PRINCIPLES OF
SYSTEM SAFETY
ENGINEERING AND MANAGEMENT
Felix Redmill
Redmill Consultancy, London
Felix.Redmill@ncl.ac.uk
RISK
(c) Felix Redmill, 2011
CERN, May '11
2
SAFETY ENGINEERING AND MANAGEMENT
It is necessary both to achieve appropriate safety and to
demonstrate that it has been achieved
• Achieve - not only in design and development, but in all
stages of a system’s life cycle
• Appropriate - to the system and the circumstances
• Demonstrate - that all that could reasonably have been
done has been done, at every stage of the life cycle
(c) Felix Redmill, 2011
CERN, May '11
3
THE U.K. LAW ON SAFETY
Health and Safety at Work Etc. Act 1974:
Safety risks imposed on others (employees and the public
at large) must be reduced ‘so far as is reasonably
practicable’ (SFAIRP)
(c) Felix Redmill, 2011
CERN, May '11
4
THE HSE’S ALARP PRINCIPLE
Increasing
risk
Unacceptable Region
(Risk cannot be justified except in
extraordinary circumstances)
Limit of tolerability threshold
ALARP or Tolerability Region
(Risk is tolerable only if its reduction is
impracticable or if the cost of reduction is
grossly disproportinate to the improvement
gained)
Broadly acceptable threshold
Broadly Acceptable Region
(Risk is tolerable without reduction. But it is
necessary to maintain assurance that it
remains at this level)
(c) Felix Redmill, 2011
CERN, May '11
5
THE HSE’S ALARP PRINCIPLE
(c) Felix Redmill, 2011
CERN, May '11
6
CALIBRATION OF THE ALARP MODEL
Recommended for Nuclear Industry
• Intolerability threshold:
– 1/10000 per year (for the public)
– 1/1000 per year (for employees)
• Broadly acceptable threshold:
– 1/1000000 per year (for everyone)
(c) Felix Redmill, 2011
CERN, May '11
7
A VERY SIMPLE SYSTEM
• Chemicals A and B are mixed in the tank to form product P
• P opens & closes input and output valves
• If emergency signal arrives, then cease operation
Emergency
signal
B
P
A
P Controller
(c) Felix Redmill, 2011
CERN, May '11
8
QUESTIONS
• How could the accident have been avoided?
– Better algorithm
• How could the software designer have known that a better
algorithm was required?
– Domain knowledge
• But we can’t be sure that such a fault won’t be made, so
how can we find and correct such faults?
– Risk analysis techniques
(c) Felix Redmill, 2011
CERN, May '11
9
A SIMPLE THOUGHT EXPERIMENT
• Draw an infusion pump
– Note its mechanical parts
• Think about designing and developing software to control
its operation
• Safety consideration: delivery of either too much or too
little of a drug could be fatal
• How would you guarantee that it would kill no more than 1
in 10,000 patients per year?
– What level of confidence - on a percentage scale - do
you have in your estimate?
(c) Felix Redmill, 2011
CERN, May '11
10
CONFIDENCE IN SAFETY IN ADVANCE
• What is safety?
• How can it be measured?
• What can give confidence that safety is high?
• Need also to demonstrate safety in advance
• Therefore, need to find hazards in advance
(c) Felix Redmill, 2011
CERN, May '11
11
NEW RISKS OF NEW TECHNOLOGY STRASBOURG AIR CRASH
• Mode: climbs and descents in degrees to horizontal
– 3.3 descent represented as ‘minus 3.3’
• Mode: climbs and descents in units of 100 feet
– 3300 feet/minute descent represented as ‘minus 33’
• Plane was descending at 3300 feet/minute
• Needed to descend at angle of 3.3 to horizontal
• To interpret correctly, pilot needed to know the mode
• ‘Mode error’ in the system
• Human error? If so, which human made it?
(c) Felix Redmill, 2011
CERN, May '11
12
CONCERN WITH TECHNOLOGICAL RISKS
• We are concerned no longer exclusively with making nature
useful, or with releasing mankind from traditional
constraints, but also and essentially with problems resulting
from techno-economic development itself [Beck]
• Risks induce often irreversible harm; not restricted to their
places of origin but threaten all forms of life in all parts of the
planet; in some cases could affect those not alive at the time
or place of an accident [Beck]
• People who build, design, plan, execute, sell and maintain
complex systems do not know how they may work [Coates]
(c) Felix Redmill, 2011
CERN, May '11
13
RISK - AN IMPORTANT SUBJECT
• Risk is a subject of research in many fields,
e.g. Psychology, Sociology, Anthropology, Engineering
• It is a critical element in many fields
e.g. Geography (climate change), Agriculture, Sport,
Leisure activities, Transport, Government policy
• It influences government policy (e.g. on bird flu) and
local government decisions (e.g. closure of children’s
playgrounds)
• An influential Sociological theory is that it is the most
influential factor in modern society
(we are in ‘the risk society’)
• Every activity carries risk
• All decisions are ‘risky’
(c) Felix Redmill, 2011
CERN, May '11
14
FUNCTIONAL SAFETY:
ACHIEVING UTILITY AND CREATING RISK
We are concerned with safety that depends on the correct
functioning of equipment
Control
system
(c) Felix Redmill, 2011
Equipment
under control
CERN, May '11
Utility
(plus risks)
15
FUNCTIONAL SAFETY 2
Functional safety depends on hardware, software,
humans, data, and interactions between all of these
Control
system
Equipment
under control
Utility
(plus risks)
Humans (design, operation,
maintenance, etc.)
(human factors)
Environment (management, culture, etc.)
(c) Felix Redmill, 2011
CERN, May '11
16
ORGANISATIONS ARE ALSO COMPLEX
(Vincristine example)
• Child received injection for leukaemia into spine instead of
into vein
• Root cause analysis showed 40 points at which the
accident could have been averted
• Complexity of modern organisational systems
• Need to identify risks in advance
– But no standard can do this for us
• Requires a risk-based approach
(c) Felix Redmill, 2011
CERN, May '11
17
SAFETY - A DEFINITION
• Safety: freedom from unacceptable risk (IEC)
• Safety is not directly measurable
– But it may be addressed via risk
(c) Felix Redmill, 2011
CERN, May '11
18
VOCABULARY
• In common usage, “Risk” is used to imply:
– Likelihood (e.g. there’s a high risk of infection)
– Consequence (e.g. infection carries a high risk)
– A combination of the two
– Something, perhaps unspecified, to be avoided (e.g.
going out into the night is risky)
(c) Felix Redmill, 2011
CERN, May '11
19
RISK - DEFINITIONS
• Risk: A combination of the probability of occurrence and the
severity of its consequences if the event did occur (IEC)
• Tolerable risk: a willingness to live with a risk, so as to secure
certain benefits, in the confidence that the risk is one that is
worth taking and that it is being properly controlled (HSE)
(c) Felix Redmill, 2011
CERN, May '11
20
SAFETY AND RISK
• Safety is only measurable in retrospect
• Safety is gauged by trying to understand risk
• Risk, being of the future, is estimable but not measurable
• The higher the risk, the lower the confidence in safety
• We increase safety by reducing risk
• Two components of risk are significant:
– Likelihood of occurrence
– Magnitude of outcome
(c) Felix Redmill, 2011
CERN, May '11
21
SOME PRINCIPLES
• Absolute safety (zero risk) cannot be achieved
• ‘Doing it well’ does not guarantee safety
– Correct functionality  safety
– We must address safety as well as functionality
• Reliability is not a guarantee of safety
• We require confidence of safety in advance, not
retrospectively
• We must not only achieve safety but also demonstrate it
(c) Felix Redmill, 2011
CERN, May '11
22
A RISK-BASED APPROACH
• If safety is addressed via risk
– We must base our safety-management actions on risk
• We must understand the risks in order to manage safety
(c) Felix Redmill, 2011
CERN, May '11
23
SAFETY AS A WAY OF THINKING
• If risk is too high, it must be reduced
• But how do we know how high it is?
– Carry out risk analysis
• But even if the risk is high, does it need to be reduced?
– We must understand what risk is tolerable in the
circumstances (in the UK, apply the ALARP Principle)
• Achieving safety demands a combination of engineering
and management approaches
(c) Felix Redmill, 2011
CERN, May '11
24
RISK - TWO COMPONENTS
• Two components:
– Probability (likelihood) of occurrence
– Consequence (magnitude of outcome)
• R = f(P.C) or f(L.C)
(c) Felix Redmill, 2011
CERN, May '11
25
A SIMPLE CALCULATION
Probability of a 100-year flood = 0.01/year
Expected damage = £50M
R (financial (Expected Value)
= 0.01 x £50,000,000 = £500,000/year
(c) Felix Redmill, 2011
CERN, May '11
26
CONFIDENCE IN RISK VALUES
 Accuracy of the result depends on the reliability of information
 Reliability of information depends on the pedigree of its source
• What are the sources of the probability and consequence
values?
• What are their pedigrees?
• What confidence do we have in the risk values?
• Why do we need confidence?
– Because we derive risk values in order to inform decisions
(c) Felix Redmill, 2011
CERN, May '11
27
CONFIDENCE IN RISK CALCULATIONS
• Where did the information come from?
– What trust do we have in the source?
– When and by whom was it collected?
– Was it ever valid? If so, is it still valid?
• What assumptions have we made?
– How valid are they?
– Are we aware of them?
(c) Felix Redmill, 2011
CERN, May '11
28
COT DEATH CASE
• A study concluded that the risk of a mature, non-smoking,
affluent couple suffering a cot death is 8543 to 1
• Prof. Meadows deduced that Pr. of two cot deaths in the
same family = 8543 x 8543 = 73 million to one
• Three cot deaths in her family resulted in Mrs Clark being
