Vous êtes sur la page 1sur 8

Proceedings of the Twenty-seventh (2017) International Ocean and Polar Engineering Conference www.isope.

org
San Francisco, CA, USA, June 25-30, 2017
Copyright 2017 by the International Society of Offshore and Polar Engineers (ISOPE)
ISBN 978-1-880653-97-5; ISSN 1098-6189

Probabilistic Engineering Critical Assessment of Circumferential Girth Weld Flaws in Sour Service

Amir Bahrami1, David Baker2, Xiaofei Cui3, Fokion Oikonomidis3, Guiyi Wu3
ExxonMobil Production Company1, ExxonMobil Upstream Research Company2, TWI3
Spring TX, USA1, 2; Cambridge UK3

ABSTRACT INTRODUCTION

Engineering Critical Assessment (ECA) is a sophisticated deterministic Engineering Critical Assessment (ECA) methods are employed to
tool based on fracture mechanics, which is mainly used to assess identify appropriate operational limits for the design and development
criticality of defects in metallic structures and particularly welds. Such of pipelines and other integrity critical structures. ECA has been
analyses are routinely undertaken in the development phases of Oil and successfully employed for decades and is considered a mature
Gas pipeline projects to develop weld defect acceptance criteria. This technique for performing structural evaluations.
enables an efficient fabrication campaign, while ensuring the integrity
of the joint are not compromised at the time of fabrication, during The vast majority of ECAs are conducted using a deterministic fracture
installation and for the intended service life. The ECA of pipeline girth mechanics approach that involves conservative estimates of the input
welds operating in inert environments is generally considered to be a parameters required for the fitness for purpose assessment. The usual
mature technology and codified through dedicated industry standards approach for conducting deterministic fracture assessments consists of
which provide a robust framework for such analysis. Although the using lower bound values for tensile properties and fracture toughness;
same fracture mechanics principles can be used, the assessment of and upper bound values for the flaw size and both primary and
defect criticality in a girth weld when in an aggressive environment secondary stresses. Although this practice is expected to yield
such as sour service requires additional considerations. This is due to conservative assessments, the results can be overly conservative
inherent uncertainties associated in defining some of the key input predicting failure when it would not actually occur. Also, expert
parameters required for an ECA, in particular the material resistance to engineering judgement is needed to evaluate associated safety margins,
crack extension (i.e. fracture toughness) in a hydrogen charging (e.g. particularly when definition of input parameters is, in some cases,
sour service) environment. As a result, the traditional deterministic subject to considerable uncertainties.
ECA, which is designed to define safe boundaries may be too
conservative or otherwise based on the values used for such inputs. Probabilistic fracture mechanics provides an alternative approach that
This is driven by the large scatter seen in toughness data which is would not simply produce a go/no-go result, but instead would enable
influenced by multitude of factors, consequently a probabilistic the likelihood of failure to be estimated taking into account
approach to ECAs may offer potential advantages. uncertainties and scatter in input data (Provan, 1987; Sandvik et al.,
2006; Mechab et al., 2014; Lee et al., 2015 and Agrell et al., 2016).
Probabilistic fracture mechanics assessment could provide an
alternative approach, which would not just define the conditions at Of particular interest is the scatter that arises for fracture toughness in
which a given defect is determined to be safe but provide a sour environments. Other parameter may also be impacted by
quantitative measure of level of uncertainty associated with a defect environment. The assessment of flaws in sour environments is
which is not deterministically safe. This would enable the probability considerably more complex given the substantial uncertainties affecting
of failure to be estimated taking into account uncertainties and scatter definition of key input parameters, in particular, the material resistance
in input data. This approach involves a large number of iterations to crack extension by hydrogen embrittlement (i.e. the fracture
where various combinations of input parameters are considered. The toughness of hydrogen-embrittled material). As a result, the resolution
choice of key variables in such assessments depends on which of a deterministic ECAs are limited when flaws in sour environment
parameters are likely to carry high levels of uncertainty and/or have a are being assessed and, consequently, probabilistic ECAs may offer
significant effect on the results. more granularity in defining failure probabilities.

The paper presents the results of a study, which was aimed at A probabilistic approach typically involves a large number of
examining applicability of a probabilistic fracture mechanics method to calculations where various combinations of input parameters are
the assessment of circumferential girth weld defects in sour service. considered. The choice of key variables in such assessments depends
The paper further discusses the approach and the challenges associated on which parameters are likely to carry high levels of uncertainty
with the determination of relevant input parameters. and/or have a significant effect on the result.

