section A2.8. Sections A2.35 are reprinted from Journal of Los's
2.
3.
4.
5.
6.
7.
75
UK).
KIds, TA., Chung, P.W.l-l., Broomfield, E. and Shen-Orr. C., 1995, Computer
Controland Human Error(Institution of Chemical Engineers, Rugby. UK).
13. Slainthorp, F., 1990, The Chemical Engineer, No. 480: 16.
14. Mooney. D.G.. 1991. An overview of the Shell fluoroaromalics plant explosion.
SymposiumSeries No. /24, 381 (Institution ofChemical Engineers, Rugby.UK).
IS. Kletz. TA., 1991, Loss PreventionBulletin, No. 100: 21.
16. Stratton, WE., 1989, A Review ofCriticality Accidents, ReportNo. DOE/NCT04
12.
letter, No.9: 3.
21. Collins. R.L., 1995, Chemical EngineeringProgress.91(4): 48.
22. Klctz, TA., 1998. What Went Wrong Case Historie,sofProces,sPlantDisasters.
4th edition. Section 2.6(a) (Gulf Publishing Company,Houston. Texas, USA).
23. Lcveson, N.G., 1995. Safeware: System Saft'ty and Computers. Appendix A
(Addison-Wesley).
76
1996.
Hazard analysis
(Hazan)
When von can measure what von are
speaking aboutandexpress it in numbers,
you know something about it.
Lord Kelvin
3.1 Objective
77
H
w
z
4:
zC
MONEY SPENT ON SAFETY
I use the term hazard analysisratherthan risk analysis as risk analysis has
beenused to describe methods of estimating commercial risks (seeReferences
I and 2) and hazard analysis because, as we shall see, an essential step is
breaking down the events leading to the hazardinto their constituent steps.
While Hazop is a technique that can, and I think should, be appliedto every
new design and major modification, Hazanis, as stated in Section 1 .1 (page 1),
a selective technique. It is neither necessary nor possible to quantify every
hazard on every plant. Unfortunately the apparentprecision of Hazan appeals
to the legislative mind and in some countries the authorities have suggested
that every hazard should be quantified.
Hazan is not,ofcourse, a technique for showing that expenditure on additional safety measures is necessary. Often it shows that the hazardis small and
that furtherexpenditure is unnecessary.
1-lazan does more than tell us the size of a risk. Especially when fault trees
(Section 3.5.9, page 113) are used, it shows how the hazard arises, which
contributing factors are the most important and which are the most effective
ways of reducing the risk. Most of all, it helps us to allocate our resources in
themost effective way. If we deal with each problem as it arises, theend result
maybe the opposite of that intended. This is common in politics28 and can also
occur in engineering. It can result in massive expenditure on preventing a repetition of the last accident while greater risks, which have not so far caused
injury, are unrecognized and ignored.
When hazard analysis was first used in the chemical industry, in the late
1960s and early l970s, it was applied mainly to well-defined (though often
complex) problems, such as those involving instrumented protectivesystems,
for which good reliability data were available(for examples see Section 3.8,
page 133). ('Good' means that the data did not vary greatly between different
plantsor industries or conditions of use.) Later, hazard analysis was extended
to much more ill-defined problems involving many sequential steps for
example, how often will a piece of equipmentleak, how big will the leak be,
howfar will it spread, howoften will it ignite, what overpressure will be devel()ped ifit does and what injuries and damage will be caused by the explosion or
heatradiation? Confidence in the accuracy was obviously lowerhut comparative values were better than absolute ones. Most of the controversy that has
been attached to hazard analysis (see Section 5.3, page 181) applies to these
studies. Those describedin Section 3.8 (page 133) are typical of the various
types of study carriedout today.
79
80
81
Hazan attempts to quantify this phrase and has therefore been accepted fairly
readily by the Health and Safety Executive and safety professionals. (ALARP
does not meanAs Low As Regulations Permit;ifit is reasonably practicable to
reduce risksfurtherwe are expectedto do so.)
In contrast, in the United States there has been much more pressure to
remove every risk and companies have been reluctant to admitthat there is a
low level of risk that is tolerable or acceptable. However, there are signs of
change in both regulatory and case law. The US Office ofNuclearand Facility
Safety. part of the Department of the Environment, usesALARA(As Low As
Reasonably Achievable) and a Supreme CourtRuling states56:
'if ... the odds are one in a billion thataperson willdiefrom cancerby taking
a drink of chlorinated water, the risk clearly could not be considered significant. On the otherhand, if the odds are one in a thousandthat regular inhala
lion qf gasoline vapors that are 2% henzenewill heflital, a reasonable person
might wellconsiderthe risk significant and takeappropriatesteps to decrease
or eliminateit.
Note that the Supreme Court make the common error of not stating their
units. Are they referring to one drink of chlorinated water and to a lifetime's
exposure to 2% benzene in gasoline?
Similarly in Germany, according to Brown57:
in German law ... one maynot legal/vpose a risk to the public from one '5
enterprise. Thispositively inhibits' the development of assessmentsthat recognize risk as an inevitableconstituent of lif'.' it makes people tread warily, and
keepslawyersrich.
The EuropeanCommunityas awholehas not accepted the use ofthe phrase
'reasonably practicable' but it has accepted a requirement to carry out risk
assessments. This shouldcome to much the same,as therewould he no point in
assessing risk unless the action required depends on the size of the risk.
However, while 'reasonably practicable' is backed by case law, there is so far
no case law on risk assessment58,
The concept of' ALARA goes hack a long way. In the 16th century Rabbi
Schlomo Cohen of Greece wrote59:
'The damage causedto thetownspeople by the vats used by the dyeingindustry
is extremely great and has to be considered as similar to smoke and bad
odours, Howevem; sincethe textile industryis the main basisforthe livelihood
82
When injury is unlikely we can comparethe annual cost of preventing an accident with the average annual cost of the accident. Suppose an accident will
83
Risks which are within a target or criterion are sometimes called 'acceptable risks', but I do not like this phrase.We haveno rightto decide what risks
are acceptable to other peopleand we shouldnever knowingly fail to act when
otherpeople's livesare at risk; but we cannot do everything at once we have
to set priorities.
More pragmatically, particularly when talking to a wider audience than
fellow technologists, the use of the phrase 'acceptable risk' often causes
people to takeexception. 'Whatrighthaveyou,' theysay, to decide what risks
are acceptable to me?' But everyone has problems with priorities; most people
realize that we cannotdo everything at once, and theyare more likelyto listen
ifwe talk about priorities.
The UK Health and Safety Executive proposes30 that the phrase 'tolerable
risk' should be used instead of 'acceptable risk'. 'Tolerable' has been
defined31 as 'that which is borne, albeit reluctantly, while "acceptable"
denotes somehigherdegreeofapprobation'.
The UK Health and SafetyExecutive alsoproposesthat instead of one level
ofrisk thereshouldbe two: an upperlevel which is never exceeded and a lower
level which should be regarded as 'broadly acceptable'.This is defined as a
level which does not worry us or cause us to alter our ordinary behaviourin
any way; it would not be reasonable to consider further improvements ifthese
involved a cost. In between theupperand lowerlevels the risk is reducedifitis
reasonably practicable to do so. Risks near the upperlevel are tolerated only
when reduction is impracticable or grossly disproportionate to the cost (see
Figure 3.2 on page 86). Cost-benefit analysis, comparing the costofreducing a
hazard with the benefits,shouldbe used to determine whetheror not an action
is reasonably practicable30'32.
We do not, of course, remove priority problems by asking for more
resources. We merely move the targetlevel to a different point.
Apart from the main uses of Hazan in helping us decide whether or not
that is, in helping us
expenditure on particularsafetymeasures is justified
set priorities it can alsohelp us to:
resolve design choices, forexample,between reliefvalvesand instrumented
protective systems(trips) (see Section 3.8.5, page 138);
decide how much redundancy or diversity (see Section 3.6.4, page 123) to
build into a protective system;
settesting, inspectionand maintenanceschedules(see Section 3.5.3, page107).
The proposals illustrated in Figure 3.2 (page 86) have been widely quoted
buttheir full implications have not yet beenrealizedoracted upon. We still fix
absolute standards for measurements such as the concentration of harmful
gases and vapours in the workplace atmosphere or the concentration of
85
Unacceptable
region
extraordinary
circumstances
practicable' or Tolerability
region
(Risk is undertaken only if
benefitis desired)
Broadlyacceptableregion
(Noneedfordetailed working to
demonstrate that the risk is as low as
Tolerable ifcost of
reduction would exceedthe
improvement gained
reasonably practicable)
NEGLIGIBLE RISK
86
If you spend your working lifetime in a factory of 1000 men, then during your
time there, if the FAR is 4, 4 of your fellow workers will be killed in industrial
accidents, but about20 will be killedin other accidents (mostly Ofl the roads and in
the home) and about370 will die from disease,including about40 from the effects
of smoking, ifpresent ratescontinue.
87
per year
Firemen in London 1940
1000
70
50
62
42
7.3
Construction
Railways
4.8
Agriculture
3.7
1.2
1.2
Vehicle manufacture
0.6
Clothing manufacture
0.05
0.05
l0 working hours
l0
100 x to
125 x io
84 x l0
14.5 x l0
10 x l0
9.6 x l0
8 X l0
7.4 x l0
2.4 x l0
2.3 x l0
1.2 x l0
0.1 x io
0.1 x l0
2000 X
140 x
The figurefor offshoreoil and gas includes the 165 people killedby the fire and
explosion on the PiperAlphaoil platform in 1988.
The figures in the first two rows are from Reference 60, the Health and Safety
Executive figures from Reference 32 and the remainder from Lees, page 2/9.
expenditure which some of its competitors do not incur. Some of the extra
expenditure can be recouped in lower insurance premiums; some can be
recouped by the greaterplant reliability which safetymeasures often produce;
the rest is a self-imposed 'tax' which has to be balancedby greaterefficiency.
Note that when estimating a FAR for comparison with the targetwe should
estimatethe FAR for the personor group at highestrisk, not the average for all
theemployees on theplant. It would be no consolation to me, if I complained
that I was exposed to a high risk, to be told, 'Don'tworry. The averagefor you
88
per year
Maximum tolerablerisk:
employees
public
public (nuclear)
l0-
50
i04
l0
75
1.5
2025
5x
0.05
Negligible risk:
employees and public
0.005
x l0-
l0
l0
See Section 5.2.6(page 176) for an explanation ofthe limitsfor ionizing radiation.
FARs are not quotedfor public risks because the number of hoursfor whichpeople
are exposedis so variable. The risk per year is a bettermeasure.
l0
(l0
(l0
89
Sincethe passage of the Health and Safety at Work, etc Act in 1974 there
has been a gradual move awayfrom prescriptive regulations, which tell people
exactly what they should do, to goal-setting ones, which set objectives to be
achieved. There is advice on how to achieve them, but it does not have to be
followed. The requirement to assess risks and the settingofrisk targets are part
of this new approach.
It is not,ofcourse, necessary or even possible to assess every risk quantitatively. Most risks are minor and can be assessed qualitatively. The Healthand
Safety Executive usually requires quantitative assessments of nuclear and
offshore risks and may ask for them in other cases. Major hazards are more
likely to be accepted by them if they are supported by a quantitative
assessment.
= 30.000 years
91
Roadaccidents (UK)
Road accidents (US)
All accidents (UK)
Murder (UK)
Smoking 20 cigarettes/day
280 X
l0-
l0
24 x
30 I0
iO5
500 X l0
75 x l0400 x l0
100 x I0
1000 0
l0
10 X
X 1
l0
Floodingofdykes (Holland)
Fall of aircraft (UK)
0.2 x
Hit by meteorite
l0''
i07
(1
in 360)
(1 in 10,000)
(1 in 4000)
(1 in 3300)
(1 in 100,000)
(I in 200)
(I in 1300)
(I in 250)
(I in 1000)
(I in 100)
(I in 10 million)
(1 in 10 million)
(I in 10 million)
(I in 50 million)
(1
in 100 billion)
Notes:
equally distributed; the veryold and the veryyoung,for example, are more likely
than others to be killedin an accident;smokers are more likely than non-smokers to
get cancer.
Table3.2 shows the risk ofdeath, per year, for a number of non-occupational
activities, including activities such as driving and smoking that we accept
voluntarily and others that are imposed on us without our permission. The
figures are approximate and should be used with caution. Nevertheless they
show that we accept voluntarily activities that expose us to risks of iO or
more per year, sometimes a lot more, while many of the involuntary risksare
much lower. We accept, with little or no complaint, a number of involuntary
risks (forexample,from lightning or falling aircraft) whichexposeus to arisk
of deathof about I
or less per year.
We thus have a possible basis for considering risks to the public at large
from an industrial activity.If the average risk to those exposed is more than
per person per year, we will eliminateor reduce the risk as a matteror
priority.If it is already less it would not be rightto spend scarceresources on
reducing the risk further. It would be like spending additional money, above
that already spent, on protecting peoplefrom lightning. There are more important hazards to be dealt with first.
The lasttwo paragraphs appeared in the earliereditionsof this book. Since
thenthe Health and Safety Executive has made the proposalsdescribed at the
end of Section 3.4.1 and summarized in Figure 3.2 and Table 3.1(b). It
suggests that a risk of 106 per yearis 'broadly acceptable',though not negligible. Itquotesthe following example to show howsmall this risk is compared
to the other risksto which we are exposed. Suppose 10,000 peoplelive near a
nuclear power station and as a result are exposed to an average risk of death
(fromcancer) of 106 per yearin addition to the normal risk; 106 per yearis
ratherless than Io per lifetime. Regardlessofwherethey live, about 2500of
the 10,000 people will die from cancer. As a result of the nuclear plant, this
number will rise to 250161. And this estimate is based on the pessimistic
assumption that the risk is proportional to the dose.
As well as considering the average risk we should consider the person at
greatest risk. A man aged20 yearshas aprobability ofdeathfrom all causes of
1 in 1000 per year. (The figure for a younger man is not much less.) An
increase of 1% from industrial risks is hardly likely to cause him much
concern, and an increase of0.1%shouldcertainly not do so. This gives arange
of
to 106 per year. The peopleatgreatestrisk are usually thosewholive
nearest to the factorybut in the case of nuclearrisk may be those whosediet
iO
l0
l0
(l0
93
I0
<10-4
('5
C
'U
C
C:'
z('5
U 106
>
>
C
U
108
Li
10
10
NUMBER OF CASUALTIES, N
Figure 3.3
a suggested criterion.
Majesty'sStationery Office)
500,000 peoplein which a chemical plantimposes somerisk on all the inhabitants, though some of them, of course, are at greater risk than others. If the
average risk is I
per year, on average one person will be killed every 20
the
time
a
second
death occurs the firstone will probablyhave been
years; by
forgotten. If the average risk is 106, on average someone will be killedevery
two years and the public would considerthis quite intolerable. In a democracy
all criteriafor risk (and everything else that affects them) must be acceptable to
thepublic (seeSection 5.3, page 181).There is adifference, ofcourse, between
94
deaths that are clearly due to an industry and a theoretical rise (of one in
several thousand) in the number ofpeopledyingfrom disease.
We have considered averagerisks and the person at greatestrisk. Another
way ofexpressing risk to the public is to draw a graphof the numberof people
killed (N) againstthe cumulative frequency of the event (F). Figure3.3 (from
Reference 30) shows an FN line for a particularchlorine installation and, for
comparison, a proposed criterion (the line AB). Both lines refer to casualties,
not deaths; Reference 30 suggests that about one third of them will result in
death. Note that the probability that 10 or 100 peoplewill becomecasualties is
higher than allowed by the criterion, but that there is a limit to the possible
numberofcasualties. Note also that the frequenciesare cumulative that is,
thepoint on the graph for N=10 (say), gives the frequency of events which
cause 10 or more casualties.
Thejagged line in Figure 3.3 isa prediction byexperts ofwhat will occur (if
the assumptions on which it is based are correct); only experts in the technology are able to derive it. (In other cases the FN line may be based on the
historical record.) In contrast, the line AB is basedonjudgement;it shows the
level of risk that people will, it is believed, tolerate.Everyonehas a right to
commenton its position, especially those exposed to the risk, and the expert
has no greaterrightto do so than anyone else (see Section 3.3, page80).
MY BOSS LIKES
TO SEE
GAPM6
ThAT RISE
95
1-IAZOP
AND HAZAN
17
102
2
14
II
N
\I'4
101
I
io
16
I\
10 -
'' N
"N N
'
\\
\\
N
N3
N
15
N
N
11
10I
1
10
100
1000
10,000
NOR MOREDEATHS
Figure 3.4
96
plants,before improvements.
4: Ditto, with recommendedimprovements.
5: Ditto, 2nd report. This wasjudged tobejust about tolerable.
6: A harbour: risk to the population onshore from a spillage ofLPG from a ship
before improvements.
7: The harbour,after improvements.
97
Value giving an
impact of0.2
Value giving an
impact of 0.6
Number of deaths
100
Number of injuredpersons
40
800
Number of evacuees
30
1000
Duration ofalarm,person-days
20,000
5,1)00,000
Number of deadanimals
20t)
8000
100
0.5
0.5
40
40
Discounted expenditure,
It),000,000
200,000,000
The numbers in the centre column arc considered to produce a similar impacton the
public (to be precise.the same membership of a fuzzy set).Those in the righthand
column produce a similarbut greaterimpact. The Impacts are combined (sec
Reference 64 for details)to produce a total impact, called a disastervalue'.