convicted of infanticide
(c) Felix Redmill, 2011
CERN, May '11
29
DEPENDENCE AND NOT RANDOMNESS
• But deaths in the same family are not independent events
(probability theory assumes independence)
• One death in the family rendered Mrs Clark “in the highest-
risk category” for another
– Probability of a second death is considerably greater
than 8543 to 1
(c) Felix Redmill, 2011
CERN, May '11
30
COMMON-MODE FAILURES
• Identifying common-mode failures is a crucial part of
traditional risk analysis
(c) Felix Redmill, 2011
CERN, May '11
31
ASSUMPTIONS
• We never have full knowledge
• We fill the gaps with assumptions
• Assumptions carry uncertainty, and therefore risk
• Recognise your assumptions (if possible)
• If you admit to them (document them)
– Readers will recognise uncertainty
– Other persons may provide closer approximations
(c) Felix Redmill, 2011
CERN, May '11
32
CONTROL OF RISK
• Risk is eliminated if either Pr or C is reduced to zero
• And risk is reduced by reduction of Pr or C or both
• In many cases we have no control over C, but we may be
able to estimate it
• It may be difficult or impossible to derive Probability, but
we may be able to reduce it (e.g. by software testing &
fixing)
(c) Felix Redmill, 2011
CERN, May '11
33
WHY TAKE RISKS?
• Progress demands risk
– e.g. exploration, bridge design
– e.g. drug development (TeGenero’s TGN1412)
• Technology provides utility - at the cost of risk
• Suppliers want cheap designs
– e.g. Ronan Point (need to foresee hazards)
• To save money
– e.g. Cutting corners on a project
• To save face
– E.g. “I’d rather die than make a fool of myself”
• We can’t avoid taking risks in every decision and action
(c) Felix Redmill, 2011
CERN, May '11
34
RISK VS. RISK
• Decision is often not between risk and no risk
– But between one risk and another
– Decisions may create risks (unintended consequences)
• Surgery, medication, or keep the illness
• Consequences of being late vs. risks of driving fast
• Solar haze vs. global warming
• Software maintenance can introduce a new defect
– Change creates a new product; test results obsolete
• Recognise both risks
• Assess both
• Carry out impact analysis to identify new hazards
(c) Felix Redmill, 2011
CERN, May '11
35
SOME NOTES ON RISK
• A single-system risk may be tiny, but the overall risk of many
systems may be high (individual vs. societal risk)
• Risk per usage may be remote, but daily use may accrue a
high risk (one-shot vs. lifetime risk)
• Beware of focusing on a single risk; a system is likely to carry
numerous risks
• Confidence in probabilities attributed to rare events must be
low
• For random events with histories (e.g. electromechanical
equipment failure) frequencies may be deduced
– Given similar circumstances, they may be predictive
• For systematic events (such as software failure) history is not
an accurate predictor of the future
(c) Felix Redmill, 2011
CERN, May '11
36
RISK AND UNCERTAINTY
• Risk is of the future - there is always uncertainty
• Risk may be estimated but not measured
• Risk is open to perception
• Identifying and analysing risks requires information
• The quality of information depends on the source
• 'Facts' are open to different interpretations
• Risk cannot be eliminated if goals are to be achieved
• Risk management activities may introduce new risks
• In the absence of certainty, we need to increase confidence
• Reducing uncertainty depends on gathering information
(c) Felix Redmill, 2011
CERN, May '11
37
RISK MANAGEMENT STRATEGIES
•
•
•
•
•
Avoid
Eliminate
Reduce
Minimise - within defined constraints
Transfer or share
- Financial risks may be insured
- Technical risks may be transferred to experts, maintainers
• Hedge
- Make a second investment, which is likely to succeed if the
first fails
• Accept
- Must do this in the end, when risks are deemed tolerable
- Need contingency plans to reduce consequences
(c) Felix Redmill, 2011
CERN, May '11
38
RISK IS OPEN TO PERCEPTION
•
•
•
•
•
•
•
•
•
•
Voluntary or involuntary
Control in hands of self or another
Statistical or personal
Level of knowledge, uncertainty
Level of dread or fear evoked
Short-term or long-term view
Severity of outcome
Value of the prize
Level of excitement
Status quo bias
 Perception determines where we look for risks and what
we identify as risks
(c) Felix Redmill, 2011
CERN, May '11
39
PROGRAMME FOR COMBATTING DISEASE
(1)
• The country is preparing for the outbreak of a disease
which is expected to kill 600 people. Two alternative
programmes to combat it have been proposed, and the
experts have deduced that the precise estimates of the
outcomes are:
• If programme A is adopted, 200 people will be saved
• If programme B is adopted, there is a one third probability
that 600 people will be saved and a two thirds probability
that nobody will be saved
(c) Felix Redmill, 2011
CERN, May '11
40
PROGRAMME FOR COMBATTING DISEASE
(2)
• The country is preparing for the outbreak of a disease
which is expected to kill 600 people. Two alternative
programmes to combat it have been proposed, and the
experts have deduced that the precise estimates of the
outcomes are:
• If programme C is adopted, 400 people will die
• If programme D is adopted, there is a one third probability
that nobody will die and a two thirds probability that 600
people will die
(c) Felix Redmill, 2011
CERN, May '11
41
UNINTENDED CONSEQUENCES
• Actions always likely to have some unintended results
– Downsizing but loose the wrong staff
 e.g. University redundancies in UK
– Staff come to rely entirely on a support system
– Building high in mountain causes landslide
– Abolishing DDT increased malaria
– Store less chemical at plant, more trips by road
• Unintended, but not necessarily unforeseeable
• Foreseeable, but not necessarily obvious
• Carry out hazard and risk analysis
– Better to foresee them than encounter them later
(c) Felix Redmill, 2011
CERN, May '11
42
RISK COMMUNICATION
• The risk analyst is usually not the decision maker
– Risk information must be transferred (liver biopsy)
• Usually only results are communicated
– e.g. There’s an 80% chance of success
• But are they correct (Bristol Royal Infirmary)?
– Overconfidence bias
• Managers rely heavily on risk information, collected,
analysed, packaged, by other staff
• But what confidence does the staff have in it
– What were the information sources, analysis methods,
and framing choices?
– What uncertainties exist?
– What assumptions were made?
(c) Felix Redmill, 2011
CERN, May '11
43
COMMUNICATION OF RISK INFORMATION
• ‘The risk is one in a million’
• One in seventeen thousand of dying in a road accident
– Age, route, time of day
• Framing
• Appropriate presentation of numbers
– 1 x 10-6
– One in a million
– One in a city the size of Birmingham
– (A launch per day for 300 years with only one failure)
(c) Felix Redmill, 2011
CERN, May '11
44
RISK IS TRICKY
• We manage risks daily, with reasonable success
• Our risk management is intuitive
– We do not recognise mishaps as the results of poor
risk management
• Risk is a familiar subject
• We do not realise what we don’t know
• We do not realise that our intuitive techniques for
managing simple situations are not adequate for complex
ones
• Risk estimating can be simple arithmetic
• The difficult part is obtaining the correct information on
which to base estimates - and decisions
(c) Felix Redmill, 2011
CERN, May '11
45
SAFETY ENGINEERING PRINCIPLES
(c) Felix Redmill, 2011
CERN, May '11
46
A SIMPLE MODEL OF SYSTEM SAFETY
Failure
or unsafe
deviation
Safe state
(in any
mode)
Accident
Danger
Disaster
Restoration
Recovery
(c) Felix Redmill, 2011
CERN, May '11
47
SAFETY ACROSS THE LIFE CYCLE
• Safety activities must extend across the life of a system
– And must be planned accordingly
• Modern safety standards call for the definition and use of a
safety life cycle
• A life-cycle model shows all the phases of a system’s life,
relative to each other
• The ‘overall safety lifecycle’ defined in safety standard IEC
61508 is the best known model
(c) Felix Redmill, 2011
CERN, May '11
48
SAFETY LIFE CYCLE MODELS
• Provide a framework for planning safety engineering and
management activities
• Provide a guide for creating an infrastructure for the
management of project and system documentation
• Remind engineers and managers at each phase that they
need to take other phases into consideration in their
planning and activities
• Facilitate the demonstration as well as the achievement of
safety
• The model in safety standard IEC 61508 is the best known
(c) Felix Redmill, 2011
CERN, May '11
49
OVERALL SAFETY LIFECYCLE
Concept
1
2
Overall scope
definition
3
Hazard and risk
analysis
4
Overall safety
requirements
5
Safety requirements
allocation
Overall planning of:
6
O&M
(c) Felix Redmill, 2011
7
Safety
validation
Realisation of:
8
Installation &
commissioning
9
Safetyrelated
E/E/PES
12
Overall installation
and commissioning
13
Overall safety
validation
14
Overall operation,
maintenance & repair
Decommissioning or
CERN, May
'11
disposal
16
10
Other tech.
safetyrelated
systems
15
11
External
risk
reduction
facilities
Overall
modification
and retrofit
50
AFTER STAGE THREE
Work done after stage 3 of the model creates additions to
the overall system that were not included in the hazard
and risk analysis of stage 3
(c) Felix Redmill, 2011
CERN, May '11
51
SAFETY LIFECYCLE SUMMARY
Understand the functional goals and design

