Engineering Failure Analysis 14 (2007) 481–488

www.elsevier.com/locate/engfailanal

Background to the derivation of partial safety factors

Received 30 August 2005; accepted 30 August 2005

Abstract

The background to the derivation of partial safety factors (PSFs) given in two standards, BS 7910 and API 579, is
described. In both cases, PSFs are provided to achieve selected target reliability levels against the failure modes of fracture
and plastic collapse in structural components. The recommended PSFs in the two standards are compared in order to
investigate differences. Example fitness for service assessments are also conducted in order to explore the potential impact
of differences on the output of an assessment. Work being carried out to review and update the current partial safety fac-
tors for the next revision of BS 7910 is discussed.
 2006 Elsevier Ltd. All rights reserved.

Keywords: Structural reliability; Partial safety factors

1. Introduction

BS 7910:1999 [1] and API 579 [2] are the only major structural integrity assessment procedures to include
partial safety factors (PSFs) where fracture and plastic collapse are the dominant failure modes. Although
limit state design codes have been in use for conventional structural design for several years, they were not
extended to fracture considerations until the publication of BSI document PD 6493:1991. BS 7910 has now
replaced PD 6493 and the original PSFs were updated based on development work carried out at UMIST
[3]. The PSFs in API 579 are the most recent ones to be published in this area and they appear to have a similar
basis to the BS 7910 work but there are differences. An overview of the background to both sets of PSFs is
presented below along with an example application.

2. Target reliabilities

A major decision required in the use of partial safety factors is the choice of target reliability. This is the
acceptable failure probability or reliability for the component under assessment. In general, the acceptable

E-mail address: [email protected].

1350-6307/$ - see front matter  2006 Elsevier Ltd. All rights reserved.
doi:10.1016/j.engfailanal.2005.08.009
482 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488

Table 1
Target reliability levels for which partial safety factors are given in BS 7910 and API 579
BS 7910:1999 API 579 2000
Failure consequences Redundant components Non-redundant components Failure consequences All components
1 3
Moderate 2.3 · 10 10 Small 2.3 · 102
Severe 103 7 · 105 Medium 103
Very severe 7 · 105 105 High 106
Expressed in terms of events per year or annual failure probabilities.

reliability to be adopted depends on the consequences of failure with target reliability increasing for increasing
failure consequences. However, there is no consensus on the appropriate target reliability levels despite several
years of research in reliability and risk engineering. It is therefore not surprising that API 579 and BS 7910
have adopted different target reliability levels on the basis of guidance provided in different documents (see
Table 1).
The target reliability levels adopted in BS 7910:1999 are based on the general principles recommended by
the Nordic Committee on Building Regulations [4] and the ISO Draft Guidelines for General Principles on
Reliability of Structures [5]. The target reliability of 7 · 103 was taken from the Eurocode 3 standard for
structural steel design [6]. The background document to the API 579 partial safety factors [7] discusses target
reliability levels citing documents dealing with methods of establishing target levels but it does not explain how
the levels used for API 579 were determined.
Target reliability levels tend to be quoted in terms of annual failure probabilities (see Table 1). However, for
failure mechanisms that are not explicitly time-dependent such as the fracture and plastic collapse considered
here, recommended annual failure probabilities are often treated as lifetime limiting values. This approach was
probably adopted in the derivation of PSFs for both BS 7910 and API 579 as the underlying documents [3,7]
made no reference to the effect of in-service period.

3. Derivation of partial safety factors

For both standards, partial safety factors were derived from the results of reliability analysis of welded
components based on a number of assumptions. In general, reliability analyses were carried out for a compo-
nent containing a surface or through thickness flaw using a limit state equation of the form:
ðK r  qÞK mat  K I ¼ 0 ð1Þ