(Others call it an indexofwoe.) For Bhopal (over2000 killed) the index is set at I.
Flixhorough (28 killed) is then 0.50, Seveso (t) killed) is 0.71 and the 986 pollution
ofthe Rhine at BasIc (0 killed)is 0.51.The authorsof Reference 64 admit that the
assignment ofrelative impacts is verysubjective but the method does allow various
factors besidesthe risk to life to be taken into account.
99
() spent to savealife
Liver transplants
Negative
Small
8K
50K
25K
125K
Road
travel
25K1OM
Industry
Agriculture (employees)
Rolloverprotection for tractors
Steelhandling(employees)
Pharmaceuticals (employees)
Pharmaceuticals (public)
Chemical industry (employees) (typicalfigure)
Nuclearindustry (employees and public)
13K
500K
I .3M
25M
70K
SM
Social
Smoke alarms
policy
700K
I25M
13M
2040M
Notes:
All figures are takenfrom Reference 36, are corrected to 1999 pricesand referto
the UK.They are approximate and somemay havebeenoutdated by changes in
technology. US figures are often higher.
A 10% increase in the tax on tobacco decreases smoking by about5% so there is
a net increase in revenue.
Ifwe spendlOMon anti-smoking propaganda and as a result 2000people (less
than I smoker in 10,000)give up smoking, S00 liveswould be savedat a cost of
20,000each.
The death rate (for almost all agesand causes) of members of socialclass 5
(unskilled occupations) is about 1.8 times that of members of social classes I
(professional occupations) and 2 (managerial occupations). It can be arguedthat, in
the longrun,a risk in income to the social class 2 level will produce a socialclass 2
lifestyle.
101
range is enormous. Doctors can save livesfor a few thousands or tensof thousands of pounds per life savedand road engineers fora few hundred thousands
per life saved, while industry spends millions and the nuclearindustry tens of
millions (even more according to someestimates) per life saved.
Most of the values in Table 3.4 are implicit
that is, unknown to the
people who authorize the expenditure, as they rarely divide the costs of their
proposals by the numberof lives that will be saved. No other commodity or
service shows such a variation, a range of 106, in the price paid. (Electricity
from watch batteries costs I O timeselectricity from the mainsbut we pay for
the convenience.)
What value then should we use in cost-benefit calculations? I suggest the
typical value for the particular industry or activity (such as the chemical
industry or road safety) in which we are engaged. Society as a whole might
benefitifthe chemical or nuclear industries spentless on safetyand the money
saved was given to the road engineersor to doctors, but there is no social
mechanism formaking the transfer. All we can do, as technologists, is to spend
theresources we control to thebest advantage. As citizens, of course, we can
advocatea transferof resources if we wish to do so.
The figuresin Table 3.4 are far from accurate. They are takenfrom various
estimates published between 1967 and 1985, corrected to 1999 prices (for
details see Reference 36), and some may have been made out of date by
changes in technology. They vary over such a wide range, however, that errors
introduced in this way are probablyunimportant (see also Section 3.8.1. page
133).
The Health and Safety Executive has published a review of the extent to
which risk assessment, including cost-benefit analysis, is used withingovernment departments65'66.It shows that these methods are often used to decide
priorities within departments but that they are not used to decide priorities
between departments. as can be seen from the figuresin Table 3.4.
This can also be seen by comparing the standard of safety required in the
Channel Tunnel with that required in the ferries which offer an alternative
method oftransportfrom England to the Continent. In the Channel Tunnel the
Health and SafetyExecutive has insisted on standards higherthan those used
on any other tunnel anywhere in the world67. But despite the disaster at
Zeebrugge in 1987 and similar incidents elsewhere in the world68, there has
been little improvement in ferry standards69.
Ofcourse, inconsistency is thepriceofprogress and we cannot expectevery
piece ofold equipment to meet the highest contemporary standards, but nevertheless would some ofthe money spenton the Channel Tunnelpotentially save
more lives ifit had been spenton the ferries?
102
in accidents seems to be at least one orderofmagnitudelowerthanthat ofordinary traffic'. If this is true, chlorine tankers are less hazardous than milk
tankers.
Theterm ALARP (As Low As is Reasonably Practical)is not used for environmental risks. The terms used instead are Best Practicable Environmental
Option(BPEO) and Best Available Technology Not Entailing Excessive Cost
(BATNEEC). A BPEOis the optionwhichprovidesthe most benefit or least
harm to the environment as a whole at an acceptable cost. BATNEEC means
that the costsofavoiding damage to the environment should bejustifiedby the
benefits. Old reportsuse the phraseBest Practicable Means instead. Reference
74 discusses the precisemeaningsof these terms.
Both BPEO and BATNEEC imply the use of cost-benefit analysis when
possible and References 38 and 53 describe attempts to apply it to environmental risks
that is, to compare the costs of pollution with the costs of
The
latter
are comparatively easy to estimate. Someof the costs of
prevention.
can
also
be
estimated; for example, the costsof cleaning, corrosion
pollution
104
Protective system
A device installed to prevent the hazard occurring; for example, a relief valve
or a high level trip.
Test interval. T
Protective systems shouldbe tested at regularintervals to see ifthey are inactive or dead'. The time between successivetests is the test interval.
Demand rate, D
If the protective
are merelyexamples.
I06
Reliefvalves and trips, however, are normally not operating and their failures
renlain latent or unrevealed until a demand occurs. Hence we haveto test them
regularly to detect failures.
Tests on relief valves show that fail-dangerfaults which will prevent them
lifting within 20% of the set pressure occur at a rate of 0.01/year(once in 100
years a typical figure).
Let test interval T = 1 year (a typical figure).
Failure occurs on average half-way betweentests. Thereforethe reliefvalve
is dead for six months ('2 T) every 100 (1/f) yearsor for 1/200 or 0.005 of the
time ('2jT). This is the fractional dead time. Suppose the demand rateD is
1/year (an example). A hazard results when a demandoccurs during the time
that the reliefvalve is dead. The reliefvalveis deadfor 1/200ofthe time, there
is one demandper year. so there is a hazard once iii 200 years.
or once in 200 years. (The more accurate formula in Section 3.5.6, page 110,
givesonce in 250 years.)
We could not determine this figureby countingthe numberof occasions on
which vessels have been overpressured becausethis occurs so rarely, but we
have been able to estimateit from the results of tests on relief valves.
Note that in this example a hazard is defined as taking a vessel more than
20% above its design pressure.Not all these 'hazards' will result in vessel
rupture or even a leak.
Relief valvefailures are discussed in detail by Maher ci ai.
3.5.3 Example2 simple trips
Assume that:
107
= 0.005
fractional deadtime
and hazardrate
= I X 0.005
= 0.005/year or I in 200 years.
= 0.02
2 x 365
0.25
= I in 4 years.
108
= D x 0.5 fT
= 100 X 0.5 X 0.5 X 0.1
= 2.5/year.
In fact, the hazard will be almost the same as the failure rate (0.5/year)
because:
therewill almost always be a demand in the deadperiod;
the fault will then be disclosed and repaired.
2.5/yearwould be the right answer if, when a hazardoccurred,we did not
repair the trip but left ii in a failed state until the next test was due.
Testing in this situation is a wasteoftime as almostall failures are followed
by a demand beforethe nexttestis due. ifyou findthis example hard to follow,
consider the brakes on a car.
Usingthe formula:
Hazard rate
= D x 0.5fT
=
x 0.5 x 0.1 x 1
= 500/year!
Not even the worst drivers have this many accidents. The true answer is
0.1/year(why?).
Thesetwo examples showhow we can get absurd answers if we substitute
figuresin a formula(or computerprogram) without understanding the reality
behind them. For another example see Reference 39. So the simple intuitive
formuladerived in Section 3.5.1 (page 105):
hazard rate = demand rate X fractional dead time
must be incorrect.
109
f(l
= 0.5fDT
f P1
1)T
ex[_ ii + I
when fT is small.
The applicability of the two equations can be understood by looking at
Figure 3.5 which shows the relationship between the hazard rate H and
demand rate D.
f
H =t(1 e J)1L2)
LU
DEMAND RATE, D
Figure 3.5 The relationship between hazard rate and demand rate
110
DT
per year
H = '2JDT
H =f(1
peryear
per year
0.1
0.2
0.001
0.00095
0.2
0.4
0.002
0.0018
0.4
0.8
0.004
0.0033
0.5
1.0
0.005
0.0039
1.0
2.0
0.0!
0.0063
5.0
lt).0
0.05
0.0099
10.0
20.0
0.!
0.0!
WhenDT = I thedifference between the two values ofH is only about25% hut for
higher values ofDT the difference increases veryquickly.
Table 3.5 shows how the methodused for calculating H becomes increasingly important as DTrises. The figuresapply to a reliefvalve; thefailureratef
is assumedto be 0.01/yearand thetest interval Tis assumedto be 2 years.
3.5.7 Two protedive systemsin parallel
Examples are two 100% relief valves in parallel or two high level trips (see
Figure 3.6).
Let FA, FB be the fractional deadtimesofsystemsA and B. The set points
of the two systems are, by accident or design.never exactly the same. Assume
A respondsfirst that is, ifA and B are two relief valves. A is set at a slightly
lowerpressure; if A and B are two high level trips, A is set at a lowerlevel.
Demand rate =D
Ill
The demandrate on A = D.
The frequency ofdemandsto which A does not respondis FAD.
This is the demandrate on B.
Therefore it seems at first sight that the fractional dead time of the
combined systemshouldbe FAFB and the hazardrate shouldbe D FAFB.
FAFB and the hazard rate is
Actually the fractional dead time is
D FAFBbecausethedemands on B tendto occur towards theend ofa proof
test interval when there is a more-than-average likelihood that B will have
failed.
Systems containing two (or more) identical devices in parallel are called
redundant. Systems containing two (or more) differentdevices in parallel are
called diverse.
Demand rate D
Pressurerises
1/year
____________________
AND
overpressured
Reliefvalve dead
Figure 3.8 Fault treeswith AND' gates. Note that afrequency is multiplied by
aprobubiluy.
113
Top event
(a)
(b
Figure 3.9 Fault trees with AND and OR' gates. Note that frequeneic.sare added
at the OR' gates.
114
115
OR meansadd
AND means multiply (as in probabilitycalculations).
As already stated, estimating hazard rates is not the only use of faulttrees.
They helpus think out all the ways in which the hazard can arise and they show
us which branchesofthe tree contribute themost towards the hazard rate.They
show us how we can reduce the hazard rate and which methods will be most
effective. For example, in the case of the free meal, we can reduce the hazard
rate,the numberoffree meals per year,by reducing the numberof visitors or
the numberof training courses or by reducing the probability that we shall be
invited. We also see that halving the numberof visitorswill he more effective
than halving the numberof trainingcourses.
In accountancy the figure produced at the end of a calculation, the bottom
line, is the one that counts. Risk assessment is different. The way the final
figure, the frequency of the top event, is derivedis as important,perhaps more
important,than the figure itself76.
To prevent confusion between rates and probabilities,always enterthe units
when drawing fault trees. If we are not clear whether the figure for the top
eventis a rate or a probability we cannot draw the tree correctly. The firsttime
Figure 3.9(a) was published the editor thought that '/year' had been omitted
from the invitation' box in error, as it appeared in every other box, so he
inserted it! Some authors suggest that we should write '/demand' after fractional dead times, as I have done in Figure 3.10.
Confusion over units is a common mistake in Hazan as a whole, notjust in
drawing fault trees. I considerthis furtherin Section 4.2, page 153.
Another common error is confusing rates and duration.In one of the Andy
Capp cartoons the eponymous hero was asked it rained during a week he
spent in the Lake District. He said ii rained twice, 'Once for three days and
once for fburdays'. The rate was low, twice per week, but the fractional dead
time for dry weather was almost 100%.
As an exercise draw a faulttree for 'car fails to start'.
Many people producefault trees like Figure 3.11. A better one is shown in
Figure 3.12. The need to take humanfailures into account as well as equipment
failures is discussed furtherin Section 3.7, page 130.
if
116
As Figure 3.11
Operator error
ffic1entQR
Operatoruntrained
Operatorerror
ta'is
Operatoruntrained
Incorrect
edure0R
L
Wrong
Operatorerror
__________________
tIonkey0R
Operatoruntrained
toruntrained
118
Table 3.6
Faults/year
Fail-safe
Fail-danger
Fractional deadtime
(simultaneoustesting)
I-out-Of-I
'2 fT
I-out-of-2
2S
JT
I-out-ot-3
3S
j3T2
'4f3T3
S2T
3/I
/27.2
2-oul-of-3
Only the measuring instruments are 2-out-of-3, not the valve. The valve may,
ofcourse, be duplicated(or even triplicated) if this is necessary to achieve the
required reliability.
Voting reduces the fail-safeor spurious trip rate and is used when spurious
trips would upset production. It does not give increased safety. A l-out-of-2
system is three times safer than a 2-out-of-3 system.
It is helpful to remember thatfail-safefaults are normally disclosed as soon
as they occur. They result in a spurious trip. But fail-danger faults remain
hidden(latent,or unrevealed) until there is a test ordemand. The formula 3S2T
for the fail-safe faults/year of a 2-out-of-3 system assumes that the faults are
not disclosed. In practice, a singlefault signal usually sounds an alarm and the
fault is thereby disclosed. this is the case, then instead ofthe test interval T
the repairtime shouldbe used in the frmula (or. more precisely,the timefrom
the alarm sounding to the completionof the repair).
On both voting and non-voting systems it is sometimes possible. by a
change in design, to turn a hidden fault into a revealed one. For example, the
failure ofan alarm bell or hooter is hidden. If it fails, it is out of action until it is
tested and repaired. We test frequentlyand accept a small chance that we may
not know when an alarm occurs. we wantgreater reliability, then instead ofa
bell that rings when an alarm is signalled we can have a device that sounds
continuallyhut becomes louderwhen there is an alarm. the sound stops, we
know something is wrong. Another example: failure of the front light on a
bicycle is noticed at once; failure of the rear light is not. the two lights are in
series, failure of eitheris noticed(hut then we have no lights at all)79.
Before installing voting systems to reduce spurious trips we shouldcheck
that the spurious trips are due to the inherentfeatures of the instrumentation
If
If
If
If
119
(a) Two of the failures in every 100 were due to the operator pressing the
wrong button. Therefore:
= 2%
OPERATOR FAILURE RATE
= 4%
MACHINE FAILURE RATE
will
therefore
Bettermechanical reliability
remove, at the most, two thirds
of the faults.To remove the otherswe would have to look at the factorswhich
affect operator error (such as better layout of the panel, locating the machine
wheredistraction is less, and so on).
(b) 98 demandsin every 100 were made on machines in the office and there
were 2 failures. The remaining 2 demandswere made on machines in a local
entertainment centre and every demand(2% of the total) resulted in a failure.
Therefore:
121
= 2%
OPERATOR FAILURE RATE
= 2%
MACHINE FAILURERATE OFFICE
MACHINE FAILURERATE
= 100%
ENTERTAINMENT CENTRE
This shows that misleading results can be obtained if we group together
widely differing data.For example, you can drown in a lakeofaverage depth 6
inches (Figure 3.13).
A similarerror was madeby a politician whosaid, ... provisionallaboratory identificationsof Salmonella infections in humans amounted to 24,000
cases in 1988 ... otherfigures suggest that half of these were due to a strain
associated with poultry and eggs', and went on to imply that action was therefore necessary to counter the infection in eggs42. However, many people
believed that nearly all the infections were due to poultry. According to one
estimate only one egg in 7000was infected.
Similarly, the former Albanian dictator Enver Hoxha was quoted in the
press43 as saying, 'Togetherwith the Chinese, the Albanians form one quarter
of the world's population.
= 2%
=I
= 1%
= 100%
LIKE
6TA11PTICIt
YOU OAN EIROVF' LW A. LAKE OI AVRA&
DEPTI4 0T
Figure 3.13
122
3.6.4 The impossibly low fractional dead time redundancy and diversity
Consider a I -out-of-3 trip systeni.
Assume that the fractional dead time of each system = 102
= 2 X (l02)3
Then the fractional dead time of the total system
= 2x
l0
I0
123
(b) Common niode failures. For example, all three instruments are from the
same manufacturer's batch and have a common manufacturing fault, all three
instruments are affected by contaminants in the instrumentair or process stream,
all three impulse linesare affected by mechanical damage or flooding ofa duct,
or all three instruments are maintained by the same nian who makes the same
error.Two orthreeprotective systems are never completely independent.
Therefore, we assume that the fractional dead time ofa redundant system is
never less than toa(that is, 1 hour peryear) and is often only
(that is, 10
hours per year). As we can get lo with two trips, a third trip is not worth
installing(exceptas part of a voting system).
For example, wearing a secondpair ofbraces attached to the same buttons
may reducethe chanceofour trousers fallingdown. Failure ofthe buttons (the
common mode) is now the biggest cause of failure and adding a third pair of
braces, attached to the same buttons, will make no further improvement.