KEY WORDS: Engineering critical assessment (ECA), sour service, Results from a probabilistic assessment would enable a decision
environmental cracking, probabilistic analysis, fracture mechanics process that is based upon acceptable levels of probability of failure

384
and it is likely that a transition phase will be needed whilst meaningful At present, there is no well-established guidance on the selection of
probabilities of failure are determined. It is expected, for example, that fracture toughness values for assessing the significance of girth weld
a whole new design philosophy (e.g., loads expressed in a statistical flaws in sour service. This is still the subject of on-going research. It is
form) and testing regime will be needed. feasible that a fracture toughness value significantly lower than that
determined at maximum load (i.e. lower than Jmax) may be more
This paper explores the applicability of probabilistic ECA for assessing appropriate (for example the fracture toughness at initiation of crack
the significance of circumferential girth weld flaws in pipelines extension, J0.2). Therefore, a couple of additional parameter studies
carrying sour hydrocarbons. It presents a summary of a study utilized a lower initiation fracture toughness value.
undertaken to investigate extending current ECA methods. The scope
of work was carried out in a staged approach for applying probabilistic Table 1 Shortlisted fracture toughness tests in sour environment for
fracture mechanics methods for the assessment of circumferential girth deterministic ECA
weld flaws in the sour environment. Side- Pre- K-rate J0.2BL Jm Average
grove charge (MPam0.5 (N/m (N/mm) (N/mm)
1) Determination of inputs to perform a suite of sensitivity studies on s-1) m)
key input parameters used for the assessment of circumferential No Yes 0.009 NA 102.8
flaws in sour environments and to adapt into probabilistic ECA No Yes 0.008 NA 94.1
techniques; No Yes 0.009 NA 81.4 ~92
2) To generate additional data on effects of hydrogen charging X65 Yes Yes 0.0077 NA 58.84
pipeline steels; Yes Yes 0.0075 NA 74.23
3) Develop an approach for conducting probabilistic sour ECA and
Yes Yes 0.005 NA 67.56
perform realistic case studies of the probabilistic ECA approach.
Yes Yes 0.005 NA 60
Yes Yes 0.005 NA 94.5
This paper focusses on the parts 1 and 3, and will briefly discuss the
additional testing efforts. Yes Yes 0.005 NA 58.5
Yes Yes 0.005 NA 59.9
DETERMINATION OF INPUTS Yes Yes 0.005 NA 64.9
Yes Yes 0.008 26.56 75.19
The objective of this activity is to assemble input data that are required Yes Yes 0.006 20.16 64.52
for assessing the significance of circumferentially oriented external and Yes Yes 0.007 21.39 65.81
internal surface flaws in girth welds in sour service. The task focusses Yes Yes 0.007 9.3 58.4
~67
on the highly important fracture toughness parameter which was Yes Yes 0.007 16.89 61.6
determined through a literature review - a number of previous papers Yes Yes 0.007 11.4 61.77
(Ali and Pargeter, 2014) and in house data on fracture toughness in sour Yes Yes 0.008 8.53 49.4
service. The results from literature review have been examined in order Yes Yes 0.005 20.7 74.69
to determine the key ECA input parameters. Yes Yes 0.009 38.95 89.12
Yes Yes 0.005 15.02 52.88
The papers and reports considered could be summarized as follows: Yes Yes 0.007 19.7 86.96
1) API 5L X65 linepipe parent material various suppliers/heats Yes Yes 0.007 10.96 61.58
2) Sour environments of 10% H2S with balance of CO2 Yes Yes 0.008 5.43 78.59
3) Solution pH of both 2.7 and 4.5 included
Yes Yes 0.009 22.03 76.55
4) Fracture toughness data from SENB specimens
Yes Yes 0.008 NA 75.19
a. Specimen geometries varied slightly, however a/W of 0.5 was
Yes Yes 0.006 NA 64.52
consistent
b. Both side grooves and no side grooves were included
5) Samples were both coated and non-coated
6) Samples were pre-charged before toughness testing KEY PARAMETER ASSESSMENT
7) Loading rates varied from roughly 0.005 to 0.042 MPam0.5s-1
8) Testing was performed at room temperature A series of deterministic engineering critical assessments (ECA) with
realistic ranges of key input parameters were carried out to establish the
In summary, in order to obtain a fracture toughness value to be used in relative influence of these parameters on the outcome of the ECA.
the deterministic engineering critical assessments, results from the These cases are intended to simulate the circumferential girth welds of
surveyed references were carefully analysed. The selected fracture offshore flowlines. All cases were assessed using CrackWISE 5
toughness results were parsed as provided in Table 1. automating the fracture and fatigue clauses of BS 7910:2013-A1 (BSI,
2013). Table 2 provides a summary of cases analysed with input data
All entries in Table 1 were manufactured from parent material. All used in each analysis. Pairs of analysis cases (i.e. Cases 1 and 2, Cases
specimens were pre-charged and tested in the sour environment, subject 3 and 4) were differentiated by their residual stress whether to allow
to an initial K-rate between 0.005 and 0.009 MPam0.5s-1. It is also noted relaxation or include PWHT. There were, in total, 50 cases analysed by
that the average toughness values from tests of side grooved specimens varying the input parameters as discussed below.
are lower than that of plane-sided specimens. The values obtained from
side-grooved specimens were selected. As a result, the toughness value Flaw type and component geometry
(Jmax) to be used in assessing key input parameters was determined to In all assessments, a pipe/cylinder containing a circumferentially
be 67 Nmm-1. It should be emphasised that this value was selected for oriented external surface flaw was considered (this geometry is suitable
use in specific work effort only it may or may not be appropriate for for assessing both external and internal flaws). The maximum tolerable
use in other efforts. flaw size was determined through sensitivity-criticality analysis. Final