where KI is the applied stress intensity factor, Kmat is fracture toughness, q is plasticity interaction factor and
Kr is the permitted fracture ratio in API 579 or BS 7910. Kr is a function of the ratio of applied load to plastic
collapse load (Lr or Sr in the case of API 579), which is in turn dependent on applied primary stress, flaw and
component geometry and material tensile properties.
To derive the PSFs recommended in both standards, first order reliability method (FORM) or second order
reliability method (SORM) was used to establish the most likely combination of the main random variables –
fracture toughness, applied stress and flaw size – leading to failure according to the limit state equation [1].
This combination gives the so-called design values of the variables (K mat , r*, a*) from which the partial safety
factors for loading effects are defined as the ratio of design value to a specified characteristic value (K mat0 ,r 0 ,a 0 )
and for resistance effects as the inverse of this. That is,
r a K mat0
cr ¼ ; ca ¼ ; cK ¼ ð2Þ
r0 a0 K mat
Thus the failure equation with partial safety factors becomes
K mat pffiffiffiffiffiffiffiffiffiffiffiffiffi
ðK r  qÞ  ðcr  Y rÞ  p  ca a ¼ 0 ð3Þ
cK
where Y is the non-dimensional coefficient to take account of the crack geometry.
 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488 483

Table 2
Summary of assumptions made in reliability analysis conducted in the production of partial safety factors for BS 7910 and API 579
Item or variable BS 7910:1999 assumption API 579 assumption
Failure assessment diagram (FAD) R6/BS 7910:1999 Level 2a (Kr, Lr) PD 6493:1991 Level 2 (Kr, Sr)
Reliability analysis method FORM SORM
Wall thickness 20 mm, 50 mm, 100 mm Not stated
Flaw geometry Surface and through-thickness Surface flaw
Flaw size Normal distribution: mean = various Lognormal (2 cases): height < 5 mm,
(small surface: 3 mm, 10 mm; 200 mm Cov = 0.3; height > 5 mm, COV = 0.2
long thru-thk); COV = 0.1, 0.2, 0.3, 0.5
Fracture toughness Weibull distribution: amean = various Weibull distribution: Normalised,
(800–6000 N mm3/2); cases: COV = 0.1, 0.2, 0.3 mean = 1.0; COV = 0.25
Yield strength Lognormal distribution: COV = 0.1 Fixed
Applied stress Normal distribution mean = various; 3 cases: Extreme value distr. Normalised,
COV = 0.1, 0.2, 0.3 mean = 1.0; 3 cases: COV = 0.1, 0.2, 0.3
a
For BS 7910:1999 PSF derivation, a range of mean fracture toughness were assumed to address different regions of the FAD.

The characteristic values were taken as the mean values of the variables in the derivation of PSFs for API
579 while for BS 7910 mean values were also adopted for applied stress and flaw size but the mean minus one
standard deviation (mean  1SD) value of fracture toughness, which is considered equivalent to the minimum
of three fracture toughness tests, was used to set cK. It should be noted that in order to apply cK as a dividing
factor in an assessment, it must have values greater than unity and the definition used here (and in BS 7910) is
the reciprocal of the value obtained from the reliability analysis as defined in [2].
Other differences in the derivation of PSFs for API 579 and BS 7910 concern the assumptions made in the
underlying reliability analysis and the form of the resulting recommendations. The API background work [7]
used the PD 6493:1991 Level 2 failure assessment diagram (FAD) which is based on the so-called strip yield
model. This FAD was considered to be inappropriate by the committee that prepared BS 7910 and is, there-
fore, not included in the current standard. In [7], the strip yield model FAD was approximated by a radial
interaction diagram to simply the reliability analysis. The BS 7910:1999 PSFs were derived using the BS
7910:1999 default Level 2a FAD for steels showing continuous work hardening [3]. In this respect, the BS
7910:1999 approach is more consistent with current thinking on the use of FADs than API 579. The API
579 derivation used the second order reliability method (SORM) while the BS 7910 derivation utilised the first
order reliability method (FORM). For non-linear limit state functions, SORM is generally more accurate.
Table 2 summarises the main assumptions made in the underlying reliability analysis. It may be noted that,
in the case of BS 7910:1999 PSF derivation work, mean fracture toughness values of 800–6000 N mm3/2 and
wall thicknesses of 20, 50 and 100 mm were employed. It was found that the most critical cases were those
involving large flaws, high wall thickness and low toughness, and these were used to set the recommended
PSFs. Other combinations would therefore have failure probabilities lower than the target values. The increase
in notional reliability when using the same partial factors for small flaws compared to large flaws was about
one order of magnitude [3]. For fracture toughness, the API 579 derivation adopted a different approach by
assuming a normalised fracture toughness distribution with a mean value of unity.

