Вы находитесь на странице: 1из 35

Guideline for Offshore Structural Reliability Analysis - General Page No.

_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
8
FEBRUARY 20, 1995
2. RELIABILITY OF STRUCTURES 9
2.1 Fundamental principles 9
2.1.1 Introduction 9
2.1.2 Safety Disciplines 9
2.1.3 Two Different Probability Concepts 10
2.1.4 Relation of QRA, SRA and ORA to the Probability Concept 13
2.1.5 Interfaces 13
2.2 Interpretation of safety 16
2.3 Safety Format of Design Codes 17
2.4 Code Calibration 18
2.4.1 Introduction 18
2.4.2 Calibration of Partial Safety Factors for a Specific Design 19
2.4.3 Code Calibration Procedure for a Class of Designs 21
2.5 Target Safety Level 24
2.5.1 General 24
2.5.2 Probabilistic Design Format 27
2.5.3 Calculation of Reliability Related to Established Design Practice 27
2.5.4 Reliability Analysis Compared with Risks Accepted by Society 28
2.5.5 Target Failure Probabilities Based on Historical Data 31
2.5.6 Target Failure Probabilities in Design Codes 32
2.5.7 Recommended Method 32
2.6 Common Mistakes in the Use of SRA 34
2.7 Standard Problems, Reasons and Remedies Relating to Codes 35
2.8 Competence Requirements 35
2.9 General Requirements to an SRA 36
2.10 Standard List of Content for an SRA Analysis Report 37
REFERENCES 38
FURTHER READING 39
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
9
2. Reliability of Structures
2.1 Fundamental principles
2.1.1 Introduction
Within the safety, risk and reliability disciplines a number of methods are used. Within different
industries the disciplines are named differently, e.g., Quantitative Risk Analysis, Probabilistic
Risk Analysis, Probabilistic Safety Assessment, Formal Safety Assessment are different names
for essentially the same methods. Within the offshore area, the most important disciplines are
usually referred to as:
Quantitative Risk Analysis (QRA)
Structural Reliability Analysis (SRA)
Organisational Reliability Analysis (ORA), including organisation, procedures, software and
human reliability
In understanding the use of SRA it is important to be aware of the basic assumptions behind an
SRA and the interfaces to other disciplines.
2.1.2 Safety Disciplines
Structural Reliability Analysis (SRA): SRA is concerned with the estimation or calculation of
the reliability or probability of failure of a structure, a structural member, or a structural element.
The methods are described in some detail in Chapter 3. The analysis should be based on state-of-
the-art physical models. SRA also constitutes the formal theory behind calibration of design
codes, and SRA together with state-of-the-art physical models may be used to calibrate design
codes where other state-of-practice models are used together with the appropriate code
formulation to achieve the targeted reliability level.
Quantitative Risk Analysis (QRA): QRA is concerned with the estimation and calculation of
overall risk to human health and safety, the environment and assets represented by the
installations. The analysis consists of the following main steps:
Hazard Identification
Assessment of probabilities /frequencies of initiating events
Accident Development (how an initiating event can develop into different accidental events)
Consequence Assessment (calculation of consequences of different accidents)
Calculation of risks
The analysis is as a tool for modifying the design and/or operation such that the risk satisfies the
targeted safety level.
Organisational Reliability Analysis (ORA): This discipline is concerned with how
organisations, their internal procedures and humans, influence reliability of processes and quality
of products. ORA thus includes the field of Human Reliability (HR), see Swain and Guttman
(1983).
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
10
2.1.3 Two Different Probability Concepts
Within the SRA, QRA and ORA disciplines two different probability concepts exists. By most
engineers this difference is not observed, even though the distinction can be very important in
many cases. We may define the two different concepts as Frequentistic probability and Bayesian
probability. The two different concepts have been in use since Bayes (1763).
Frequentistic Probability: The probability is interpreted as the frequency of occurrences of
outcomes of stochastic experiments. The frequentistic probability is sometimes referred to as
being objective, since the probability may be known to whatever precision we want if the
experiment is repeated a sufficient number of times.
Bayesian Probability; Bayes (1763): The probability is an expression of degree of belief,
Keynes (1921), Lindley (1985), De Groot (1988), Finetti (1974), and betting odds, Savage
(1954). The probability is therefore often referred to as being subjective, since two different
decision makers may have different knowledge and therefore their respective, assessed
probabilities are different. This interpretation of probability is not based on relative frequencies
and does not require many identical trials.
It should be noted that both probability concepts or interpretations satisfy the standard axioms of
the mathematical theory of probability. In practice, this may also explain that the existence of
two different probability concepts is seldom observed by engineers.
Some authors, see Kaplan and Garrick (1981), think of the two probability concepts as
fundamentally different and use the words frequency for frequentistic probability and probability
for Bayesian probability. In this guideline we use the word probability in the Bayesian tradition.
SRA belongs to the Bayesian tradition. One good reason is that two identical structures, situated
in the same environment and constructed by the same materials, are almost never built and
probably never will be. A number of the uncertainties that have to be accounted for in the
decision process are thus impossible to base on a frequentistic interpretation of uncertainties.
Therefore each experiment is only carried out once. For the aleatory uncertainties, however,
there is no difference between their modelling in an SRA and in a frequentistic probability
interpretation. The epistemic uncertainties, on the other hand, are not properties of the structure
and its environment. Therefore the probabilities and reliabilities that result from an SRA are not
structural properties. They are rather properties of our assessment and level of knowledge.
A simple example will illustrate this: If a response parameter is monitored during the service life,
this may lead to a reduced uncertainty in this response parameter. If we model this in the SRA,
and the mean value is unchanged, the result will be an increase in the reliability. However, the
structure is the same. It is our knowledge about the structure and its environment that has
changed. This knowledge has changed the reliability, our degree of belief that the structure is
safe, or the betting odds that the structure is safe.
Arguments for QRA and ORA to change from the frequentistic to the Bayesian tradition may be
found in Apostolakis (1990). This paper gives good reasons why the frequentistic interpretation
of probability should be adopted also within QRA. In the communication between the QRA,
ORA and SRA specialists it may be very important to be aware of the distinction between these
two different probability concepts. In particular this relates to questions regarding which
uncertainties are actually included in the different analyses. This, in turn, is of particular
relevance to statistical and model uncertainties and other epistemic uncertainties. Which
uncertainties are included and which are disregarded should have implications for the target
reliability level, see Section 2.5.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
11
Within the frequentistic approach the calculated probabilities may be associated directly with
uncertainties. However, within the Bayesian approach the probabilities are a consequence of the
uncertainties and, e.g., to ask for uncertainty in a calculated probability is not relevant. In SRA
the relevant question will relate to the sensitivities of the results with respect to, e.g., assumptions
and models of the uncertainties. In the Bayesian tradition, probabilities are subjective, they are
not a property of the structure or component, but of our knowledge about the structure and its
environment. In the frequentistic approach, the probabilities are objective, in the sense that
they supposedly come from observed data. However, this frequentistic interpretation is in
contradiction with all experience in practical use of reliability analysis. For new structures,
materials and environmental conditions failure data do generally not exist, decisions must be
based on models with inherent uncertainties, etc.
The difference between a reliability index
A
, calculated by including only the aleatory
uncertainties, and the reliability index , with all uncertainties included as is advocated in this
guideline for SRA, may be calculated by the omission sensitivity factors, see Section 3.5.10. For
example, if all epistemic uncertainty variables are mutually independent and represented by their
median values the relation is

A
i
i
I
=

1
2
1
(2.1)
where
i
2
denotes the uncertainty importance factor of the ith epistemic stochastic variable
[ ] x i I
i
, , 1 . This relation would then represent the difference between the frequentistic
probability and the Bayesian probability. In an attempt to compare historically observed
frequencies with probabilities calculated by SRA, the comparison should be made with
P
A A
= ( ) , and only data representing what is actually checked by the SRA should be
included in the data base.
This guideline advocates the view that we should build state-of-the-art models with state-of-
knowledge representations of epistemic uncertainties, such that these are included in the resulting
probabilities. It is thus meaningless to ask an SRA analysts to say something about the
uncertainty in the resulting probabilities. The uncertainties are included in the analysis. An SRA
analyst should, however, be prepared to document the sensitivity of the probability with respect
to all assumptions. This task is made easy with the available tools used in SRA.
The alternative would be to represent only the aleatory uncertainties as stochastic variables in the
model, and then perform sensitivity studies for changes in the fixed-valued epistemic
uncertainties. If, however, the epistemic uncertainties were quantified by distributions we could
propagate the epistemic uncertainties through the model. This would lead to a distribution
function for the probabilities.
Relating to Bayes theorem we should note that it is the epistemic uncertainties that are updated as
empirical evidence is gathered. The aleatory uncertainties cannot be removed.
There are a few technical problems involved with treating the epistemic uncertainties different
than the aleatory uncertainties. Some authors have proposed to do this, and then to treat the
aleatory uncertainties as frequentistic, and the epistemic uncertainties as Bayesian, see Kaplan
and Garrick (1981), Parry (1988). This distinction, from a conceptual point of view, is really
unnecessary and may lead to theoretical problems (Apostolakis, 1990). Treating epistemic and
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
12
aleatory uncertainties differently should reflect a different risk aversion against these two types
of uncertainties. If, at all, such a case of different risk aversion could be presented, it would be an
alternative to split the stochastic variables x representing the uncertainties into an aleatory set
x
A
and an epistemic set x
E
, i..e., x (x , x
A E
= ) and calculate the expected value of the aleatory
probability P
A
[ ]
E P f f d d
A E
R
A E
g
A E
E
E
=

