Evaluating Fitness-for-Service of Corroded Metal

Pipelines: Structural Reliability Bases

Naiyu Wang, Ph.D., A.M.ASCE 1; and Mehdi S. Zarghamee, Ph.D., P.E., F.ASCE 2

Abstract: This paper presents a methodology for evaluating fitness-for-service (i.e., safety, serviceability, and reliability) of corroded metal
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.

pipelines. A model for predicting pipe failure pressure is developed from a series of nonlinear finite-element analyses of metal pipes of
different diameters and wall thicknesses with corrosion defects of different sizes and depths. The strength limit state (burst) and service
limit states (yield and leakage) both are investigated in the pipe fitness-for-service analyses. Uncertainties associated with in situ measurement
of flaw size, pipe geometry, pipe material properties, operating pressures, and corrosion in exposure environments are considered in a time-
dependent reliability analysis of pipe performance over a service life of 50 years. Engineering variables that have the most significant impact
on the reliability of a corroded pipe are identified through Monte Carlo–based sensitivity studies. Finally, a set of simple reliability-based
evaluation tools are presented for assessing pipeline safety. The procedure and the results presented in this paper can be used to support
decisions related to continued safe operation of pipes, optimal pipe inspection intervals, and maintenance schedules. DOI: 10.1061/(ASCE)
PS.1949-1204.0000148. © 2013 American Society of Civil Engineers.
Author keywords: Civil infrastructure; Corrosion; Fitness-for-service; Pipelines; Probability; Reliability; Structural engineering.

Introduction various uncertainties that may impact the performance of a pipeline

In-service inspections of existing metal pipelines have indicated that
the most common causes for pipeline failure are mechanical damage
and corrosion (Cosham and Hopkins 2004; Teixeira et al. 2008; Xu
and Cheng 2011). Following the detection of corrosion in a pipeline,
maintenance actions, in the forms of repair, replacement, or reduc-
tion of the working pressure, usually cannot occur immediately and
often are deferred due to pipeline operational requirements and
financial constraints. Without a proper guideline for maintenance,
pipelines could be at high risk of failure, and the consequences
of pipeline failure, including economic, political, and social costs,
could be unacceptable. Consequently, there is a need to better under-
stand whether a pipe with a limited amount of localized corrosion is
still fit for service (i.e., can resist applied pressure loads safely) and if
it is “safe” at the present time (i.e., what its remaining service life
before major maintenance action must be taken).
Existing predictions of retained strength of corroded steel pipes
are usually performed in a deterministic manner using allowable
stress methods (ASME 2009; Kiefner and Vieth 1989, 1990; DNV
2004). Even though the allowable stress method with a sufficient
safety factor works well in designing a new pipeline, it may not
result in optimal pipe maintenance strategies because it cannot
properly account for the numerous uncertainties associated with
performance of existing metal pipes in their unique exposure
environments. Variations in pipe geometry, mechanical properties,
exposure conditions, and time/exposure-dependent corrosion prop-
agation can cause major deviations of pipe behavior from what
might be predicted by a deterministic evaluation. In the face of the
1
Structural Engineer II, Simpson Gumpertz and Heger, 41 Seyon St.,
Waltham, MA 02453 (corresponding author). E-mail: NWang@sgh.com
2
Senior Principal, Simpson Gumpertz and Heger, 41 Seyon St.,
Waltham, MA 02453. E-mail: MSZarghamee@sgh.com
Note. This manuscript was submitted on September 16, 2012; approved
on May 10, 2013; published online on September 12, 2013. Discussion per-
iod open until February 12, 2014; separate discussions must be submitted for
individual papers. This paper is part of the Journal of Pipeline Systems En-
gineering and Practice, © ASCE, ISSN 1949-1190/04013012(9)/$25.00.
susceptible to active corrosion, the evaluation of pipe safety and
remaining service life of a corroded pipe as well as decisions
regarding pipeline maintenance schedules should be based on
modern structural reliability principles. Recent probabilistic studies
on pipeline safety have focused either on investigating the dif-
ferences between various available methods for assessing pipe
reliability (e.g., first order second moment method, Monte Carlo
simulation, or Taylor series approximations) or on making compar-
isons of pipe reliabilities computed using different burst pressure
prediction models (Caleyo et al. 2002; Hasan et al. 2011; Anghel
2009). An integrated approach to pipeline safety assessment with
practical reliability-based engineering evaluation tools has yet to be
developed.
This paper is aimed at providing practical tools for engineers
to evaluate pipe safety based on structural reliability principles.
First, a methodology to determine the strength of corroded pres-
sure pipes is presented. The impact of corrosion defects of various
geometries on pipe failure pressure is investigated using nonlinear
finite-element (FE) analysis, leading to a simple analytical ex-
pression for estimating retained pipe strength. A series of time-
dependent reliability analyses of corroded pipelines considering
both strength (burst) and service (yield and leakage) limit states
is then performed. Uncertainties associated with in situ measure-
ment of flaw size, pipe geometry, pipe material properties, oper-
ating pressures, and exposure environments are included in the
probabilistic analyses. Variables that have the most significant
impact on the reliability of a corroded pipe are identified through
Monte Carlo-based sensitivity studies. Finally, a set of simple and
practical reliability-based evaluation tools are presented for pipe
safety assessment.
Burst Pressure Prediction Model of Corroded Steel
Pipes
Among the widely accepted methods for assessing the remaining
strength of steel pipes with localized defects are those contained