4. Comparison of partial safety factors

Other than the actual values of the PSFs, there are differences between API 579 and BS 7910 in the form of
the recommendations. API 579 includes a set of factors for fracture toughness (cK), applied stress (cr) and flaw
size (ca) for specified coefficient of variation (COV) in applied stress of 0.1, 0.2 and 0.3, while BS 7910 provides
a factor for each of the three variables depending only on the COV for that variable and the specified target
failure probability. Rather than provide PSFs for different uncertainty levels in fracture toughness and flaw
size, the API 579 approach assumes a single COV of 0.25 for fracture toughness and a flaw is either considered
to be shallow with a COV of 0.3 or deep with a COV of 0.2. In this respect, the BS 7910 recommendation
covers a wider scope as factors are provided for a range of uncertainty levels in fracture toughness and flaw
size.
484 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488

BS 7910:1999 recommends appropriate PSFs for yield strength based on the Eurocode 3 for structural steel
design. Where these are applied, the PSFs for the other variables (cK, cr, ca) were derived to achieve the
required target failure probabilities. API 579 does not give any PSF for yield strength and the underlying reli-
ability analysis assumes yield strength to be constant on the basis that the COV would typically be about 5%
or less.
Although, some sort of equivalence in consequence categories between BS 7910 and API 579 may be estab-
lished, direct comparison in terms of PSFs can only strictly be made for the target failure probability of 103.
The differences in PSFs for this target reliability level can be seen in Table 3 for the case of a shallow flaw
(a < 5 mm) and in Table 4 for a deep flaw (a > 5 mm). Unlike in the case of API 579, it is noted that BS
7910 PSFs for fracture toughness (cK) and flaw size (ca) are specified independently of changes in the applied
stress COV. The BS 7910 cK and ca values given are those appropriate for the fixed levels of uncertainty
adopted in the derivation of API 579 values in order to facilitate the comparison. The cK is therefore interpo-
lated between a COV of 0.2 and 0.3 as the API 579 work assumed a COV of 0.25.
The PSFs given in Tables 3 and 4 for API 579 are those provided for the general failure mode of elastic–
plastic fracture/plastic collapse. For the cases where failure can be assumed to be predominantly by plastic
collapse, PSFs are provided only for applied stress. A transition point is also specified in terms of the ratio
of mean fracture toughness to yield strength for all combinations of COV and target reliability addressed
in API 579.
In general, the API 579 PSFs on applied stress (cr) are higher than the corresponding values recommended
in BS 7910 while the reverse trend is seen in terms of both fracture toughness and flaw size. In particular, the
BS 7910 PSF on fracture toughness are very much higher than the corresponding values in API 579 especially
for low uncertainty level in applied stress. The net position is likely to be that assessments conducted to the BS
7910:1999 PSFs would be much more conservative than those to API 579 PSFs. It is however, difficult to gen-
eralise without a thorough examination over a wider range of analysis p cases.
The Tables also include figures for the product relationship (cK Æ cr Æ ca) and the significance of these fig-
ures is considered further in the discussion section below.

Table 3
BS 7910 partial safety factors compared with those in API 579 for target failure probability of 103 for the case of a shallow flaw
(a < 5 mm)
COVr API 579 PSF BS 7910:1999 PSF
p p
cr cK ca cK Æ cr Æ ca cr cK ca cK Æ cr Æ ca
a
0.1 1.4 1.43 (1.07) 1.20 1.64 1.2 2.3 1.5 3.38
0.2 1.5 1.82 (1.37) 1.1 2.16 1.25 2.3a 1.5 3.52
0.3 2.0 2.0 (1.5) 1.05 3.07 1.4 2.3a 1.5 3.94
Notes: ( ) gives the API 579 cK converted for application to a mean minus one standard deviation for comparison with the corresponding
BS 7910 value.
a
Value interpolated between BS 7910 cK values for COV values of 0.2 and 0.3.