With a diverse system (that is. one in which the approach to a hazardous
condition is measured in differentways
say by a change in an analysis. a
change in pressure and a change in temperature), b5 (6 minutesper year)
may he possible with an extremely complex protectivesystem44.For example,
belt and bracesare betterthantwo pairs ofbraces. This example illustrates the
perils of using thorough mathematics and ignoring practicalities.
Another example of a common mode failure is shown in Figure 3. 14(a), (b)
and (c). A pressure switch installed on a firewater main switcheson a pump
when the pressurefalls. The failure rate is 0.8/year, the test interval T is 0.1
year and the demand rate D is 10/year. The hazard rate H. the frequency with
which the pump failsto starton demand,
l0
= D x 0.5 fT
= 10 x 0.5 x 0.8 x 0.1
= 0.4/yearor once in 2.5 years
or oncein 3.2 years if we use the more accurate formulain Section 3.5.6(page
I It)).
The systemshown in (b) was therefore installed. The hazardrate fell to only
once in 4 years as the most likely reason for failure of the pressure switch is
choking ofthe impulseline. The system shown in (c) has a hazardrate of once
in 77 years.
Watch out for phoney redundancy parallel or series systems that look as
if they are duplicated but the duplication is ineffective. Here are three
examples.
Two bursting discs were installed in series so that the failure of one (below
the intended failure pressure) would not interrupt production. The upstream
124
Firewatermain
(a)
Fire watermain
*
(h)
Fire watermain
Figure 3.14 A commonmodefailure; (h) is little more reliable than (a); (c) is better
one was accidentally installed upside down and it rupturedat a low pressure.
The second disc was then ruptured by the shock wave and pieces of the first
disc8'
The casingof the Challengerspace shuttle was madein two parts. with an
0-ring seal between the two parts. Realizing that the 0-rings were weak
features, the designers decided to duplicate them. However, this was ineffective as one ring in a pair is liable to be grippedmore tightly than the other83.
If twodevices,connected in seriesor parallel, are tested as a pairthen failure
is not detected until both have failed. For example, if there are two valves in
series and we wish to check that they are isolating, we should check them
125
individually. Ifwe check themas a pair we are not getting thefull advantage of
redundancy. Two valves in parallel can, of course, be tested as a pair if we
wish to check that both are isolating, but not ifwe wishto check that neither is
blocked. Several incidents have occurred on US nuclear power stations
becauseduplicate systems were tested as a unit84.
Redundancy and diversity are effective when failures are random. They are
less effective when failures are due to wear (see Section 3.6.7, page 130) and
least effective when failures are systemic. For example, if failure is due to
corrosion two identical systems will corrode at the same rate. Two diverse
systems made, say, from different materials of construction, may give extra
protection but they may both corrode. The ultimate example of a systemic
failure is an error or ambiguity in an instruction (to people or computers).
People may (and often do) say, 'This can't be right, whoever wrote it must
meansomething else'; computers can't.
3.6.5 More about common mode failures
What is wrong with the trip system shown in Figure 3. 15?.
The pressurein the vessel is measured by the pressuretransmitter (PT) and
controlled by the pressure indicatorcontroller (PlC) which adjuststhe setting
/1
II
Process
Electric
Pneumatic
on the motor valve. If this control systemfails to work and the pressure rises
above the set point, then the high pressure switch and trip (PSZ) operate to
close the motor valve. At the same time the high pressure alarm (PA')
operates.
This trip system is almost useless. The most likelycauses ofthe pressurein
the vessel gettingtoo high are:
(I) Failureofthe pressure transmitter (PT) or choking of the impulseline. If
eitheroccursthe trip will not know thereis a high pressurein the vessel.
(2) Motor valve sticks open. In this case the trip will know that there is a high
pressure in the vessel and will send a signalto the motor valve, but the motor
valve will not respond.
(3) Failure of the pressureindicator(PlC). In this case the trip will work.
(3) is less likely than (I) or (2) as the PlC is in the clean atmosphere of the
control roomwhile the PT and valve are out on the plant. The trip will therefore
operate on less than one thirdofthe occasions when we want it to operate. Such
a trip is not worth having. It is neither'nowt nor summat'.It maydo more harm
than good, as we mayexpect itto operate and notwatch thepressure so closely.
The system shown in Figure 3.16 has a high reliability. The high pressure
trip and alarm(PSZAHI)has an independent connection to the vessel and operates a separate motor valve. There is a cross-connection to the control valve.
Pre-alarm
A high pressure switch (PS) and pre-alarm (PA) give a warning that the
pressure is approaching the trip setting and allow the operatorto take action.
This pre-alarm will operateif the rise in pressure is due to failureof the pressure indicatorcontroller (PlC) or motor valvebut not if it is due to failure of
the pressure transmitter (PT). if a high pressure occurs the pre-alarm will
operate on about two occasions out of three and the trip on almost all
occasions.
The system shown in Figure 3.16 is expensive. That shown in Figure 3.15
mayhave beena compromise between no trip and the design shown in Figure
3. 16, but it is a compromise that is worsethan eitherextreme.
Another example of common mode failure: a group of chemical factories
believed that power failure was impossible as their supply was duplicated.
They did not realizethat both supplies came from the same 132 kV overhead
power lines. A fire in a warehouse underneath the power lines caused a
complete loss of power and several incidents in the chemical factories,
including a fire51.
3.6.6 Designer's intentions not followed
The tank shown in Figure 3.17 was filled once/day. Originally the operator
switchedoff the pump when the tank was full. After 5 years the inevitable
happened. One day the operator allowed his attention to wander and the tank
was overfilled. A high-level trip was theninstalled. To everyone'ssurprise, the
tank was overfilled again after 1 year.
The trip had been used as a process controller to switch off the pumpwhen
thelevel rose to the set point. The operatorno longer watched the level. The
managerknew this and thought that better use was being made of the operator's time. When the trip failed, as it was bound to do after a year or two,
another spillage occurred.
It is almost inevitable that the operator will use the trip in this way. We
should either remove the trip and accept an occasional spillage or install two
trips one to function as a process controller and one to act when the
controller fails. The singletrip increased the probability of a spillage.
In this example and the last one we saw that no trip was a reasonable soluti()n and so was a good trip. The compromise solution was a wasteof money.
On occasions eitherof two extremes makessensebut a compromise does not.
(Because this is true of instrumentation do not assume it is true elsewhere.)
A similar incident occurred on a plant in which a delivery tank was filled
frequently from a suction tank. To reduce effort, the operators switched offthe
pump between transfers but did not close any valves. They relied on a
non-return valve to prevent reverse flow. Inevitably, one day the non-return
valve failed (a piece of wire had become trapped in it). and reverse flow
occurred from the delivery tank, backwards through the pump to the suction
tank,which was overfilled.
If we increase the demand rate on a protective system we increase the
failure rate. When more protective systems are addedto a plantthere maybe a
tendency for operators to increase the demandrate on them and if they do we
may soon be back with the old failure rate. For example, suppose a high
temperature alarm is added to a reactor. The operator may say. 'There is no
need to watch the temperature now. The alarm will do it for me'. The extra
equipmenthas then achieved nothing except more expense and more equipment to maintain. It is a useful exercise to calculate the hazardrates ofour trip
systems, from failure rates, demand rates and test intervals (as described in
Section 3.5.3, page 107). We mayfind that to get an acceptable hazard rate we
haveto assume that nine out often deviations are spotted by operatorsbefore
the trip operates. Do operatorsrealizethis? Do managersrealizethis?
If we comment Ofl a design and thedesigner says, Don't bothermewith it
now. Bring it up at the Hazop?', we are increasingthe demandrate on the
Hazop. The chance that the meeting will miss something increases. Hazop
should he a final check that nothing has been missed, not an occasion to
discuss known weaknesses in the design (see Section 2.4.7, page 33).
129
A new plant had two 100% compressors (one working, one spare). The failure
rate and the time required for repair were known.Calculation showed that if
failures are random, the off-line time would be 0.04% (3 hours per year). The
actual off-line time was 1.8% (144 hours per year). Why?
The failure ratesand repair timeswere as expectedbut the failures were not
random; most occurred soonaftera compressor had been put on line. This may
have been due to wrong diagnosis of the fault, installation of wrong parts or
incorrect re-assembly.
Mathematical techniques (Weihull analysis) for handling non-random
failure are available ifthe needto use them is recognized5.
Most machinery, perhaps all equipmentwith movingparts, seems to fail in
a non-random way. One study showedthat valve failure is due to wear45.
Motor cars provide another example of non-random failure they are more
likely to require attention during the week after servicing than at any other
time. If you had two cars (one working, one spare) and one had just been
serviced, would you leave it unused until the other broke down or required
servicing? Equipment after repair is asbad as new,rather thanas goodas new.
Non-random incidents can he due to non-random demands as well as
non-random failures of equipment. A study showed that bank cash machines
failed to operate when required on 17% of the occasions on which they were
used. The banks said that the non-availability of the machines was only half
this figure. The banks quoted an average availability round the clock but the
trials measured the availability at the time ofuse. Usageis heavy at weekends
when thereis usually no-one available to repair or refill the niachines46.
There is another example ofnon-random demands in Section 3.5.7, page 111.
Alarm
Reliability
Easy to improve?
Valve
Known accurately
Known roughly
Yes
Yes
saidthat sooner or later all operatorsmake errors and therefore we need fully
automatic equipment.
Both these extremes are unscientific. We should not say, 'The operator
always should' or 'The operator never will' but ask why he does not always
close the right valve in the required time and how often he will do so. The
failure to close the valve in the required time may be due to lack oftraining or
instructions (mistakes
he does not know he should do so), to a deliberate
decision not to do so (violations), to lack of physical or mental ability or (and
this is the most likely reason) to a momentary slip or lapse of attention. It is
difficult to estimate the probability of the first three causes (but see later),
though we can assume that failures for these reasons will continue in an
organization at the same rate as in the past, unless there is evidence ofchange.
Violations would be better called non-compliances as many (and perhaps
most) ofthemare due toa genuinebeliefthat therules are unnecessary or inappropriate and that thereis a better method of doingthe job.
The probability ofa slip or lapse of attention can he estimatedroughly. The
answerwill depend on the degreeof stress and distraction and the suggestions
in Table 3.7 (page 132) mayhelp us make ajudgement.
En carrying out a familiar routine, such as starting up a batch reactor, a
typical failure rate is I in 1000 ftr each operation (frr example, close valve).
Some of these failures will be immediately apparent hut others will not9
Note that the figuresin Table3.7 assume that the operators are welltrained,
capable and willing. As already stated, it is difficult to give a figure for the
probabilitythat this assumption is correct; it can vary from 0 to I depending
on the policy ofthe company.We can howevermake a rough estimateof the
as we all do in
probabilitythat a man will have a moment's aberration
life
and
to
out
a
task
(see Section 4.7,
everyday
forget carry
prescribed
page 162).
131
HAZOP
iNi) HAZAN
Table3.7 Suggestedhumanfailurerates
I in
I in 10
I in 100
It must alsobe remembered that not all tasks can he prescribed. Sometimes
the operatorhas to diagnosethe correctactionfrom the alarmand other instrument signals and maynot do SC) correctly, particularly ifthe instruments are not
reading correctly. This happened at Three Mile Islandt0.
Poor management may result in neglect and a high rate of equipment
failure. A method proposed for allowing for this is to multiply generic hardware failure rates by a factor between 0.1 and 10 which is a measure of the
competence of the management. The factor is derived from an audit using a
standard set of questions85.In a more advanced method developed by Hurst et
a!86 a detailedanalysis of the underlying causes of various types of failure is
used to weightthe audit factor. For example,according to the authors 24% of
vessel failures could be prevented by human factor reviews. In deriving the
audit factor for vessel failures the auditmarksfor human factors are weighted
accordingly.
This method does providea possible way ofmaking someallowance for the
fact that employees may be poorly trained,instructed or supervised, lack motivation, or do not have the necessary ability. It is rough justice, however, as
managers may not be uniformly weak in all these areas. More importantly,
bettermanagement will havelittle effecton slips and lapses ofattention, which
are due to innate weaknesses in human nature. To prevent them,or makethem
less likely, we havett) remove or reduceopportunities for human error, a task
for designers as well as managers. We can estimatethe frequency of slips and
lapsesof attention from data such as those in Table3.7.
Like all Hazans, data derived from these studies may not he accurate but
may pinpoint the areas in which improvement will be most effective.
132
The descriptionswhich follow are typical of Hazans carried out today. They
include well-defined problems using good data, mainly on instruments (for
example, Sections 3.8.2 and 3.8.5 and those referenced in 3.8.9). and less
well-defined problems where order-of-magnitude accuracy is the bestthat can
be expected (for example. Sections 3.8.4 and 3.8.6), though conclusions
should err on the safe side. Sections 3.8.1 and 3.8.3 lie between these two
extremes (see alsoSection 6.3. page 196).
Stainlesssteel
Mild steel
Highlevel trip
LC
Level controller
Liquid
This cost is a notional one that is, spending the money would make an
already low risk even lowerbut it is very unlikely that anyone will be killed if
the money is not spent. In contrast, many of the costsof saving a life listed in
Table 3.4 are not notional real lives will be savedif more money is spent on
health or road safety.
Note that the decision might have been differentif the hazard had been
identified during design. Unfortunately no Hazop was carriedout.
Nitrogen
Kill signal to
solenoidvalves
Tocatchpot
Designoption
Failurerate
(freq/yr)
Probability offailure
compared to Case 4
Single valve
(fail closed)
1.6 X
102
.95
Seriesvalves
(fail closed)
2.6 N
102
3.17
Single valve
(fail open)
1.1 N 102
1.34
Single valve
(fail open)
(includes operator action)
8.2 x I
1.0
Parallel valves
6.6 X
l0
0.8
(fail open)
(includes operator action)
became necessary to improve the reliability ofthe kill system. Table 3.8 shows
several cases that were considered. Case 2 was the existingsystem. It can be
seen that the kill systemwould be over threetimesmore reliable ifthe two 'fail
closed' valveswere replaced by a single 'fail open' valve (Case4). If the site
coolingwater supplyfailed, the operatorwould haveto activate the killsystem
and an allowance was madeforthe probability that he would fail to do so.
Installing two parallel kill valves (Case 5) makes only a slight improvement
in reliability. If a Hazan had not been carried out, this optionwould probably
have been adoptedon the philosophy that 'ifone is good, two must be better'.
The Hazan showed that the least reliable component ofthe kill system was the
solenoid valve that actuated the kill valve. Duplication of the solenoid valve
gave almost the same reliability as Case 5.
the risk imposed by a petrochemical site on its neighbourhood. The first study
was madewhenthreetenders were received for a new unit.Two of the designs
required a bufferzoneof 100 iii between the unit and the nearest houses while
the third design required 300 m. The difference between the estimates was
more significant than the actual figuresand detailedexamination of the calculations drew attention to a feature in the third design which had been
overlooked.
The model was then used to look at the total risk from all the units on the
site. It was about three times the target that the company had set itself, though
within the margin of error. This confirmed the gut feeling of the staff that the
nearest houses were rather closer than they would haveliked but not so close
as to be demonstrably unsafe. The model was thenused to pinpointthe features
that contributed most to the risk. When another new unit was planned the
model was run oncemore.Againit showed a risk on the wrongsideofborderline and changes were made to the design and layout to reduce the risk.
Without the results of these calculations the project team would have found it
hard to justify the extra cost.
However, the studies assumed good standards of management and operation. Tweeddale comments that perhaps the studies shouldhave assessed the
probability that this would continue to he the case. Management standards, like
hardware, can fail.
137
Ellis87 (of the Health and Safety Executive) has described a similar but
simpler study of a hypothetical application for planning permission for a
130-bedroom hotel. The proposed site was 500650 m from a water treatment
plantcontaining two 40 tonne chlorine tanks and a road tanker offloading hay.
Calculation showed that the contourrepresenting a risk of 106 per personper
year passed through the hotel. This is just on the limit of acceptability (see
Section 3.4.1,page87) but,as a large number ofpeoplemight he in the hotelat
the same time, the Health and Safety Executive would suggest that the hotelbe
moved further away (see Section 3.4.3. page90).
3.8.5 Use ofslam-shut valves instead of relief valves
In the UK, naturalgas is distributed ata gaugepressureof 70 bar and letdown
to 35 bar and thento 7 bar and 2 bar for customer use. Ifreliefvalves were used
to protect against failure of the let-down control system there would be noisy
discharges of gas in built-upareas and the releases might catchfire or explode.
Slam-shut ball valves, powered by high-pressure gas or bottled nitrogen, have
therefore been used for over 20 years instead of relief valves. They isolate the
high-pressure gas if the pressure downstream of the let-down valves rises
above a pre-set value.
The use ofinstrumented protective systemsinstead ofrelief valves has been
advocated within ICI since the early 1970s4 (for example, in a paper called
'Are safety valves old hat?'88), hut many engineers were at first reluctant to
use them. (They were, however, used to protect against explosions as relief
valves could not operate quickly enough.)Startingin 1985, a detailed study
was made of the use of slam-shut valves in place of relief valves on an
ammonia plant.Nine valves were needed. To achieve the reliability required it
was necessary to have two pressure switches, made by different manufacturers, sending electrical signals to a one-out-of-two voting system (that is,
either signal trips the valve shut). The output from the voting system triggers
three solenoid valves; two of them vent compressed air from the cylinder
which is holding the isolation valve open and the third sends air to the other
side of the piston. There is a spring on this side of the piston hut the air
provides diversity (Figure 3.21). The valvesare tested every three months and
the probability that any one will fail to operate is less than I in 1000 per
demand or about 1.5 X
per year. A single valve failing to operate is
to
cause
of
unlikely
rupture
equipment. The design was discussed with the
Health and Safety Executive which raised no objection.