© ASCE 04013012-1 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract. 2014.5.

 in ASME Manual B31G (ASME 2009) and RSTRENG (Kiefner
and Vieth 1989, 1990). These methods have been developed from
experimental tests using steel pipes with low-strength but high-
toughness properties. Studies have shown that these methods can
be overly conservative when applied to certain grades of pipeline
material and corrosion environments and may lead to unnecessary
repair and replacement of in-service pipes (Batte et al. 1997;
Teixeira et al. 2008; Chauhan, et al. 2009; Xu and Cheng 2011).
Netto et al. (2005) proposed a relatively simple formula for predict-
ing the burst pressure of corroded pipes, but the formula was only
based on a study of limited pipe geometry parameters and defect
configurations. The applicability of this formula to a wide range of
pipe dimensions and defect sizes has not been established.
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.
In the remainder of this section, a nonlinear FE model of the
low-carbon steel pipe with corrosion defect that was tested by Netto
et al. (2005) is developed. After validating this FE model against
Netto et al.’s test results, the validated model is used to perform a
parametric study of the retained strength of different carbon steel
pipes with different corrosion defect geometries. Finally, a general
expression for retained strength of a corroded pipe with either
external or internal corrosion defect is developed.
Finite-Element Analysis
In the present study, a localized corrosion defect was assumed to
have a rectangular projected surface area, and the defect penetration
was assumed to have elliptical profiles in both longitudinal and
circumferential directions, as shown in Fig. 1. Nonlinear FE models
of corroded pipe segments were developed using ABAQUS version
6.10. The pipe, including the corroded region, was idealized with
three-dimensional, eight-node linear brick elements (designated
C3D8 in ABAQUS), with three translational degrees of freedom at
each node. Due to the symmetric nature of the pipe cross section
and loading conditions with respect to planes X-Z and X-Y in Fig. 1,
only one-quarter of the actual pipe was modeled, as shown in Fig. 2.
The overall length of the pipe section was selected at ten times
its outside diameter to eliminate the effect of the boundary condi-
tions from the stresses in the corrosion area. To trace the pipe
behavior over the entire loading range and to replicate experimental
observations, this FE model included material nonlinearity due to
plasticity of the metal and geometric nonlinearity due to large de-
formations. The true stress-true strain curve of the AISI 1020 mild
steel used in the FE analysis was constructed from a mean (average)
experimental engineering stress-strain relation (Ling 1996). The
Fig. 1. Schematic view of corroded pipe with a single corrosion defect
© ASCE 04013012-2
J. Pipeline Syst. Eng. Pract. 2014.5.
steel has an elastic modulus of approximately 208,000 MPa