<

( ) ( )
( )
x x | x x x
x x |x
A E
0
(2.2)
in which f
E
( ) x is the joint probability density function for x
E
, and f
A E
( | ) x x is the joint
probability density function for x
A
conditioned on x
E
. Higher order moments of P
A
may be
calculated accordingly. This allows for the possibility to deal with two different utility functions
or risk aversions, one for aleatory uncertainties and one for epistemic uncertainties. The
distribution of P
A
could also be presented, see Kaplan and Garrick (1981). However, these
methods are not recommended, and may lead both to theoretical problems (Apostolakis, 1990,
1993) and to communication problems, as decision makers may find uncertainties in probabilities
difficult do deal with.
The distinction between aleatory and epistemic uncertainties may in many cases seem unclear. An example may
illustrate this. When we perform fatigue crack growth experiments in the laboratory and assume a linear elastic fracture
mechanics model
da
dN
C K
m
= ( ) (2.3)
the regression analysis lead to establishment of a joint distribution of ( , ) C m , and due to small sample sizes there will
also be some epistemic uncertainty, while the joint distribution of ( , ) C m represent the aleatory uncertainty. When
we perform more experiments we may remove the epistemic uncertainty while the aleatory uncertainty remains. These
data may later be used in the prediction of the failure probability of a structure by the limit state function
g x
da
aY a
C S
m
a
a
i
m
i
I c
( )
( ( ))
=

0
1
(2.4)
where the crack of depth a grows from the initial depth a
0
to the critical a
C
within I stress cycles, the ith of which
having stress range S
i
. Y a ( ) is a geometry function modifying the stress intensity factor K as the crack grows.
The inspection event, in case a crack is not found, may be formulated as
h x
da
aY a
C S h x
m
a
a
i
m
i
J D
( )
( ( ))
, ( ) = >

0
1
0 (2.5)
where a
D
is the distribution of the smallest detectable crack size, the POD, see Section 7.6 after J stress cycles. In
the Bayesian updating, the failure probability after this inspection may be calculated by
P g h
P g h
P h
( ( ) | ( ) )
( ( ) ( ) )
( ( ) )
x x
x x
x
< > =
< >
>
0 0
0 0
0
(2.6)
Is this an updating of aleatory uncertainty? No, the uncertainty is epistemic, since we take the joint distribution of
(C,m) to represent our knowledge of the material parameters. We have no knowledge of which outcomes of (C,m)
should be expected at a specific location in the real structure.
This unclear distinction between aleatory and epistemic uncertainties has also been noted in
QRA applications. It has become quite common in QRA to use the variability or aleatory
uncertainty of a parameter value from plant to plant to characterize the epistemic uncertainty
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
13
about the parameter value for a plant not in the original population. Thus, in this case and many
similar cases, a variability or an aleatory uncertainty in one context becomes an epistemic
uncertainty in another context. Such consideration would therefore be very important in models
where the two types of uncertainties are treated differently.
2.1.4 Relation of QRA, SRA and ORA to the Probability Concept
In SRA, probabilities are calculated from analysis models and a detailed knowledge of the
environment, the loads, the response, the detailed design, the material performance etc. The
Bayesian probability concept is used.
In QRA, component failure probabilities are usually not calculated. Component failure
probabilities are based on collected data, usually organised in data bases. Hundreds of such
data bases exist for different types of components in various industries. System failure
probabilities may be calculated from component probabilities, or may be found in data bases.
The data represent historic averages, and new components will usually not be represented in
the data base to the extent they are represented in a new design. When analysing older
installations, the historical component failure probabilities are sometimes modified based on
-Experience data from the plant
-A review of the technical standard (wear, corrosion etc.) and the
maintenance programs that have taken place over the operational life.
The probability concept used is frequentistic in most applications. However, this seems to be
changing.
In ORA, probabilities may to some extent be estimated on the basis of statistics (e.g.,
WOAD). However, no generally applicable accepted method for calculation of probabilities
exists. The probabilility concept used is frequentistic.

2.1.5 Interfaces
In design there is no input from ORA to SRA: This is called The Fundamental Principle of
SRA, see Ditlevsen (1981). SRA does not consider the effect of human mistakes. It is
expected that Quality Assurance systems work and Gross Errors (which form the common
classification of organisational, procedural, software and human errors in SRA context) are
excluded from the analysis. The reason for this exclusion is that SRA is primarily concerned
with optimisation of design solution. A change in a structural dimension might be costly and
not be an adequate safety measure to safeguard against gross errors. Redundancy or
robustness are, however, used as such a safety measure and SRA are used to calculate the
effect of such redundancies. Some variations in the quality of products from organisations,
procedures and humans are accounted for in SRA, such as variations in properties that may be
analysed by testing (e.g., material data, geometrical data). In operations, ORA may be used in
connection with inspections. A Probability Of Detection (POD, see Section 7.6) curve is a
quantitative measure of human, organisational, procedural and sometimes also software
performance.
Input from ORA to QRA: QRA attempts to include ORA. However, no generally accepted
methods exist. Within the offshore industry this is observed as a lack of adequate technology,
since QRA is used extensively to estimate risks.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
14
Input from QRA to SRA: The accidental loads that exceed certain criteria (probabilities
and/or severity) are estimated by QRA, and SRA is used to design the structure in such a way
that the structural reliability is maintained under such conditions. QRA today does not provide
this information in terms of probability distributions, which is the format required by SRA.
The only known exception is when probabilistic tools are used, see PROEXP (1993).
Input from QRA to ORA: In risk estimation there is none. However, QRA gives quantitative
measures of importance of components. This may be used to focus quality assurance efforts to
such important components.
Input from SRA to ORA: SRA gives quantitative measures of importance of components or
parameters. This may be used to focus quality assurance efforts to such important components
or parameters. Otherwise there is no interface.

Input from SRA to QRA: Calculated failure probabilities from SRA are sometimes used in
QRA. However, in such cases SRA can only provide data for hazards initiated by structural
failures. Therefore, probabilities calculated by SRA cannot be used without including the
probabilities from 'gross errors' or ORA. In some few cases there is agreement between P
SRA
and P
QRA
. In general, the relation may be complicated. However; in many situations the
relation is simply P P P
QRA SRA ORA
= +

QRA
SRA ORA
Dotted: No communication
Dashed: Some communication
Hairline: Communication
Figure 2.1 Interfaces between QRA, SRA, and ORA
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
15
Responsibility of QRA Responsibility of SRA
Identify and Characterize
Accidental Events
Identify Strucural Failure
Modes
Determine Structural
Redundancy and Consequences
Determine Acceptable
Structural Risks
Determine Acceptable
Determine Structural Failure
Mechanisms and Determine
Governing Load Effect
Determine Design Accidental
Event
Determine Design Requierments
and
Load Specifications
Validation
Validation
Structural
Design Specification
Criteria
Concept Risk < Acceptance
Structural Failure Frequency
Figure 2. 2 Structural Reliability Regarding Accidental Events; Interface QRA-SRA
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
16
2.2 Interpretation of safety
Structural reliability analysis (SRA) is one part of a total safety concept, and safety factors used
for design of structures is only one item governing the overall safety. This is so because also
accidents and gross errors are major failure causes. Thus, in order to achieve a high safety level
of a structure, it is not sufficient to prescribe high safety factors that give the nominal safety level
aimed for, but in addition one has to control the possibilities of accidental events and reduce the
possibilities for gross errors during design, construction and installation.
Accidental events are controlled through use of risk analysis (QRA). Gross errors are attempted
controlled by requirements to organisation of the work, verification of design, and quality
assurance during fabrication and construction.
In-service inspection is another measure for control of the safety of the structure. For example,
the requirements to fatigue design is closely linked to the extent of in-service inspection and
possibilities for repair during service life.
Some items are difficult to include in a reliability analysis. Data may not always be available to
support such an inclusion, or relevant theories or solutions of the problems may not always be at
hand. Examples of such items which still have an impact on the total safety are:
Material selection and documentation, Charpy V-requirements and fracture toughness.
Welding procedure specifications and qualifications including testing requirements.
Requirements to non-destructive examination during fabrication.
Corrosion protection systems.
All items of significance for the total safety of the structure should be considered. For a selected
target safety level, it should therefore, in principle, be possible to derive a balanced set of safety
measures including those indicated here in addition to the safety factors used in design. Thus,
engineering judgement will be required for evaluation of the results from reliability analysis in
relation to the problem to be solved.
The calculated value of the reliability index is a function of analysis methods and distributions
assumed for the analyses, see also Section 2.5.3. This means that it is difficult to calculate the
absolute value of the inherent safety of design codes by reliability analyses. For this reason, use
of reliability analyses does not necessarily imply abrupt changes in a design code although the
results from reliability analyses may justify such a modification. However, such use may enable
inconsistencies within a particular specification to become eliminated and allow more uniform
reliabilities to be attained over a range of situations.
Structural failures from inspection reports and accidents should be filed in a format that might
support long-term calibration of structural reliability models. Reliability analysis models should
have such objectivity properties that long-term revisions of underlying assumptions in principle
can be made based on empirical evidence. In other words, the sequence of failure probabilities
computed from the sequence of updated analysis models should, in principle, asymptotically
reflect the relative frequency of failures in the corresponding sequence of analysed structural
problems when failures owing to gross errors are excluded from the pertinent data base.
A gross error may be defined as a human mistake during the design, construction and installation
process that may lead to a safety level far below that normally aimed for in design by use of
normal load and resistance factors. This human mistake may be due to lack of understanding of
the actual behaviour of the considered structure, ref., e.g., Takoma bridge, by employing
inexperienced personnel for performing the design, or by failure of the control procedures during
design, construction or installation. It is generally accepted that gross errors cannot be controlled
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
17
by the load and resistance factors. However, the robustness of a structure may be a function of
the safety factors used, and the consequences of a gross error may therefore be a function of the
values of these load and resistance factors. This illustrates some of the problems of developing
design rules that are optimal with respect to total cost, i.e., an optimal relation between
robustness of the structure, load and resistance factors, and control during the design,
construction and installation to reduce possibilities for gross errors.
2.3 Safety Format of Design Codes
The design format most frequently used for development of new design codes is that based on
partial safety factors. This form of a safety format results from requirements to an easy and
economic design.
A safety format based on partial safety factors can be expressed as
E R
d d
< (2.7)
where
E
d
= design load effect
R
d
= design resistance
The design load effect is calculated by the equation
E E
d f k
= (2.8)
where

f
= load factor
E
k
= characteristic load effect
The design resistance is calculated by the equation
R
R
d
k
m
=