385
assessment output was the limiting (or critical) flaw height (a) versus 48 450 535 360 1.5 6.35 67
flaw length (2c) curve. The following input data were assumed for the Material properties
pipe geometry: Two sets of tensile properties were to be considered in the deterministic
ECA. The first set of tensile data is from actual test results reported in
Section thickness: 25.4mm (-10% to 12.5% of thickness surveyed literature while the second set is the minimum required tensile
representing manufacturing tolerance that will be considered for strength by API 5L (API, 2007). Option 1 assessment route of
probabilistic efforts); BS 7910:2013-A1 was adopted in all deterministic ECAs (BSI, 2013).
Outside radius, ro: 178mm. In summary, the tensile properties considered were:

Table 2. Deterministic ECA matrix Yield strength: 511MPa (actual), 450 MPa (API 5L);
Yield Primary Fracture UTS: 600MPa (actual), 535 MPa (API 5L);
Case strength UTS stress e L toughness Youngs modulus: 207GPa;
No. (MPa) (MPa) (MPa) (mm) (mm) (Nmm-1) Poissons ratio: 0.3.
1 511 600 102.2 0.5 63.5 67
2 511 600 102.2 0.5 63.5 67 The above tensile properties were obtained from tests carried out in air.
3 450 535 90 0.5 63.5 67
4 450 535 90 0.5 63.5 67 As discussed above, the fracture toughness value (Jm) to be used was
3a 450 535 90 0.5 63.5 15 chosen as 67 Nmm-1. A lower value of 15 Nmm-1 will be applied to
4a 450 535 90 0.5 63.5 15 represent the J0.2BL.
5 511 600 408.8 0.5 63.5 67
6 511 600 408.8 0.5 63.5 67 Misalignment
7 450 535 360 0.5 63.5 67 The presence of misalignment, either axial, angular or both, at a welded
8 450 535 360 0.5 63.5 67 joint may cause an increase in stresses acting on this joint when it is
9 511 600 102.2 1.5 63.5 67 loaded, due to the introduction of local bending stresses. The bending
10 511 600 102.2 1.5 63.5 67 stresses arising from misalignment affect both the stress intensity factor
11 450 535 90 1.5 63.5 67 and reference stress. Further details regarding misalignment are given
12 450 535 90 1.5 63.5 67 in Annex D of BS 7910:2013-A1 (BSI, 2013).
13 511 600 408.8 1.5 63.5 67
14 511 600 408.8 1.5 63.5 67 For this study, an upper and lower bound for girth weld misalignment
15 450 535 360 1.5 63.5 67 of 1.5mm and 0.5mm respectively was used.
16 450 535 360 1.5 63.5 67
17 511 600 102.2 0.5 25.4 67 Loading conditions
For primary stresses, 20% and 80% of the yield strength was assumed
18 511 600 102.2 0.5 25.4 67
to represent the lower band and upper band of the possible axial stress
19 450 535 90 0.5 25.4 67
during operation. In terms of residual stresses, both as-welded
20 450 535 90 0.5 25.4 67
condition and post weld heat treatment (PWHT) were considered. The
21 511 600 408.8 0.5 25.4 67 PWHT condition was selected purely as an approximate and pragmatic
22 511 600 408.8 0.5 25.4 67 way for considering cases where the magnitude of girth weld residual
23 450 535 360 0.5 25.4 67 stress is significantly lower than the default yield-magnitude value. The
24 450 535 360 0.5 25.4 67 latter is normally assumed to ensure conservative assessments but it can
25 511 600 102.2 1.5 25.4 67 lead to pessimistic results if the girth weld flaw is in a region of low
26 511 600 102.2 1.5 25.4 67 residual stresses. During the assessment of the girth weld in as-welded
27 450 535 90 1.5 25.4 67 condition, relaxation of residual stresses due to the application of
28 450 535 90 1.5 25.4 67 mechanical loads was allowed. According to BS 7910:2013-A1 (BSI,
29 511 600 408.8 1.5 25.4 67 2013), the residual stress for the weld after PWHT is 20% of its yield
30 511 600 408.8 1.5 25.4 67 strength.
31 450 535 360 1.5 25.4 67
32 450 535 360 1.5 25.4 67 Environmental tensile properties
33 511 600 102.2 0.5 6.35 67 The effect of environment on the tensile properties was examined.
34 511 600 102.2 0.5 6.35 67 Due to the limited amount of testing no results are provided here.
35 450 535 90 0.5 6.35 67 Rather a qualitative discussion of the activity is provided below.
36 450 535 90 0.5 6.35 67
37 511 600 408.8 0.5 6.35 67 The tensile testing was carried out on X65 grade steel (API 5L: 2015)
38 511 600 408.8 0.5 6.35 67 in air, hydrogen and sour environments. Round tensile specimens were
39 450 535 360 0.5 6.35 67 machined from the parent material of the specified pipe in the
40 450 535 360 0.5 6.35 67 longitudinal orientation. Tensile testing was carried out at room
temperature at a strain rate of 10-6 sec-1 in the following environments:
41 511 600 102.2 1.5 6.35 67
42 511 600 102.2 1.5 6.35 67
Air;
43 450 535 90 1.5 6.35 67
Gaseous Hydrogen at 250bar, 99.9% pure;
44 450 535 90 1.5 6.35 67
Sour environment, modified NACE A solution (standard
45 511 600 408.8 1.5 6.35 67
TM0177:2005), purged with a mixture of 10% H2S balance CO2.
46 511 600 408.8 1.5 6.35 67
47 450 535 360 1.5 6.35 67