Identify the hazards

Analyse the hazards

Determine and assess the risks posed by the hazards

Specify the risk reduction measures and their SILs

Define the required safety functions (and their SILs)

Carry out safety validation

Operate, maintain, change, decommission, dispose safely
(c) Felix Redmill, 2011
CERN, May '11
52
THE SAFETY-CASE PRINCIPLE
• The achievement of safety must be demonstrated in advance
of deployment of the system
• Demonstration is assessed by independent safety assessors
• Safety assessors will not (cannot)
– Examine a complex system without guidance
– Assume responsibility for the system’s safety
• Their assessment is guided by claims made by developers,
owners and operators of the system
• The claims must be for the adequacy of safety in defined
circumstances, considering the context and application and
the benefits of accepting any residual risks
(c) Felix Redmill, 2011
CERN, May '11
53
THE SAFETY CASE
• The purpose: Demonstration of adequate and appropriate
safety of a system, under defined conditions
– To convince ourselves of adequate safety
– The basis of independent safety assessment
– Provides later requirements for proof
(It can protect and it can incriminate)
• The means
– Make claims of what is adequately safe
o And for what it is adequately safe
– Present arguments for why it is adequately safe for the
intended purpose
– Provide evidence in support of the arguments
(c) Felix Redmill, 2011
CERN, May '11
54
GENERALIZED EXAMPLE
• Claim: System S is acceptably safe when used in Application A
• Claims are presented as structured arguments
– The use of System S in Application A was subjected to
thorough hazard identification
– All the identified hazards were analysed
– All risks associated with the hazards either were found to be
tolerable or have been reduced to tolerable levels
– Emergency plans are in place in case of unexpected
hazardous events
• The safety arguments are supported by evidence
– e.g., Evidence of all activities and results, descriptions of all
relevant documentation, and references to it
(c) Felix Redmill, 2011
CERN, May '11
55
THE CASE MUST BE STRUCTURED
• Demonstration of adequate safety of a system requires
demonstration of justified confidence in its components
• Some components may be COTS (commercial off-the-shelf)
– Don’t wait until too late to find that they cannot be justified
• Evidence is derived from different sources, and different
categories of sources
• The evidence for each claim must be structured so as to
support a logical argument for the “top-level” claim
(c) Felix Redmill, 2011
CERN, May '11
56
PRESENTATION OF SAFETY CLAIMS
• The claims must be structured so that the logic of the overall
case is demonstrated
• The principal claims may be presented in a “top-level”
document
– With references out to the sources of supporting evidence
(gathered throughout the life cycle)
• Modern software-based tools are available for the
documentation of safety cases (primarily: GSN (goal-structured
notation))
(c) Felix Redmill, 2011
CERN, May '11
57
THE NATURE OF EVIDENCE
• Evidence may need to be probabilistic
– e.g. for nuclear plants
• It may be qualitative
– e.g. for humans, software
• In applications of human control it may need to include
– Competence, training, management, guidelines
• It is collected throughout the system’s life
– Importantly during development
(c) Felix Redmill, 2011
CERN, May '11
58
EVIDENCE PLANNING - 1
• Any system-related documentation (project, operational,
maintenance) may be required as evidence
• It should be easily and quickly accessible to
– Creators of safety arguments
– Independent safety assessors
• Numbering and filing systems should be designed appropriately
• Principal safety claims should be identified in advance
– Activities may need to be planned so that appropriate
evidence is collected and stored
• The safety case should be developed throughout a development
project
– And maintained throughout a system’s life
(c) Felix Redmill, 2011
CERN, May '11
59
EVIDENCE PLANNING - 2
• The safety case structure should be designed early
• Evidential requirements should be identified early and
planned for
• Safety case development should be commenced early
• Evidence of software adequacy may be derived from
– Analysis
– Testing
– Proven-in-use
– Process
(c) Felix Redmill, 2011
CERN, May '11
60
VALIDITY OF A SAFETY CASE
Validity (and continued validity) depends on (examples only)
• Observance of design and operational constraints, e.g.:
– Use only specified components
– Don’t exceed maximum speed
• Environment, e.g.:
– Hardware: atmospheric temperature between T1 & T2C
– Software: operating system remains unchanged
• Assumptions remain valid, e.g.:
– Those underlying estimation of occurrence probability
– Routine maintenance carried out according to spec.
(c) Felix Redmill, 2011
CERN, May '11
61
HAZARD AND RISK ANALYSIS
(c) Felix Redmill, 2011
CERN, May '11
62
NEED FOR CLARITY
• Is a software bug a risk?
• Is a banana skin on the ground a risk?
• Is a bunker on a golf course a risk?
(c) Felix Redmill, 2011
CERN, May '11
63
THE CONCEPT OF A HAZARD
• A hazard is the source of risk
– The risks that may arise depend on context
• What outcomes could result from the hazard?
• What consequences might the outcomes lead to?
• Hazards form the basis of risk estimation
(c) Felix Redmill, 2011
CERN, May '11
64
GOLF-SHOT RISK
• What is the probability of hitting your golf ball into a bunker?
• What is the consequence (potential consequence) of hitting
your ball into a bunker?
• Should I “take the risk”?
– In this case: What level of risk should I take?
• What’s missing is a question about benefit!
(c) Felix Redmill, 2011
CERN, May '11
65
WHAT IS RISK ANALYSIS?
• If we take risk to be a function of Probability (Pr) &
Consequence (C)
– Then, in order to estimate a value of a Risk, we need to
derive values of Pr and C for that risk
• Values may be qualitative or quantitative
• They may need to be more or less “accurate”, depending on
importance
• Thus, risk analysis consists of:
– Identifying relevant hazards
– Collecting evidence that could contribute to the
determination of values of Pr and C for a defined risk
– Analysing that evidence
– Synthesising the resulting values of Pr and C
(c) Felix Redmill, 2011
CERN, May '11
66
PROBABILITY OF WHAT EXACTLY?
• Automobile standards focus on the control of a
vehicle (and the possible loss of it)
• Each different situation will
– Occur with a different probability
– Result in different consequences
– Carry different risks
(c) Felix Redmill, 2011
CERN, May '11
67
CHOICE OF CONSEQUENCE ESTIMATES
• Worst possible
• Worst credible
• Most likely
• “Average”
(c) Felix Redmill, 2011
CERN, May '11
68
DETERMINATION OF RISK VALUES
Threat to
security
Safety hazard
Causal
analysis
Threat of
damage
Potential for
unreliability
Risk
Consequence
analysis
Potential for
unavailability
(c) Felix Redmill, 2011
CERN, May '11
69
PRELIMINARY REQUIREMENTS
• Knowledge of the subject of risk
• Understanding of the current situation and context
• Knowledge of the purpose of the intended analysis
– The questions (e.g. tolerability) that it must answer
 Such knowledge and understanding are essential to
searching for the appropriate information
(c) Felix Redmill, 2011
CERN, May '11
70
BOTTOM-UP ANALYSIS
(Herald of Free Enterprise)
Bowsun asleep in his cabin when ship is due to depart

