-
RECENT INNOVATIONS IN PIPELINE SEAM WELD INTEGRITY
ASSESSMENT
Ted L. Anderson
Quest Integrity Group
2465 Central Avenue
Boulder, CO 80301 USA
ABSTRACT. The integrity of pipelines with longitudinal seam
welds has received renewed interest by
operators and regulators, due primarily to a number of
high-profile incidents. The pipeline industry
currently assesses planar flaws in seam welds with methodology
that dates back nearly 40 years. The
traditional crack assessment models can lead to gross errors in
the prediction of burst pressure. This
article points out the problems with these models, using both
theoretical analysis and a comparison with
burst test data. Improved flaw assessment methods are described,
along with a new software application
for automated pressure cycle fatigue analysis.
BACKGROUND
There have been a number of catastrophic failures in seam-welded
pipelines in recent
years, the most notable of which was the 2010 explosion of a
Pacific Gas & Electric
(PG&E) pipeline in San Bruno, California. There is
increasing pressure on operators to
demonstrate that they are taking steps to improve the integrity
of their pipelines. Simply
adhering to the status quo in the form of existing integrity
plans is no longer an option.
The pipeline industry currently relies on flaw assessment
methods that are nearly 40
years old, but improved models are available. There have been
significant advances in
fracture mechanics, fitness-for-service assessment, and
remaining life models in the past 40
years. The pipeline industry can benefit by adopting
methodologies that have been
successfully applied in other industries, including oil &
gas production, refinery, chemical,
petrochemical, and power generation.
This paper presents a sample of innovative technology that can
be applied to the
integrity management of seam-welded pipe. The focus of this
article is on cracks and other
planar flaws, but innovative approaches for other anomaly types
are also available.
Traditional methods for assessing flaws in pipelines,
particularly crack assessment models,
have a number of series serious shortcomings. These models are
re-evaluated against burst
test data in light of our improved knowledge of fracture
mechanics. Also, a new software
application for automated pressure cycle fatigue analysis is
described.
-
OVERVIEW OF FAILURE OF PIPELINES CONTAINING CRACKS
The presence of cracks and other planar flaws can significantly
lower burst pressure in a
pipe. The magnitude of the reduction is a function of crack size
and material toughness.
Figure 1 illustrates the relationship between burst pressure and
toughness for a given crack.
The toughness in this example is characterized by the critical
stress intensity factor, Kc,
which is a fracture mechanics parameter [1].
Figure 1 shows three distinct regimes. At low toughness values,
the pipe exhibits
linear elastic material behavior, and burst pressure is linearly
related to toughness. At the
opposite extreme, the burst pressure is independent of
toughness, and is instead governed
by tensile properties. In the ductile rupture regime, metal loss
models such as the B31G
equations can be used to predict burst pressure. In the
elastic-plastic regime, burst is
governed by a combination of toughness and tensile
properties.
Fracture mechanics [1] is a relatively new engineering
discipline that provides
mathematical relationships between flaw size, material
toughness, and burst pressure.
Linear elastic fracture mechanics (LEFM) is suitable when the
toughness is low. Elastic-
plastic fracture mechanics (EPFM) is a more robust model that
encompasses both the linear
elastic and elastic-plastic regimes in Fig. 1.
For the past 40 years, the pipeline industry has relied on a
crack assessment method,
the ln-sec model [2], which predates the development of modern
elastic-plastic fracture
mechanics. Moreover, LEFM was in its infancy when ln-sec model
was developed.
Consequently, the ln-sec model contains fundamental errors that
often lead to highly
inaccurate predictions of burst pressure. This issue is explored
further below.
FIGURE 1. Relationship between burst pressure and material
toughness.