(2.9)

m
= material factor
R
k
= characteristic resistance
Normally, a number of different load effects have to be combined into a resulting load effect.
The different load coefficients should be determined such that the probability of exceeding the
design load effect corresponds to the same level of reliability as that otherwise achieved by the
design code.
Whereas several load effects have to be combined into one design load effect, it may be efficient
with respect to a uniform reliability level (and structural weight) to consider a number of
combinations where the largest value will be governing for the design. Thus more than one
combination of load effects may be required for design. A calibration of one combination should
be performed within its region of validity. An example of this is the DNV rules (1977) and the
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
18
NPD regulations (1994) where two sets of equations are used to determine the design load effect
for the ultimate limit state:
E P L E D
d P k L k E k D k
= + + +

(2.10)
where the load coefficients are given in Table 2.1.
P = permanent loads
L = variable functional loads
E = environmental loads
D = deformation loads.
Table 2.1 Load Factors for the Ultimate Limit State in the DNV Rules
Load combination
P

L

E

D
a
b
1.3
1.0
1.3
1.0
0.7
1.3
1.0
1.0
The following considerations related to an efficient design format can be listed:
For common structures the design values for load effects should be independent of the design
values of the resistance.
There should only be a small set of load factors and load combinations.
One material factor should be sufficient for each material property. An efficient use of
characteristic values may reduce the need for different material factors for the resistance for
different failure modes.
Use of characteristic values can be considered as an efficient means to achieve a uniform
safety level as function of a design parameter with the use of a constant material factor, even
if different scatter in test data as function of this design parameter occurs. This can be
illustrated by test data for buckling strength of columns: Test data show different scatter as
function of slenderness, but by defining a characteristic value close to the design point value
for the material variable, as achieved from a probabilistic analysis, a rather uniform safety
level as function of slenderness can be achieved even if the same material factor is used for
the whole range of slendernesses.
2.4 Code Calibration
2.4.1 Introduction
The purpose of a code calibration is to determine a vector of partial safety factors for use in the
code checks that are executed during structural design in order to verify that the design rules as
set forth by a structural design code are fulfilled. An example of such a design rule is the simple
inequality
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
19
R L
D D
(2.11)
in which R R
D c m
= / is the design resistance and L L
D c f
= is the design load. The code
specifies how to arrive at the characteristic resistance R
C
and the characteristic load L
C
and also
specifies values of the partial safety factors
m
and
f
to be used. The design equation is a
special case of the design rule obtained by turning the inequality into an equality, i.e., R L
D D
=
in the example.
Structural reliability analysis results, obtained as outlined in Chapter 3, play an important role in
codified practice and design. Their application to calibration of partial safety factors for use in
structural design codes is of particular interest. With the first-order reliability method available, it
is possible to determine sets of equivalent partial safety factors which, when applied with design
rules in structural design codes, will lead to designs with a prescribed reliability.
A design case is formed by a specific combination of type of structure, geographical location,
material, and limit state. For a particular design case, which can be analyzed by a structural
reliability method, a set of partial safety factors can thus be determined that will lead to a design
which exactly meets the specified reliability. Different sets of partial safety factors may result for
different design cases. Calibration of partial safety factors for a specific design case is dealt with
in Section 2.4.2.
A structural design code usually has a scope that covers an entire class of design cases, formed
by combinations among several types of structures, multiple geographical locations with
individual environmental loading regimes and soil conditions, different structural materials, and
possibly a series of potential limit states or failure modes. The design code will usually specify
one common set of partial safety factors to be applied regardless of which design case is at stage.
This practical simplification implies that the prescribed reliability will usually not be met exactly,
but only approximately, when designs are carried out according to the code. Calibration of partial
safety factors for a class of design cases is dealt with in Section 2.4.3.
2.4.2 Calibration of Partial Safety Factors for a Specific Design
A specific design case is considered in the following. The safety of the associated structure is
governed by N stochastic variables which are denoted by the vector X. A design parameter
which is characteristic for the structure is chosen. This is a quantity that can be controlled during
the design, e.g., a geometrical quantity such as a length, a wall thickness, or a cross-sectional
area. In principle, there can be more than one design parameter, but one is sufficient, and this is
assumed in the following. By standard deterministic design methods, the structure is designed by
choosing an appropriate value of the design parameter such that the structure fulfills the design
rule that pertains to the limit state or failure mode in question.
A relevant limit state function for the failure mode is established in compliance with
requirements specified in Section 3.1. Structural reliability analyses of the structure are then
carried out for a series of trial values of the design parameter , each resulting in a reliability
index , a set of importance factors
2
, and a design point x
D
*
which is the most likely
realisation of X at failure. From these analyses, the particular value of is found that yields a
reliability index equal to the prescribed target reliability index
t
.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
20
A vector of characteristic values x
C
for the stochastic variables X is chosen as quantiles of the
respective distribution functions, e.g., a high quantile for a load variable and a low quantile for a
resistance variable. A set of partial safety factors is introduced. In principle, contains one
factor
i
per stochastic variable X
i
. However, one may wish to prescribe some of the safety
factors in , e.g., to a value of 1.0, in order to limit the total number of safety factors of the
code. If fewer partial safety factors are introduced than the number of stochastic variables
indicate, then these factors should be introduced as factors on those of the stochastic variables
that are most important as determined by the reliability analysis. Depending on the set-up of the
design rule, one may also prefer to apply a safety factor on a function of the characteristic values
of one or more of the stochastic variables rather than on the characteristic values of the variables
themselves. For example, in fatigue (Section 2.1.3), the material factor would be applied on the
intercept loga rather than on a itself. A total of only two partial safety factors may suffice in the
minimum case, e.g., one for load and one for resistance.
By a design value format, the design values x
D
of the stochastic variables are derived from the
characteristic values x
C
by multiplication of these values by the pertinent partial safety factors,
x x
D i i C i , ,
= , i=1,...N. Substitution of these design value expressions into the design equation,
together with the value of that is found to yield the target reliability index
t
, leads to an
equation in the partial safety factors . The solution of this equation can be interpreted as the
requirement to the partial safety factors in order to achieve a design with the prescribed target
reliability. However, when contains more than one element, which is usually the case, there
will be an infinite number of such solutions and thus an arbitrariness in selecting a particular
solution to form a unique requirement to the set of partial safety factors. This arbitrariness can be
remedied by taking into account the importance information from the reliability analysis, e.g., by
requiring that the design values x
D
shall be equal to the design point x
D
*
as determined from the
reliability analysis. This will eliminate possible ambiguities and reduce the number of solutions
for to one particular set. The design point is the most likely outcome of the stochastic
variables X at failure.
Example 2.1: Axially Loaded Steel Truss
The example given here is of a simple calibration of partial safety factors for an axially loaded steel truss. In this
example the design of the axially loaded steel truss is governed by the extreme axial force Q in its lifetime where Q
follows a Gumbel distribution
F q a q b
Q
( ) exp( exp( ( ))) = (2.12)
in which a=0.641 and b=79.1 correspond to a mean value E[Q]=80 MN and a standard deviation D[Q]=2 MN. The
yield strength of steel
F
follows a lognormal distribution with mean value E[
F
]=400 MPa and standard deviation
D[
F
]=24 MPa. The cross-sectional area of the steel truss is A. For design of the truss, the load Q shall not exceed
the capacity
F
A, so the design rule is
F D D
A q
,
and the design equation becomes
F D D
A q
,
= . The subscript
D denotes design value. The limit state function is correspondingly chosen as
G A Q
F
= (2.13)
and the area A is chosen as the design parameter.
Analysis by a first-order reliability method leads to determination of A = 0 257 . m
2
in order to meet a target
reliability index =3.719. This corresponds to a failure probability P
F
=

10
4
. The characteristic value of the extreme
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
21
axial force is taken as the mean extreme value, q E Q
C
= = 80 MN. The characteristic value of the yield strength
is taken as the 5% quantile,
F C F , , %
. = =
5
361 8 MPa. One partial safety factor
1
is introduced on the
characteristic force, and another
2
on the characteristic capacity. Substitution of the expressions for the design force
and the design capacity in the design equation yields

2 1 2 1
361 8 0 257 80 0
F C C
A q
,
. . = = (2.14)
which gives a requirement to the ratio of the partial safety factors
2 1
0 86 / . = . The reliability analysis gave the
design point values q
D
*
=86.3 MN for the force and
F D ,
*
=335.9 MPa for the strength. The partial safety factors are
therefore selected as