Bow doors not closed

Ship puts to sea with bow doors open

Water enters car deck

As ship rolls, water rushes to one side

Ship capsizes

Lives lost
(c) Felix Redmill, 2011
CERN, May '11
71
TOP-DOWN ANALYSIS
(Herald of Free Enterprise)
Ship puts to sea with bow doors open
Bosun did not close doors
Bosun not available
to close doors
Bosun not
on ship
Bosun on board
but not at station
Bosun asleep
in cabin
(c) Felix Redmill, 2011
Problem with doors
& bosun can’t close them
Problem
with closing
mechanism
Door or
hinge
problem
Bosun in
bar
CERN, May '11
Problem
with power
supply
72
RISK MANAGEMENT PROCESS
• Define scope of study
• Identify the hazards
• Analyse the hazards to determine the risks they pose
• Assess risks against tolerability criteria
• Take risk-management decisions and actions
• It may also be appropriate to carry out emergency
planning and prepare for the unexpected
– If so, we need to carry out rehearsals
(c) Felix Redmill, 2011
CERN, May '11
73
FOUR STAGES OF RISK ANALYSIS
(But be careful with vocabulary)
• Definition of scope
– Define the objectives and scope of the study
• Hazard identification
– Define hazards and hazardous events
• Hazard analysis
– Determine the sequences leading to hazardous events
– Determine likelihood and consequences of hazardous
events
• Risk assessment
– Assess tolerability of risks associated with hazardous
events
(c) Felix Redmill, 2011
CERN, May '11
74
DEFINITION OF SCOPE
• Types of risks to be studied
– e.g. safety, security, financial
• Risks to whom or what
– e.g. employees, all people, environment, the company,
the mission
• Study boundary
– Plant boundary
– Region
• Admissible sources of information
– e.g. stakeholders (which?), experts, public
(c) Felix Redmill, 2011
CERN, May '11
75
SCOPE OF STUDY
• How will the results be used?
– What questions are to be answered?
– What decisions are to be made?
• What accuracy and confidence are required?
• Define study parameters
– What effort to be invested?
– What budget afforded?
– How much time allowed?
(c) Felix Redmill, 2011
CERN, May '11
76
OUTCOME SUBJECTIVELY INFLUENCED
• Defining scope is subjective
– Involves judgement
– Includes bias
– Can involve manipulation
• Scope definition
– Influences the nature and direction of the analysis
– Is a predisposing factor on its results
(c) Felix Redmill, 2011
CERN, May '11
77
VOCABULARY - CHOICE OF TERMS
• Same term means different things to different people
• Same process given different titles
• There is no internationally agreed vocabulary
• Even individuals use different terms for the same process
• Beware: ask what others mean by their terms
• Have a convention and define it
(c) Felix Redmill, 2011
CERN, May '11
78
SOME TERMS IN USE
Hazard identification,
Risk identification
Risk analysis,
Risk
assessment,
Risk
m anagement
Hazard analysis,
Risk analysis
Risk assessm ent,
Risk evaluation
Risk mitigation,
Risk reduction,
Risk managem ent
(c) Felix Redmill, 2011
CERN, May '11
79
AN APPROPRIATE CONVENTION?
Scope definition
Risk
analysis
Hazard identification
Hazard analysis
Risk assessment
Risk com munication
Risk m itigation
Risk
management
Em ergency planning
(c) Felix Redmill, 2011
CERN, May '11
80
HAZARD AND RISK ANALYSIS
• Obtaining appropriate information, from the most
appropriate sources, and analyzing it, while making
assumptions that are recognized, valid, and as few as
possible
(c) Felix Redmill, 2011
CERN, May '11
81
HAZARD IDENTIFICATION
• The foundation of risk analysis
- Identify the hazards (what could go wrong)
- Deduce their causes
- Determine whether they could lead to undesirable
outcomes
• Knowing chains of cause and effect facilitates decisions
on where to take corrective action
• But many accidents are caused by unexpected
interactions rather than by failures
(c) Felix Redmill, 2011
CERN, May '11
82
HAZARD ID METHODS
• Checklists
• Brainstorming
• Expert judgement
• What-if analysis
• Audits and reports
• Site inspections
• Formal and informal staff interviews
• Interviews with others, such as customers, visitors
• Specialised techniques
(c) Felix Redmill, 2011
CERN, May '11
83
WHAT WE FIND DEPENDS ON WHERE WE
LOOK
• We don’t find hazards where we don’t look
• We don’t look because
– We don’t think of looking there
– We don’t know of that place
– We assume there are no hazards there
– We assume that the risks are small
• Must take a methodical approach
– And be thorough
(c) Felix Redmill, 2011
CERN, May '11
84
CRACK IN “TIRELESS” COOLING SYSTEM
(c) Felix Redmill, 2011
CERN, May '11
85
EXTRACT FROM ‘FILE ON 4’ INTERVIEW (12/12/00)
•
•
•
•
•
•
•
•
•
•
(O'Halloran) For the Navy Captain Hurford concedes that the possibility that this critical
section of pipework might fail was never even considered in the many years that these 12
submarines of the Swiftsure and Trafalgar classes have been in service.
(Hurford) "This component was analysed against its duty that it saw in service and was
supposed never to crack and so the fact that this crack had occurred in this component in the
way that it did and caused a leak before we had detected it, is a serious issue.”
(O'Halloran) How big a question mark does this place over your general risk probability
assumptions … about the whole working of one of these nuclear reactors.
(Hurford) "It places a question on the surveillance that we do when the submarines are in
refit and maintenance, unquestionably”
(O'Halloran) How long have these various submarines been in service ?
(Hurford) "The oldest of the Swiftsure class came into service in the early seventies”
(O'Halloran) So has this area of the pipework ever been looked at in any of the submarines,
the 12 hunter killer submarines now in service ?
(Hurford) "No it hasn't, because the design of the component was understood and the
calculations showed and experience showed that there would be no problem.”
(O'Halloran) But the calculations were wrong ?
(Hurford) "Clearly there is something wrong with that component that caused the crack and
we don't know if it was the calculations or whether it was the way it was made and that what
is being found out in the analysis at the moment"
(c) Felix Redmill, 2011
CERN, May '11
86
REPEAT FORMAL HAZARD ID
• Hazard identification should be repeated
when new information becomes available and when
significant change is proposed. e.g.
– At the concept stage
– When a design is available
– When the system has been built
– Prior to system or environmental change during
operation
– Prior to decommissioning
(c) Felix Redmill, 2011
CERN, May '11
87
HAZARD ANALYSIS
• Analyse the identified hazards to determine
– Potential consequences
 Worst credible
 Most likely
– Ways in which they could lead to undesirable outcomes
– Likelihood of undesirable outcomes
• Sometimes rough estimates are adequate
• Sometimes precision is essential
– Infusion pump (too much: poison; too little: no cure)
– X-ray dose (North Stafford)
(c) Felix Redmill, 2011
CERN, May '11
88
NOTES ON HAZARD ANALYSIS
• Two aspects of a hazard: likelihood and consequence
• Consequence is usually closely related to system goals
– So risk reduction may focus on reduction of likelihood
• Usually numerous hazards associated with a system
• Analysis is based on collecting and deriving appropriate
information
• Pedigree of information depends on pedigree of its source
• The question of confidence should always be raised
• Likelihood and consequences may be expressed
quantitatively or qualitatively
(c) Felix Redmill, 2011
CERN, May '11
89
NOTES ON HAZARD ANALYSIS - 2
• For random events (such as hardware component failure)
– Historic failure frequencies may be known
– Probabilistic hazard analysis may be possible
• For systematic events (such as software failures)
– History is not an accurate predictor of the future
– Qualitative hazard analysis is necessary
• When low event frequencies (e.g. 10-6 per year) are desired,
confidence in figures must be low
(c) Felix Redmill, 2011
CERN, May '11
90
THE NEED FOR QUALITATIVE ANALYSIS
AND ASSESSMENT
• When failures and hazardous events are random
– Historic data may exist
– Numerical analysis may be possible
• When failures and hazardous events are systematic, or
historic data do not exist
– Qualitative analysis and assessment are necessary
• Example of qualitative techniques is the risk matrix
(c) Felix Redmill, 2011
CERN, May '11
91
HAZARD ANALYSIS TECHNIQUES
• FMEA analyses the effects of failures