-
THE LN-SEC MODEL
Theoretical Evaluation of the Model
Kiefner, et al [2] developed the ln-sec model in the early
1970s. It is based on a simplified
fracture mechanics approach, and was calibrated to a series of
burst data. In 2008, Keifner
[3] introduced a modified version of the ln-sec model, which was
intended to address
known problems in the original method. The original ln-sec model
starts with the linear
elastic fracture mechanics equation for failure in a pipe with a
longitudinal crack of length
2c:
c T FK M c (1)
where F is the hoop stress at failure, and MT is known as the
Folias factor, which is a
function of the crack length relative to the pipe diameter and
wall thickness. The Folias
factor can be viewed as a magnification factor on hoop stress
due to the presence of the
flaw. For fully ductile rupture (Fig. 1), MT is inversely
proportional to the remaining
strength factor (RSF), which is the ratio of the hoop stress at
failure in the flawed pipe to
the flow stress of the material:
1F
flow T
RSFM
(2)
where the flow stress is taken to be the hoop stress at failure
in a pipe without a flaw. Now
consider a part-through surface crack, as illustrated in Fig. 2.
Keifner et al introduced a
surface correction factor, Ms, which is equal to the remaining
strength factor for this
configuration:
1
11
o
s
T o
A
AM RSF
A
M A
(3)
Where A is the area of the flaw and Ao is the rectangular area
2t c . Equation (3) forms
the basis of the B31G remaining strength factor equations and
similar standard methods for
assessing metal loss.
Keifner et al [1] observed that Eq. (3) over-predicted the burst
pressure for pipes
with axial cracks, except when the material toughness was very
high (as one would expect
from Fig. 1). They adjusted for toughness by incorporating Eq.
(3) into a simplistic elastic-
plastic fracture mechanics model, which was the only such model
available at the time:
-
FIGURE 2. Part through surface flaw.
2
2 2
12
lnsec8 8 2
c Fc
flow flow s flow
CVNE
K A
c c M
(4)
Where CVN is Charpy energy in ft-lbs, Ac is the area of the
Charpy specimen (in2), and E is
Youngs modulus in psi. The above expression, which is the
original ln-sec model,
assumes the following relationship between Kc and Charpy
energy:
12
c
c
CVNK E
A (5)
Although Eq. (4) predicts the correct shape of the burst
pressure versus toughness
curve (Fig. 1), the predicted slope in the linear range differs
from LEFM. Part of the
problem is the surface correction in Eq. (3), which appears to
work well for fully ductile
rupture, does not apply to the linear elastic range. Figure 3 is
a plot of the error in the
original ln-sec model in the linear elastic range, relative to a
rigorous LEFM model. Note
that the largest error occurs for long and shallow cracks.
Keifner became aware of the problems with Eq. (4) for long &
shallow cracks,
based on outliers in the original set of burst test results, as
well as through subsequent
application the model to new burst test data and hydrostatic
test failures. In 2008, Keifner
introduced a modified version of the ln-sec model [3]. The
modified model included a
correction factor that attempted to address the problems with
long, shallow cracks.
-
FIGURE 3. Error in the original ln-sec model in the linear
elastic range. Equation (4) was compared with a
rigorous LEFM model for the case where toughness 0.
Figure 4 is a plot of predicted burst pressure versus toughness,
which is represented
as CVN in order to retain the linear relationship with pressure
at low toughness. The
original and modified ln-sec models are compared with the LEFM
prediction. As stated
earlier, the original ln-sec model predicts the correct shape
(Fig. 1), but the slope in the
linear elastic range is off by roughly a factor of 2 in this
specific case. The modified
model, however, does not even predict the correct overall shape.
This model predicts that
burst pressure is insensitive to toughness, and that a non-zero
burst pressure would be
achieved with zero toughness. In other words, the error in the
modified ln-sec model is
infinite as toughness approaches zero.
-
FIGURE 4. Predicted burst pressure versus toughness for LEFM, as
well as the original and modified ln-sec
models. The original model has the correct shape (Fig. 1) but
the slope in the linear elastic range is off by a
factor of 2 in this case. The modified model predicts that burst
pressure is insensitive to toughness, which is
fundamentally incorrect.
Comparison with Burst Test Data
The shortcomings in both the original and modified ln-sec
models, which were identified
above through theoretical analysis, can also be observed in
burst test data. The original
Keifner et al article [2] includes burst test data for both
blunt and sharp notches, where the
latter was intended to represent sharp cracks. A subset of 35
burst tests in this article,
which consisted of pipes with sharp surface notches for with
Charpy data were available,
are suitable for comparing with both Eq. (4) and Keifners recent
modification.