386
The 10mm gauge diameter specimens were tested in air and in the sour
environment. The 5mm gauge diameter specimens were tested in the
gaseous hydrogen environment; the specimen size was dictated by the
limitations of the hydrogen vessel dimensions. All environmental tests
were exposed to their relevant test environment prior to testing.
Hydrogen concentration measurements were performed on the tested
tensile specimen and compared to reference specimens that
accompanied the tests specimens for environment exposure but were
not strained.

It was observed that the yield and tensile strengths were roughly 10%
lower for the tests performed in hydrogen gas than the corresponding
in-air tests, whereas the sour environment yield and tensile strengths
were roughly equivalent to the in-air results. More variability was
realized for the sour environment test results than the other two
environments.

As soon as the tensile tests were completed, the broken halves of the
specimens, and the exposed coupons, were subjected to hydrogen Figure 1. 2c versus a curves for all cases in deterministic assessment
measurement. The diffusible hydrogen was measured at 400oC using
the Nitrogen gas carrier method. Hydrogen content was measured by
thermal conductivity in accordance with standard BS EN ISO
3690:2012.

The hydrogen absorption in the sour environment was generally higher


than that in the gaseous high pressure hydrogen environment. In both
cases, the strained material absorbed more hydrogen than the
unstrained material. Particularly for the hydrogen content in unstrained
material, the measured values were almost zero. This suggests there is
little need to pre-charge tensile specimens without load prior to testing
in hydrogen gas, as the main effect of the hydrogen is experienced once
the specimen is under load.