(30,200 ksi), 0.2% strain offset yield stress of about 262 MPa
(38 ksi) and a tensile strength of 393 MPa (57 ksi).
To validate the FE modeling approach summarized earlier, one
of the defect configurations studied by Netto et al. (2005) was an-
alyzed. The specimen was a steel pipe with an outside diameter of
42 mm (1.65 in.) and wall thickness of 2.7 mm (0.108 in.), with a
machined elliptical defect 2.1 mm (0.084 in.) in depth, 42 mm
(1.65 in.) in length in the pipe longitudinal direction, and 13 mm
(0.51 in.) in width in pipe circumferential direction. In the FE
analysis herein, internal pressure was applied gradually until the
burst strength of the pipe was reached. A significant amount of
plastic deformation and localized necking were observed prior to
burst, which is consistent with experimental studies reported by
other researchers (Cronin and Pick 2000; Netto et al. 2005). The
modeling approach predicted a burst pressure of 24.1 MPa
(3,492 psi), defined conservatively as the pressure that produces
20% plastic strain in the necking area (Manjoine 1976), which
is about 90% of the burst pressure of 26.8 MPa (3,881 psi)
measured in the laboratory test (Netto et al. 2005). This difference
can be attributed to the definition of the failure criterion in terms
of plastic strain, the assumptions made in the FE modeling, the
idealization of the defect shape (Fig. 2), and various uncertainties
associated with experimental testing. The FE model approach is
believed to be conservative and is adequate for predicting the burst
pressure of corroded pipe.
Parametric Study
Following validation, the FE modeling approach was used in a
parametric study to investigate the effects of pipe diameter and wall
thickness and corrosion defect configurations on the burst pressure
of corroded pipes. The burst pressures of one hundred pipes with
pipe diameters, D, ranging from 51 to 2,286 mm (2 to 90 in.), pipe
wall thicknesses, t, ranging from 3.42 to 12.7 mm (0.135 to 0.5 in.),
ratios of defect depth to wall thickness, d=t, ranging from 2.54 to
Fig. 2. Finite-element model of pipe and a single corrosion defect
J. Pipeline Syst. Eng. Pract.
 Table 1. Pipe and Defect Geometries Used in Parametric Study
D t
Pipe ID mm (in.) mm (in.) d=t l=D c=D
Pipe 1 610 (24) 3.4 (0.13) 0.5; 0.85 0.14; 0.28; 0.56 0.14
Pipe 2 610 (24) 6.4 (0.25) 0.5; 0.85 0.26; 0.52; 1.04 0.14
Pipe 3 1,524 (60) 6.4 (0.25) 0.5; 0.85 0.10; 0.21; 0.42 0.09; 0.11; 0.14; 0.18
Pipe 4 1,524 (60) 12.7 (0.50) 0.5; 0.85 0.21; 0.42; 0.83 0.14
Pipe 5 2,286 (90) 9.5 (0.38) 0.1; 0.2; 0.3; 0.4; 0.5; 0.6; 0.7; 0.85 0.21; 0.42; 0.5; 0.62 0.14
Pipe 6 2,286 (90) 12.7 (0.50) 0.1; 0.2; 0.3; 0.4; 0.5; 0.6; 0.7; 0.85 0.21; 0.3; 0.42; 0.62 0.16
Pipe 7 406 (16) 12.7 (0.50) 0.1; 0.2; 0.3; 0.4; 0.5; 0.6; 0.8 0.5; 1.0; 1.5; 2.0 0.14
Pipe 8 51 (2) 3.9 (0.15) 0.5; 0.85 0.14; 0.28; 0.56 0.14
Pipe 9 102 (4) 6.0 (0.24) 0.5; 0.85 0.26; 0.52; 1.04 0.14
Pipe 10 152 (6) 7.1 (0.28) 0.5; 0.85 0.10; 0.21; 0.42 0.14
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.