Table 1 summarizes the comparison between predicted and actual
burst pressure in
the 35 experiments. The original ln-sec model is conservative on
average, while the
average predictions of the modified model are close to reality.
The standard deviation in
the predictions is slightly higher for the modified model.
Merely looking at average predictions is not sufficient to
identify systematic errors
in the two models. Figures 5 to 7 are plots of predicted/actual
burst pressure versus flaw
length, flaw depth, and Charpy energy, respectively. According
to Figures 5 and 6, the
original ln-sec model grossly under-predicts burst pressure for
long and shallow cracks,
respectively. This result is consistent with Fig. 3. The
modified model appears to
correct the under-predictions, which was Keifners intent, but
there are also outliers
-
where the burst pressure is grossly over-predicted. The reason
for these unconservative
outliers is evident in Fig. 7, where predictions are plotted
against Charpy energy. Most of
the unconservative predictions correspond to low-toughness
materials.
TABLE 1. Summary of the comparison of the original and ln-sec
models with 35 burst test results from the
original Keifner et al article [2].
Predicted/Actual Burst Pressure
Original Ln-Sec Model Modified Ln-Sec model
Average of 35 Tests: 0.871 0.980
Standard Deviation: 0.129 0.135
FIGURE 5. Ln-sec model predictions versus flaw length. The
original ln-sec model significantly under-
predicts burst pressure for long flaws.
-
FIGURE 6. Ln-sec model predictions versus crack depth/thickness
(a/t). The original ln-sec model
significantly under-predicts burst pressure for shallow
flaws.
FIGURE 7. Ln-sec model predictions versus Charpy energy. The
modified ln-sec model significantly over-
predicts burst pressure for materials with lower toughness.
-
STATE-OF-THE-ART FRACTURE MECHANICS ANALYSIS
Significant advances in the field of fracture mechanics have
occurred over the past 40
years, when the original ln-sec model was developed. For
example, the API 579-1/ASME
FFS-1 Fitness-for-Service Standard [4] includes a simplified
fracture mechanics model that
is vastly superior to the ln-sec model, both the original and
modified versions. A more
advanced fracture mechanics model is described below.
The J-Integral and Finite Element Analysis
The most rigorous (and accurate) method to predict the effect of
cracks on burst pressure is
3D elastic-plastic finite element analysis. Figure 8 shows a
typical finite element model of
a longitudinal crack in a seam weld. The preferred fracture
mechanics parameter for
elastic-plastic analysis is called the J-integral. Fracture
toughness in the elastic-plastic
regime is characterized by a critical value of J. Refer to
Reference [1] for the detailed
background of the J-integral.
FIGURE 8. Typical finite element model of a longitudinal crack
in a seam weld. The model is
symmetric, meaning that the above picture represents of the
pipe, with sections taken longitudinally
through the seam weld and circumferentially at the mid-point of
the crack.
-
While 3D finite element analysis is the most accurate method to
predict burst
conditions in pipes that contain cracks, it is not practical for
widespread use. Finite
element analysis requires special software and expertise which
may not be easily obtained
by a pipeline operator for standardized assessment. Recently,
PRCI undertook a research
program that addressed this concern. Southwest Research
Institute, who was the contractor
for the PRCI project, performed a large number of 3D
elastic-plastic finite element
analyses of pipes with seam weld cracks. They curve fit these
results and produced a series
of equations and tables that are presented in the final report
[5].
Using Charpy Data in a Crack Assessment
Traditionally, the pipeline industry has relied on Charpy
testing to characterize toughness.
However, fracture mechanics testing that quantifies toughness as
a critical stress intensity
factor (Kc) or critical J-integral provides a more direct
measurement. Figure 9 compares
the Charpy test specimen geometry with a typical fracture
mechanics specimen. Both were
notched on the ERW bond line. With the fracture mechanics
specimen, however, a fatigue
crack is introduced on the bond line. As a result, the specimen
shown in Fig. 9(b) is more
representative of a seam weld crack. Stated another way, a J
test provides a direct measure
of fracture toughness, while the Charpy test is an indirect
measurement, analogous to
ultrasonic wall thickness (UT) versus magnetic flux leakage
(MFL) inline inspection
technologies.