1
1 079 = =
q
q
D
C
*
. and

2
0 928 = =
F D
F C
,
*
,
. (2.15)
According to current design practice, a load coefficient
f
is used as a factor on the characteristic load to give the
design load, and a material coefficient
m
is used as a divisor on the characteristic strength to give the design strength.
Hence, these coefficients become

f
= =
1
1 079 . and

m
= =
1
1 077
2
. (2.16)
2.4.3 Code Calibration Procedure for a Class of Designs
In the following, the various terms and steps involved in carrying out a reliability-based code
calibration for a class of designs are defined and presented. The procedure for such a calibration
capitalises much on the procedure for calibration of partial safety factors for a specific single
design.
The scope of code is selected. As described above, the scope of the structural design code
consists of a class of design cases formed by the possible combinations of
- structures
- materials
- geographical locations, including their individual environmental loading regimes and
soil conditions
- failure modes or limit states
that the code is meant to cover. The scope of code is also referred to as the data space, and is
sometimes characterised by substitute quantities such as live-to-dead load ratio, slenderness ratio,
axial vs. bending ratio, etc., or combinations thereof.
The code objective is specified. The code objective is the target reliability index
t
corresponding to the safety level aimed at in the design, see Section 2.5. For simplicity in the
following, the same
t
is assumed for all limit states covered by the scope of code. In practice,
however,
t
may vary from one limit state to another, if the consequences of the associated
failures are different.
The demand function expresses the frequency of occurrence of a particular point in data space,
i.e., of a certain combination of structure, material, geographical location, and limit state. The
demand function is used to define weighting factors w for the various combinations of structures,
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
22
materials, geographical locations, and limit states within the scope of code. The weighting factors
thus represent the relative frequency of the various design cases within the scope, and their sum
is 1.0. The weighting factors are taken as those that are representative for the expected future
demand. For this purpose it is common to assume that the demand seen in the past is
representative for the demand in the future.
Because the code cannot be calibrated so as to always lead to designs which exactly meet the
target reliability, a closeness measure needs to be defined. This can be expressed in terms of a
penalty function for deviations from the target reliability. Several possible choices for the penalty
function exist. One that penalizes over- and underdesign equally on the scale may be
M w
i j k l
l k j i
i j k l t
=
, , , , , ,
( )
(structures) (materials) (locations) (limit states)

2
(2.17)
in which M denotes the penalty, w
i j k l , , ,
is the weighting factor for the design case identified by
the index set (i,j,k,l), and
i j k l , , ,
is the reliability index that is obtained for this design case by
design according to the code. This expression for the penalty function M may be interpreted as
the expected squared deviation from the target reliability over the scope of design cases. Other
choices for M, e.g., some that penalise underdesign more than overdesign, may be preferred in
some cases. Reference is made to Lind (1973) for suggested choices for penalty functions.
Consider in the following one particular design case within the scope of code. As in Section
2.4.2, the safety of the associated structure is governed by N stochastic variables X, and a design
parameter is chosen to characterize the structure. A relevant limit state function for the failure
mode in question is established. Structural reliability analyses of the structure are then carried out
for a series of trial values of the design parameter , each resulting in a reliability index , a set
of importance factors
2
, and a design point x
D
*
which is the most likely realisation of X at
failure. This gives the reliability index as a function of the design parameter , = ( ).
Characteristic values x
C
for the stochastic variables X are chosen as outlined in Section 2.4.2.
The characteristic values are an integral part of the design code and must be specified in the code
to ensure proper results by use of the code. A set of partial safety factors is introduced and
applied to the characteristic values x
C
to give design values x
D
as described before.
For the same series of trial values of that was used for the reliability analyses, deterministic
structural analyses are carried out by means of the design equation with x x
D i i C i , ,
= , i=1,...N,
substituted for the design values of X. This gives the partial safety factors as a function of the
design parameter , = (), when possible ambiguities in the design equation solutions for
have been eliminated. This function thus gives the requirement to in order to achieve a
structural design with the reliability index that has been found by the reliability analysis for the
current value of .
The result of the structural reliability analyses, = ( ), and the result of the deterministic
structural analyses, = (), are combined by elimination of the design parameter to give the
reliability index as a function of the set of calibrated partial safety factors , =( ).
This procedure is repeated for all design cases within the scope of code, such that one function
=( ) results for each design case, or rather
ijkl
=
ijkl
( ) in which the indices i, j, k, and l refer
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
23
to limit state, location, material, and type of structure, respectively. With these usually different
functions available, the calibration of one common set of partial safety factors to be used for
the entire scope of code is ready to commence. Because of the differences between the functions

ijkl
=
ijkl
( ) for different design cases, use of a common set of partial safety factors will lead to
designs with a higher reliability than target in some cases, and a lower reliability than target in
others. A prime requirement to the calibration of a common set of safety factors for the entire
scope of code is then that, over the scope of code, the calibrated set of safety factors shall lead to
designs with safety levels as close as possible to the target. The common set of safety factors is
therefore determined as the set that minimizes the penalty function
M w
i j k l
l k j i
i j k l
=
, , , , , ,
( (
(structures) (materials) (locations) (limit states)
) )
t
2
(2.18)
over the scope of code, if desirable with a constraint
ijkl

min
, in which
min
is some minimum
acceptable reliability index. This can be achieved by means of an optimisation technique, and
this applies also if another choice of penalty function is made, such as one that is more heavily
biased against underdesign than against overdesign. For an example of reliability-based optimal
code calibration, reference is made to Hauge et al. (1992). Figure 2.3 demonstrates how such a
code calibration leads to designs with achieved safety levels of improved uniformity as compared
to the situation prior to the calibration.
CODE CALIBRATION
DIFFERENT DESIGNS
BEFORE RULES
CALIBRATION
AFTER RULES
CALIBRATION
0.00000001
UNNECESSARILY
SAFE AND COSTLY
DESIRABLE
LEVEL
TARGET
RELIABILITY
0.1 UNACCEPTABLE,
UNSAFE
U
N
I
F
O
R
M
L
E
V
E
L
N
O
N
U
N
I
F
O
R
M
L
E
V
E
L
F
A
I
L
U
R
E
P
R
O
B
A
B
I
L
I
T
Y
MINIMUM
ACCEPTABLE
LEVEL
Figure 2.3 Achieved safety levels by design before and after an optimal code calibration
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
24
2.5 Target Safety Level
2.5.1 General
The target safety level is a function of some fundamental items which are defined and explained
in the following:
Notional Failure Probability: The calculated reliability is not a property of the structure. It is a
property of our knowledge of the structural reliability of a structure, e.g., new information about
the uncertainty of a parameter will change the calculated reliability. This implies that there is no
contradiction when two analysts calculate two different reliabilities for the same structure, since
they may have different information. However, two analysts with the same information should
arrive at the same reliabilities. Since uncertainties related to knowledge is modelled and since
they contribute to the calculated probabilities of failure, reliability analysis will tend to
overpredict the failure probabilities from failure/accidental data when gross errors are removed
from the data base.
Reference period: The structural reliability is calculated for a certain reference period, usually
one year or a design lifetime. It is recommended to use a one-year reference period in all cases.
This is most relevant for personnel safety evaluations, for inspection and maintenance planning,
and for requalification. For all these applications the annual reliability in year no. i may be
conditional on the survival up to year no. i. This conditional failure probability is then identical
to the annual hazard rate.
Scale effects: It is important to clearly define what a limit state function represents. In most cases
it defines the failure of a structural part in one particular failure mode. There may be other failure
modes for the same structural part. The size of the part, represented by the limit state, may be
implicitly defined by the size of test specimens used to generate the capacity data from tests. The
failure of a real structural element may be a function of the failure of any one or all of a number
of such parts, represented by a series or a parallel systems of such parts. This implies that the
correlation between the limit states representing the failure of such small parts may dominate the
reliability of the element. Examples of structures and limit states where the effect may be
important are:
Fatigue limit state in a tension leg. (The load process would tend to 'pick' the worst capacity.)
Fatigue limit state in a wire, rope, or anchor line. (The load process would tend to 'pick' the
worst capacity.)
Ultimate load capacity in a foundation. (Here the capacity would tend to be an average over a
large area and a mean value may be used.)
In most situations, correlation data are missing since this information would require more testing.
However, in many situations judgement on approximations of these parameters in the limit states
representing such a part have to be performed, such as considerations of either fully correlated
from part to part or fully independent from part to part. A conservative model would be to
assume independent strength and fully dependent loads. An example of scale effects and their
importance is included towards the end of the current Section 2.5.1.
Mechanical models: It is always assumed that state-of-the-art mechanical models are used in
SRA. However, simpler models may be used if they have been verified by use of more accurate
or advanced models. This would in most cases lead to simpler models that have larger
uncertainties and also possible bias. The bias and the uncertainty may then be represented by the
statistical model with a bias factor with given uncertainty and expected value different from one.
If mechanical models with different accuracy are used for developing design codes, the result is
therefore different sets of partial safety factors, if, e.g. the same fractiles in the model
distributions are used to define the characteristic values.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
25
Stochastic Models: In SRA it is expected that all relevant uncertainties and variabilities are
represented in the stochastic models. Such uncertainties may be epistemic as well as aleatory.
Example 2.2: Scale Effects in Fracture Mechanics Model by Series System Approach
The influence of scale effects in fracture mechanical models is important and is illustrated here through a simple example,
based on a series system approach. It is assumed that the Paris Equation is valid for describing the crack growth from an
initial crack size a
0
to a critical crack size a
C
.
where C and m are material parameters. The stress intensity factor K is described by the far field stress range and the
geometry function Y(a) accounting for the local geometrical effects
In this simple case the limit state function may be described as:
where N is the number of stress cycles.
The random variables in this reliability problem are in a typical case: C, m, , Y(a), a
0
and a
C
. For simplicity we deal with
a weld where each point is subjected to the same axial load. Furthermore we assume that material test data exists for
volumes of dimension v, of the same material under the same axial loading, and that the volume of the weld is V. A
reliability model for this full scale problem would then be a series system of dimension M=v/V. Under the described
assumptions, the random variables , Y(a) and a
C
could be assumed to be fully correlated between volumes v
i
, and C ,m
and a
0
. would be independent or uncorrelated from one such volume to the next. Since the marginal distributions of each of
these parameters are identical from one volume to the next, we may assume with no loss of generality that the defined
problem represents a system of equi-correlated components i of size M. The equi-correlation would have to be estimated
from data and models. The reliability of each volume element v
i
would be identical
i
=. The probability of failure of such
a system can be determined as:
and for
i
=:
The parameters governing the problem is thus the scale parameter M, the equi-correlation and the reliability index for
each of the M volume element v
i
. The equi-correlation will be dominated by the uncertainty in the fully correlated and
da
dN
= C( K )
m
(2. 19)
K = aY(a) (2. 20)
g(x) =
da
( aY(a) )
- CNE[( ) ]
0
C
a
a
m
m