• FMECA analyses the risks attached to failures
• HAZOP analyses both causes and effects
• Event tree analysis (ETA) works forward from identified
hazards or events (e.g. component failures) to determine
their consequences
• Fault tree analysis (FTA) works backwards from identified
hazardous events to determine their causes
(c) Felix Redmill, 2011
CERN, May '11
92
USING THE RESULTS
• The purpose is to derive reliable information on which to
base decisions on risk-management action
• ‘Derive’ - by carrying out risk analysis
• ‘Reliable’ - be thorough as appropriate
• ‘Information’ - the key to decision-making
• ‘Decisions’ - risk analysis is not carried out for its own sake
but to inform decisions (usually made by others)
• Hazard identification and analysis may be carried out
concurrently
(c) Felix Redmill, 2011
CERN, May '11
93
RISK ASSESSMENT
• To determine the tolerability of analysed risks
– So that risk-management decisions can be taken
• Need tolerability criteria
• Tolerability differs according to circumstance
– e.g. medical
• Tolerability differs in time
– e.g. nuclear
• Tolerability differs according to place
– e.g. oil companies treatment of staff and environment in
different countries
(c) Felix Redmill, 2011
CERN, May '11
94
TOLERABLE RISK
• Risk accepted in a given context based on the current
values of society
• Not trivial to determine
• Differs across industry sectors
• May change with time
• Depends on perception
• Should be determined by discussion among parties,
including
– Those posing the risks
– Those to be exposed to the risks
– Other stakeholders, e.g. regulators
(c) Felix Redmill, 2011
CERN, May '11
95
THE HSE’S ALARP PRINCIPLE
(c) Felix Redmill, 2011
CERN, May '11
96
RISK TOLERABILITY JUDGEMENT
• In the case of machinery
– We may know what could go wrong (uncertainty is low)
– It may be reversible (can fix it after one accident)
• In the case of BSE or genetically modified organisms
– The risk may only be suspected (uncertainty is high)
– It may be irreversible
• Moral values influence the judgement of how much is enough
• If we don’t take risks we don’t make progress
(c) Felix Redmill, 2011
CERN, May '11
97
HAZARD AND RISK ANALYSIS
TECHNIQUES
(c) Felix Redmill, 2011
CERN, May '11
98
TECHNIQUES
Techniques support risk analysis
– They should not govern it
(c) Felix Redmill, 2011
CERN, May '11
99
TECHNIQUES TO BE CONSIDERED
•
•
•
•
•
•
•
•
Failure (fault) modes and effects analysis (FMEA)
Failure (fault) modes, effects and criticality analysis (FMECA)
Hazard and operability studies (HAZOP)
Event tree analysis (ETA)
Fault tree analysis (FTA)
Risk matrices
Human reliability analysis (HRA - mentioned only)
Preliminary hazard analysis (PHA)
(c) Felix Redmill, 2011
CERN, May '11
100
A SIMPLE CHEMICAL PLANT
P1
Fluid A
V1
V2
Vat R
P2
Fluid B
V3
(c) Felix Redmill, 2011
CERN, May '11
101
FAILURE MODES AND EFFECTS ANALYSIS
• Usually qualitative investigation of the modes of failure of
individual system components
• Components may be at any level (e.g. basic, sub-system)
• Components are treated as black boxes
• For each failure mode, FMEA investigates
– Possible causes
– Local effects
– System-level effects
• Corrective action may be proposed
• Best done by a team of experts with different viewpoints
(c) Felix Redmill, 2011
CERN, May '11
102
FAILURE MODES AND EFFECTS ANALYSIS
• Boundary of study must be clearly defined
• Does not usually find the effects of
– Multiple faults or failures
– Failures caused by communication and interactions
between components
– Integration of components
– Installation
(c) Felix Redmill, 2011
CERN, May '11
103
EXAMPLE FMEA OF A SIMPLE CHEMICAL
PROCESS
Study
No.
1
It em
Pump
P1
2
3
Failure
mode
Fails to
st art
Burns out
Valve
V1
4
Sticks
closed
Sticks
open
Possible
causes
1. No power
2. Burnt out
1. Loss of
lubricant
2. Excessive
t emperatur e
1. No power
2. Jammed
1. No power
2. Jammed
(c) Felix Redmill, 2011
CERN, May '11
Local
effe cts
Fluid
does
flow
Fluid
does
flow
A
not
A
not
Fluid A
cannot
flow
Cannot
st op
flow of
Fluid A
Syst emlevel
effe cts
Excess of
Fluid B in
Vat R
Excess of
Fluid B in
Vat R
Monitor
pump
operat ion
Add alarm
to pump
monito r
Excess
Fluid B
Vat R
Danger
excess
Fluid A
Vat R
Monitor
Valve
operat ion
Int roduce
addit ional
valve in
series
of
in
of
of
in
Proposed
correct ion
104
FAILURE MODES, EFFECTS AND
CRITICALITY ANALYSIS
• FMECA = FMEA + analysis of the criticality of each failure
• Criticality in terms of risk or one of its components
– i.e. severity and probability or frequency of occurrence
– Usually quantitative
• Purpose: to identify the failure modes of highest risk
– This facilitates prioritization of corrective actions,
focusing attention and resources first where they will
have the greatest impact on safety
• Recording: add one or two columns to a FMEA table
(c) Felix Redmill, 2011
CERN, May '11
105
HAZARD AND OPERABILITY STUDIES
(HAZOP)
• The methodical study of a documented representation of a
system, by a managed team, to identify hazards and
operational problems.
• Correct system operation depends on the attributes of the
items on the representation remaining within design intent.
Therefore, studying what would happen if the attributes
deviated from design intent should lead to the identification
of the hazards associated with the system's operation. This
is the principle of HAZOP.
• 'Guide words' are used to focus the investigation on the
various types of deviation.
(c) Felix Redmill, 2011
CERN, May '11
106
A GENERIC SET OF GUIDE WORDS
Guide word
Meaning
No
The complete negat ion of t he design int ent ion. No part of t he
int ent ion is achieved, but nothing else happens.
More
A quant it at ive increase.
Less
A quant it at ive decrease.
As well as
A qualit at ive increase. All t he design int ent ion is achieved, t ogether
wit h somet hing addit ional.
Part of
A qualit at ive decrease. Only part of t he design int ent ion is achieved.
Reverse
The logical opposit e of t he design int ent ion.
Other t han
A complete subst it ut ion where no part of t he design int ent ion is
achieved but somet hing diff erent occurs.
Early
The design int ent ion occurs earlier in t ime t han int ended.
Late
The design int ent ion occurs lat er in t ime t han int ended.
Before
The design int ent ion occurs earlier in a sequence t han int ended.
Af t er
The design int ent ion occurs earlier in a sequence t han int ended.
(c) Felix Redmill, 2011
CERN, May '11
107
HAZOP OVERVIEW
Introductions
Presentation of design
representation
Examine
design representation
methodically
Possible
deviation from design intent
?
No
Yes
Examine
consequences
and causes
Document
results
Define follow-up
work
No
T ime up, or
completed study?
Yes
Agree documentation
Sign off meeting
(c) Felix Redmill, 2011
CERN, May '11
108
HAZOP SUMMARY
• HAZOP is a powerful technique for hazard identification and analysis
• It requires a team of carefully chosen members
• It depends on planning, preparation, and leadership
• A study usually requires several meetings
• Study proceeds methodically
• Guide words are used to focus attention
• Outputs are the identified hazards, recommendations,
questions
(c) Felix Redmill, 2011
CERN, May '11
109
EVENT TREE ANALYSIS
• Starts with an event that might affect safety
• Follows a cause-and-effect chain to the system-level
consequence
• Does not assume that an event is hazardous
• Includes both operational and fault conditions
• Each event results in a branch, so N events = 2N branches
• If event probabilities can be derived
– They may be assigned to the branches of the tree
– The probability of the final events may be calculated
(c) Felix Redmill, 2011
CERN, May '11
110
A SIMPLE EVENT TREE
Valve
operates
Valve
monitor
O.K.
Alarm
relay O.K.
Claxon
O.K.
Operator
responds
Yes
Yes
Outcome
Safe
outcome
No
Yes
No
Yes
No
Unsafe
outcomes
No
No
(c) Felix Redmill, 2011
CERN, May '11
111
REFINED EVENT TREE
Valve
operates
Valve
monitor
functions
Alarm
relay
operates
Claxon
sounds
Yes
Yes
Operator
responds
Yes
Yes
No
No
(c) Felix Redmill, 2011
CERN, May '11
112
EVENT TREE: ANOTHER EXAMPLE
Fire
starts
Fire
spreads
quickly
Sprinkler
fails to
work
People
cannot
escape
Yes (P=0.4)
Yes (P=0.2)
No (P=0.6)
Yes (P=0.1)
No (P=0.8)
Resulting
event
Multiple
fatalities
Damage
and loss
Fire
controlled
Yes
No (P=0.9)
(c) Felix Redmill, 2011
Fire
contained
CERN, May '11
113
BOTTOM-UP ANALYSIS
(Herald of Free Enterprise)
Bowsun asleep in his cabin when ship is due to depart