It is possible to use Charpy data in an advanced crack
assessment that applies either
finite element analysis or the PRCI method. The trade-off is
that the correlation between
Charpy energy and the critical J-integral exhibits scatter and
uncertainty, as discussed
below.
Wallin [6] has correlated Charpy energy with critical J-integral
data by evaluating
over 1000 data sets that encompass a range of steels. While
there is a reasonable
correlation between the two tests, it is not perfect. Equations
(6a) to (6c) give the Wallin
relationship, in US customary units, for the mean correlation,
as well as the 5% lower
bound and 95% upper bound. The original Wallin correlation is
based on full-size Charpy
specimens, but the expressions below include an adjustment for
subsize specimens.
1.282
1 mm
0.124 in6.248 Mean Correlation
c
J CVNA
(6a)
1.282
1 mm
0.124 in4.482 5% Lower Bound
c
J CVNA
(6b)
1.282
1 mm
0.124 in8.013 95% Upper Bound
c
J CVNA
(6c)
-
(a). Charpy V-notch specimen.
(b) Fracture mechanics specimen.
FIGURE 9. Laboratory specimens for measuring toughness at an ERW
seam.
ERW Seam
Fatigue
Crack
-
Where J1 mm is the critical J-integral at 1 mm (0.039-in) of
crack growth (ductile tearing).
Despite the large dataset Wallin had at his disposal when
developing his correlation,
it is unlikely that the data included tests on ERW pipe. One
would expect the base material
(typically an API 5L steel) to fit the correlation because the
Wallin dataset undoubtedly
included steels with similar chemistry and microstructures. The
ERW bond line and heat-
affected zone are another matter, however. One concern is that
the Charpy notch is wider
than the bond line, and might not give a true reflection of the
toughness.
Recently, the present author conducted Charpy and J testing on
two different
samples of ERW pipe. Both Charpy and fracture mechanics
specimens were notched in 3
locations: the fusion line, 1 mm from the fusion line, and in
the base metal. The results are
plotted in Fig. 10, along with the predictions from the Wallin
correlation (Eq. (6)). Each
data point corresponds to the average of 3 Charpy tests, as well
as a single J test. Five of
six points fall within the 90% confidence band, but there is one
outlier for tests on the bond
line. Further testing on additional ERW seams will be necessary
to confirm whether or not
the Wallin correlation applies to the bond line, but the results
in Fig. 10 are encouraging.
FIGURE 10. Correlation between Charpy data and critical
J-integral values for two samples of ERW pipe.
The curves correspond to predictions from the Wallin correlation
(Eq. (6)).
-
Even if the Wallin correlation applies to ERW seams, using
Charpy data to infer
toughness comes at a price. Namely, the uncertainty band in
critical J values translates to
an uncertainty in predicted burst pressure. Figure 11 shows an
example of burst pressure
predictions from Charpy values. The Wallin correlation was used
in conjunction with the
PRCI crack model described earlier. In this particular case, the
90% confidence band in the
Charpy-J correlation corresponds to approximately a 200 psi
uncertainty band in failure
pressure.
Figure 12 is a repeat of Fig. 11, but predictions from the
original and modified ln-
sec models are overlaid on the graph. The original ln-sec model
is biased toward the
conservative side, which is consistent with the burst test
comparison in Figs. 5 to 7. The
burst pressure prediction from the modified model is insensitive
to toughness, so it is
unconservative at low toughness values and falls within the
predicted scatter band (based
on the PRCI J-integral model) for moderate toughness levels.
FIGURE 11. Effect of Charpy energy on burst pressure, as
predicted from the PRCI J-integral model,
combined with the Wallin Charpy correlation.
-
FIGURE 12. Same as Fig. 11, but with the ln-sec models overlaid
for comparison.
AUTOMATED PRESSURE CYCLE FATIGUE ASSESSMENT
Seam welded pipelines that are in cyclic service can experience
fatigue failure if not
properly managed. Planar flaws that are introduced at
manufacture can grow over time due
to pressure cycling. Eventually, a growing crack will lead to a
leak or rupture if it is not
remediated.