The observed reduction in area for the tensile coupons was in line with
the measured hydrogen contents. The effect of hydrogen on yield or Figure 2. Critical flaw sizes for Case 8 by defining the stress intensity
tensile strength is more complicated and nuanced. Due to the limited factor taken from the maximum value around the crack tip and from the
test data and other considerations, the difference on yield or tensile deepest point of the crack.
strength should not be broadly interpreted as a real difference.
Nevertheless, it is possible to address a reliable conclusion after a lot In Figure 3, the effect of secondary stresses on the maximum tolerable
more data being generated. Due to the lack of sufficient tensile data, the flaw sizes can be seen. From Table 2, it can be seen that the difference
tensile properties obtained from this effort were not considered as between cases 1 and 2 (or cases 3 and 4, 5 and 6, 7 and 8) is the level of
distribution parameters adopted in the probabilistic ECA effort. Instead, residual stress. For cases 1 and 2, the applied stress equals to 20% of
the tensile properties were adopted as deterministic values. yield strength. However, the residual stress for case 1 is assumed to be
equal to yield strength. The residual stress for case 2 is equal to 20% of
Results and discussion yield stress as a result of PWHT. Since residual stresses contribute to
The results of the deterministic ECA cases tabulated in Table 2 are the crack driving force and a larger value of residual stress was
presented in Figure 1. In general, the maximum tolerable flaw height considered for case 1, smaller critical flaw heights were obtained at the
decreases with the increase of flaw length. However, there are some same flaw length compared to case 2. This is also supported by results
exceptions such as Case 8 (shown in Figure 2) where the maximum of cases 3 and 4. However, when the primary stress increases to 80% of
tolerable flaw height increases with the increase of flaw length but yield strength, the difference between the critical flaw heights reduces
decreases when the flaw length is greater than 8mm. The reason for this dramatically. This can be concluded from the results of cases 5 and 6 in
is that the maximum value of stress intensity around the crack tip was Figure 3. The reason is that the residual stresses relaxed considering the
selected for the assessment and the location of maximum stress interaction with large primary stresses. The residual stress level in the
intensity factor may change as the crack grows. When the flaw height is presence of primary stress with a value of 80% of yield strength is
less than 8mm, the maximum stress intensity factor was found at the about 50% to 60% of yield strength for cases 5 and case 7 while the
surface point of the crack. It was observed that the stress intensity residual stress considered for case 6 and 8 was 20% of yield strength.
factor decreased at the surface point when the flaw length increased to The difference between residual stresses decreases. Hence, primary
about 130mm. When the flaw length is greater than 130mm, the stresses contributed more to the final crack driving force.
maximum stress intensity factor was found at the deepest point of the Consequently, a smaller difference of critical flaw height between cases
flaw. Figure 2 shows the critical flaw size when only the deepest point 5 and 6 or between cases 7 and 8 was observed compared to cases 1
of the flaw is concerned. For this case, the critical flaw height and 2 or cases 3 and 4. It is also indicated that if the primary stress is
decreased with the increase of flaw length. For the results shown in large, the critical flaw sizes are not very sensitive to the residual stress.
Figure 1 and the discussion below, the maximum stress intensity factor In other words, the influence of residual stress on critical flaw sizes is
found at the crack tip was used for the deterministic studies. small when the primary stress is large.

387
17 when the flaw size is about 40mmX10mm. The stress concentration
affects the crack driving force for a shallow flaw but the effect
decreases gradually for deeper flaws. Therefore, it was found that the
same critical flaw sizes in cases 17 and 33 when the flaw length is
greater than 40mm.

Figure 3. Critical flaw sizes for cases 1, 2, 5 and 6

In Figure 4, the influence of misalignment on the critical flaw sizes is


exhibited. It can be seen that when the misalignment increases from
0.5mm to 1.5mm, the maximum tolerable flaw height decreases for a Figure 5 Critical flaw sizes for cases 1, 2, 17, 18, 33 and 34.
low primary stress (20% of yield strength). This can be found by
comparing the results of cases 1 and 9, cases 2 and 10, cases 3 and 11 The effect of fracture toughness on critical flaw sizes is presented in
or cases 4 and 12. However, when the primary stress increases from Figure 6. For cases 3 and 4, the J-integral fracture toughness assumed
20% of yield strength to 80% of yield strength, the effect of was 67kJ/mm2 while this value was 15kJ/mm2 for cases 3a and 4a. By
misalignment becomes smaller and can be ignored (see cases 5 and 13). comparing the results from cases 3 and 3a or cases 4 and 4a, it can be
This may arise as the bending stress due to misalignment has small seen that the maximum critical flaw sizes decrease with the decrease of
contribution to the crack driving force and limit load. It is also found fracture toughness.
that for small primary stress if the residual stress is high, a large portion
of contribution to the crack driving force may be from residual stress
and the bending stress due to misalignment is very small. Therefore,
the critical flaw size may be not sensitive to misalignment if the
primary stress is high or if the primary stress is low but the secondary
stress is high.

Figure 6 Critical flaw sizes for cases 3, 4, 3a and 4a.