Table 4
BS 7910 partial safety factors compared with those in API 579 for target failure probability of 103 for the case of a deep flaw (a > 5 mm)
COVr API 579 PSF BS 7910:1999 PSF
p p
cr cK ca cK Æ cr Æ ca cr cK ca cK Æ cr Æ ca
a
0.1 1.4 1.67 (1.25) 1.15 1.88 1.2 2.3 1.5 3.38
0.2 1.8 1.43 (1.07) 1.0 1.93 1.25 2.3a 1.5 3.52
0.3 2.3 1.43 (1.07) 1.10 2.58 1.4 2.3a 1.5 3.94
Notes: ( ) gives the API 579 cK converted for application to a mean minus one standard deviation for comparison with the corresponding
BS 7910 value.
a
Value interpolated between BS 7910 cK values for COV values of 0.2 and 0.3.
 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488 485

5. Implications for flaw acceptability

To explore the potential impact of adopting the PSFs described here in a fitness-for-purpose assessment, the
case of a propane sphere containing a surface-breaking flaw was considered. The analysis conducted is based
largely on a previous fitness for purpose assessment involving a storage sphere, which was initiated after low
impact energy values were reported during welding procedure qualification testing. The objective of the assess-
ment was to determine if a surface-breaking flaw that could be readily detected by ultrasonic testing would be
tolerable despite the low impact energy. The target probability of failure was set at 103 for this example
mainly to allow a comparison of the API 579 and BS 7910 PSFs.
In the present exercise, a surface-breaking flaw measuring 3 mm high by 20 mm long is assumed and three
sets of analyses are carried out. The first assumes worst case estimates of all variables without any safety fac-
tors, the second employs appropriate characteristic values with API 579 PSFs while the third utilises the BS
7910 PSFs with appropriate characteristic values. Fracture toughness assumption is based on the BS
7910:1999 Charpy-toughness correlation derived from the master curve approach. This assumes that fracture
toughness can be described by a 3 parameter Weibull distribution. It was assumed that the ultrasonic measure-
ment error would be within ±1.8 mm and that this band represents ±2 standard deviations from actual flaw
height. The calculations were carried out using the TWI CRACKWISE 3 software with the PSF entered
through the user-defined PSF feature.
In selecting the primary membrane stress for the calculations, the value of 160 MPa considered in the
assessment without PSFs was assumed to correspond to a mean + 2 standard deviation value. The COV
was taken as 0.1 assuming that primary stress levels arising mainly from pressure loading could be estimated
fairly accurately. These assumptions result in a mean value of 133.3 MPa to which the respective PSFs are
applied for API 579 and BS 7910:1999 cases.
Details of the input data for the three analysis cases and the outputs obtained are given in Table 5. The
corresponding assessment points are also plotted in Fig. 1. The results show that the assumed flaw (3 mm
nominal height by 20 mm long) will be considered acceptable using either worst-case estimates or mean values
with API 579 PSFs. However, if the BS 7910:1999 PSFs are used the flaw is found to be unacceptable, exceed-
ing the minimum tolerable height by a significant margin. By far, the most important reason for the more
severe assessment is the much higher partial safety factor for fracture toughness in this particular example.

Table 5
Example fitness-for-purpose calculations with and without PSFs
Case 1 No PSFs Case 2 API 579 PSFs Case 3 BS 7910 PSFs
Input
Wall thickness, mm 48 48 48
Sphere radius, mm 7500 7500 7500
Flaw length (2c), mm 20 20 20
Membrane stress (Pm), MPa 160 Mean = 133.3; Cov = 0.1 Mean = 133.3; Cov = 0.1
(assumed); cr = 1.4 (assumed); cr = 1.2
Misalignment SCF 1.18 1.18 1.18
Residual stress (Qm), MPa 79.6 79.6 79.6
Thermal stress (Qb), MPa 120 120 120
Flaw height (a), mm 3 + 1.8 = 4.8 Mean = 3.0;) Cov = 0.3; Mean = 3.0 Cov = 0.3; ) ca = 1.5
) ca = 1.2
Fracture toughness (Kmat), N mm3/2 1500 (5th percentile) Mean = 2861; cK = 1.43 Mean  1SD = 2235;
COVK = 0.22; cK = 2.3
Yield strength, MPa 345 345 345
Tensile strength, MPa 485 485 485
Output
Kr 0.92 0.83 1.59
Lr 0.60 0.70 0.60
Result Acceptable Acceptable Not acceptable
Max tolerable height (ac), mm 5.0 4.0 0.7
All analyses based on the BS 7910:1999 Level 2a FAD.
486 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488