(2. 21)
F
-
i
M
i
P
= (u)[1 (
+ u
1-
)]du

1
(2. 22)
F
-
M
P
= (u)[1- (
+ u
1-
) ]du

(2. 23)
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
26
uncorrelated random variables in each volume v
i
. The most important factor is most likely a
0
amongst the "uncorrelated"
variables and amongst the fully correlated variables. The correlation in thus low in case the uncertainty in a
0
dominates,
and the equi-correlation high in cases where is the dominating uncertainty. The effect is illustrated for different 's and
's as a function of M in the figures below.
Assume that the correlation is =0.50, the scale difference is M=20 and the reliability index disregarding scale effects is
=3.0 corresponding to a failure probability P
f
=1.310
-3
. Corrected for scale effects, Figure 2. 5 gives =2.07
corresponding to P
f
=1.910
-2
, an increase in failure probability by more than a factor of 10.
-1,5
-1
-0,5
0
0,5
1
1,5
2
0 20 40 60 80 100
M
B
E
T
A
1,00
0,95
0,90
0,80
0,70
0,60
0,50
0,40
0,30
0,20
0,10
0,00
Figure 2. 4 Series system, dimension M, =2
1
1,2
1,4
1,6
1,8
2
2,2
2,4
2,6
2,8
3
0 20 40 60 80 100
M
B
E
T
A
1,00
0,95
0,90
0,80
0,70
0,60
0,50
0,40
0,30
0,20
0,10
0,00
Figure 2. 5 Series system, dimension M, =3
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
27
2,5
2,7
2,9
3,1
3,3
3,5
3,7
3,9
4,1
0 20 40 60 80 100
M
B
E
T
A
1,00
0,95
0,90
0,80
0,70
0,60
0,50
0,40
0,30
0,20
0,10
0,00
Figure 2. 6 Series system, dimension M, =4
2.5.2 Probabilistic Design Format
A probabilistic design can be performed by using the following design format

calculated
or P P
target f calculated f target
< (2.24)
where

calculated
is the reliability index calculated by reliability analysis and
target
is a target
value that should be fulfilled for the design to be found acceptable.
Different procedures may be used to determine the target reliability level:
Calculation of reliability related to established design practice
Reliability analysis compared with risks accepted by the society
Economic considerations
Economic considerations may be used to determine optimal safety levels for a structure including
the different costs involved in construction, possibility for repair and the consequence cost of a
possible failure.
2.5.3 Calculation of Reliability Related to Established Design Practice
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
28
The calculated value of the reliability index is a function of analysis methods used and
distributions assumed in the analysis. Therefore, one should not directly compare reliability
indices as obtained from different models and sources.
A calculation of
calculated
and

target
should be based on the similar assumptions about
distributions, statistical and model uncertainties. It is thus understood that Eq. (2.24) is not con-
sidered to be unique when using different distribution assumptions and computational
procedures. The value of the target reliability level may be dependent on how the reliability is
calculated.
It should be noted that the industry is in a transition period from pure experience based to
reliability based design codes. As the physical models describing the failure mechanisms are
improved and more calibration studies are carried out, the knowledge of target reliabilities in
excising codes will be improved. Today the inherent target level is not precisely known or it
should hardly be presented without a number of assumptions used for its calculation due to lack
of precise knowledge of physical models and lack of data. Furthermore, the use of reliability
analysis is new to the industry and the reliability models are updated from time to time. If an
author claim that the target reliability in a specific code is
target
this will relate to the models
used. Another model refined during the years would give a different claimed
target
.
Owing to dependence on assumptions and analysis models used for reliability analysis the word
"reliabilities" in a frequency interpretation of observed structural failures cannot be used in a
narrow sense. And due to the unknown deviations from ideal predictions the computed failure
probabilities are often referred to as nominal failure probabilities. This is due to the recognition
that it is difficult to determine the absolute value of the inherent safety in e.g. design codes by
reliability analyses. The requirements to absolute reliability levels may be avoided by performing
analysis on a relative basis. I.e., a similar model and distributions are used for calculation of

target
as for
calculated
. By relating the reliability analysis to relative values it may be possible to
use probabilistic analysis for design without a specific requirement to an absolute target
reliability level. Such considerations are included in the NPD regulations accepting probabilistic
analysis by stating: "The design may be based on a more complete probabilistic analysis,
provided it can be documented that the method is theoretically suitable, and that it
provides adequate safety in typical, familiar cases." Reference is made to NPD (1994).
2.5.4 Reliability Analysis Compared with Risks Accepted by Society
Probabilities of failure from structural reliability analysis may be compared with risks associated
with other activities in the society. These risks are, in some way, experienced to be acceptable by
the society. For a comparison between the risk levels obtained from statistics on the one hand
and probabilities of failures obtained from reliability analyses on the other, the following
differences should be kept in mind:

Reliability analyses do not normally account for gross errors which are the most frequent
reason for structural failures as reported by statistics.

Statistical and model uncertainties are normally included in the reliability analyses. These
uncertainties may lead to larger failure probabilities than those due to the physical uncertainty
alone.

Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
29
The results from the reliability analysis depend on the modelling and the description of the
tails of the distributions, which are difficult to assess in most cases.
Table 2.2 Statistics for Canada in 1985
Cause or activity Annual risk/10,000 persons
All causes
All causes by age
20-24
40-44
60-64
Heart disease
Cancer
Lung cancer
All accidents
Motor vehicles
Airplane (crew)
71.5
9.0
19.3
133.4
23.0
18.3
4.5
5.3
1.7
1-2

According to Jordaan (1988), the annual risk in accidents involving a motor vehicle is 1.7x10
-4
per annum in Canada, whereas the risk for airplan accidents for a crew member is in the range 1-
2x10
-4
per year, see Table 2.2. The present level of air safety is generally perceived as
acceptable and air travel has been a well regulated industry. The risk associated with motor
vehicles is seen in society as high, but it is in some way accepted. For more details on risk
considerations see, e.g., Lotsberg (1990b).
In Norway, mortality statistics are issued annually by the Central Bureau of Statistics (1991).
Causes in 48 categories are listed, given by age groups. Some key figures are extracted and given
in Table 2.3.
Table 2.3 Death by age and cause in Norway 1990
Causes \ Age
Of
100,000
<1 1-6 7-14 15-24 25-34 35-44 45-54 55-64 65-74 75-
All Causes
1086 428 111 92 383 537 957 1525 3864 10187 27966
Car accidents 7.9 2 7 7 106 47 38 26 34 30 37
Deaths by accidents are given in Table 2.4, from Gaarder (1992).
A common logical way of defining an acceptance criterion based on individual safety would be:
From Table 2.3 it is noted that the mortality rate in Norway is lowest for the age group 7-14
years, corresponding to 0 92 10
3
.