Anyone considering a similar installation should consult the original
paper89 which gives details of the design, the testing arrangements, code
requirements, and so on. The paper shows Hazan at its best: the problem is
l0
138
clearly defined; goodquality data are available; the assumptions, including the
testing necessary, are clearly set out: the model of the process is realistic.
Whateverone'sreservations about the application ofHazan to risks from a site
as a whole, there is no reason to doubt the valueof studies such as this one (see
the last paragraph of Section 3.2, page 79).
3.8.6 Fermi estimates and electrical area classification
The physicistEnrico Fermi had a reputation for making quick numerical estimales of the answer to a problem or query90. For example, how many piano
tuners are therein the area coveredby my telephone directory? The population
is about a million, say250,00() households. If one in five owns a piano which is
tuned every five yearstherewill be about 50,000tunings per year. Ifeach tuner
tunesfive pianos per day for 250 days per year,or 1250 per year,there will be
about eight tuners. But many piano tuners are part-time
they tune other
139
l0
comply with fire precaution regulations to save about a fifthofa life per year;
'We don't think that'sgood value for money'. After the fire London Underground had brought London virtually to its kneesby attacking every escalator
and tearing out all the wood'. Intuitively, that had seemed a good idea but
calculations showed that this would reduce the probability of a serious escalator fire from once in six yearsto once in nine, while installation of sophisticated sensors and automatic sprinklers would reducethe probability to once in
a thousand years.
The managing director also praised QRA for compelling people to face the
setting of safety spending priorities and the valuation of human life and
accused media persons, politicians and others of publicly implying infinite
value foreach life. Yet motoring, flying,and indeed all activity,would ceaseif
we did not accepta trade-offbetween risk and benefit. Nevertheless, QRA did
not supersede judgementbut should lie alongsideit93.
Similar criticisms were made in a report produced for the Health and Safety
Executive following an incident in 1992 when two suspect briefcases were
found in a train. Seven trains were stopped in tunnels during a morning rushhour
as there were more trains on the line thantherewere stations to stop them at. It
took five hours to evacuate all 6000 passengers, 70 of whom were taken to
hospital with heat exhaustion. Smoke from a short circuit on one of the trains
added to the confusion and if it had developed into a fire the result might have
disastrous. The briefcases turned out to be harmless pieces oflost luggage.
The report says that closure and evacuation of stations may not alwaysbe
theright response. It recommends that railway staff are given training,similar
to that given to airportstaff, to helpthemassessthe seriousness of bombwarnings.On fireprevention thereport is morepositive. It saysthat as a result of the
actiontakensince 1987 the situationhas been transformed and fire prevention
shouldno longerclaim a lion's shareofresources. Instead QRAshouldbe used
to assess priorities. The existing legislation, based on regulations whichmust
he followed, shouldbe replaced by one basedon the quantitative assessment of
risk94.
If the inventoryin aplant or storage area is reduced, the maximum size ofa
leak will be less and so the consequences will be less but the probability of a
leak will not be changed. Reducing the numberof leak points such as valves,
drains, pumps, and soon, maybe more effective than reducing the inventory in
the existingequipment. If it is possible to take a vessel out of service, however,
thenthere will he fewerplacesfrom whichleaks can occur and both the probability and maximum size ofa leak will be lower52.
Is it betterto enclose equipmentthat handles chlorine in a building, so that
any leaks are confined, or would the money be better spent on reducing the
probability and/or the size of leaks?Detailed examination of a particularcase
showedthat containment was very expensive, had disadvantages and did not
greatlyreducethe risk95.
Liquefied petroleum gas (LPG) had to be piped across country for storage
in a well. Two options were considered: pumping at high pressure(about 100
bar) so that the LPG could go straight into the well, and pumpingat low pressure (35 bar), when another pump would be needed near the well. With the
secondoptionthere would be more sources of leaks and more leaks,as pumps
leak far more often than pipes. However, with the first option, if the pipe did
rupture the leak would be larger. As the pipelinefollowedopen country the
first option was chosen, as it was cheaper,but the decision would havebeen
different ifpeoplehad lived near the pipeline96.
3.8.9 Other examples
Lawley'1"2'3 hasdescribed three hazard analyses in detail, showing fault
treesand explainingthe derivation of each item ofdata used. The first11, which
is quoted by Lees, Chapter 9, analyses the precautions taken to prevent a series
of crystallizers overflowing, the second12 analyses the precautions taken to
prevent a pipeline gettingso cold that it becomes brittleand might fail, and the
third13 analyses the precautions takento prevent loss of level in the base of a
distillation column and discharge of high pressuregas intoa low pressure tank.
Reference 24 describes how the methods of Hazan have been applied to a
numberof other high-technology industries.
The subject ofthis chapter is discussed more fully in References 1317 and in
Lees. Chapter 9. References 16 and 17 deal particularly with risks to the public.
Reference 17 reviews the various targets or criteria that have been proposed.
There is an enormous literature on the philosophy of risk acceptability,
most of which deals with the more philosophical difficulties, and does not
offer much advice to the practitioner. References 1822 and 26 are typical of
these publications while References 23, 98 and 99 are more practical in their
approach.
142
(1) Failureto foresee all the hazards or all the ways in which a hazard can
arise (seeSection 3.5.9,page 113).
(2) Errorsin the logic (see Sections 3.5.4and3.6.5,pages 108 and 126).
(3) Failure to foresee that protection may not be fully effective becauseof
poor design (see Section 3.6.4, page 123) or becausetime of action has been
ignored.
(4) Design assumptions not correct; for example,less testing, more demands,
failures not random (seeSection 3.6.7, page 130), differentmode ofoperation
(seeSection 3.6.6, page 128).
(5) Common mode failures (seeSections 3.6.4and 3.6.5,pages 123 and 126).
(6) Wrong data (see Sections 3.6.13.6.3, pages 120121).
Some other errors are discussed in Chapter4.
References in Chapter 3
7.
8.
9.
10.
II.
12.
13.
14.
IS.
16.
17.
18.
19.
UK).
20. The RoyalSociety, 1992, TheAssessment and Perception of Risk (London, UK).
21. Schwing. R.C. and Albers, W.A. (eds), 1980. Societal Risk Assessmeni (Plenum
Press, NewYork, USAand London, UK).
22. The Royal Society, 1992, Risk:Analysis, Perception and Management Report
26. Risk Ana( .sis in the Process Industries Report ofthe International StudyGroup
on Risk Analysis, 1985 (Institution ofChemical Engineers. Rugby,UK).
27. KIet,, T.A., 1998. Process Plants: A Handbook jir Inherently Saft'r Design
(Taylor& Francis, Philadelphia, Pennsylvania, USA).
28. Mann, M.. 1986, Journal ofthe RoyalSociety ofArts, 134 (5358): 396.
29. Withers. J., 198$, MajorindustrialHazards,8597 (Gower, Aldershot. UK).
30. Health and Safety Executive, 1989, Risk Criteria
Land-usePlanning in the
Vicinityof MajorindustrialHazard.r (HMSO, London, UK).
3!. Barnes, M., 1988. The Hincklev PointPublic Inquiry: Report,Chapters 34 and 35
.tr
32. Health and Safety Executive, 1992, The Tolerability of Risk from Nuclear Power
Stations, 2nd edition (HMSO. London. UK).
144
UK).
54. Pithlado, R. and Turney. R. (eds), 1996, Risk Assessment in the Process'industries',
2nd edition,Chapter3 (Institution ofChemical Engineers, Rugby, UK).
55. US Department of the Environment, 1998. Operating Experience Weekly
Summary, No. 9826, 6 (Washington, DC, USA).
56. DNV, 1998, Techin/. Winter/Spring,3.
145
1-IAZOP
AND 1-IAZAN
58. Everley, M., 1996, Health & Safely at Work, 18 (10): 18.
59. Hoffman, R. and Schmidt. S.L., 1997, old Wine New Flasks, 49(Freeman, New
York, USA).
60. Hambly. E.C.. I May 1992. PreventingDisasters, RoyalinstitutionDiscourse.
61. Health and Safety Executive Nuclear Safety Division, 1995, NuclearSafety Newsletter.7: 3.
62. Health and Safety Executive, 1989, Quantified Rick Assessment: its Inputto Decision Making (HMSO, London, UK).
63. Cohen,A.V. and Pritchard, D.K., 1980, Comparative risk ofelectricity production
systems: a critical survey of the literature. HSE Research Paper
(1-IMSO,
London, UK).
64. Christen, P., Bohnenhlust, H. and Seitz, S., 1994. Proces.s SafrtyProgress, 13 (4):
234.
65. Interdepartmental Liaison Group on Risk Assessment. 1996. Use of Risk Assessnientin Government Departments (Health and Safety Executive, London, UK).
66. McQuaid. J.. 1995, Transaction.co/the institution of Chemical Engineers, Part B,
ProcessSafi'iv and 1:nvironmenlal Protection,73 (B4): S39.
67. Mortcn, A., 1995, Eliminating Risksfirthe Travelling Public (Royal Academy of
Engineering, London. UK).
68. Kletz. TA.. 1994. Learning from Accidents, 2nd edition, Chapter 20
(Butterworth-Heinemann, Oxford, UK).
69. Ro-Ro Ferries and the Safety of the Travelling Public, 1997 (Royal Academy of
Engineering, London. UK).
70. Philley, JO., 1992, Plant/Operations Progress, 11(4): 218.
71 . Health and Salety Commission, 1991, Major HazardAspects ofthe Transportof
DangerousSub,stance,s (HMSO, London, UK).
72. The Risks of Fuel Transport, 1982, Proceedings of a conference sponsored by
HazardousCargo Bulletin (Oyez. London, UK). Quoted by Clifton,ii., April
1984, The effect of wall thickness on the behaviourof aluminium and steel road
tankers carrying flammable liquidswhen they are engulfedin flames, ReportNo.
SRD R 29/. page6 (UKAEA).
73. Department of the Environment. 1995, A Guide to Risk Assessment and Risk
Management/orEnvironmental Protection(HMSO, l..ondon,UK).
74. Goats, G.C., 1996, The Safrtv & Health Practitioner, 14 (12): 20.
75. Withers, J., 1988. Major Industrial Hazards,208 (Gower. Aldershot, UK).
76. Tweeddale, H.M., 1992. Tran,saction,sof the Institution of Chemical Engineers,
PartB, Process .Sa/etv and Environmental Protection,70 (B2): 70.
77. Lindley, J., 1997, Ls,s Prevention Bulletin, No. 136: 7.
78. Inquiryinto the Fire on Heavy Goocl,s VehicleShuttle 7539 on 18 November /996,
1997 (1-IMSO. London, UK).
79. Rushton. AG.. 1997. private communication.
ii
146
81.
82.
83.
84.
85.
9(3): 169.
86. Hurst, NW.. Bellamy. Li. and Wright, M.S., 1992, Research models of safety
management of onshore majorhazards and their possible application to offshore
safety,SymposiumSeriesNo. J30, 129 (Institution ofChemical Engineers. Rugby,
UK).
87. Ellis, A.F. and Pokorny. B., 1992, Continuous and episodic risksThe assessment link, Center/icrChemical Process Scitetyinternational Conforence on Risk
Analysis, Human Factors and Human Reliability in Process Safety.
88. Kletz, T.A.,September 1974, Chemical Processing, 77.
89. McConnell. R.A., 1997. Process Sati'tv Progress, 16 (2): 61.
90. von Bayer, H.C.. 1988, The Sciences, 28 (5): 2.
91. Benjaminsen, J.M. and Wiechen. RH., 1968, HydrocarbonProcessing. 47: 121.
92. Fennell. D., 1988, investigation into the King's CrossUndergroundFire (HMSO,
London, UK).
93. Conway, A., 1992, Atom, No 420:9.
94. Appleton, B., 1992, TheAppleton Report (HMSO, London, UK).
95. Purdy, G. and Wasilewski, 1994,JournalofLo,rsPreventionin the Process indus-
99. Williams, D.R.. 1998, What isSafo?, The Risks ofLiving in a NuclearAge(Royal
Society of Chemistry, Cambridge, UK).
147
Appendix to Chapter 3
Belt and braces
eliminated completely.
The accident we wish to prevent is our trousers falling down and injuring
our self-esteem. Braces are liable to break and the protection they give is not
considered adequate. Assume that breakagethrough wear and tear is prevented
by regular inspection and replacementand that we are concernedonly with
failure due to inherent weaknesses or faults in manufacture which cannotbe
detected beforehand and which are randomevents.
Experience shows that, on average, each pair of braces breaks after ten
years' service. Experience also shows that belts fail in the same way and as
frequently as braces. Collapse of our trousersonce in ten years is not considered acceptable.
How often will a belt and braces fail together? Ifone failsthen itwill not be
detecteduntil the item is removed at the end ofthe day. Assuming it is worn for
sixteen hours per day, then, on average,every manis wearing a broken belt for
eight hours every ten yearsand brokenbracesfor eighthours every ten years.
The fractional deadtime (fdt) ofthe braces is
16
I
10
365
=0.000137
dead' is:
=2X
10
148
APPENDIX TO CHAPTER 3
= 0.1 x 'f2T2
0.1
365)
= 7.5 X l0/year
or oncein 133,000,000 years.
The calculations are approximate as they do not make any allowance for
commonmodefailures (see Sections3.6.4and 3.6.5, pages 123 and 126).
149
1-IAZOP
ANt) HAZAN
(I) The risk can be reduced to any desired level by duplication ofprotective
ISO
APPENDIX TO CHAPTER 3
Prince Alfred,
IDuke of Edinburgh
(Son 0f Queen
Victoria)
At the Clontarf Picnic Grounds
on the i2th lvtarch, 8G8, one
Henry
0'Farrell attempted to
151
A manager's guide
to hazard analysis
'Aristotle maintainedthat women havefewer
tee/h than men;althoughhe was twice marriedit
neveroccurredto himto verify this statement by
examining his wives' mouths.
BertrandRussell
4.1 Introduction
During the last 100 years managershave become increasingly dependenton
the adviceofexperts of all sorts. The days have long gone when one man
George Stephenson could survey and construct a railway line, design and
construct the engineand drive it on the firstjourney. Perhaps an unconscious
desire to be such an engineer is shown by those who display one of
Stephenson'sengineson their ties!
it is always temptingfor a busy person, whether he is managing a plant,
workshop or design team, to simplylook at the last pageof the expert's report
and accept his conclusion. The managercannot, as a rule, check the whole
report and, even given the time, such reports often contain incomprehensible
mathematics. This chapter is intended to help managers locateand check a few
and in the course of it the managershould ask the questions below. Nevertheless, on someoccasions a seniormanagermay be presented with an analysis as
the justification for a proposal to spend (or not spend) some money, and in
these cases he will be questioning a finished or draftreport.As a rule the first
issuesofHazanreports should be drafts.
The following, for ease of style, is addressed to managers. The firstpoint to
check is that the three questions in Section 3.3 (page 80) havebeen answered.
Does the
report:
Say how often
the incident will occur?
l0'.
amateuranalysts.
153
l0.
'54
5C. The analyst then fails to include frozen pipelines in the list of initiating
events which can cause a pipeline to block. Similarly, an analystmaydecide to
estimate the leak rate from a circulating gas system in the eventofpipe failure.
The analyst asks for the flow rate and is told that it is, say, 10,000 m3/h. He
does not ask and is not told that the total amount of gas in the system is only
1000 m3. It is. ofcourse, an advantage to employ analysts with experience of
design and/or production.
In checking an analysis, the managershouldtherefore ask:
Haveany unusual propertiesof the process materials beenconsidered?
Haveany limitations on flow rates, heat inputs, etc, providedby the inventory orequipmentbeenconsidered?
Have alternative methods of operation, such as regeneration of catalyst
beds, been considered?
Havestart-upand shutdown been considered?
Does automatic protection protect against all demands or only someofthem?
Has the model been discussed with the maintenance organization (particularly the instrument maintenance organization) as well as the operating team?
If the model is buried in a computer program it will not be transparent and
the managerwill haveto dig deepwhenhe questions the analyst.
An example of a sophisticated error in the model is provided by the
anti-growth movementand theircalculations of impending doom:
'In eflect, what the Club of Rome report did was to assume that all "had.s
such as pollution, demand for ftod and raw materials, and so on, would
increase exponentially tar everand ever, and all "goods ", suchas techniques
to reducepollutionper unil o/ output, or supplies oftbod and raw materials,
could only in 'i-ease by finite a/flaunts.
'Clearly, however generous are these finite amounts, it does not need a
computerto show that, one day, the "hads" must exceedthe 'goods
in the words of Lord Ashhyt "if we fed doom-laden assumptions
into computers it is notsurprisingthat theypredict doom
Thomas Malthus made the same error in 1816 in his book Essay on the
Principle ofPopulation. He forecastthat the production of food would rise
arithmetically while the population would rise geometrically. If he had been
correctEuropeans would havebegun starving to deathin a few generations. In
fact, agricultural production increased substantially as a result ofthe discovery
by Justus von Liebig that minerals were essential to plantgrowth21.