PROBABILISTIC ENGINEERING CRITICAL


ASSESSMENT

The purpose of performing a probabilistic ECA analysis is to assess the


Figure 4. Critical flaw sizes for cases 1, 2, 5, 9, 10 and 13 component taking into account uncertainties in the understanding of
loading conditions, flaw dimension and material properties, which will
In Figure 5, the influence of Mk factor on the critical flaw sizes is help to mitigate the conservatism of the deterministic ECA. It is also
illustrated by changing the weld cap width. The weld cap width required for a risk-based inspection (RBI) assessment to provide
assumed for cases 1 and 2 is 63.5mm, 25.4mm for cases 17 and 18 and guidance for operators to schedule the inspection plan economically
12.7mm for cases 33 and 34. It can be seen that for cases 2, 18 and 34, and effectively. An inspection should be scheduled when the predicted
the maximum tolerable flaw height decreases as the weld cap width probability of failure (PoF), i.e. the probability that a flaw is not
increases. Considering that the maximum value of stress intensity acceptable, exceeds the target PoF.
factor was observed at the surface point of this flaw, the contribution of
Mk factor to stress intensity is more evident. However, the maximum Considering the nature of the assessment procedure and the equations
value of stress intensity factor is observed at the deepest point in case given for the stress intensity factor and reference stress in

388
BS 7910:2013-A1 (BSI, 2013), Monte Carlo Simulation (MCS) were tried on the available fracture toughness data generated in the sour
technique (performing a large number of iterations with different environment to fit the test results sample shortlisted in Table 1. It was
combinations of input data) was found to be more suitable compared to found that the Kolmogorov-Smirnov (K-S) value for this sample was
various other techniques including FORM (First Order Reliability 0.144 (log-normal distribution) and 0.174 (Weibull distribution). K-S
Method) or SORM (Second Order Reliability Method) for probabilistic test is one of the most useful and general nonparametric methods for
ECA. Alternatively, a sampling method can also be applied in order to comparing a sample with a reference probability distribution to
accelerate the Monte Carlo Simulation. determine the most suitable distribution type for this sample. Figure 7
and Figure 8 show the R plot for each distribution. Regarding the K-S
To support the probabilistic ECA analysis, a development version of test, the lower the K-S value the better the fitting is observed. The exact
CrackWISE5 was built. This builds upon the interfaces already critical value for a K-S test is a function of sample size and the
present in the software and incorporates a probabilistic interface option. preferred significance level. In general, if the K-S value of a fit is over
In this CrackWISE5 probabilistic version, MCS and Latin Hypercube 0.05 (which is a conservative value giving the significance level of 0.05
Sampling (LHS) techniques were automated. for the sample size over 600), the fitting needs to be reviewed. In this
project, the high K-S value is most probably due to the small size of the
In summary, to execute a probabilistic fracture assessment, it is sample considered.
required to set up a normal deterministic fracture assessment case,
select appropriate parameters (geometry, material, loading)
representing the distributions. The distributions that can be chosen are
normal, log-normal and Weibull distributions. If normal distribution is
selected, the distribution parameters are mean value and standard
deviation. If log-normal distribution is selected, the distribution
parameters are location and scale. If Weibull distribution is selected,
the distribution parameters are shape and scale. After setting up the
parameters, the probability of failure can be calculated in the results
interface that allows the user to plot the failure assessment diagram.

CrackWISE5 probabilistic version allows majority of input


parameters of ECA to be defined as distributions. For the case studies,
parameters related to component/flaw dimensions and fracture
toughness were considered as distributions. In Table 3, the assessment
matrix of the three case studies is presented with the parameters
assumed to be normal, log-normal or Weibull distributions.

Table 3 Assessment matrix of probabilistic ECA case studies Figure 7. R plot of Log-normal distribution Fitting parameters of
Case Distributed Distribution fracture toughness. (Jm).
Study parameter type Distribution parameters
Weibull Scale: 73.01 Shape: 5.95
Log-mean: 4.2, Log-std:
1 Toughness Log-normal 0.165
Thickness Normal Mean:25.4, std: 1.296
2 Toughness Weibull Scale: 73.01 Shape: 5.95
Flaw length Normal Mean:50.56, std: 1.29
Flaw height Normal Mean: 6.078, std: 0.155
Thickness Normal Mean:25.4, std: 1.296
3 Toughness Weibull Scale: 73.01 Shape: 5.95

Case study 1
The base case of this initial case study is Case 8. The example flaw
dimensions are: 2c is equal to 15mm and a is 3mm. This flaw size is
recommended by BS 7910 as the minimum flaw dimensions can be
detected by Non-destructive Test (NDT) (e.g. manual Ultra-sonic).
Under the given conditions, the assessment point of this analysis lies
under the Failure Assessment Line (FAL) on the FAD, which will be
considered as acceptable in a deterministic assessment. Figure 8. R plot of Weibull distribution Fitting parameters of fracture
toughness. (Jm).
In this case study, it was assumed that fracture toughness can be
expressed as Weibull or log-normal distributions. The fracture The scale parameter of Weibull distribution is 73.01, and shape
toughness data in Table 1 was processed using statistical software for parameter is 5.95. For log-normal distribution, the log-mean is 4.2 and
fitting a distribution using maximum-likelihood estimation (mle) log-standard deviation is 0.165. These two different distribution types
approach. have been adopted in Case study 1 and the results are as follows.