. The causes are dominated by natural and accidents related to
leisure activities. An acceptance criterion of an industrial activity could be to state that an
activity contributing 1% to 10% to this individual risk is acceptable to the workers, thus giving
annual target reliabilities 10 10
4 5
.
Arguments like this may seem acceptable. However, for large accidents the public does not in
general accept this way of "individual safety thinking".
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
30
Table 2.4 Occupational deaths/10,000 persons, Norway, 1986-1990
1986 1987 1988 1989 1990
Land Transport 5.00 4.20 3.62 3.31 3.62
Railway 3.42 0.85 0.86 0 0.28
Sea Transport 4.11 4.34 3.09 3.53 1.87
Fisheries 5.78 8.88 4.24 8.02 4.52
Shipping 3.45 2.22 2.55 1.85 1.03
Oil activities 0.56 1.12 0 1.08 1.10
Aviation 4.04 10.28 6.67 10.28 12.96
Agriculture & forestry 1.23 1.09 0.95 9.8 0.91
Mining and quarry 0 1.44 1.52 0 1.59
Industry 0.32 0.22 0.29 0.09 0.13
Other 0.27 0.31 0.20 0.20 0.19
Total 0.61 0.60 0.46 0.45 0.42
80-90 70-79
0
2
4
6
8
10
12
80-90 70-79
Total
Industry
Aviation
Oil
Shipping
Fisheries
Landtrnsp
Figure 2.7 Deaths per 100 worklives of 40 years, 1970-1990, from Gaarder (1992)
If the target reliabilities are based on individual risk, the targets should be corrected for the
expected number of fatalities. Such societal risk criteria are given for a large number of
regulatory bodies as FN Curves, see Table 2.5. (F=Frequency=The accepted frequency of an
accident and N=Number of fatalities). The most used curves are linear in lnF and lnN or
F N
m
= .
The slope in the FN curve between 1 and 2 may also be found in observed data from Litai et al.
(1981). As a guide, it is therefore suggested to use such a risk aversion factor in acceptance
criteria with m = 1.
Table 2.5 Official societal risk criteria
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
31
FN
Curve
slope
(m)
Maximum
Tolerable
intercept with
N=1
Negligible
intercept
With N=1
Limit on N
VROM, The Netherlands
(New Plants) -2
10
3
10
5
-
Hong Kong Government
(New Plants)
-1
10
3
10
5
1000
Health&Safety
Commision,UK
(Existing ports)
-1
10
1
10
4
-
2.5.5 Target Failure Probabilities Based on Historical Data
Target reliabilities may be decided on historic accidental statistics. The decision will involve the
following steps
Table 2.6 Degree of Damage vs. Type of Accidents with Offshore Units
Number of Accidents. All Units, Worldwide, 1980-93 (WOAD, 1994)
DEGREE OF DAMAGE
TYPE OF
ACCIDENT
Total
Loss
Severe
Damage
Significant
Damage
Minor
Damage
Insignific.
Damage
Total
Anchor failure - - 17 28 6 51
Blowout - 4 12 24 76 116
Capsize 49 60 1 1 1 112
Collision 4 24 23 21 3 75
Contact 1 10 60 70 8 149
Crane accident - - 3 8 6 17
Explosion - 3 16 19 25 63
Falling load 1 2 28 18 28 77
Fire 19 37 55 61 135 307
Foundering 11 2 3 2 - 18
Grounding 5 9 14 5 1 34
Helicopter acc. - - - 14 2 16
Leakage 2 3 9 6 2 22
List 3 5 12 6 5 31
Machinery
Failure
- - - 5 6 11
Off position - - 1 1 7 9
Spill/release - 6 27 120 116 269
Structural
Damage
6 24 105 24 2 161
Towing accident - 2 1 - 36 39
Well problem - - - 2 17 19
Other - 1 5 14 24 44
Total 101 192 392 449 506 1640
1. Find relevant data on the limit state in question
2. Compare this to other causes
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
32
3. Decide on acceptability of present statistics. Is it acceptable that the offshore industry has
higher accident rates than onshore?
4. The calculated risk should not contribute significantly to this risk level. A reduction of 10%
compared to the target from all causes could be based on accidental statistics, see Table 2.6.
Table 2.6 shows accidents in the offshore industry filed in the WOAD (1994) data base. It is seen
that approximately 6-8% (structural damage and leakage) of the total losses are due to reasons
accounted for, or that could be accounted for, in SRA. Among the severe damages, 12-14% of the
reasons may be accounted for in SRA. As a guideline, one might therefore say that, if the
accident statistics is judged to be acceptable, the target reliability in SRA should be such that the
structural failure probability does not exceed a probability which is a factor 10
1
times the
overall failure probability according to such statistics. Hence, this should lead to quantification of
the target reliability of the complete offshore unit as resulting from all significant structural
failure modes. The target reliabilities for the individual failure modes should then, accordingly,
be even higher.
Some comments that should be made to the table we concluded on:
When thinking about the severity of the potential accident it is legitimate to think in societal risk
terms. For example in cases where consequences would differ a factor of ten in potential
fatalities, the target reliability should also change a factor 1/10. If accidents are compared where
the accidental development scenario indicates that most personnel would have high chances of
successful evacuation the safety target may be decreased. The base line in the figure is that
accidents which may lead to fatalities should have a target reliability of 10
5
. Severe accidents
with small but non-negligible risks of fatalities should have target reliabilities of 10
4
.
In cases of progressive collapse the target should be based on the risk concept, accounting for the
probability of the accidental event and the severity of the consequence of the accidental load.
P
SRA Tabulated
Accident
QRA
Target Target
P P = / (2.25)
where
P
SRA
Target
= target safety level for SRA
P
Tabulated
Target
= target safety level from Table 2.7
P
Accident
QRA
= probability of accident determined from QRA
2.5.6 Target Failure Probabilities in Design Codes
In the Norwegian part of the North Sea a limiting target value in risk analysis of 10
4
per annum
has been used the last decennium. According to Fjeld (1978), the theoretical annual probability
of defined failures at the ultimate limit state of the NPD regulations is in the order of 10
4
to
10
5
, somewhat lower for concrete structures than for steel structures. This was also the basis for
the DNV rules for construction of fixed offshore structures in 1977. Reference is made to
Appendix B for the Eurocode, CSA and NKB.
2.5.7 Recommended Method
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
33
It is recommended to relate the target reliability to that of well-known cases, i.e., the procedure
outlined in Section 2.5.3 is recommended. In cases where it is difficult to follow this
recommendation, Table 2.7 may be resorted to.
The target failure probabilities in Table 2.7 are frequently referred to. This table originally refers
to NKB (1978). Although some adjustments of definitions of columns and rows in Table 2.7 have
been performed, the target values are the same.
The target failure probabilities in Table 2.7 are given for a reference period of 1 year, and should
be treated as operational or notional probabilities and not as relative frequencies.