The manager should look out for features in a model which make the
answers inevitable, regardless of the data (seealso Section 6.6, page 199).
155
C)
Cs
C)
C)
C
5)
C)
C)
55
5C
4C
C)
5)
55
5)
Cs
Level ofexposure
(b) People with high blood pressureare more likely to havea stroke thanthose
with lower blood pressure. Nevertheless most strokes occur to people with
bloodpressures in the middlerange as thereare many more peoplewith blood
pressure in this range. The most successful way to reduce the number of
strokes would be to lowerthe blood pressure of the population as a whole, to
move from curve A to curve B in Figure 4.1. A 5% reduction in bloodpressure
would prevent 75,000 strokes per year (30% of the total) while targetingthe
5% of the population with the highest bloodpressurewould prevent only half
as many''. (This illustrates the dilemma discussed in Section 3.4.6 on page99:
shouldwe try to save the most livesper million pounds spentor shouldwe try
to protectthepeopleat greatest risk?)
It is possible that similar arguments might apply to, say, corrosion
prevention.
(c) Students at the FrenchGrande Ecoles cost threetimes as much, per term,
as those at French universities. But the Grand Ecoles have virtually no
drop-outs while the universities havea 60% drop-out ratet2.
(d) Railway companies are under pressure to improve timekeeping. As a
result connections leavebeforethe connecting train has arrivedand passengers
reach their destination later than if the connection had waited. The correct
parameter is not the lateness ofthe trains but the lateness of the passengers.
1981/82
% change
167
210
+26
84
67
20
10
16
+60
12,48!
14,643
+ 17
5674
4414
22
Mi!esat>lOOmph
689
1157
+68
157
page 151.
In general, ask what methods havebeen used to identify all the hazards. Has
a Hazop been carriedout? Ifnot, what other methods have been used to identify hazards?
158
temperature rises.
159
if assumptions
satisfactory.
more important15.
In some cases it may be possible to assess the probability that an assumption
will cease to be true. For example, in my firstpaper on Hazant6I assumed that
for 10% of the time the nitrogen blanketing on a storage tank would not he in
operation. At the time I had experience of a factory in which moribund nitrogen
blanketing equipment had been brought back into operation. The operators (and
some more senior people) wondered why this was necessary as there had been
no incidents that nitrogen blanketing could haveprevented. Standards ofcompliance were therefore poor. (This changed after an explosion in a storage tank,
described elsewhere'7.) Actions imposed by authority rather than conviction
soonceasewhen the boss moves or loses interest. When aHazanassumes higher
standards of management than have been usual, perhaps it should assess the
probability that theywill he maintained (seeSection 3.S.4, page 137).
Working closely with clients, a hazard analystmay 'go native' and accept
uncritically their estimates of the adequacy of existingand proposed procedural safeguards.
4.6 Data
Errors can arisebecausedata are inapplicable or misinterpreted
(see Sections
pages 120121). The managershould therefore look at the data
used to see if they seem about right. For instruments the data are well
3.6.13.6.3,
160
Two examples
ofinapplicable data:
byjudgement.
Estimated by comparison with previous cases for which fault tree assessments havebeen made.
be independ-
33%
Dustand ozone 7%
Surface (average) 5% (it can vary from 3% foroceans to 80% forsnowfields)
He then continues18:
'it really isn't very sensible to make approximations like those and then to
perthrna highly complicated computercalculation, while claiming the arithmeticalaccuracyofthe computeras the standard for the whole investigation.
Once the precise detail of the Earth's reflectivity has beenlost, the investigation has been so degradedthat meritcannot be recoveredby attention toarithJfleliC.'
The Victorian biologist, Thomas Henry Huxley, said much the same in
l876:
what you gel out depends on what you put in; and as the grandest mill in
theworld will not extractwheatflour from peascods, sopages offormulaewill
not get a frtinite result outof loosedata.'
It is important to distinguish between those data that affect the final result
4.7 Humanreliability
systems.
162
Figure 4.2
IO
Is the result ofthe Hazanin accordance with experience and common sense? If
not the Hazan must be wrong. This is obvious, of course, and would not be
worth saying if analysts had not, on a number of occasions, been so carried
away by enthusiasm for their calculations that they forgot (like Aristotle)to
comparethem with experience. For example, a numberof theoretical studies
of chlorine and ammonia releaseshave forecastlarge numbersof casualties.
When releaseshave actuallyoccurred, the casualties have been few. Yet the
studies do not say this. It was alwaysrealizedthat casualties could be high if
164
Should the managers and the designers call in experts to carry out hazardanal-
yses for them (a closed shop policy) or should managers and designers make
their own analyses (an open shoppolicy)? To quote Kelly et a13:
As the level of detail required by the reliabilityanalyst increases, so do his
demands on the designer's time and experience. At somepoint it becomes
more effective to train the designer in reliability techniques than to train the
reliability analyst in design techniques.
Hazan is not so esoteric that it can be practised only by an eliteband of the
initiated. Engineers engaged mainly in design or operations can be trainedto
apply it. It should be our long-term objective for design teams tocarry out their
ownstudies. The experts in Hazan should train,check,helpand encourage, but
not necessarily do all the work, They shouldbe sharersofthe tools,not keepers
of the tools. At the same time we should remember the words of Thomas
Hobbes(15881679):
in Arithmetic, unpractisedmenmust, and Professorsthemselves ,nay often,
165
Old styleadviser
Newstyle adviser
Waitsfor requests
Dealswithadhoc problems
Does all the work
Uses jargon
Issuesa report
Works in his own department
Sharerofthe tools
technology
Works on technically challenging
projects
assumptions uncritically
References in Chapter 4
2.
3.
4.
5.
6.
7.
166
8.
9.
10.
11.
12.
13.
14.
15.
16.
17.
167
Objections to
Hazop and Hazan
She hadonemajor jailing in thatshe tended to
quantify benefits. Thus areas ofendeavour which
This chapter discusses some of the objections that have been raised to the
methods discussed in Chapters 2 and 3, mainlyChapter3.
doctorbecausewe do not havetime to do so. If we waituntil we become seriouslyill we may losemore time in theend. Experiencehas shown that the time
spent in carryingout a Hazop. thoughit maydelaycompletion of the design,is
well repaidin asmootherstart-up, earlierachievement of flowsheet output and
trouble-free operation. One survey of four Hazop studies showed that apart
from an increase in safetythe financial savings werebetweenfive and 80 times
the additional cost29.
A thirdobjection, that 'goodpeople' are a substitutefor Hazop,is discussed
in Section 2.4.4, page 30.
One company has suggested that to save time a Hazop shouldlook only for
departures from its design standards5. This maybe acceptable if the process is
a familiar one in which all hazardshavebeenrecognized and allowed forbut if
we arc innovating, and thereis usually someinnovation, new hazardsmay not
be recognized. Also, in most companies, standards lag behindthe latest informationand ideas.
1)escription
of risk
Hazard
category
1
Minor
Appreciable
Major
Severe
Total
<3000
<t30.00()
<e300.000
<3M
destruction
>53NI
Minor
injuriesonly
Injuries
I in 10 chance
Fatality
personnel
Multiple
fatalities
Works
Damage
None
None
Minor
Appreciable
Severe
Business
Business
loss
None
None
Minor
Severe
Total loss of
Damage
Noire
Very ninor
Minor
Appreciable
Severe
Eltectson
NI nor
Sortie
I in Ill
Fatality
people
None
Ismells
Reaction
None/mild
Plant
Danrage
Eflect on
Public
ofa fatality
business
hospitalrtalion
chanceof
public
fatality
Minorlocal
Considerable
outcry
local and
nationalpress
Severe local
and
considerable
rational
reaction
Relative
guide
Severe
national
reaction
press
reaction
(pressure to
strip
business)
ll)'
l0
l0
l0
I/yr
1/lO yrs
1/101)yrs
1/1000yrs
Ill))1 yrs
Irequcircy if
occurrence
Typical
judgrrieirtal
values
a
fir
plant/snaIl
works
lvB These typical comparative Irguresaregiven Or illustration rindshould not he takenasapplicableto all
situationsnor takento iridiculeabsolute levelsot acceptability.The cash figures have been increased in Iinc
with inflation.
cateoory
(see Table
Smaller()
Same (=)
Greater(+)
Uncertain (U)
5.1)
frcquency/
potential raised
to B at team's
discretion
hazard as those
below A hut if
lower end of
frequency/
potential cculd
be lowered to B
at team's
discretion
Frequency
estimates should
not be difficult
at this category
may be a lack of
fundamental
knowledge
which requires
research
A/B at team's
Major hazard
discretion. Such
potential should
he better
understood
4 and 5
B/C at team's
B, but can be
discretion
raised to A at
team's
discretion
Major hazard
Such potential
should he better
understood
used to derive prioritiesbetween A, the highest, and D, the lowest6. Note that
this is not a technique for rapid Hazan but merely a technique for helping us
decide which hazards shouldbe analysed first.
Somewhat similar techniques have been devised for the rapid assessmentof
less serious hazards when the size of the risk makes a full Hazan unnecessary
(or the sparsity of data makes it impossible)7'332. Table 5.3, from a draft
standard for safety-related systems52, is a good example of such systems. The
'risk classes' are the same as those shown in Figure 3.2 (page 86). The
numbers in part (c) of the table (page 174) are not part of the standard and are
merely suggestions. They maydiffer for different industries.
172
Table 5.3(a) A semi-quantitative method for the classification of risks: risk classes
(FromReference 52)
Risk class
Interpretation
Note
Class I
Intolerable
Unacceptable region
Class II
ALARPregion
Class Ill
Class IV
Broadly acceptable
Not quitenegligible
Frequency
Catastrophic
Consequence
Critical
Marginal
Negligible
Frequent
Probable
Occasional
Remote
Improbable
III
III
IV
IV
Incredible
IV
TV
IV
IV
Extracts from draft IEC 6I508.-l: 1998 are reproduced with the permission of
BSI underlicence number PD\ 1999 0567. Complete copies ofthe standard can
be obtained by post from BSI Customer Services, 389 Chiswick HighRoad,
London W4 4AL,UK.
173
Manypeople killed
Marginal
Broadly acceptable
Minorinjuries only
Frequency
Frequent
Probable
Occasional
Remote
Improbable
Incredible
100010,000 years
10,000100,000 years
Table 5.3(c) is not part ofthe draftinternational standard IEC 61508I (Reference
52) from whichTables 5.3(a) and (b) havebeentaken.It is merely apossible
interpretation oftheconsequence and frequencies categories shown in (a) and (b).
The 'critical' definition is consistent withFigure3.2 and Table3.1(b). assuming the
people killedare employees and the factory is small. Negligible' in the original has
beenchanged to 'broadlyacceptable'for consistency with Figure 3.2.
Descriptor
Score
if
if
otheropenings or is suitablyprotected to
allow at leasthalfan houror more for
escape
if
if
4
3
2
without reversing
4
2
I
site
people.
175
effects with thosewhich produce long-term effects. The results indicate that the
allocation of resources between the two sorts ofhazards is not out by more than
an orderof magnitude. This may not sound very good but is not bad for problems ofresource allocation. As shown in Section 3.4.7 (page 100), the financial
resources spenton saving a life can vary over a range of a million to one.
As an example consider ionizing radiation. Ifwe havemoreresources available for saving life, should we spend them on preventing accidents which kill
people quickly, or on reducing exposureto radiation?
The International Committee on Radiological Protection recommend that
the maximum dose for an employee should not exceed 50 millisieverts
(mSv)/year. Formany yearsit was believed that this would givea risk ofdeath
of 5 x l0 per year or a FAR of 25 (see Section 3.4.1, page 87). Very few
people are actually exposed to the maximum dose but nevertheless it does
seem rather high when we bear in mind that the average FAR for manufacturing industry in the UK is about 1. Much of the UK chemical industry
regards 2 as an upper level for all chemical risks (Section 3.4.1, page 89), but
workers in the chemical industry are also exposed to health risks. People
exposed to ionizing radiation are exposed to other risks as well.
However, the radioactivity dose limits tare not to be taken as a target, but
rather as the lower limit of values that are not acceptable ... a properly
managed practiceshouldneverexposeworkersor the public to anywhere near
the limit'25. The figure of 50 mSv should be compared with the maximum
tolerable risk (FAR 50 or
per year) shown in Figure 3.2 (page 86) and
Table3.1 (pages 8889).
There is nowevidencethat the risk from radiationmay be as much as three
times higherthanwas originally thought, butto compensate for this the nuclear
industry in the UK has set 15 mSv/year as the maximum tolerable level.
Average doses are now about a tenth of this figure and only a few employees
are exposed to 10 mSv.
Similarcomparisons are madein Reference 3 for coal dust,asbestos, chemicals as a wholeand industry as a whole.
In considering these comparisons, remember that acute risks such as fires,
explosions, falls and some toxic chemicals kill peopleimmediately while radiation (and many toxic chemicals) kill them 2040 years later. Many people
argue that a higherdeath rate from these long-term risks is therefore tolerable.
On the other hand industrial disease may produce many years of illness and
reduced quality of life followedby death at the time of retirement whenone is
looking forward to well-earned leisure. Perhaps these effectscan be offset and
l0
176
Pltblado arguesthat to improve the qualityoftheir hazardanalyses consultants should expend more effort on the activities that come before and afterwards(Figure 5.1, page 178) rather than on the Hazan itself35.
5.2.8 The results do not agree with thoseobtained by other
methods of calculation
Relativeeffortexpended
TOP
Establish clientneeds
Educate clientin
details ofstudy Collect relevantinformation
Identify whatcan go wrong
Undertake consequence
calculations
MIDDLE
Estimate failure
ftequencies
Calculate risk results
TAIL
Investigate mitigation
alternatives
Developcost-effective
solutions
Communicateresults
effectively
Past QRAstudies
Betterbalanced QRAstudy
1-IAZAN
someone'sjudgement? If you can put your feelingsinto words you are more
likelyto convinceothers.
What are the alternative methods you can use if you decide to ignore a
Hazan? The first is to rely entirely on gut feeling'. Unfortunately different
guts feel differently and a dialogueis difficult. Numerical methods do allow a
dialogueto take place. If one person saysthat risk A is high and anotherthat it
is not, a dialogue is difficult. If we havea scale formeasuring risksa dialogue
becomes possible (seeSection 2.9, page50).
In making a decision in matters that affectthe public a managermust take
public opinion into account. Ultimately, in a democracy, governments mustact
in accordance with public opinion. They may have to take action that their own
judgementtells them is incorrect. This is part of the democratic process. The
advocates of Hazan do not seek an alternative to public opinion; they seek to
persuade it. They look for an alternative to a policy of giving the most to those
whoshoutthe loudest. Publicopinionshouldnot be confused with the opinion
of the mediaor of self-appointed pressuregroups.
As an example of the way that governments feel compelled to take other
factors into account, consider the introduction ofunleaded petrol.According to
Everest36. whowas in the Department of the Environment (DoE) at the time:
179
A more philosophical objection to Hazanis that deaths from industrial accidents,smoking, sportand contaminants in food are not the same and therefore
cannot be compared. However, comparing differentthings is what managementis about. Resources are not unlimited and we have to decide how to allocate them between safety, protection of the environment, improving working
conditions, increasing the wealth ofthe community, and so on. Information on
the relativesizes ofvariousrisksand the costsof removing themwill helpus to
makebetterdecisions. Of course, we alsohaveto takeinto account the public's
aversion to differentrisks, as discussedin Section 5.3. Andwhile deathsfrom
differentcauses are undoubtedly different, theyare probablyless differentthan
most of the alternatives we have to choose between, at work and in everyday
life. We are just as dead whichever way we die.
Some writers, notably Cotgrove4, havesuggested that much of the opposition to Hazan comes from people who have a differentparadigm or set of
valuesto thosewhoadvocatetechnological advance. They are moreconcerned
with protection of the environment, for example, than with output or efficiency.They opposethe values oftechnologists rather thanthe systematic allocation of resources but the two are linked in their minds. In fact, though it
shows that sonic risksare trivialand hardly worth botheringabout.Hazanhas
probably resulted in a largeincrease in expenditure on safety.
Accountants try to quantify everything financially and thus, according to
Malpasand Watson26,overlookwhat they call 'Options for the future' that
is, expenditure which does not show a good rate of return but nevertheless
makes it possible to pursuepromising linesofdevelopment.
5.2.11 Changes in technical objections to Hazan
The Study Group that prepared the firsteditionwas set up at the suggestion
of people from Germany and Holland who were opposedto quantitative risk
assessment (QRA) or rather to the way it was being misused by the authorities in theircountries andwanted to kill it. They found,to theirsurprise, that
people from UK industry were in favour. The report concluded that QRA
could help or hinder, depending on the legislative environment.
The new (1996) edition is very different. Since 1985. churchmen have
become more liberal about sex and those engineers who had doubts about
QRA havebecomemore willingto accept it. The new edition contains lots of
soundadviceon how to carry out risk assessment it is twice as long and
has none of the 'be careful, just stick a toe in the water' approach of the old
edition. Problems and limitations are described but not carriedto excess.