1.8
BS 7910:1999 Level 2a FAD
1.6 Case 1-NO PSFs

Fracture ratio, Kr
1.4 Case 2-API 579 PSFs
1.2 Case 3 -BS 7910 PSFs

0.8
0.6
0.4
0.2
0
0 0.2 0.4 0.6 0.8 1 1.2 1.4
Yield ratio, Lr

Fig. 1. Results of example calculation showing assessment points.

6. Discussion

When the form of Eq. (3) is considered, it is clear that for a given set of input variables, r, a and Kmat may
be considered with and without partial safety factors: if the partial safety factors are to have the same overall
effect, then the following relationship must be satisfied:
pffiffiffiffiffiffi pffiffiffiffiffiffi
cK1  cr1  ca1 ¼ cK2  cr2  ca2 ð4Þ
The higher this product relationship of partial safety factors becomes, the higher the notional reliability will
also become.
When the two sets of partial safety factors for BS 7910 and API 579 shown in Tables 3 and 4 are consid-
ered, it can be seen that the product relationship of the partial safety factors is significantly higher for the BS
7910 figures than for the API 579 figures. This again implies that the BS 7910 figures will give a significantly
higher reliability than the API 579 figures.
When the pattern of the partial safety factors is considered, the BS 7910 figures are constrained so that
when uncertainty in stress levels is considered (as in Tables 3 and 4) it is only the partial safety factor on stress
which varies and the partial safety factors on toughness and defect size remain constant. The basic analyses on
which the BS 7910 figures were based [4] did consider all combinations of uncertainty levels for stress level,
defect size and toughness, but the values of partial safety factors selected were chosen to be safe for all cases.
On the other hand, the API 579 figures appear to have been chosen for specific combinations of uncertainty
and hence may have been able to optimise results more effectively. The trends for increasing partial safety fac-
tor on stress with increasing uncertainty in stress estimates (increasing COVr) are clearly correct, but the API
579 picture is less clear because the other partial safety factors are also varying.
What is perhaps of greater concern is the difference in partial safety factors for toughness. Fracture tough-
ness results are notorious for high scatter and when only a few results are available, there will be a very sig-
nificant effect of the choice of the appropriate combination of lower bound toughness results with estimates of
stress level and defect size to give the required target reliability level.
A point which may be of some significance is the question of the choice of type of distribution in the reli-
ability analyses. The intention behind the choice of normal distributions for stress and defect size in BS 7910 is
that these should represent likely variations in a single estimate of the values for a specific analysis. Thus the
estimate of stress is intended to be the estimate of the worst case loading likely to occur and the uncertainties
in that estimate due to errors/variations, rather than an estimate of all possible variations over a period of
time. Similarly, the defect size to be used in the BS 7910 analysis is intended to be the best estimate for a spe-
cific defect and the normal distribution is intended to represent the potential errors in sizing that particular
defect. On the other hand, the use of the extreme value distribution for stresses and the log normal distribution
for defect sizes in the API 579 method suggests that perhaps these are seeking to cover possible occurrences of
stresses over a period of time and of multiple defect distributions.
It is of interest to consider in more detail the underlying reasons for the differences in results of the example
analyses presented in Section 5 and summarised in Table 5. The target reliability has been assumed to be 103
 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488 487