calculated
should be derived from an analysis model considered to give results on the conservative
side when used together with target values from Table 2.7.
The failure consequences may be interpreted as follows:
Table 2.7 Target annual failure probabilities and corresponding reliability indices
Failure consequences
Failure
development
Not serious Serious Very serious
Ductile failure
with reserve
strength capacity
p = 10
-3
= 3.09
p = 10
-4
= 3.71
p = 10
-5
= 4.26
Ductile failure
with no reserve
capacity
p = 10
-4
= 3.71
p = 10
-5
= 4.26
p = 10
-6
= 4.75
Brittle behaviour
in terms of
fracture or
instability
p = 10
-5
= 4.26
p = 10
-6
= 4.75
p = 10
-7
= 5.20
Not serious:
A failure implying small possibility for personal injuries. The possibility for a pollution is small
and the economic consequences are considered to be small. These failure consequences are not
considered to be a safety or environmental threat, the target reliability may therefore be decided
based on cost-benefit analysis.
Serious:
A failure implying possibilities for personal injuries/fatalities or pollution or significant
economic consequences.
Very serious:
A failure implying large possibilities for several personal injuries/fatalities or significant
pollution or very large economic consequences.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
34
Note that for the serious and very serious failure consequences a cost-benefit analysis may result
in higher target reliabilities than those implied by the safety and environmental considerations.
The evolution of a failure should be considered in view of the actual failure mode. For example,
the evolution of a fatigue failure may be classified different from that of a failure in the ultimate
limit state. An example is given by Lotsberg (1991), in which the redundancy of a tether group of
a Tension Leg Platform is considered to be less for the ultimate limit state than for the fatigue
limit state and the progressive collapse limit state.
2.6 Common Mistakes in the Use of SRA
Some errors in code calibration and probabilistic design relating to target reliability are known
and quite common.
Safety factors are not updated when new models are used
Safety factors are changed by studying the effect of a change in one parameter, without
realising that the load model, the response model, the capacity function and the target
reliability are dependent
Sometimes older codes do not state explicitly the purpose of the code check, e.g. the fatigue
check is implicit in an ultimate limit states (The code writers were confident that if the
ultimate limit state code was fulfilled there would not be a fatigue problem). When such codes
are calibrated such non-stated objectives is easy to forget.
The reference period is not correct, e.g., lifetime vs. one year (Section 2.5.1)
The scale effects are not properly dealt with (Section 2.5.1)
The existence of multiple failure modes, using from the same target reliability budget is not
properly considered (Section 2.5.5)
Some of the uncertainties are not included in the analysis, e.g., statistical uncertainties or
model uncertainties.
Historical data are accepted as basis for target reliabilities without realising that some
historical records are unacceptable
Implicit target reliabilities from other previous codes are used without realising that the
reliabilities are unacceptable or conservative or relates to incomparable limit states or
consequence classes.
Characteristic values are given an uncertainty, instead of a distribution for the variable the
characteristic value represent
The models are biased in competence, e.g., very detailed material models with no uncertainty
on the loads or vice versa
The distribution representing a stochastic variable in a limit state is confused with a variable
describing the populations in the definition of the scope. The following is a small example
Example: The long-term distribution of the 10-minute mean wind speed is often represented by a
Weibull distribution
F u
u
a
U
b
10
1 ( ) exp( ( ) ) = (2. 26)
Owing to spatial variation of the wind loading regime, the coefficients a and b may vary from
point to point within a geographical area considered for the scope of a code. This variability in a
and b is not to be represented as a probability distribution in the reliability analyses that are used
for the code calibration. A structure to be designed by the code is supposed to be subject to a site-
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
35
specific design, such that local values of a and b are to be used in the wind speed distribution, and
the only uncertainty in a and b that possibly ought to be represented by a probability distribution
in the reliability analyses is statistical uncertainty when available data for estimation of the local a
and b values are limited. The point-to-point variability of a and b is only to be reflected through
the selection of representative design cases in accordance with the demand function for the scope
of code.
2.7 Standard Problems, Reasons and Remedies Relating to Codes
When a design code is found to give unsafe designs and/or unneccesarily conservative and costly
designs, some of the reasons may be described and remedied by use of SRA. Such cases are
listed in Table 2.8.
2.8 Competence Requirements
A practical application of SRA, for probabilistic design and reliability based code calibration,
will generally require special competence that cannot be expected to be covered by one single
specialist. A typical situation will be that competence on a high level is required both for the load
modelling, the structural strength representation, as well as the reliability formulation.
It is therefore recommended that personnel responsible for a reliability analysis shall have
relevant training and competence to undertake or supervise the execution of such an analysis as
well as have relevant competence in the particular area of application. At least five years of
relevant experience should be required.
Table 2.8 Reasons and remedies for unsafe or overly safe designs
Problem Reason Remedy
It has been found that the There is an error in the limit Correct limit state. The target
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
36
original code give unsafe
design solutions in a number
of cases
state formulation reliability cannot be reused in such
cases.
There is an error in the
stochastic models/data.
Correct stochastic models or collect
new data. The target reliability can
not be reused in such cases
The models are correct, the
target reliability is to low.
Recalibrate with the existing model
and with acceptable target reliability
The code has never been
calibrated
Calibrate the code as described in the
guideline
Listed or other reasons Consider probabilistic design
It has been found that the
original code give unsafe
design solution in special
cases
The code may have been
developed for a scope not
covering this special case
Recalibrate the code with an extended
scope, e.g. other materials,
environments, design concepts
(different response characteristics).
Make sure that the limit state
formulation and statistical
information is correct for the new
scope. Target reliabilities may be
reused.
In this case: Consider for simplicity
to do probabilistic design
The code has never been
calibrated
Calibrate the code as described in the
guideline
Listed or other reasons Consider probabilistic design
It has been found that the
code gives conservative/costly
design solutions in a number
of cases
There is an error or conservative
assumption in the formulation of
the limit state
Remove error and/or conservative
assumption. Beware, the target
reliability may originally have been
selected to be on the safe side. The
target reliability should not be reused,
without good documentation.
There is an error in the
stochastic models/data.
Correct stochastic models or collect
new data. The target reliability can
not be reused in such cases
The models are correct, the
target reliability is unnecessarily
high.
Recalibrate with the existing model
and with acceptable target reliability,
Chapter 2.
The code has never been
calibrated
Calibrate the code as described in the
guideline
Listed or other reasons Consider probabilistic design
It has been found that the
code gives conservative/costly
design solutions in one
particular case
The code may have been
developed for a scope not
covering this particular case
Recalibrate the code with an extended
scope, e.g. other materials,
environments, design concepts
(different response characteristics).
Make sure that the limit state
formulation and statistical
information is correct for the new
scope. Target reliabilities may be
reduced. In this case: Consider for
simplicity to do probabilistic design.
2.9 General Requirements to an SRA
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
37
The results obtained from an SRA and the interpretation of these results should not relieve the
engineer of his responsibility to exercise good engineering judgement. In any given SRA, the
following evaluation methodology is to be undertaken and satisfactorily documented:
1. Failure criteria are to be formulated in terms of limit state functions for each relevant failure
mode. In order to verify that a limit state function is correctly defined, one deterministic
calculation by means of the defined limit state function shall normally be carried out for a
particular set of the governing input variables.
2. When more than one limit state govern the reliability of a structural component, or when more
than one component constitute a structure being analysed, the corresponding system reliability
is to be evaluated, in addition to the individual component reliabilities.
3. Relevant stochastic parameters of each uncertain basic variable in the limit state function are
to be identified. In cases where the asymptotic behaviour of a distribution is not well defined
the stochastic characteristics are to be selected based upon conservative considerations.
4. Considerations shall be given to time-dependent degradation of the resistance of the structure
or component analysed, e.g. corrosion, creep, shrinkage, fatigue etc.
5. For all relevant failure modes the reliability is to be analysed and results presented.
6. The type of reliability analysis undertaken is to be suitable for the limit state under
consideration. In this context, analytical reliability methods (FORM/SORM) are not to be
applied for probability figures greater than 0.05. Further, when simulation techniques are used
to estimate the probability of failure, the number of simulations required is to be appropriate
for the expected order of magnitude of the probability of failure. The choice of simulation
method and number of simulations should always be documented.
7. The resulting reliability is to be evaluated for sufficiency. To the extent possible, target
reliabilities should be established according the principles set forth in Section 2.5.7.
8. The most probable realisation of the stochastic variables at failure, referred to as the design
point in SRA, is to be evaluated to ensure that it has a sound physical interpretation.
9. The results of the reliability analysis are to be evaluated with respect to sensitivity
considerations. Such sensitivity evaluations are to include relevant considerations of
parametric sensitivities for changes in design parameters. Uncertainty importance factors
should always be documented for all stochastic variables.
2.10 Standard List of Content for an SRA Analysis Report
It is recommended that an SRA analysis, its assumptions and results are documented in a report.
The contents of such a report should contain the following items:
Introduction to problem of analysis with assumptions and provisions.
Probabilistic analysis method.
Theory of physical or mechanical model for representation of physical problem.