A number ofwritershave analysed the factorsthat determinethe public's attitude to risks and the following is based on the work of Lee10, Slovic et
Sandman12and Kauffman18.The probability of an incident is, of course, one
of the factors that the public take into account but not theonly one, and even
here the public's knowledge of the relative size of different probabilities is
often far removed from their actual sizes. Their knowledge of the numbers
killed by differenthazards is not too far out but their knowledge of relative
rates bears little relation to reality. For example,the risk from pesticideresidues in food,a subject ofpopularconcern, is far less thanthe risk from natural
poisons. Otherfactors that affectthe public's attitude are discussed below.
5.3.1 Voluntary or imposed?
We acceptwithoutcomplaintrisks such as smoking or rock-climbing that we
choose to follow, but object to risks such as those from industry that are
imposed on us withoutour permission. For this reason many writersbelieveit
may be counterproductive to use cigarettes as a unit of risk (Figure 5.2, page
182).
Figure 5.2
We acceptmore readily natural risks such as those from floods, storms, radon
and natural foods and drugs than man-made risks such as those from industry,
nuclearpower stations, pesticides, food additives and synthetic drugs. This is
oneofthe less defensible ofthe public'sviews,in part,it isdue to the mistaken
beliefthat little can he done about Acts of God,as they are sometimes called;
in fact, floods,droughts and faminesare dueto mismanagement ratherthan too
much or too little rain while the effects of earthquakes, volcanos and hurricanes areoften magnified by mismanagement'3. In part,the public's attitude is
due to an equallymistaken belief that natural foods and drugs are alwaysgood
for us. In fact, the average US diet contains about 1.5 g/day of natural pesticides but only about 0.15 mg/day (10,000 times less) of synthetic pesticides.
Many ofthe naturalpesticides present in food would neverbe approvedifthey
were tested in the same way as synthetic pesticides4I5. Similarly, natural
drugs can be sold without goingthrough therigoroustestingnecessary fornew
synthetic drugs. Plants contain natural pesticides becausethey cannot pull up
theirrootsand run awayorfightbackwith tooth and claw; their only defenceis
to poison(or prick) their enemies.
182
I-IAZAN
5.3.4 Familiarity
We readily acceptfamiliar risks such as those of driving, long-established
drugs such as aspirin and traditional industries such as farming, but are less
ready to acceptunfamiliar riskssuch as thoseofnew drugs and nuclearpower.
We know the size of familiarhazards (Figure 5.3). Road accidents kill about
5000peopleperyear in the UK. This is terrible but at least the extentis known;
we are confident that the number killed this year will not be 10,000. In
contrast, although we may agree that nuclearpowerand the chemical industry
will probably kill no-one this year, we donotfeel sure therewill not be another
Bhopal or anotherChernobyl.
5.3.5 Experience
we have personal experience ofa risk, we are waryof it in future. Ifshellfish,
say, have made us ill we may avoid them in the future even though we know
that we are unlikely to he offered another contaminated batch. Similarly, ifthe
local factory has caused pollution in the past we tendnot to believe assurances
that all will be well in the future.
If
Figure 5.3
183
5.3.6 Dread
Heart diseasekilts about twice as many people as cancer but nevertheless
many peoplewould support the expenditure ofgreatersums on cancer prevention as cancer inspires so much more dread. This is not a decision made in
ignorance as almost every family has experience of both.
5.3.7 I benefit
We acceptthe risk of driving becausethe benefits of the car are clear and
obvious. The benefitsof the chemical industry are not obvious. All it seems to
do is to produceunpleasant chemicals with unpronounceable namesin orderto
increase its sordid profits. At best, it provides employment and exports. Most
people do not realizethat it providesthe essentialsfora standard oflivingthat
has vastly improved the length and quality oflife.
We acceptrisks from which we earn a living or deriveother benefits. When
thegovernment was considering whether ornot to allow British NuclearFuels'
new plant for reprocessing spent nuclearfuel (THORP) to start up, the leader
of the local Council was reportedas saying, 'A delay in thestart of THORP
will lead to 20% unemployment in this area by next December. That would
have a devastating effect. The effects on unemployment, on the health of our
people, their morale, the crimerate, dietaryhabits, infant mortality, and so on,
would likewise be devastating'40.
5.3.8 Morality
Far more people are killedby cars than are murdered, but murderis still less
acceptable. We would be outraged if the police stopped trying to catch
murderers, or child abusers, and looked for dangerous drivers instead, even if
more lives would be savedin that way.
5.3.9 Numbers moreimportantthan rate
How many people could be killed? To the public the number of people that
could be killed is more important than the probability that they will be killed.
The airlines realized 20 or more years ago that as the number of flights
increased the numberof accidents could not be allowedto increase in proportion or there would be a public outcry. They found it possible to decrease the
rate so that the number remainedroughly constant. We find the death of 10
people at a time less acceptablethan the death of one person per year for 10
years (seeSection 3.4.3, page 90).
Similarly, the public seem to believe that the consequence of an action
measures the degree ofnegligence. It does not,if a car is parked on ahill without
the brakes on, and rolls down, the negligence is the same whatever the result.
184
5.3.10 Associations
between
We still meet peoplewho say that smoking cannotcause lung cancer because
their grandfather smoked heavily and lived to be 90. Perhaps they would
understand us betterif, instead of usinglong words, we saidthat someeffects
alwaysoccur, while othersoccur sometimes, that somethingsmayhappen, but
others will happen (seeSection 3.3, page 80).
185
OBJECTIONS TO
5.3.15 Conclusion
Sandman admitsthat real peopledie becausewe are more concerned about the
factors discussed herethan about the actual probability ofbeing killed. But, he
adds, we also valuefairness, moral values and individual freedom, sometimes
more thanlife itself.
It is not sufficient therefore to present the facts and hope that in time the
public will acceptthem; the power ofa beliefdoes not depend on its truth. We
should also try to answer the public'sconcerns, rational and irrational. Unfortunately most of these concerns tend to makethe man in the street oppose the
chemical and nuclearindustries (the risks are imposed, not under his control,
man-made, unfamiliar and dreaded;past experience has been unpleasant; the
industries do not obviously benefithim: and the spokesmen for the industries
are often outsiders) and this is reinforcedby the media's desire for disaster,
their daily bread (every reporterhas Jeremiah as a middle name). There is no
easy solution butthe improvement in the image ofBritish NuclearFuelsduring
the I990s shows what can be done, thoughit still has a long way to go.
'Individualism used to mean the right to act as one wished providedit did not
harm othersand the right tohold views radicallyat odds with the consensus
188
It is now asserted as the right to decide what is and is not true ... a steady
increase in expenditure on tbrmal education has been accompanied by a
decline in faith in the possibility ofauthoritativeknowledge ... The inevitable
consequence is relativism, not just in matters of behaviour we have long
passed thepoint ofbeing abletoagree on howwe shouldbehave but nowin
the realms of knowledge. The judgeinentsof any group of experts can be
dismissed with
flippant assertion of partiality. "they would say that,
wouldn't they? ". Any layman who can read can claim to understand the
origins ofthe world or the causesofdepression ... the very idea ofauthoritative knowledge is underattack in such spheresas science and medicine, where
it seems obvious that there is a vast gulf between the expert and the lay
person.'
Ultimately, however, ifthe Health and Safety Executive and other experts
cannotconvincethe public that a risk is negligible, they will haveto remove or
reduce it. This, after all, is democracy in action. In 1983 Fremlin24 wrote,
'When little children are afraid of the dark, you put a light there, eventhough
you know there is nothingto be afraid of. It would therefore be sensible ifthe
Government insisted now on getting the amounts lof radioactive materiall
dispersed from Windscale reduced, not because this is faintly necessary to
reducecancer,but in orderto show peoplethat theycare, and to put their minds
at rest'. Since then the Governmenthas done just this.
Finally, let us remember that the experts do not always agree with each
other (seeFigure 6.2, page201) and are not alwaysright(see Section 4.3, page
154). They are, however, more likely to be right than thosewhose knowledge
comesonly from the newspapers and television.
Further reading
189
13 August:
14 August:
15 August:
18 August:
23 August:
11 September:
the true figure, no-one seems to havecared very much or commented on the
discrepancies. Why are peopleso much more concerned about chemical engineering disasters thancivil engineering disasters? Perhaps becausedams have
pleasant associations, reminding us of summer days in the country, but chemical factoriesdo not.
References in Chapter 5
I.
2.
3.
4.
5.
6.
7.
S.
9.
10.
II.
12.
13.
14.
IS.
190
9(3): 169.
21. Hurst, N.W., Bellamy, Li., Geyer, T.A.W. and Astley, iA., 1991, Journal of
HazardousMaterials,26: 159.
22. Hurst, N.W., Bellamy, L.J. and Wright,M.S., 1992, Research models of safety
management of onshoremajor hazardsand their possible application to offshore
safety,SymposiumSeriesNo. 130, 129 (Institution of Chemical Engineers, Rugby.
23.
24.
25.
26.
27.
28.
29.
30.
31.
32.
33.
34.
35.
36.
37.
38.
39.
40.
UK).
Joko.simovic,V. and Vesely, W.E., 1980, Reliability Engineering, 1(1): 72.
Fremlin, J.H., 21 November1983, quoted in The Daily Telegraph.
Kovan, R. and Conway, A., 1991, Atom, No. 416: 20.
Malpas, R. and Watson, S.J.J., 1991, Technology and Wealth Creation (Fellowship ofEngineering, London, UK).
Luton Industrial College, 1991, Energy A Matter ofLife and Death (Merlin
Books, Braunton, Devon, UK).
Schneider, S., 1996, quoted in The GlobalWarming Debate,editedby R. Bate, 253
(European Science & Environment Forum, London, UK).
Sanders, R., 1998, A view of what we do: Making safetysecond nature,Mary Kay
O'ConnorProcess Saft'ty CenterSymposium: Beyond RegulatoryCompliance
Making SafetySecondNature, College Station, TX, USA, 3031 March.
Tweeddale, H.M.. Cameron. R.F. and Sylvester, S.S., 1992, Journal of Loss
Preventionin the Processindustries,5 (5): 279.
Gillett, i.E., 1996, Hazard Study and Risk Assessment in the Pharmaceutical
Industry,Appendix 5 (Interpharm Press, Buffalo Grove, Illinois, USA).
Maddison, T.E. cud, 1995, LossPrevention Bulletin, No. 125: 11.
Health and Safety Executive, 1996, Dispensing Petrol (HSE Books, Sudbury,
UK).
Griffiths, R.F.. 1994, Journal of LossPrevention in the Process industries,7 (5):
432.
Pithlado, R., 1994, Journal of Loss Preventionin the Process industries,7 (4):
360.
Everest, D., 1997. in What Risk?Science, Politicsand Human Health,editedby R.
Bate, 242 and 253 (Butterworth-Heinemann, Oxford, UK).
Risk Analysis in the ProcessIndustries Reportofthe InternationalStudyGroup
on RiskAnalysis, 1985 (Institution of Chemical EngineersRugby, UK).
Pitblado, R. and Turney, R. (eds), 1996, Risk Asse,s.sment in the ProcessIndustries,
2nd edition (Institution of Chemical Engineers, Rugby, UK).
Kletz,T.A., 1985. The Chemical Engineer,No.414: 72.
Sellafield Newsletter, 22 December 1992.
191
41. Sandman. P.M., 1996, in What Risk? Science, PoliticsandHuman Health, edited
by R. Bate,276 (Butterworth-Heinemann, Oxford. UK).
42. Peschel, R. and E., 28 April 1990, British MedicalJournal, 1145.
43. Dow Chemical Company, undated, Lifein the Balance (DowChemical Company,
Midland, Michigan, USA).
44. BritishMedical Association, 1987, Living with Risk, 141 (Wiley, Chichester, UK).
45. Darymple, T., 1998, MassLisieria, 132 (Deutsch, London,UK).
46. Bate, R. (ed), 1996, What Risk? Science, Politics and Human Health, Preface
(Butterworth-Heinemann, Oxford, UK).
51. Bruce, S.. 1995, Religion in Britain, 122 (Oxford University Press, Oxford,UK).
52. International Electrotechnical Commission, 1998, Draft international Standard
IEC 6/508i: Functional Safety of Electrical/Electronic/Programmable
Electronic Safety-related Systems (IEC, Geneva, Switzerland).
192
Appendix to Chapter 5
Limitations on the application
of quantitative methods to
railway travel
The following letter appeared in Reliahilit}'Engineering, 1981, 2: 77. It shows
generally applicable method for calculating the time required for a railway
journey and the probable starting and finishing times. However, experience
over a number of years has shown that this optimism is not justified and the
limitations of the method are such as to render it unsuitable for widespread
application, though it may be useful in a few limitedareas.
l'he serious limitations on the use of railway timetables result from the
following well-established facts:
The answers obtainedassume that all possible routes between the starting
and finishingpoints are known and have therefore been investigated. In fact,
this is often not the case and routes which have not been thought of provide
possible pathways, particularly under abnormal operating conditions such as
Sundays, BankHolidays and nights.
The timetable is an expression of intention or, at the best, of past performance, ratherthan offuture performance. It is not unknown for trains to fail to
run or to run late.
The railways are subjectto human error on the part ofthe drivers, signalmen
and station staff. Numerous detailed reports, over many years, have established this beyond reasonable doubt. There is no satisfactory way of making
allowance forthese errors in estimating journey times, despitethe considerable
effort expended in recent yearson the studyofhuman reliability.
The complexity of the timetables is such that extensive, detailed and
time-consuming studies are necessary to evaluatejourney times. The necessary resources of manpowerand time are rarely available.
Timetable data are usually shown to a degreeofaccuracythat is untrueand
misleading. Times of arrival and departure are shown to the nearest minutefor
193
journeysthat may take 10 hours or more. Users are misled into thinking that a
degreeof accuracy is attainable that is not, in fact, the case.
It is clear that the use of railway timetables for the estimation of journey
durations and arrival and departure times cannot be recommended and that
they should not be used for this purposejust turn up at the station and hope
therewill be a train.
194
confidence limits
Errorsusing inadequaledata are muchless
than thoseusing no data atall.
Charles Babbage(17921871)
(pages 120121) and 4.6 (page 160). This section provides a few notes on
sources of data.
The bestsource ofdata, especially forinstruments and electrical equipment,
is the Data Bank operated for the SRD Association by AEA Technology,
Warrington, UK. Member organizations pay an annual subscription and are
expected to contribute data. In return theyhave access to the data providedby
AEA Technology and by other subscribers. The American Institute of Chemical Engineers (AIChE) has published a book of data4 and guidelines on its
useu. Dhillon and Viswanath5 havelisted 367 sources ofdata.
Dataare discussed by Lees. Chapter7, Section 20. while his Appendix 14
lists much published data and gives references to other sources. References 6
and 7 also providesomedata and Reference 6 has a chapteron data banks.
Many large companies have produced their own data books which summarize data obtained from AEA Technology, the literature and internal sources.
Unfortunately these are often misused. The intention of the compilers is that a
readerwill look in the data book to see ifthereare any data on, say, relief valve
failure rates and will then consult the original references for details. Unfortunately many users take a figurefrom the data book, do notbotherto consult the
original source and maymiss important qualifications.
For example, there is a well-known report on pressure vessel failures1
which gives a 'catastrophic failure rate' of 4.2 X
per vessel-year. It
defines 'catastrophic failure' as destructionof the vessel or component, or a
failure so severeas to necessitate major repairs or replacement'. The definition
thus includes defects which are found during inspection or test and do not
result in a leak. The figureis often quoted withoutthe definition. Readers who
do not take the troubleto referto the original paper assumethat 'catastrophic'
meansdestruction in servicewith releaseofthe contents, and are misled.
Companies which collect and publish data may he more responsible and
l0
195
If there are many components in a system and many of them have never
failed, it is straining credulityto assumethat they will all fail next year.
Sometimes no failure data are available and an estimate has to be supplied
by an experienced person. Somepeople maythen ask, 'If we have to estimate
thefailure data, why not estimate the answerto the whole problem?'.
If we break problems down into their component parts, answering them
with factswhenpossible and with opinion only whenno factsare available, we
are more likely to get a correct answer than ifwe try to guess the answer to the
wholeproblem.
Fault tree calculations are not 'series' calculations in which a 10% error in
the input is carried through to the output. They are 'parallel' calculations in
which differentstreams are combined and most errors in the data have little
effecton the final answer. If we put 10% impurityin the water entering a long
pipelinewithout branches, therewill be 10% impurity in the output, However,
if we put 10% impurityin one of the streams feeding a river, there will not be
10% impurity in the water reaching the sea.
uncommon (see Section 3.8, especially 3.8.6 on page 139). Estimates are
usually conservative as analysts prefer to err on the safe side. Relatively few
estimates havebeen validated by experience; inevitably so, as most deal with
rare events. One study3 looked at the estimatedreliabilities of 130 different
engineering systems and pieces of equipment and showed that 10% of the
observed values were within a factor of two of the estimate, 90% within a
factor of four.These were well-defined systems.
Hazard analysts could well place estimates on the accuracy of their data
(seeSection 4.6, page 160) and the final result. But the meaning ofsuch confidence limitsshouldbe made clear. They can allow for uncertainties in the data
but not for errors in the logic, for failure to identify all the ways in which
hazards can occuror forerrors in estimates ofhumanreliability. In practice the
firsttwo are usually much more importantthanerrors in the data.