and should be assumed to relate to the specific case of the particular defect, stress level and material properties.
Having applied the PSFs where used, the effective flaw heights for the three cases considered are 4.8 mm (worst
case), 3.6 mm (API 579) and 4.5 mm (BS 7910), respectively. The corresponding membrane stress levels are
160 MPa (worst case), 187 MPa (API 579) and 160 MPa (BS 7190), with values of residual and thermal stress
common to all three analyses. The effective toughness levels for the three cases are 1500 N mm3/2 (worst
case), 2000 N mm3/2 (API 579), and 972 N mm3/2 (BS 7910). The mean fracture toughness was given as
2861 N mm3/2 with a COV of 0.22. It can immediately be seen that the main reason for the much more con-
servative result from the BS 7910 analysis is the low value of the fracture toughness assumed. In fact the effec-
tive toughness, upon application of the PSF cK, for the BS 7910 analysis is equivalent to the mean  3SD level.
It should also be noted that this assessment is for a small defect whilst the BS 7910 partial safety factors
would be expected to give about one order of magnitude conservatism on target reliability as they cover
for the more severe large defect cases. The original analyses on which the BS 7910 partial safety factors were
based suggest that a change in target reliability of one order of magnitude at the 103/104 level would
correspond to a change in partial safety factor of about 0.5. If a partial safety factor of 1.8 was applied
to the fracture toughness for the BS 7910 analysis, this would give an effective fracture toughness of
1242 N-mm3/2.

7. Concluding remarks

A review of the background to the derivation of partial safety factors for BS 7910:1999 and API 579 has
been given. Differences in the reliability analysis methods and input assumptions made in deriving the partial
safety factors have been discussed. In particular there may be differences because of the assumptions of
extreme value and log normal distributions for stress and defect size in API 579 compared to corresponding
normal distributions in BS 7910.
It is noted that differences in the target failure probabilities adopted in the PSF derivations mean that com-
parison can only strictly be made for the target failure probability of 103 which is used in both BS 7910 and
API 579.
For the cases compared, the API 579 PSFs on applied stress (cr) are higher than the corresponding values
recommended in BS 7910 while the reverse trend is seen in terms of both fracture toughness and flaw size. In
addition, the BS 7910 PSFs on fracture toughness are very much higher than the corresponding values in API
579 especially for low uncertainty level in applied stress. The net position is likely to be that assessments con-
ducted to the BS 7910:1999 PSFs would be much more conservative than those to API 579 PSFs. This is con-
firmed by an example assessment drawn largely from the details of a propane storage sphere. It is however,
difficult to generalise without a thorough examination over a wider range of analysis cases and it is to be
expected that the BS 7190 figures will give about one order of magnitude conservatism for small defect sizes
as used in the example considered.
Overall, the API 579 partial safety factors are comprehensive in certain aspects but then limited in other
areas such as considering only fixed uncertainty levels in fracture toughness and flaw size. Also, the API
579 PSFs are essentially based on an obsolete failure assessment diagram but the impact of this may only
be significant in the plastic collapse regime where API gives a separate set of PSFs. The BS 7910:1999 partial
safety factors cater for the limitations suffered by the API 579 in the treatment of uncertainties in fracture
toughness and flaw size but some of the derivations may be covering too broad a range of parameters resulting
in overly conservative recommendations in some cases.
Further work is now required to combine the best aspects of the two approaches in order to produce a revi-
sion of the existing recommendations.

Acknowledgements

The author would like to thank Professor Michael Burdekin for very useful discussions on the derivation
of the BS 7910:1999 partial safety factors and related issues pertinent to this paper. I am also grateful to
Dr. Henryk Pisarski of TWI for reviewing the initial manuscript of this paper and for providing valuable
comments.
488 A. Muhammed / Engineering Failure Analysis 14 (2007) 481–488

References

[1] BS 7910:1999 (incorporating Amendment No.1): Guide on methods for assessing the acceptability of flaws in metallic structures,
British Standards Institution BSI. London, October 2000.
[2] American Petroleum Institute: API 579-Recommended Practice for Fitness-for-service, API, January 2000.
[3] Burdekin FM, Hamour W. Derivation of partial safety factors for BS 7910:1998, UMIST Report CIV/SAG/98/2, 1998.
[4] Nordic Committee on Building regulations Recommendations for Loading and Safety Regulations for Structural Design. NKB-
Report No. 36, November 1978.
[5] International Organization for Standardization: Draft of new edition of ISO 2394;1986, Geneva, ISO 1986.
[6] Sedlacek G. Use of safety factors for the design of steel structures according to the Eurocodes, Paper No. 2 TAGSI Symposium 2003.
[7] Wirsching PH, Mansour AE. Incorporation of Structural Reliability Methods into Fitness-for-Service Procedures, The Materials
Properties Council Inc., May 1998.