Bow doors not closed

Ship puts to sea with bow doors open

Water enters car deck

As ship rolls, water rushes to one side

Ship capsizes

Lives lost
(c) Felix Redmill, 2011
CERN, May '11
114
TOP-DOWN ANALYSIS
(Herald of Free Enterprise)
Ship puts to sea with bow doors open
Bosun did not close doors
Bosun not available
to close doors
Bosun not
on ship
Bosun on board
but not at station
Bosun asleep
in cabin
(c) Felix Redmill, 2011
Problem with doors
& bosun can’t close them
Problem
with closing
mechanism
Door or
hinge
problem
Bosun in
bar
CERN, May '11
Problem
with power
supply
115
FAULT TREE ANALYSIS
• Starts with a single 'top event' (e.g., a system failure)
• Combines the chains of cause and effect which could lead
to the top event
• Next-level causes are identified and added to the tree and
this is repeated until the most basic causes are identified
• Causal relationships are defined by AND and OR gates
• One lower-level event may cause several higher-level
events
• Examples of causal events: component failure, human
error, software bug
• Each top-level undesired event requires its own fault tree
• Probabilities may be attributed to events, and from these
the probability of the top event may be derived
(c) Felix Redmill, 2011
CERN, May '11
116
EXAMPLE FTA OF SIMPLE CHEMICAL
PROCESS
(c) Felix Redmill, 2011
CERN, May '11
117
THE PRINCIPLE OF THE PROTECTION SYSTEM
Required event
frequency
10 -7 per hour
AND
Dangerous failure
frequency of
equipt.: 10 - 3 / hour
(c) Felix Redmill, 2011
Reliability of safety
function
10 -4 / hour
CERN, May '11
118
COMPLEMENTARITY OF TECHNIQUES
• Compare results of FMEA with low-level causes from FTA
• Carry out HAZOP on a sub-system identified as risky by a
high-level FMEA
• Carry out ETA on low-level items identified as risky by FTA
(c) Felix Redmill, 2011
CERN, May '11
119
A RISK MATRIX
(An example of qualitative risk analysis)
Likelihood
or
Frequency
Consequence
Negligible
Moderate
High
Catastrophic
High
Medium
Low
(c) Felix Redmill, 2011
CERN, May '11
120
RISKS POSED BY IDENTIFIED HAZARDS
Likelihood
or
Frequency
Consequence
Negligible
High
Medium
Moderate
Catastrophic
H2
H1, H5
H6
Low
(c) Felix Redmill, 2011
High
H4
H3
CERN, May '11
121
A RISK CLASS MATRIX
(example only)
Likelihood
or
Frequency
Consequence
Negligible
Moderate
High
Catastrophic
High
C
B
A
A
Medium
D
C
B
A
Low
D
D
C
B
(c) Felix Redmill, 2011
CERN, May '11
122
TOLERABILITY OF ANALYSED RISKS
• Refer the cell of each analysed hazard in the risk matrix to the
equivalent cell in the risk class matrix to determine the class
of the analysed risks
– So, Hazard 1 poses a D Class risk, Hazard 2 a B, etc.
• Risk class
– Defines the (relative) importance of risk reduction
– Provides a means of prioritising the handling of risks
– Can be equated to a defined type of action
• Risk class gives no indication of time criticality
– This must be derived from an understanding of the risk
• What is done to manage risks depends on circumstances
(c) Felix Redmill, 2011
CERN, May '11
123
THOUOGHTS ON THE USE OF TECHNIQUES
• Techniques are intended to support what you have decided
to do
– They should not define what you do
• Each is useful on its own for a given purpose
– Thorough analysis requires a combination of techniques
• Risk analysis is not achieved in a single activity
– It should be continuous throughout a system’s life
• Early analysis (PHA) is particularly effective
• Techniques may be used to verify each other’s results
• And to expand on them and provide further detail
(c) Felix Redmill, 2011
CERN, May '11
124
IMPORTANCE OF EARLY ANALYSIS
Root causes
1000s
Hazards
<100
Accidents
<20
• Too many causes to start with bottom-up analysis
– Effort and financial cost would be too great
– Would lead to over-engineering for small risks
– Would miss too many important hazards
• Fault trees commence with accidents or hazards
• Carry out Preliminary Hazard Analysis early in a project
(c) Felix Redmill, 2011
CERN, May '11
125
PRELIMINARY HAZARD ANALYSIS
• At the ‘Objectives’ or early specification stage
– Potential then for greatest safety effectiveness
• Take a ‘system’ perspective
• Consider historic understanding of such systems
• Address differences between this system and historic
systems (novel features)
• Consider such matters as: boundaries, operational intent,
physical operational environment, assumptions
• Identify accident types, potential accident scope and
consequences
• Identify system-level hazards
• If a checklist is used, review it for obsolescence
• Create a new checklist for future use
(c) Felix Redmill, 2011
CERN, May '11
126
HUMAN RELIABILITY ASSESSMENT
• Human components of systems are unreliable
• Hazards posed by them should be included in risk analysis
• Several Human Reliability Assessment (HRA) techniques
have been developed for the purpose
– These will be covered later in the degree course
(c) Felix Redmill, 2011
CERN, May '11
127
RISKS POSED BY HUMANS
• We need to
– Extend risk analysis to include HRA
– Pay more attention to ergonomic considerations in
design
– Consider human cognition in interface design
– Produce guidelines on human factors in safety
• Safety is multi-dimensional, so take an interdisciplinary
approach
– Include ergonomists and psychologists in development
and operational teams
(c) Felix Redmill, 2011
CERN, May '11
128
RISKS POSED BY SENIOR MANAGERS
• Senior managers (should)
– Define safety policy
– Make strategic safety decisions
– Provide leadership in safety (or not)
– Define culture by design or by their behaviour
• They predispose an organisation to safety or accident
• Accident inquiry reports show that the contribution of
senior management to accident causation is considerable
• Yet their contributions to risk are not included in risk
analyses
• Risk analyses therefore tend to be optimistic
• Risk analyses need to include management failure
• Management should pay attention to management failure
(c) Felix Redmill, 2011
CERN, May '11
129
RISK ANALYSIS SUMMARY
• Risk analysis is the cornerstone of modern safety engineering
and management
• It can be considered as a four-staged process
• The first stage, hazard identification, is the basis of all further
work - risks not identified are not analysed or mitigated
• All four stages include considerable subjectivity
• Subjectivity, in the form of human creativity and judgement, is
essential to the effectiveness of the process