Even the uncertainties in the data allowedfor in the confidence limits are
not the complete range of uncertainties. The confidence limitsallow for uncertainties due to sample size but not, of course, to errors due to changes in
design. use of inapplicable data, and so on.
Suppose a Hazan shows that an event will occur on average once in 100
years. Ifthe eventoccurs next year (or nextweek)this does not prove that the
estimatewas wrong (though it may be).Iftheeventoccurs randomly,then it is
equallylikely to occur in any year in the next 100 years.This point is misunderstoodby many people.
Onthe other hand, few accidents occur becausethe unlikely oddsofone in
so many thousand years actually come off (see Section 4.8, page 163). More
often, after an accident has occurred,it is found that some ofthe assumptions
on which the analysis was basedare incorrect. Forexample,testing ofprotective equipment has lapsed or is not thorough, or the faults found are not
promptly rectified.
Different estimatesof consequences may differ greatly, particularly where
gas dispersion is involved, hut in recent years the estimates have converged
(see Section 6.6, page 199).
page 120).
6.5 Chaos
Dataon consequences are usually relatedto data on inputs in a consistentway.
If we increase theheat input to a vessel we expect the contentsto get hotter,
though the actual rate will depend on many factors and can be very fast if a
runawayreaction starts. In chaotic systems the consequences are apparently
patternless. In some cases the system is unstable and a small change in the
input can produce a big change in the output. In other cases the output is
related to the input in a complex way which may be difficultto unravel. The
system is deterministic but it is hard to predict the results.
198
In Figure 6.1 the horizontal axis is a measure of a dose or action and the
vertical axis is a measure of the response or effect.
When the dose is I, theresponse is 2.
When the dose is 2, theresponse is 4.
When the dose is 3, theresponse is 6.
When the dose is 4. can we say thattheresponse will be 8?
RESPONSE
:/
6_0
z
2
DOSE
:
4
/
/
,,
NUMBER OF
ENGINE
FAILURES
Many peoplewould say that we can. To estimatethe effectsof low concentrations ofradiation or toxicchemicals, we measurethe effectsofhigh concentrations and thenextrapolate to low concentrations.
In Figure 6.2 I have added meanings to the figures. The horizontal axis
gives the numberof enginesthat have failedon a four-engine aircraft and the
vertical axis gives the delay in arrival at the destination.
If one engine failstheplanewill be 2 hours late.
If two enginesfail the plane will he 4 hours late.
If threeengines fail the plane will be 6 hours late.
But if all four enginesfail the planewill not be 8 hours late!
S.J. Gould has shown that over the yearsHershey chocolate bars havegradually got smaller. By extrapolating the figures he has calculated the date at
which the bars will have zeroweight'5.
10
0. a
U
C
V.
C
-.
0. .0
0.1
a
0CU
t).01
TT
0.001
EPA
EPA
CDC
FDA
NY
WHO
.:
. CDC
Cahiomia
FDA
Carada
Germany
WHO
Netherlands
= EnvironmentalProtectionAgency (US)
= CenterforDisease Control (US)
is not so low. A more scientific approach is to divide the country into areas (or
into population groups, by occupation. social class, dietary habits, and so on)
and then compare the incidence of disease in them17.
References in Chapter6
2.
3.
4.
5.
6.
7.
8.
Green, A.E. (ed), 1982, HighRisk Safi'tv Technology (Wiley, Chichester, UK).
Young, R.S., 1986, Risk analysis appliedto refinery safetyexpenditure, American
Petroleum Institute Committee on Safety and Fire Protection Spring Meeting,
8li April.
9. Kletz, TA., 1985. Reliability Engineering, 11(4): 185.
10. Kletz, TA., 1998, Process Plants: A Handbook ofInherently Saft'rDesign,2nd
edition (Taylor & Francis, Philadelphia, Pennsylvania, USA).
11. Center for Chemical Process Safety, 1998, Guidelines thr Improving Plant Reliability through Data Collection and Analysis (American Institute of Chemical
Engineers, NewYork,USA).
12. Tweeddale, H.M., 1994, Conducting a peer reviewof a safetystudy,Chemeca94
Conference, Perth, Australia,September.
for use in risk assessments ofmajor hazard sites, SymposiumSeries No. 14/, 317
(Institution of Chemical Engineers, Rugby, UK).
14. Gould,J.H.. 1996, Loss PreventionBulletin, No. 127: 12.
IS. Gould, Si.. 1984, Hen's Teeth and Horse's Toes, 313 (PenguinBooks, London,
UK).
16. Silbergeld. E.K.,Nov/Dec 1995, Scientific American Science and Medicine, 48.
17. Mahesuaran. R. and Staines. A., 7 April 1997, Chemistry andindustry,254.
202
The history of
Hazop and Hazan
No revolutionary'idea arises withouta
pedigree.'
SJ. Gould6
while (Leonardoda Vinci's) mechanics and
engineering are, fortheir breadthanddepth of
experience, unique and at times aheadoftheir times,
they are not afruit ripenedalone in a desert.
M. Cianchi7
7.1 Hazop
Reference
WHY?
HOWis it
Alternatives
Selectionfor
development
What SHOULD be
achieved?
be achieved?
achieved?
Page . . . Date
achieved?
WHY
THAT
WAY?
WHEN is it
achieved?
WHY
THEN?
When ELSEcould
it be achieved?
WHERE is it
achieved?
WHY
THERE?
WHY
THAT
PERSON?
204
When SHOULD it
be achieved?
Who SHOULD
achieve it?
life wasjust too short. After a good many tries we came up with an approach
which has much ofthe principle ofcriticalexamination butwas somewhat bent
in style'. The essence of the new approach was that a technique designedto
identify alternatives was modified so that it identified deviations8. It was
recognizably Hazop as we know it today though it was further modified during
later studiesto the form describedin Chapter2.
The following are a few of the safety points that came out of this early
Hazop (though that term was not used then; the exercise was described as a
method studyor hazardinvestigation). Someof the points are now includedin
design specifications but were not included at the time.
Bypasses around control valves which areconnected to safety trips should
be deleted. Use of a bypassrendersthe safetytrip useless.
Nitrogen should be used for vacuum breakingto preventthe ingress of air
into a hot system.
Breaktanksshouldhe fittedin the townwater supply to preventcontamination by reverse flow.
The relief valve system should be checked for places in which liquid could
collect.
A slip.-plateshould be fittedin the feed line to [vessel Xl to prevent liquid
leaking in beforeconditions are correct.
Vent valvesshould be fitted to all blowing points so that the pressurecan be
blown offbefore hoses are disconnected.
A ventvalveshould be fittedto a high pressure filter sothat the pressure can
be blown offbeforethe filter is opened for cleaning.
Extended spindles shouldbe fitted to the valveson acid tanks to reducethe
risk that operators maybe splashedby leaks.
Special equipment should be designed forcharging and discharging catalysts
and otherauxiliary materials, to remove thedangers thatgowith improvisation.
Note that all these points are written as recommendations. Today most
Hazop teams would not say 'should' but simply 'Delete bypasses... etc'.
More operating points than safety ones came out of the study. This was
expected. The remit of the team was 'To devote themselves full-time to
obtaining and studying information from all sources and to take any necessary
decisions on broadplant design aimedat ensuringthat the phenol plantwould
start up quickly and satisfactorily that it will produce its design output and
quality of products; that it will operate safely and its effluents will be satisfactorily treated'. Today many, perhaps most, Hazops produce more operating
points than safety ones.
A few monthsbefore the phenol study was undertaken in ICI HOCDivision
at Billingham, the Mond Divisionat Runcorn carried out a similar but very
205
much shorterstudy (it occupieda team of four for 21 hours, a fortieth of the
time taken by the HOC study) on a semi-technical plant. The remit for this
studywas 'To evaluatethe process for hazardswhichmay ariseduringoperation of the semi-technical plant. Particularattention to be paid to the effectof
impurities in raw materials, build-up of products in recycle systems,
maloperation and equipment failures'.
In 1968 D.M.Elliott and J.M. OwenofMondDivisiondescribedthe use of
critical examination for generatingalternatives in theearly stages ofdesign,as
Even earlier, in 1960, D.S. Binsted
suggested in Section 2.7 (page
described a similar application in ICI Organics Division. However, these
applications of critical examination never became as popular as Hazop,
perhaps because they were before their time but more probably because,
compared with Hazop, they were too cumbersome and time-consuming.
The ICT Central Work Study Department in London played a part in integrating the Mond and HOC forms of the developing Hazop technique and
spreading knowledge of it throughout the company. A report by G.K. Cooper
dated November 1964 brings out clearly the difference between Hazop and
critical examination:
4l).
206
Reading this report over 30 years later, the need for a better Company
policy seems equallyobvious.
ICT Pharmaceuticals Division adoptedHazop enthusiastically and the first
use ofthe technique outside ICI occurred in 1967 when R.E. Knowlton (then in
Central Work Study Department) led a study for Ilford Ltd8. The flrst
published paper on Hazop was H.G. Lawley's 1974 paper from which the
example in Section 2.5 (page 34) has beentaken. It was presentedat the American Institute of Chemical Engineers LossPrevention Symposium in Philadelphia the previous year (held, incidentally, in the hotel which later became
famous as the site of the first recognized outbreak of Legionnaire's disease)
and aroused interest from the outset. Gradually other companies adopted
Hazop. The first contractor to do so was probably ChemeticsInternational,
then part-owned by ICI.
Mond Division later integrated Hazop into a six-stage hazard study
programme extending from the early stages of design through to
post-commissioning1 . Hazop is the thirdstage (see Section 1 .1, page 1).
7.2 Hazan
7.2.1 Early history
Who determines our future: the fates, the gods or ourselves?Greeks, early
Christians and Muslims believed that everything was in the lap ofthe gods(or
God). Probability theory and risk management did not emerge until people
came to believe that they were to some degreefree agents. This did not occur
until the Renaissance when people developed freedom of thought, a wish to
experiment and a desire to control the future17. Even today, not everyone
believes that he or she is a free agent. Cupitt writes, In every civilization from
the Bronze Age to the present day, the common people have been gamblers,
believers in fortune who know that life is a lottery
and we should not be
surprised. That, indeed, is how it looksto them'18.
Tait has reviewed the application ofprobabilistic methods to engineering19.
In any engineering structure the loadL and strength S are notpreciselydefined
but vary about a mean value. Failure may occur ifLis a maximum whenS is a
minimum. However largewe make S, complete safety is never achievedbut is
approached asymptotically (Figure 7. I, page 208). If we knowL and the variation in L and S and can define an acceptable failurerate, we can fix a design
value for S.
The firstuse ofstatistical techniques in this way was Chaplin'sstudy ofiron
chains in 188013 but the methods were not widely used until about 60 years
later. Why was this? On a visit to Wigan PierI saw two magnificent preserved
207
STRESS
l0
l0
208
Figure 7.2 Trencherfield cotton mill. Wigan Pier. Withthese two steamengines
and six boilers,the reliability of the powersupply was not a problem.
209
THE HISTORY OF
or vapouris not likelyto occur under normal conditions and, ifit does occur,
will existforonly a shorttime. Lord suggested that a Zone 2 area shouldhe one
in which flammable gas or vapour is present for less than 10 hours per year
and, if so, it can be shown that the FAR for a plant operatorfrom this risk is
tolerable (see Section 3.8.6, page I39). Otherworkers arrived independently
at similarconclusionst2.
During the 1970s Hazan was applied to many chemical industry problems
by many workers, outstanding among whom were S.B. Gibson and HG.
Lawley. During the 1980s and 1990s growth continued and many computer
programs were developed, particularly for the calculation ofconsequences.
Although cost-benefit analysis was mentioned in my firstpaper5 and some
examples of the cost of saving a life were listed (see Sections 3.4.67, pages
99102), there has been less interest in them than in the calculation of probabilities and consequences.
References in Chapter
121.
13. Pugsley, AG., 1966, The Safety of S/rue/ores (Arnold. London, UK) (quoted by
Tail. N.R.S., 1987. Endeavour, 11(4): 192).
14. Pugsley, AG. and Fairthorne, R.A.. May 1939. Note on Airworthiness Statistics
(FIMSO, London. UK).
211
HA/OP AND
1-IAZAN
15. Blackett, P.M.S., 1962, StudiesofWar, 169, 173 and 210 (Oliver and Boyd,Edinburgh, UK).
16. The Dour Telegraph, 23 April 1991.
17. Bernstein, P.L., 1996, Against the Gods The Remarkable Story of Risk, 35,
4344, 54 (Wiley, New York, USA).
18. Cupitt. D.. 997, After God The Future of Religion, 30 (Weidenfeld and
Nicolson, London, UK).
19. TaiL, N.R.S., 1993, ReliabilityEngineeringand System Safety. 40: 119.
20. Jones, R.V.. 13 April 1968, Chemisir and Industry, 470.
212
Conclusions
All human activities involve some risk. It can be reducedbut not eliminated
completely.
Hazard and operability study(Hazop) is now a maturetechnique for identifying hazards withoutwaiting for an accident to occur (Chapter 2).
Hazard analysis (Hazan) is now amaturetechnique for estimating the probability and consequences of a hazard and comparing them with a target or
criterion (Chapters 35).
Takentogether, the two techniques allow us to allocate our resources so that
we deal with the biggest problemsfirstand in the most effective way. Neither
technique will be effective,however, unless there is a commitment to safety at
all levels (Chapter 1).
Cost-benefit analysis is less well-established so far as safety is concerned,
but nevertheless has a part to play (Sections 3.4 and 3.9, pages 83 and 143).
Hazard analysis and cost-benefit analysis are difficultsubjects to explainto
the public but nevertheless we shouldtry to do so. The hazards oftechnology
should be balancedagainst the benefits (Sections 3.4 and 5.3, pages 83 and
181).
'The most versatileand ambitious species are thosewhich have evolvedmechanisms capable of recognizing and facing threais heftre they have had a
chance to inflict expensive andpossiblyirreparable damage.'
Jonathan Miller. 1978, The Body in Question, 216 (Cape)
213
Addendum 1
An atlas of safety thinking
MORE OF VIBRATION
LESSOF FRICTION?
MORE OF ANGLE?
(ie, table may not
be level)
The firstand most importantstage in any hazard study is to identify the things that
can gowrong and produce accidentsoroperatingproblems.Itis little use studying
small hazards ifwe have failed to realize thatbigger ones are round the corner.
214
ADDENDUM
We need to know how often the hazard will occur. Again, the best way is to
look at past experience but sometimes there is no experience and we have to
use synthetic methods.
215
(4) Prevention
We should comparethe risk (that is, the probability times the consequences)
with generally accepted codesand standards or with the other risksaroundus.
216
ADDENDUM
We should also compare the cost ofprevention with the cost of the accidentin
orderto see ifthe remedy is 'reasonably practicable' or ifwe shouldlook for a
cheapersolution.
(7) Prevention 2
Plastic
What are the disadvantages of our solution? Gluing the tumbler to the table
may be acceptable if it is used to store pencils but not if it is used for drinks.
Perhapswe can think of a better method. A plastictumblerwill not break but
thecontentscan still spill. We shouldanswerthis question before the table is
made or the tumbler ordered.
217
Addendum 2
Myths of Hazop and Hazan
I/a company usesHazopand Hazan we don't need 10 worry about the compe
tencc 0/the management.
All Hazop and Hazan can do is apply people's knowledge and experience in a
systematic way so as to make the most of it. If peoplelack knowledge, experience or commitment, Hazop and Hazaii are a wasteoftime (seeSection 2.4.4,
page 30).
Top rate people don 't need Hazop and Hazan.
This is the opposite ofthe last myth.Modernplantsare so complicated that no
one can see all that can go wrong and assess its consequences unless they
follow a systematic procedure. Without Hazop and Hazan (or similar techniques), no one will achievetheir full potential (seeSection 2.4.4. page30).
ADDENDUM
and
accident.
They will not for at least three reasons: (a) Being human, we will not spot
everything that might go wrong; (b) People will not always act in the way we
have assumed theywill; and (c) Hazop cannot, by its nature, preventmechanical accidents, such as people bumpinginto equipment which has been badly
located.
Since Hazop will discoverall the hazards, we don't need to learn aboutpast
accidents.
Techniques discover nothing; only people discover hazards; techniques can
help them do so. Hazop helps people apply their knowledge and experience,
and much of that comes from learning and remembering the lessons ofthe past.
Hazop and Hazan can he carried out atany time.
In fact,the window of opportunity for a Hazop of the line diagrams is small. It
cannotbe carriedout until the line diagrams are ready but if it is left too late
then detailed design, even construction, will have started and expensive
changes will be necessary. Similarly, Hazops of flowsheets must be made at
the appropriate time (see Sections 2.3 and 2.7, pages 26 and 41). Hazans
cannotbe carriedout until the hazards are recognized. Some of them may be
obviousfrom the starthut othersmaybe brought to lightby the Hazop and then
the answers will be wanted as soon as possible.
The differences between Hazop and Hazan are well-known.
They are wellknownto those whouse them regularly but many people are not
awareof the difference. If someone asksyou to carry outa Hazop (or a Hazan),
before you do so, make sure that it is really what they want (see cartoon on
page 95).