Pressure cycle fatigue analysis (PCFA) is a technique that has
been used by the
pipeline industry to manage the risk associated with seam weld
flaws that may grow in
service. Pressure data are typically collected at pumping
stations (in liquid lines) and
stored in a PI data historian or similar system. Periodically,
pressure readings at discrete
time intervals are exported to a csv file or spreadsheet. These
data are processed through a
rainflow cycle counting algorithm, which quantifies the number
and magnitude of pressure
cycles in the form of a histogram. The histogram is then input
into a fracture mechanics
model to predict the growth of actual or postulated flaws in the
pipeline. The PCFA is used
to make decisions on the retest interval or re-inspection
interval in cases where the integrity
management plan calls for hydrostatic testing or ILI,
respectively.
The PCFA process is fairly time consuming and labor intensive.
In a typical case, a
pipeline operator sends pressure data to a consultant, who then
submits a report to the
-
operator 2 or 3 months later. A PCFA is usually performed
annually because more
frequent intervals are not practical.
Quest Integrity is has developed a software system,
PACIFICATM
, for automatically
performing PCFA. Figure 13 illustrates the system architecture.
At initial setup for a given
pipeline, the user enters basic data, such as pumping station
locations, pipe dimensions,
elevations, and material properties. Once the system is online,
it periodically imports
pressure data from the PI data historian, and then processes it
through the rain-flow and
fracture mechanics algorithms. Both the pressure data input and
the processed output are
stored in a database. Reports are generated at regular intervals
based on user settings.
Because the system is automated, it is possible to obtain
virtually real-time updates on
pressure cycling. For example, an operator may choose to
generate PCFA reports on a
weekly or monthly basis. It is also possible to track the growth
of thousands of flaws in
multiple pipelines.
FIGURE 13. Overall system architecture of PACIFICATM
.
-
CONCLUSIONS
1. The original ln-sec failure model, which was developed in the
early 70s, contains
fundamental errors when compared to modern fracture mechanics
theory. As a
result, significant errors in burst pressure predictions can
result. Based on a
comparison with burst test data, the model tends to be
conservative, but is
particularly conservative for long and shallow flaws.
2. The modified ln-sec model was published in 2008, and was
intended to address the
problems in the original equation. Although the modified model
reduces the level
conservatism for long and shallow flaws, this fix comes at a
high price. Namely,
the model is highly unconservative for low-toughness materials.
The model
predicts that burst pressure is insensitive to toughness, which
is simply incorrect.
3. The most accurate predictions of failure of pipes with
longitudinal cracks can be
made with 3D finite element analysis, but this requires special
software and
expertise. A new PRCI model overcomes this difficulty by
providing parametric
equations that are a curve fit of finite element solutions.
4. The J-integral test provides the most direct measurement of
fracture toughness.
Inferring toughness from Charpy testing is acceptable, but will
result in a greater
uncertainty in the burst pressure prediction.
5. A new software system has been developed for pressure cycle
fatigue analysis
(PCFA). This system assists in the integrity management of seam
welded pipelines
in cyclic service.
REFERENCES
1. Anderson, T.L. Fracture Mechanics: Fundamentals and
Applications. Third
Edition, Taylor & Francis, Boca Raton, Florida, USA,
2005
2. Kiefner, J. F., Maxey, W. A., Eiber, R. J., and Duffy, A. R.,
Failure Stress Levels
of Flaws in Pressurized Cylinders. ASTM STP 536, American
Society for Testing
and Materials, 1973.
3. Kiefner, J.F., Modified Equation Helps Integrity Management.,
Oil and Gas
Journal, October 6, 2008, pp. 64-66.
4. API 579-1/ASME FFS-1, Fitness-for-Service, jointly published
by the American
Petroleum Institute and the American Society for Mechanical
Engineers, June 2007.
5. Chell, G.G., Criteria for Evaluating Failure Susceptibility
due to Axial Cracks in
Pressurized Line Pipe. PRCI Project MAT-8 Final Report, December
2008.
6. Wallin, K., Fracture Toughness of Engineering Materials:
Estimation and
Application. EMAS Publishing, Birchwood Park, Warrington, UK,
2011.