Generally, Weibull distribution and log-normal distribution are Weibull distribution: CrackWISE5 probabilistic version calculated
believed to be the most appropriate distribution types that can be that the probability of failure for this case is 0.0243, based on
applied for fracture toughness values (Tuma et al, 2006). Both of them Monte Carlo with 106 iterations, and 0.025 using LHS method;

389
Log-normal distribution: CrackWISE5 probabilistic version
calculated that the probability of failure (PoF) for this case is
0.00066, based on Monte Carlo with 106 iterations, and 0.001 using
LHS method.

As discussed above and shown in Figure 6, the critical flaw size will
decrease by reducing material fracture toughness. Therefore, a
distribution considering J0.2BL value as a lower band of the material
fracture toughness was also fitted and the same assessment was
repeated to study the effect of lower toughness in probabilistic ECA.
The J0.2BL values corresponding with the data in Table 1 testing were
fitted with statistical software.

Figure 9 and Figure 10 show the R plot for each distribution. The scale
parameter of Weibull distribution is 20.91, and shape parameter is 2.21
with K-S value of 0.138. For log-normal distribution, the Log-mean is
2.79 and log-standard deviation is 0.508 with K-S value of 0.143.
Adopting these two distribution types in this case study resulting in: Figure 10. R plot of Weibull distribution Fitting parameters of fracture
toughness. (J0.2BL).
Weibull distribution: CrackWISE5 probabilistic version calculated
that the probability of failure for this case is 0.9821, based on Case study 2
Monte Carlo with 106 iterations, and 0.982 using LHS method; The base case of Case study 2 is Case 8 as well. The example flaw
Log-normal distribution: CrackWISE5 probabilistic version dimensions are: 2c is equal to 15mm, and a is 3mm. This case is
calculated that the probability of failure (PoF) for this case is designed to investigate the effect of pipe manufacturing tolerance (i.e.
0.9585, based on Monte Carlo with 106 iterations, and 0.958 using wall thickness) on PoF considering that fracture toughness is
LHS method. represented as a Weibull distribution as was the case in Case Study 1.

The K-S values for both Log-normal distribution and Weibull


distribution are more than 0.05, which means it is hard to decide which
distribution type is more suitable. Considering that Weibull distribution
has been more commonly used (Vadholm, 2014) and for J m values, the
Weibull distribution seems more conservative as shown in Case study
1, it will be used in Case studies 2 and 3, to represent the fracture
toughness for Weibull distributed Jm:

Scale: 73.01;
Shape: 5.95.

As suggested by ExxonMobil, the thickness of the pipe can also be a


distribution due to the manufacturing tolerance. The thickness was
assumed to be normally distributed and the mean value is 25.4mm with
a standard deviation of 1.423mm which allows a -10% and +12.5% of
manufacturing tolerance with 95% confidence interval.

Figure 9. R plot of Log-normal distribution Fitting parameters of CrackWISE5 probabilistic version calculated that the probability of
fracture toughness (J0.2BL). failure for this case is 0.0248, based on Monte Carlo Simulation with
106 iterations, and 0.024 using LHS method.
Comparison between different distribution types showed in this
particular case that the Weibull distribution seems more conservative. Comparison with the assessment results between Case study 1 and 2,
Moreover, by reducing fracture toughness, the probability of failure the influence of manufacturing tolerance to the total PoF is not very
will increase, which is in line with the discussion in Section 4.5. The significant in this case.
benefit of probabilistic assessment in this case is that it can represent
the influence of toughness between its lower and upper band, rather Case study 3
than just deterministically show two points on the FAD for the worst The base case of this case study is Case 8. Case study 3 is designed to
and best cases. investigate the contribution of inspection error to PoF considering that
fracture toughness and thickness are expressed as distributions.
Besides, it is important to emphasise that the sample size for either J m
or J0.2BL was too small to generate an accurate distribution model. The assumptions made on the thickness and fracture toughness in the
previous case studies are valid for this case as well. It is also assumed
that the flaw size is normally distributed as a result of inspection error.