The Hazop guide words (see Section 2.1. page 9) are inviolate, should not he
changedandneednot he added to.
They are based on long experience hut, nevertheless, if you find other words
are useful, by all means use them, particularly if you are applying Hazop
outside the typeofactivity for which it was originally designed. For example,
219
ifa plant is computer-controlled, then the control system need not be/cannot
he Hazoped.
ADDENDUM 2
they will not know why the plant was designedthe way it is and the designers
will not know about problems of concern to the operating staff (see Section
2.4.5, page33).
Anyweaknesses in the designcan he left until the Hazop.
Hazop is a check on the design to spot points that have been overlooked. It
should not replace the normal consultations and discussions that take place
while a design is being developed (seeSection 2.4.7, page33).
Computers can carry out a Hazop.
Computersare now widely used for recordingresultsand remindingteams of
thecommon causesof deviations. Programshave alsobeen developed, but so
far little used, forproducinga listoftechnicalproblemsfor consideration.A
computeris highly unlikely to be ableto identify the problemsthat arise out
of interactions, or failures to interact,betweenpeople (see Section 2.6, page
37).
We shouldfirst remove (or reduce) the risks that are cheapestto remove (or
reduce).
At first sight this seems reasonable as it will save the most lives for a given
expenditure. However, on moral grounds we usually prefer to remove (or
reduce) those risks that are considered intolerably high. We can use costs to
help us decide between different ways of removing (or reducing) a risk and
also to help us decide whetherfurther reduction in a tolerable but still significant risk is justified (seeSection 3.4.6, page99).
Tripsystems shouldalwavsftuil safr.
When trips fail safe they operate even though there is nothing wrong with
conditions on the plant. Suppose a trip is designedto shut down a plant when
the temperature rises abovea presetvalue. If the trip develops a fault the plant
is shut down unnecessarily. This maybe safer thanletting the plantgettoo hot,
but may still be hazardous as the sudden shutdown may cause leaks. The
phrase 'fail-safe' misleads people into thinking that the action is not
hazardous. In fact, there are cases where the 'fail-safe' action may be less safe
than the alternative (see Section 3.5.10, page 118).
If spurious operation is frequent, operators are tempted to bypass the trip.
We can reduce the number of spurious trips by usingvoting systems. Before
doing so we should look for other possible causes of frequent spurious trips,
such as poor maintenance (see Section 3.5.10,page 118).
Whenthe Hazopmeetings havebeenheld and the Hazan report issued, the job
is done.
No, thejob is not over until actionshave been agreed and carried out. Unless
actions are given to a named personnothingwill happen. All actions should be
tracked until they are complete and have been inspected by the originator to
make sure that they havebeen completed correctlyand look right. What does
not look rightis usually not rightand shouldat least be checked.
Morepowerful computers can compensate for deficiencies in models or data.
Stated as bluntly as this, the myth is obvious nonsense. Nevertheless, people
sometimes act as if it was true. More and more complex models, requiring
greater computing power, are devised for estimating the dispersion of gas
leaks, for example, and the pressures developed if they ignite. Yet the size of
the leak is determined by an arbitrary rule, such as assuming that the largest
pipe will break, producing two open ends (see Section 4.6, page 160 and the
quotations therefrom F. Hoyle and T.H. Huxley, page 162).
There are some further mythsin Section 5.2, page 169.
ADDENDUM
References in Addendum 2
1.
2.
223
Index
A
67
Abbeystcad
audits
automatic start
131132
ability
acceptability criteria(see tolerability
criteria)
B
accident content(of products)
104 backfiow (see reverse flow)
87103 batch processes
accidents, fatal
1620,52, 5556,
acetone
203
6566, 13I
actions,typesof
28 BATNEEC (see BestAvailableTechActs of God
182
nology Not Entailing Excessive Cost)
aircraft
208
bellows
1920, 93, 184, 200,
123, 198
alarms
3. 22, 3436. 42, 51, belt and braces
148150
6566, 9394, 108, 116, benzene
82
118I 19. 126 128, 133. 163
Best Available TechnologyNot
81, 6
Albania
122
153
algebra
alternatives
4445
103104, 138139, 158
comparing
ambulance crews
156
ammonia
54. 61, 164
214217
analogies
185
(of
victims)
anonymity
2829
appearances, a problem
arithmetic
153, 161162,222
artificial intelligence
38
asbestos
6, 176
assessment of hazards (see hazard
analysis)
associations
185, 190
assumptions
62. 159160
224
Entailing ExcessiveCost
(BATNEEC)
104
BestPracticable Environmental
104
Option (BPEO)
190
2.4647,62, 165, 183,
Bhopal
119
bicycle lights
'blackboxes'
66
blastwalls
210
blinds(see slip-plates)
blood pressure
157
blowdown
74
BPEO (see BestPracticable
Environmental Option)
braces
124, 148151
brakes(on cars)
British NuclearFuels (see also
Sellafield)
'broadlyacceptable'
109
184, 187
85
INDEX
bufferzones
burstingdiscs
137
112, 124125, 135
4!
bypasses
C
calculation, errors in
compressors
153154,
161162,222
75, 163, 184186,
189. 200201, 210
161
cancer
CanveyIsland
cars
cash machines
130
catchpots
causes,imaginery
certainty
chains
185
207
125
Challenger
Channel Tunnel
102. 118
chaos
198199
74
charcoal
67
Chazop
check valves (see non-return valves)
check-lists
2, 54
chemicalindustry
184, 187
183
Chernohyl
chilblains
186
34
chimneys
chips
chlorine
186
chocolate
chokes
clocks
closed shop
Club ofRome
clusters (of data)
coal dust
codes ofpractice
coincidences
commissioning
common modefailures
66
165
155
201202
176
2,34, 170
73
4
124128
66
180
3
23
50, 1 30
126129
compromise solution
2()2 I, 6567
computer control
3741
computers
for recording Hazops
24, 26
software errors
75
errors
66
specification
concentration changes, effects of
37
186
concentration.,small
condensate
64
conflict
21
connections, wrong
163
8081, 197,
consequence analysis
200201
deterministic and
probabilistic compared
80, 185,
198199
6
consequences
of deviations 10, 28, 38, 40, 52. 6870
as a measure of negligence
184
consequences and probability
141142
compared
construction
4. 27, 55
contamination
62
contractors
21, 28, 33, 220
118
contradictory instructions
corrosion
120
cost-benefit analysis 85, 100, 103104,
141,211
costs
8384, 87, 99100,
143, 169, 188, 196, 203
of pollution
of safety
of saving a life
costsand safety
cranes
creativity
104105
78
100103, 133,
135, 141
47
34
23. 38
225
HAZO AND
1-IAZAN
critical examination
43.4546,51.
203206
72
criticality
crystallizers
cumene
142
203
D
dams
185, 189190
79, 160165,
169, 172, 193202
accuracy of
appliesto past
confidence limits
effectof maintenance
and operating policy
estimating
inapplicable
on instruments
on mechanical equipment
pitfalls in
types of
databases
definitions
demandrates
dread
184
earth. age of
156
eggs
electrical area classification
122
161162
121
196197
121122,
170. 198
196
120
197
121,
197198
120123, 198202
161
139140,
21C21 I
75
electricity supplies
isolation
valves
169
emergency
emulsionbreaking
72
environment
104105, 179180
ETA (see event tree analysis)
ethylene
1,43, 61, 133,210
ethyleneoxide
1,61,170,210
event tree analysis (ETA)
54
experience
21, 3033
comparison with calculation 164, 177
learning from
30, 80, 121
expertsystems
3839
41
57, 105106
71I 17. 159
on 1-lazop
129
designintention not followed 128129
experts
dependence on
management of
qualities needed
34
developing countries
deviations (see also consequences
ofdeviations)
10. 1617, 23, 2728,
3538, 40. 44, 52, 6870, 205
from design standards
34
dirnerization
1316, 3438,40
dioxin
200201
disease, industrial
176177
222
dispersion (of leaks)
distillation
37, 46
118, 124126, 130
diversity
33
downsizing
drains
165
34, 36,72.
226
explosions
81, 152
152155
165
FNcurves
9597, 196
INDEX
failures,non-random
fairness
169, 176
124
2
firewater
Flixborough
10. 6870
flow
4. 42
flowsheets
FMEA (see failure mode and
effectanalysis)
51
food processing
foods
182 183
fractional dead time
71117, 133. 148
'free lunches'
113116
gases, liquefied
gaskets
genetically modifiedorganisms
Germany
181
133142, 148ISO
207211
4, 219
5, 193194, 219
143
morality of
168190
objections to
108109, 115, 120130,
pitfalls in
153165. 199202.218222
177
quality
rapid
172174
reasonsfor
7779
163164
recommendations from
8083
stages of
79
types ofproblem examined
hazard (definition)
57
hazard and operability studies
175. 129, 158, 203207
(Hazop)
26
acceptability
aids
2426, 37, 40
3, 30, 62
assumptions
audits of
56, 220
of hatch plants
1620
benefits of
5657
82
by computer
3741
coarse-scale
law
guidewords
12
52
159160, 197
196197
contrastwithexperience
136 137
criteria
80105
assumptions
confidence limits
206. 219220
'gut feeling' (see judgement)
handrails
1-lazan (see hazard analysis)
hazard analysis (Hazan)
84
17, 21, 28,
I80181
acceptance of results
176
accuracy
137
agreement withexpectation
176
177
application to disease
ofcomputer-controlled
6567, 220
plants
4647
ofconcepts
21
conflictin
54
ofdefencesystems
27
duration
6
175
1319,3437.
examples
3437. 6175
experts in
flowsheets
of
4246, 206, 221
in food industry
51
203207
227
introduction of
oflaboratory design
51
techniques compared
incident rate (see hazard rates)
'index of woe'
India
inherently safer design
innovation
insecticide
5556
98
189190
47,65. 163
2.34.169
47
55
inspections
instructions
1618, 33, 131132, 163
blanket
66
instruments (see also alarms,protective
systems, trips)
costsof
35
interlocks (see trips)
70
Trwell, River
120
joints
judgement
124130, 148150
hazard studies
4, 207, 214217
hazard
and
(see
Hazop
operability studies)
Health and Safety at Work.etc Act
90
Health and Safety Executive 33, 83, 85.
8990,93, 100, 102, 138. 141,158.
187189
heartdisease
184, 186
heat exchangers
28, 33
Heinrich
98
history
Holland
hoses
hotels
human error
203211
181
198, 205
138
identification ofhazards
175, 115,
158, 168170
228
Kelvin. Lord
kick-back
kill systems
King's Cross
knock-on effects
knowledge
156
35, 4849
6364, 135136
140141
33
21, 3033, 56. 218219
ofothers' activities
practical
ragbagof
specialized
typesof
in wrong place
L
laboratories
74
I89
74
74
72
23, 67
73
3840
6774
51
2
INDEX
lapsesofattention
131132
7,8182. 181
law
7172
41
179
layering
layout
lead
leaks
compared
size of
life, cost ofsaving
135, 141
93
lightning
liquefied gases (see gases,liquefied)
36
liquid hammer
207
load and strength
140141
London Underground
154
lost-time accident rate
198
lubrication
73
lutes
M
maintenance
preparation for
Malthus
3536,74. 120123.
management
competence of
effecton failurerates
46, 62
1920
12
154158. 170
2830, 65, 73
myths
33
184, 187
9092
218222
natural gas
Netherlands
138
181
74. 205
nitrogen
160
nitrogen blanketing
non-random demands
112, 130
non-random failures
126, 130, 159
non-return valves
48, 129
nuclearpower(see also criticality,
93, 120, 126, 182,
radioactivity)
183, 185, 187
decide
numbers cannot
178180
everything
bolts
13
nuts and
155
132, 159160,
163, 170, 198
management failure
123, 137, 160, 163
probability of
70
Manchester Ship Canal
materials of construction
121. 123,
133 134
measurements
meaningless (see also
153154
parameters, choiceof)
medical equipment
54
186187
messengers
67
methane
0
0-rings
old plants
open shop
operations research
operatorerror (see humanerror)
'optionsfor the future'
125
13
165
208
180
184
outrage
2728
over-enthusiasm
198
128129,
170.
overfilling
113114
overpressuring
33
ownership ofproblems
210
1,74,
oxygen
229
p
parallel systems 111112, 125126, 196
parameters, choiceof (see also
measurements, meaningless) 156158
181182, 186
pesticides
petrol
174
delivery
unleaded
179
phenol
203
pianotuners
139140
pipe branches, small
50
pipelines
9, 6770, 72
failure of
36. 64, 120, 123,
133, 142, 161
hazard
pitfalls (see
analysis, pitfalls in)
137138
plants,effecton public
72
plutonium
185
pollution
155, 179, 181, 183,
43
polyethylene
136137
ports
122
poultry
failures
128
power
PRA (see probabilistic risk
assessment)
premises, false
47
pressure
10, 18
priorities
85, 87, 91, 141, 171, 177
probabilistic risk assessment (PRA)
(see also hazard analysis)
3. 5
probability
ofan incident
8083. 115120,
ofhuman error
148150. 210211
131132, 165
230
effecton operators
129
failureof
6364
in placeofreliefvalves 84, 138139
public opinion
179, 188
185
publicity
35, 37, 40, 42, 4748, 64, 67,
pumps
120, 124, 128129, 161, 169
Q
QRA(see quantitative risk assessment)
quantitative risk assessment(QRA)
(see also hazard analysis)
3, 5
182
railways
54, 118, 181
timetables
157158, 193194
transport of chemicals
103. 158
underground
140141
171172
rapid ranking
rate and duration confused
116
reactionkill systems
6364, 135136
reactors
42,6166,71,74,135
'reasonably practicable'
redundancy
8 182, 85.
100, 168,221
phoney
124
regulators
187189
reliability,
ofelectricity supplies
75
resources
allocating
reverse flow
4849,61,129
INDEX
acceptable
acuteand longterm
of alternatives
cheapest to remove
conceptual problemswith
definition
experience of
familiar
from radioactivity
levels
natural
8590
176
103105
99100
185186
57. 78
183
183
89, 176
8590
182
8486
negligible
of
181189
91, 96,
perception
88, 93,97
person at greatest
size and probability
9092, 184
compared
to employees
8790, 176
to the environment
104105
to the public
9397
tolerable
8597,99, 105
under ourcontrol
83, 181
184
versus benefits
and
voluntary
imposed
93, 181
compared
road(see also cars)
accidents
103104, 158
100
safety
of
chemicals
103104,
transport
158. 198
181
rock-climbing
51. 6566, 71
runaway reactions
discs
(see
discs)
rupture
bursting
S
sabotage
Salmonella
seal pots
62
122
73
'secondchance' design
Sellafield (see also Bdtish
NuclearFuels, Windscale)
separation, of liquids
series systems
service lines
68
6770
46, 7172
112, 124126, 196
settling
ofsolids
Seveso
sex
ships, collisions of
shutdown
Sizewell
slip-plates
slips
smoking
spectacle plates
spokesmen
stand-by equipment
start-up
steam
steamengines
Stephenson, George
stirring
STOPHAZ
storage
intermediate
strokes
sulphur
systems, limitations of
136137
208
152
72
55
12, 35, 40
46, 62
157
73, 159
30. 218219
1
tanks
1819
tea-making
tO. 18
temperature
tests
3, 3536.41, 120, 123, 125126,
138, 159, 163164, 197
of protective equipment
71Ill
Texas sharpshooter
Therac
231
2, 83105. 210
tolerability criteria
events
113, 116
top
toxicchemicals (see also specific
chemicals such as chlorine)
176, 200
training
35, 131132
transport
6,41
trips
3,22, 28, 35, 48, 50SI, 64,
8485, 106109, III, 118120,
124, 126129, 133134,
159, 170,205,221,222
demands
on
108109
frequent
118120
spurious
tunncls
turbines
67
64
two-phase systems
62, 7172
United States
law
units
188
82
vending machines
vents
vessels
failurerates
operations in
vinyl acetate
violations
121, 198
voting systems
W
water
resultsofadding
6264,67, 74
62, 7172
sprays
Weibull analysis
'what-if' analysis
Wigan Pier
Windscale (see also Sellafleld)
witches
120
130
5455
207
189
186
116, 153154
x
X-rays
V
valve by-passes
36
valves
1213, 34, 130131
169
emergency isolation
non-return
48, 129
relief
118, 138139,
159, 169, 205
232
75
Zeebrugge
102
Ic
HAZOP
HAZAN
4:IIAZ
ND
OCESthDJSTRY
Trevor Kletz
The techniques ofHazopand Hazan were developed to identify
and assess hazards in the process industries.Their use leads to
safer plant, the eliminationofmany operational problems at the
design stage and improvement in reliability. They will only be
effective,however,ifthereis a commitment to safety at all levels
in an organization.Understandingthe practical issues involved
in the correctimplementationof these techniques is the theme
of this book.
1-lazopandIfazan:1k'ntljjingandAssessirig Process Indusiry
IChemE
ISBN 0852954212
Da'is Building
hrracu
IQ. I K
16189 Railwa'i
Rugby CV2 1
'Iuk11
uiiu
Iii:::
nt(-!tii1I(Iu(I
(11785
82
-f t 55
(-(
9 780852 954218