• Subjectivity also introduces bias and error
• HRA techniques need to be included in risk analyses
• We need techniques to include management risks
• Often the greatest value of risk analysis is having to do it
(c) Felix Redmill, 2011
CERN, May '11
130
SAFETY INTEGRITY LEVELS
(SILs)
(c) Felix Redmill, 2011
CERN, May '11
131
SAFETY AS A WAY OF THINKING
• Reliability is not a guarantee of safety
• If risk is too great, it must be reduced
(the beginning of 'safety thinking')
• But how do we know if the risk is too great?
– Carry out risk analysis
(safety engineering enters software and systems
engineering)
• But even if the risk is high, does it need to be reduced?
– Determine what level of risk is tolerable
(this introduces the concept of safety management)
(c) Felix Redmill, 2011
CERN, May '11
132
SAFETY INTEGRITY
• If risk is not tolerable it must be reduced
• High-level requirement of safety function is 'to reduce the
risk'
• Analysis leads to the functional requirements
• The safety function becomes part of the overall system
– Safety depends on it
• So, will it reduce the risk to (at least) a tolerable level?
• We try to ensure that it does by defining the reliability with
which it performs its safety function
– In IEC 61508 in terms of Pr. of dangerous failure
• This is referred to as 'safety integrity'
(c) Felix Redmill, 2011
CERN, May '11
133
SAFETY INTEGRITY IS A TARGET
REQUIREMENT
• Safety integrity sets a target for the maximum tolerable rate of
dangerous failures of a safety function
– (This is a ‘reliability-type’ attribute)
• Example: S should not fail dangerously more than once in 7
years
• Example: There should be no more than 1 dangerous failure in
1000 demands
• If the rate of dangerous failures of the safety function can be
measured accurately, achievement of the defined safety
integrity may be claimed
(c) Felix Redmill, 2011
CERN, May '11
134
WHY SAFETY INTEGRITY LEVELS?
• A safety integrity target could have an infinite number of
values
• It is practical to divide the total possible range into bands
(categories)
• The bands are 'safety integrity levels' (SILs)
– In IEC 61508 there are four
• N.B. Interesting that the starting point was that
safety and reliability are not synonymous
and the end point is the reliance of safety
on a reliability-type measure (rate of dangerous failure)
(c) Felix Redmill, 2011
CERN, May '11
135
IEC 61508 SIL VALUES (IEC 61508)
Safety
Integrity
Level
Low Demand Mode of
Operation
(Pr. of failure to perform its
safety functions on demand)
Continuous/High-demand Mode
of Operation
(Pr. of dangerous failure per
hour)
4
>= 10-5 to 10-4
>= 10-9 to 10-8
3
>= 10-4 to 10-3
>= 10-8 to 10-7
2
>= 10-3 to 10-2
>= 10-7 to 10-6
1
>= 10-2 to 10-1
>= 10-6 to 10-5
(c) Felix Redmill, 2011
CERN, May '11
136
RELATIONSHIP (IN IEC 61508) BETWEEN
LOW-DEMAND AND CONTINUOUS MODES
• Low-demand mode:
'Frequency of demands ... no greater than one per year'
• 1 year = 8760 hours (approx. 104)
• Continuous and low-demand figures related by factor of 104
• Claims for a low-demand system must be justified
(c) Felix Redmill, 2011
CERN, May '11
137
APPLYING SAFETY INTEGRITY TO
DEVELOPMENT PROCESS
• When product integrity cannot be measured with confidence
(particularly when systematic failures dominate)
– The target is related, in IEC 61508, to the development
process
• Processes are equated to safety-integrity levels
• But
– Such equations are judgemental
– Process may be an indicator of product, but not an
infallible one
(c) Felix Redmill, 2011
CERN, May '11
138
IEC 61508 DEFINITIONS
• Safety integrity:
Probability of a safety-related system satisfactorily
performing the required safety functions under all the
stated conditions within a stated period of time
• Safety integrity level:
Discrete level (one out of a possible four) for specifying the
safety integrity requirements of the safety functions to be
allocated to the E/E/PE safety-related systems, where SIL 4
has the highest level of safety integrity and SIL 1 the lowest
(c) Felix Redmill, 2011
CERN, May '11
139
SIL IN TERMS OF RISK REDUCTION (IEC 61508)
• The necessary risk reduction effected by a safety function
• Its functional requirements define what it should do
• Its safety integrity requirements define its tolerable rate of
dangerous failure
(c) Felix Redmill, 2011
CERN, May '11
140
THE IEC 61508 MODEL OF RISK
REDUCTION
Utility + risks
EUC
Control
system
Safety functions
(c) Felix Redmill, 2011
Protection
system
Safety functions
CERN, May '11
141
PRINCIPLE OF THE PROTECTION SYSTEM
Event frequency
10 -7
AND
EUC dangerous
failure frequency
Reliability of
safety function
10 -3
10 -4
(c) Felix Redmill, 2011
10 -4
10 -3
CERN, May '11
142
BEWARE
• Note that a ‘protection system’ claim requires total
independence of the safety function from the protected
function
(c) Felix Redmill, 2011
CERN, May '11
143
EACH HAZARD CARRIES A RISK
Residual
risk 1
T olerable
level of
risk 1
Risk
2
T olerable
level of
risk 2
Necessary reduction of risk 1
Risk
1
Increasing
risk
Actual reduction of risk 1
(c) Felix Redmill, 2011
CERN, May '11
144
EXAMPLE OF A SAFETY FUNCTION
•
•
The target SIL applies to the entire safety instrumentation system
– Sensor, connections, hardware platform, software, actuator, valve,
data
All processes must accord with the SIL
– Validation, configuration management, proof checks, maintenance,
change control and revalidation ...
(c) Felix Redmill, 2011
CERN, May '11
145
PROCESS OF SIL DERIVATION
• Hazard identification
• Hazard analysis
• Risk assessment
– Resulting in requirements for risk reduction
• Safety requirements specification
– Functional requirements
– Safety integrity requirements
• Allocation of safety requirements to safety functions
– Safety functions to be performed
– Safety integrity requirements
• Allocation of safety functions to safety-related systems
– Safety functions to be performed
– Safety integrity requirements
(c) Felix Redmill, 2011
CERN, May '11
146
THE IEC 61508 SIL PRINCIPLE
(c) Felix Redmill, 2011
CERN, May '11
147
BUT NOTE
• IEC 61508 emphasises process evidence but does not
exclude the need for product or analysis evidence
• It is impossible for process evidence to be conclusive
• It is unlikely that conclusive evidence can be derived for
complex systems in which systematic failures dominate
• Evidence of all types should be sought and assembled
(c) Felix Redmill, 2011
CERN, May '11
148
SIL ALLOCATION - 1 (from IEC 61508)