Generally speaking, the inspection results may incorporate 5% error


which means that the flaw size determined through the inspection will

390
be normally distributed with a 5% error and 95% confidence interval, scatter should result in a K-S value of up to 0.05 in general.
i.e. for 2c, the mean is 15mm with 0.383 standard deviation while for a,
the mean is 3mm with standard deviation of 0.077. This error can be ACKNOWLEDGEMENTS
mitigated if inspection can be repeated or if the inspection technique
can be improved. The authors wish to acknowledge the management of ExxonMobil
Production Company, ExxonMobil Upstream Research Company and
CrackWISE5 probabilistic version calculated that the probability of TWI for their support of this publication. The authors also thank
failure for this case is 0.0254, based on Monte Carlo with 106 Richard Pargeter (TWI) and Andreea Crintea (TWI) for the work effort.
iterations, and 0.026 using LHS method.
REFERENCES
Comparison among the above three case studies, the influence of
distribution of fracture toughness to PoF is the most significant in this Ali M and Pargeter R (2014): Techniques for determining the effect of
project. However, it should be noted that due to the small size of the a sour environment on fracture toughness of steel, Steely Hydrogen.
sample, the fitting of this distribution has a relatively wider scatter 2nd International Conference on Metals and Hydrogen. 5-7 May 2014,
band. This will also result in a relatively more scattered plot when Gent, Belgium.
applying simulation approach for estimating probability.
Agrell C, Ostby E, Levold E, Hauge M and Bjerke S (2016): The use
CONCLUSIONS of probabilistic fracture assessment procedures in design of pipelines
subjected to large strains, The 26th International Ocean and Polar
This paper summarizes a work effort to investigate the development of Engineering Conference, 26 June 2 July, Rhodes, Greece.
probabilistic ECA approach, with particular emphasis on utility in sour
environments. Key parameters were identified and then utilized in both API, 2007: API Specification 5L Specification for Line Pipe.
deterministic and probabilistic assessments. The findings are presented
below. BSI, 2013: BS 7910-A1 Guide to methods for assessing the
acceptability of flaws in metallic structures British Standards
1. From literature review, for the specific material(X65) Institution, London.
/environment combination, an appropriate initial K-rate for SENB
fracture toughness tests was found to be approximately Lee J, Kim K, Cho W, Shim S and Huh N (2015): On probabilistic
0.008MPam0.5s-1. For this specific combination, the value of fracture assessment of defective pipeline based on target failure
67Nmm-1 is close to the average value of Jm results considered and probabilities, ASME 2015 34th International Conference on Ocean,
a value of 15Nmm-1 is close to the average value of J0.2 results Offshore and Arctic Engineering, Vol. 5B, Canada, May 31- June 5,
considered. 2015.

2. The deterministic ECA analyses carried out in this report for the Mechab B, Serier B, Kaddouri K and Bouiadjra B (2014):
pipes containing circumferentially oriented surface flaws reveal Probabilistic elastic-plastic analysis of cracked pipes subjected
that: to internal pressure load, Nuclear Engineering and Design,
Vol. 275, pp. 281-286.
For the same loading ratio (i.e. same applied stress/yield
stress), the influence of yield strength and ultimate tensile
Provan J (1987): Probabilistic fracture mechanics and reliability,
strength on critical flaw size is small;
Springer Science and Business Media Dordrecht.
The critical flaw size may be insensitive to misalignment if
the primary stress is high or if the primary stress is low but Sandvik A, Ostby E and Thaulow C (2006): Probabilistic fracture
the secondary stress is high; assessment of surface cracked pipes using strain-based approach,
If the other conditions are the same and stress relaxation is Engineering Fracture Mechanics, Vol. 73, pp. 1491-1509.
allowed, the residual stress has relatively smaller influence
on critical flaw sizes when the primary stress is large; Vadholm T (2014): Investigation of low temperature toughness and
The weld cap width affects the crack driving force for a crack initiation in welded structural steels, Materials Science and
shallow flaw but the effect decreases gradually as the flaw Engineering, Norwegian University of Science and Technology.
becomes deeper.

3. The effect of hydrogen on yield or tensile strength is more


complicated and a number of factors need to be taken into
account. The difference in the tensile properties observed in air
and environment (sour/hydrogen gas) was certainly significant
and further investigation is warranted.

4. The probabilistic studies indicated that the distribution of fracture


toughness played the most significant role on the determination
of PoF in this project compared with inspection error and
manufacturing tolerance.

5. It should be noted that due to the small size of the samples, the
distribution of toughness fitted had a relatively wider scatter
band. A good fitting of the samples in terms of both size and

391

Vous aimerez peut-être aussi