Limit state function formulation.
Probabilistic and deterministic modelling, i.e., input to analysis in terms of stochastic
variables and deterministic parameters, e.g., based on data.
Reliability analysis.
Results of reliability analysis, including reliability index, failure probability, uncertainty
importance factors, and parametric sensitivity factors.
Discussion and conclusions.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
38
References
Apostolakis, G. (1990), "The Concept of Probability in Safety Assessment of Technical
Systems", Science, Vol. 250, Article 1359, pp. 1360-1364.
Apostolakis, G. (1993), "Commentary on Model Uncertainty", Workshop on Model Uncertainty:
Its Characterization and Quantification, Annapolis, Maryland, October 20-22, 1993.
Bayes, T. (1763), "An essay towards solving a problem in the doctrine of chance", Philosophical
Transaction of the Royal Society, 53, pp. 370-418.
Central Bureau of Statistics in Norway (1991), Statistical Yearbook of Norway, Oslo, Norway.
De Groot, M.H. (1988), in Accelerated Life Testing and Expert Opinions in Reliability, C.A.
Clarotti and D.V. Lindley, Eds. North Holland, Amsterdam, 1988, pp. 3-24.
Ditlevsen, O. (1981), "Fundamental Postulate in Structural Safety", Journal of Engineering
Mechanics, ASCE, Vol. 109, No. 4, August 1983. pp 1096-1102.
DNV (1977), Rules for the Design, Construction and Inspection of Offshore Structures, Det
Norske Veritas, Hvik, Norway.
DNV (1994), WOAD, World Offshore Accident Databank, Statistical Report 1994, DNV
Technica.
DNVR (1993), "PROEXP Probabilistic Explosion Modeling", DNV Research Report No. 93-
2081.
Finetti, B.D., (1974), Theory of Probability, Wiley, New York, 1974, Vol. 1 and 2.
Fjeld, S. (1978), "Reliability of Offshore Structures", Journal of Petroleum Technology, pp.
1486-1496. OTC Paper No. 3027, May 1978.
Gaarder, S. (1992), "Ddsrisiko i skipsfart og andre yrker 1970-1990", DNV Research Report.
Hauge, L.H., R. Lseth, and R. Skjong (1992), "Optimal Code Calibration and Probabilistic
Design", Proceedings, 11th International Conference on Offshore Mechanics and Arctic
Engineering, Vol. II, pp. 191-199, Calgary, Alberta, Canada.
Jordaan, I.J. (1988), "Safety levels implied in Offshore Structural Design Codes: Application to
CSA Program for Offshore Structures." CSA Standard Division February 1988.
Kaplan, S. and B.J. Garrick (1981), On the Quantitative Definition of Risk, Risk Analysts, Vol 1.,
pp. 11-27.
Keynes, J.M. (1921), A Treatise on Probability, St. Martins Press (1952), New York.
Kinchin, G.H. (1978), Assessment of Hazards in Engineering Work Proc. Instn. Civ. Engrs. 64:
431-38.
Lind, N.C. (1973), "The Design of Structural Design Norms", Journal of Structural Mechanics,
Vol. 1, No. 3, pp. 357-370.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
39
Lindley, D.V. (1985), Making Decisions, Wiley-interscience, New York, ed. 2, 1985.
Litai, D., D.D. Lanning, and N. Rasmussen (1981), "The Public Perception of Risk", Society for
Risk Analysis International, Workshop on the Perceived Risks, Washington, D.C., pp. 213-223.
Lotsberg, I. (1990), "Target Reliability Index. A Literature Survey." Veritas Research Report No.
90-2023, April 1990.
Lotsberg, I. (1991), "Probabilistic Design of the Tethers of a Tension Leg Platform," OMAE,
Houston, Texas, February 1990, Journal of Offshore Mechanics and Arctic Engineering, Vol.
113, May 1991.
Murley (1994), "Perspectives on Reactor Safety", Lecture Note from Massachusetts Institute of
Technology Nuclear Power Safety Course: Part 1. MIT July-18-22, 1994.
NKB (1987), "Guidelines for Loading and Safety Regulations for Structural Design", Report No.
55E, June 1987.
NKB (1978), (NKB=The Nordic Committee on Building Regulations), Recommendations for
Loading and Safety Regulations for Structural Design. NKB-Report No. 36, Copenhagen,
Denmark, Nov. 1978.
Norwegian Petroleum Directorate (1994), Regulations Relating to Loadbearing Structures in the
Petroleum Activities.
NRC (1975), Nuclear Regulatory Commission, Reactor Safety Study, WASH-1400, Washington,
D.C. , Government Printing Office (1975).
Otway, H.J. and J.J. Cohen (1975), "Revealed Preferences: Comments on the Starr Benefit-Risk
Relationships", International Institute for Applied System Analysis. Vienna, Austria.
Parry, G.W., (1988), On the Meaning of Probability in Probabilistic Safety Assessment,
Reliability of Probability on Probabilistic Safety Assessment, Reliability Engineering and System
Safety, Vol 23, pp 309-314.
Rowe, W.D. (1977), An Anathomy of Risk, Wiley, New York
Savage, L.J. (1954), The Foundation of Statistics, J.Wiley, New York.
Starr, C. (1969), 'Social Benefits vs. Technological Risk', Science, 165: 1232-38.
Swain, A., and Guttmann (1983), Handbook of Human Reliability Analysis with Emphasis on
Nuclear Power Plant Applications, U.S. Nuclear Regulatory Commission, Technical Report
NUREG/CR-1278.
Further Reading
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
40
Major hazard aspects of the transport of dangerous substances (1991), HMSO, ISBN 0 11
885699 5.
Risk assessment: a study group report (1983), The Royal Society, ISBN 0 85403 208 8.
Risk criteria for land use planning in the vicinity of major industrial hazards, HMSO, ISBN 0 11
885491 7.
The development of risk acceptability criteria for the BP Group (1990), CSS Report No 04-90-
0014.
The use and value of quantitative risk assessment (QRA) and its development within the BP
Group (1990), GSC Report No-04-90.0005.
Anderson, E.A. (1989), "Scientific trends in risk assessment research", Tox Ind Health, Vol. 5,
pp. 777-790.
Chemical Industrial Associated (1987), A guide to hazard and operability studies.
Cooper, M.G. (ed.) (1985), Risk: man-made hazards to man, Oxford University Press, ISBN 0 19
854155 4.
Dawson, J. (1987), Living with risk: the British Medical Association guide, John Wiley & Sons,
ISBN 0 47191598 X.
Dept. of Energy (1991), Interpretation of major accidents to the environment for the purpose of
the CIMAH Regulations.
Dept. of Environment (1990), Circular 20/90. (Welsh Office Circular 34/90). EC Directive on
protection of groundwater against pollution caused by certain dangerous substances (80/68/EEC)
Douglas, M. and A. Widawsky (1982), "How do we know the risk we face? Why risk selection is
a social process", Risk Analysis, Vol. 2, No. 2, pp. 45-52.
Ecetox (1991), "Emergency exposure indices for industrial chemicals", Technical Report No. 43,
European Chemical Ecology and Technology Center, Brussels, Belgium.
HSE (1991), A guide to the Control of Industrial Major Accident Hazards Regulations 1984,
HS(R)21, (Rev) HMSO 1991, ISBN 0 11 885579 4.
Kleindorfer, P.R. and Kunreuther, H.C. (eds.) (1987), Insuring and managing hazardous risks.
From Seveso to Phopal and beyond, Springer Verlag, Berlin, Germany, ISBN 3 5401 17732 9.
Lord Zukerman (1980), "The risks of a no-risk society", The Year Book of Word Affairs 1980,
Vol. 34, Stevens & Sons, London, England, ISBN 0 420457305.
Lowrance, W.W. (1976), Of acceptable risk: science and determination of safety, William
Kaufman, Los Altos, California, ISBN 0 913232 31 9.
Press, W.P. and A.M. Ehrlich (1990), The EPAs risk assessment guidelines.
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
41
RCEP (1988), Twelfth report, Best practiable environmental option, HMSO, CM 310, ISBN 0
101031 025.
RCEP (1991), Fourteenth report, GENHAZ. A system for the critical appraisal of genetically
modified organisms to the environment, CM 1557, HMSO, ISBN 01 01155727
Rimington and Pape (1988), Nuclear power performance and safety, Volume 3; Safety and
international cooperation, International Atomic Energy Agency, Vienna, Austria, IAEA-CN-
48/35. Available from HMSO, ISBN 9 20050281.
Slovic, P. (1986), "Informing and educating the public about risk", Risk Analysis, Vol. 6, No. 4,
pp. 403-415.
Star, C. (1985), "Risk management, assessment and acceptability", Risk Analysis, Vol. 5, No. 2,
pp. 57-102.
Stringer, D.A. (1990), "Hazard assessment of chemical contaminants of soil", Ectox Technical
Report No. 40, European Chemical Ecology and Technology Center, Brussels, Belgium.
Weinberg, A.M. (1985), "Science and its limits: the regulators dilemma", Issues in Science and
Technology, Vol. II, No. 1, pp. 59-72.
Wilson, R. and E.A.C. Crouch (1987), "Risk assessment and comparisons: an introduction",
Science, Vol. 236, pp. 267-270.
A
Accident statistics, 32
Accidental load, 14, 32
Aleatory uncertainty, 12
B
Bayes theorem, 11
Bias, 24
Buckling, 18
C
Characteristic load, 17, 19, 21
Characteristic resistance, 17, 19
Characteristic value, 18, 20, 22, 24, 34
Code, 18, 19, 21
calibration, 18, 21, 23, 34, 35
objective, 21
scope, 21, 22, 23, 35
Crack growth, 12, 25
Current, 21, 22, 24
D
Damage, 32
Data space, 21
Dead load, 21
Demand function, 21, 35
Design code, 9, 16, 17, 18, 19, 21, 22, 24, 28, 32, 35
Design format, 17, 18, 27
Design life, 24
Design point, 18, 19, 20, 21, 22, 37
Design value, 18, 20, 22
E
Epistemic uncertainty, 11, 12
Error
gross, 13, 14, 16, 24, 28
human, 13
F
Failure consequences, 33, 34
Failure mode, 18, 19, 21, 22, 24, 32, 34, 37
Failure probability, 12, 20, 24, 26, 32, 37
conditional, 24
Fatigue, 12, 16, 20, 34, 37
Fatigue crack growth, 12
FORM (First-order reliability method), 37
Fracture mechanics, 12
Fracture toughness, 16
Guideline for Offshore Structural Reliability Analysis - General Page No.
_____________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________________
DNV Report No. 95-2018 Chapter 2
Skjong,R, E.B.Gregersen, E.Cramer, A.Croker, .Hagen, G.Korneliussen, S.Lacasse, I.Lotsberg, F.Nadim,K.O.Ronold (1995)
Guideline for Offshore Structural Reliability Analysis-General, DNV:95-2018.
9
G
Geometry function, 12, 25
Gross error, 13, 14, 16, 24, 28
H
Hazard identification, 9
Human error, 13
gross, 13, 14, 16, 24, 28
Human reliability, 9
I
Importance factor, 11, 19, 22, 37
Inspection, 12, 16, 24
L
Limit state, 12, 18, 19, 20, 21, 22, 23, 24, 25, 31, 32,
34, 35, 36, 37
function, 12, 19, 20, 22, 24, 25, 37
ultimate, 18, 32, 34
Load, 14, 16, 17, 18, 19, 20, 21, 24, 25, 31, 32, 34, 35
dead, 21
factor, 17, 18
permanent, 18
wind, 34
Load combination, 18
Load process (Excitation process), 24
Lognormal distribution, 20
M
Model, 10, 11, 12, 24, 25, 28, 33, 34, 36, 37
O
Omission sensitivity factor, 11
Organizational reliability analysis, 9, 10, 13, 14
P
Parametric sensitivity factor, 37
Partial safety factor, 17, 18, 19, 20, 21, 22, 23, 24
Penalty function, 22, 23
Potential, 19, 32
Probability of failure, 9, 12, 20, 24, 25, 26, 32, 37
Q
QRA (Quantitative risk analysis), 9, 10, 12, 13, 14, 15,
16, 32, 40
Quality assurance, 14, 16
Quantitative risk analysis, 9
R
Redundancy, 34
Regression analysis, 12
Reliability, 14, 16, 19, 22, 24, 28
system, 37
Reliability analysis, 28
Reliability index, 11, 16, 19, 20, 21, 22, 23, 25, 26,
27, 28, 37
target, 19, 20, 21
Reliability method, 19, 20, 37
first-order (FORM), 19, 20
Requalification, 24
Resistance factor, 16
Response, 10, 13, 34, 36
S
Safety factor, 16, 17, 18, 19, 20, 21, 22, 23, 24
partial, 17, 18, 19, 20, 21, 22, 23, 24
Safety format, 17
Safety index (reliability index), 11, 16, 19, 20, 21, 22,
23, 25, 26, 27, 28, 37
Scale effects, 24, 25, 26, 34
Scope, 19, 21, 22, 23, 34, 36
Scope of code, 21, 22, 23, 35
Sensitivity factors, 11, 37
omission, 11
parametric, 37
Simulation, 37
Simulation methods, 37
SORM (Second-order reliability method), 37
SRA (Structural reliability analysis), 9, 10, 11, 13, 14,
15, 16, 24, 25, 32, 34, 35, 36, 37
State, 9, 11, 12, 18, 19, 20, 21, 22, 23, 24, 25, 29, 31,
32, 34, 35, 36, 37
Statistical uncertainty, 35
Stress range, 12, 25
Structural reliability analysis, 28
System
parallel, 24
series, 25
System reliability, 37
T
Target reliability, 10, 19, 20, 21, 22, 27, 28, 32, 33,
34, 35, 36
Target safety, 16, 24, 32
U
Updating, 12
W
Weibull distribution, 34
Wind, 34
Wind speed
mean, 34
Y
Yield strength, 20, 21

Вам также может понравиться