• Functions within a system may be allocated different SILs
• The required SIL of hardware is that of the software safety
function with the highest SIL
• For a safety-related system that implements safety
functions of different SILs, the hardware and all the
software shall be of the highest SIL unless it can be shown
that the implementations of the safety functions of the
different SILs are sufficiently independent
• Where software is to implement safety functions of different
SILs, all the software shall be of the highest SIL unless the
different safety functions can be shown to be independent
(c) Felix Redmill, 2011
CERN, May '11
149
SIL ALLOCATION - 2
(from IEC 61508)
• Where a safety-related system is to implement both safety
and non-safety functions, all the hardware and software
shall be treated as safety-related unless it can be shown
that the implementation of the safety and non-safety
functions is sufficiently independent (i.e. that the failure of
any non-safety-related functions does not affect the safetyrelated functions). Wherever practicable, the safety-related
functions should be separated from the non-safety-related
functions
(c) Felix Redmill, 2011
CERN, May '11
150
SIL — ACHIEVED OR TARGET
RELIABILITY?
• SIL is initially a statement of target rate of dangerous failures
• There may be reasonable confidence that achieved reliability =
or > target reliability (that the SIL has been met), when
– Simple design
– Simple hardware components with known fault histories in
same or similar applications
– Random failures
• But there is no confidence that the target SIL is met when
– There is no fault history
– Systematic failures dominate
• Then, IEC 61508 focuses on the development process
(c) Felix Redmill, 2011
CERN, May '11
151
APPLICATION OF SIL PRINCIPLE
• IEC 61508 is a meta standard, to be used as the basis for
sector-specific standards
• The SIL principle may be adapted to suit the particular
conditions of the sector
• An example is the application of the principle in the Motor
Industry guideline
(c) Felix Redmill, 2011
CERN, May '11
152
EXAMPLE OF SIL BASED ON
CONSEQUENCE
(Motor Industry)
• In the Motor Industry Software Reliability Association (MISRA)
guidelines (1994), the allocation of a SIL is determined by 'the
ability of the vehicle occupants … to control the situation
following a failure'
• Steps in determining an integrity level:
a) List all hazards that result from all failures of the system
b) Assess each failure mode to determine its controllability
category
c) The failure mode with the highest associated controllability
category determines the integrity level of the system
(c) Felix Redmill, 2011
CERN, May '11
153
MOTOR INDUSTRY EXAMPLE
Controllability
category
Acceptable failure rate
Integrity
level
Uncontrollable
Extremely improbable
4
Difficult to control
Very remote
3
Debilitating
Remote
2
Distracting
Unlikely
1
Nuisance only
Reasonably possible
0
(c) Felix Redmill, 2011
CERN, May '11
154
SIL CAN BE MISLEADING
• Different standards derive SILs differently
• SIL is not a synonym for overall reliability
– Could lead to over-engineering and greater costs
• When SIL claim is based only on development process
– What were competence, QA, safety assessment, etc.?
• Glib use of the term ‘SIL’
– No certainty of what is meant
– Is the claim understood by those making it?
– Be suspicious until examining the evidence
• SIL X may be intended to imply use in all circumstances
– But safety is context-specific
• SIL says nothing about required attributes of the system
– Must go back to the specification to identify them
(c) Felix Redmill, 2011
CERN, May '11
155
THREE QUESTIONS
• Does good process necessarily lead to good product?
• Instead of using a safety function, why not simply improve
the basic system (EUC)?
• Can the SIL concept be applied to the basic system?
(c) Felix Redmill, 2011
CERN, May '11
156
DOES GOOD PROCESS LEAD TO GOOD PRODUCT?
• Adhering to “good” process does not guarantee reliability
– Link between process and product is not definitive
• Using a process is only a start
– Quality, and safety, come from professionalism
• Not only development processes, but also operation,
maintenance, management, supervision, etc., throughout the
life cycle
• Not only must development be rigorous but also the safety
requirements must be correct
– And requirements engineering is notoriously difficult
• Meeting process requirements offers confidence, not proof
(c) Felix Redmill, 2011
CERN, May '11
157
WHY NOT SIMPLY IMPROVE THE BASIC
SYSTEM?
•
•
•
•
IEC 61508 bases safety on the addition of safety functions
This assumes that EUC + control system are fixed
Why not reduce risk by making EUC & control system safer?
This is an iterative process
– Then, if all risks are tolerable no safety functions required
– But (according to IEC 61508) claim cannot be lower than
10-5 dangerous failures/hour (SIL 1)
• If there comes a time when we can't (technological) or won't
(cost) improve further
– We must add safety functions if all risks are not tolerable
• We are then back to the standard as it is
(c) Felix Redmill, 2011
CERN, May '11
158
CAN THE SIL CONCEPT BE APPLIED TO THE
BASIC SYSTEM?
• Determine its tolerable rate of dangerous failure and translate
this into a SIL (as in the MISRA guidelines)
• But note the limit on the claim that can be made (10-5
dangerous failures per hour) to satisfy IEC 61508
(c) Felix Redmill, 2011
CERN, May '11
159
DOES MEETING THE SIL GUARANTEE SAFETY?
• Derivation of a SIL is based on hazard and risk analyses
• Hazard and risk analyses are subjective processes
• If the derived risk value is an accurate representation of the
risk AND the selected level of risk tolerability is not too high
THEN meeting the SIL may offer a good chance of safety
• But these criteria may not hold
• Also, our belief that the SIL has been met may not be justified
(c) Felix Redmill, 2011
CERN, May '11
160
SIL SUMMARY
• The SIL concept specifies the tolerable rate of dangerous
failures of a safety-related system
– It defines a “safety-reliability” target
• Evidence of achievement comes from product and process
– IEC 61508 SIL defines constraints on development
processes
• Different standards use the concept differently
– SIL derivation may be sector-specific
• It can be misleading in numerous ways
• The SIL concept is a tool
– It should be used only with understanding and care
(c) Felix Redmill, 2011
CERN, May '11
161