Pipe 11 203 (8) 8.2 (0.32) 0.5; 0.85 0.21; 0.42; 0.83 0.14
Pipe 12 254 (10) 9.3 (0.37) 0.5; 0.85 0.21; 0.42; 0.62 0.14

21.59 mm (0.1 to 0.85 in.), ratios of defect length to pipe diameter,
l=D, ranging from 2.54 to 26.42 mm (0.1 to 1.04 in.), and ratios of
defect width to pipe diameter, c=D, ranging from 2.29 to 4.57 mm
(0.09 to 0.18 in.), as summarized in Table 1, were computed using
FE analysis.
As shown in Fig. 3, the parametric study revealed that d=t has
the strongest influence on the burst pressure retention ratio, defined
as the ratio of degraded strength, Pb , to the original uncorroded
pipe strength, Pbi ; furthermore, the burst pressure decreases as
l=D increases. Such observations are consistent with those made
by Netto et al. (2005). Furthermore, the ratio of pipe wall thickness
to outside diameter, t=D, was also found to have only a slight im-
pact on the retained burst pressure, and the ratio c=D had almost
negligible effect. Netto et al. (2005) proposed the burst pressure
retention ratio can be expressed as a function of ðd=tÞa ðl=DÞb
based on the study of a 406.4 mm (16 in.) diameter pipe. In the
section entitled “Retained Strength of Corroded Steel Pipe,” this
expression will be extended to pipes of different diameters and wall
thicknesses with a wide range of defect geometries.
Retained Strength of Corroded Steel Pipe
The burst pressure retention, Pb =Pbi , versus the “defect shape
factor” proposed by Netto et al. (2005) in the form of ðd=tÞa ðl=DÞb
are plotted for all cases analyzed in Figs. 4(a and b) for pipes with
D < 610 mm (24 in.) and for pipes with D ≥ 610 mm (24 in.), re-
spectively. The best fitting curve from a regression analysis of burst
strength retention ratios as a function of defect shape factor is
(
Pb 1 − 0.886ðdtÞ1.00 ðDl Þ0.30 ; for D < 610 mm
¼ ð1Þ
Pbi 1 − 1.120ðdÞ1.15 ð l Þ0.30 ; for D ≥ 610 mm
t D
where Pbi ¼ 1.1 · σy 2t=D as defined in ASME B31G (ASME
2009) and σy = the nominal yield stress of the steel in the pipe.
Note that the assumption of the ultimate strength σu ¼ 1.1 · σy em-
bedded in the expression of Pbi may not be appropriate for other
steels and other metals. The coefficient of the defect shape factor
and the two exponents in Eq. (1) are 0.886, 1.00, and 0.30 for pipes
with D < 610 mm, comparing to 0.943, 1.6, and 0.4, respectively,
recommended by Netto et al. (2005). The coefficient of determina-
tion, R2 , in these figures is a measure of the degree of linearity of
the fit, a value of 1.0 indicating perfect linearity.
To determine whether Eq. (1) might be used to predict the burst
pressure of pipes other than those considered in this study, 48 pipes
tested to burst by Batte et al. (1997), Cronin and Pick (2000), and
Chauhan et al. (2009) were analyzed. Each pipe included a natural
or manufactured corrosion defect with known geometric and
material properties. Fig. 5 compares the burst pressure predicted
by Eq. (1) and the experimental test results for the 48 pipes. Fig. 6
shows a comparison between burst pressures predicted by Eq. (1)
and those predicted by ASME B31G (ASME 2009). The ratio of
measured burst pressure to the prediction from Eq. (1) has a mean
of 1.14 and standard deviation of 0.17, whereas the ratio of
measured burst pressure to ASME B31G prediction has a mean
of 1.48 and standard deviation of 0.48. Even though both analytical
methods provide conservative predictions of the experimental data,
those from Eq. (1) are much closer to the test results, with signifi-
cantly smaller scatter than the predictions based on the ASME
B31G method.

Internal Corrosion Defect versus External Corrosion

Defect
Corrosion may occur on either the inside or the outside of a pipe.
To investigate whether the location of a defect has a significant
Fig. 3. Effect of corrosion defect geometry on retained strength of
impact on the burst pressure, both internal and external corrosion
corroded pipe
defects were modeled for the 16 pipes listed in Table 2. These pipes

© ASCE 04013012-3 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract. 2014.5.

Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 4. Burst pressure retention of corroded steel pipe, Pb =Pbi , versus defect shape factor for (a) small pipes with D < 24 in: (610 mm); (b) large pipes
with D ≥ 24 in: (610 mm)
were selected to cover a wide range of pipe designs and defect
geometries. For each of these pipe configurations, two FE models
were made, one with the defect located inside the pipe, and the
other with the same defect located outside the pipe. Burst pressures
for both cases are tabulated in Table 2. The ratio of burst pressure of
the two counterpart cases presented in the last column of Table 2
has a mean of 0.995 and a standard deviation of 2.5%. Therefore, it
may be concluded that the burst pressure of the pipe is essentially
the same, regardless of whether the defect is located inside or
outside of the pipe.
Corrosion defects may occur in clusters. Each corrosion defect
introduces disturbances in the pipeline stress and strain fields that
Fig. 5. Comparison between burst pressure measured in experimental
tests and the burst pressure predicted by Eq. (1) (1 psi ¼ 6.9 kPa)
© ASCE 04013012-4
J. Pipeline Syst. Eng. Pract. 2014.5.
spread out beyond the border of the defect. When adjacent defects
are considered to interact, the burst pressure with a cluster of in-
teractive defects can still be computed using Eq. (1) with the overall
dimensions of the combined defect region (dclus , lclus ), where the
cluster depth, dclus , is equal to the depth of the deepest defect of all
defects included in the interactive group, and the cluster length,
lclus , is the overall length of the cluster in pipe longitudinal
direction. This method will result in a conservative estimate of burst
pressure.
Fig. 6. Comparison of burst pressure of tested corroded steel pipe with
predicted pressure using Eq. (1) and with the pressure predicted using
ASME B31G (1 psi ¼ 6.9 kPa)
J. Pipeline Syst. Eng. Pract.
 Table 2. Comparison of Retained Strength of Pipes with Corrosion Defects Located on the Inside or on the Outside of the Pipe
D t Burst pressure
Internal defect External defect
External/internal
Pipe ID mm (in.) mm (in.) d=t l=D kPa (psi) kPa (psi) ratio
Pipe 1 610 (24) 3.4 (0.13) 0.50 0.28 3,443 (499) 3,409 (494) 0.99
610 (24) 3.4 (0.13) 0.85 0.28 1,601 (232) 1,580 (229) 0.99
Pipe 2 610 (24) 6.4 (0.25) 0.50 0.26 6,576 (953) 6,562 (951) 1.00
610 (24) 6.4 (0.25) 0.85 0.26 3,554 (515) 3,450 (500) 0.97
Pipe 3 1,524 (60) 6.4 (0.25) 0.50 0.21 2,767 (401) 2,808 (407) 1.01
1,524 (60) 6.4 (0.25) 0.85 0.21 1,366 (198) 1,463 (212) 1.07
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.

Pipe 4 1,524 (60) 12.7 (0.50) 0.50 0.21 5,630 (816) 5,617 (814) 1.00
1,524 (60) 12.7 (0.50) 0.85 0.21 3,250 (471) 3,160 (458) 0.97
Pipe 5 2,286 (90) 9.5 (0.38) 0.50 0.21 2,822 (409) 2,822 (409) 1.00
2,286 (90) 9.5 (0.38) 0.85 0.21 1,525 (221) 1,504 (218) 0.99
Pipe 6 2,286 (90) 12.7 (0.50) 0.50 0.21 3,767 (546) 3,754 (544) 1.00
2,286 (90) 12.7 (0.50) 0.85 0.21 2,111 (306) 2,029 (294) 0.96
Pipe 9 102 (4) 6.0 (0.24) 0.50 0.18 40,600 (5,884) 39,958 (5,791) 0.98
102 (4) 6.0 (0.24) 0.85 0.18 31,050 (4,500) 31,788 (4,607) 1.02
Pipe 12 254 (10) 9.3 (0.37) 0.50 0.18 25,323 (3,670) 24,937 (3,614) 0.98
254 (10) 9.3 (0.37) 0.85 0.18 18,747 (2,717) 18,513 (2,683) 0.99

Reliability of Corroded Pipes the statistics of k and α, as summarized in Table 3. The study
also indicated that k and α can be best described by normal
Corrosion Model for Underground Pipes distributions.

The growth of a corrosion defect depends primarily on the character-

istics of the pipeline material and on the surrounding corrosion Reliability Analysis
environment. As corrosion of a pipe continues with time, the defect Pipe fitness-for-service evaluation requires prediction of future
length and depth, which are the critical parameters that influence pipe performance of corroded pipes. Variation in mechanical properties,
safety, continue to grow. Defect growth in low-carbon steel is com- soil conditions, and depths and sizes of corrosion defects can
monly modeled with a power law function that relates the average cause significant deviations of the actual in situ pipe performance
maximum defect depth, d, to the exposure time (Romanoff 1957): from what might be predicted deterministically. In the face of nu-
merous uncertainties that might have an impact on the performance
dðtÞ ¼ kðt − T 0 Þα ð2Þ of a pipe that is susceptible to active corrosion, probability-based
structural reliability principles provide a powerful means for inte-
in which k and α = the growth constants and T 0 = the time required to grating these uncertainties into risk-informed decision-making
initiate corrosion. In most corrosion studies, k and α are assumed to tools for evaluating pipeline safety, future service, and maintenance
be constant, with α ranging from 0.3 to 1.0 (Zapp 1996). schedules.
In a study recently reported by Caleyo et al. (2009), data In structural reliability analysis, a limit state function describing
describing maximum defect depth, local soil conditions, and pipe the state of performance is defined as
geometrical parameters were collected over a 3-year period from
250 excavated pipeline sites located across southern Mexico. The GðXÞ ¼ GðX 1 ; X 2 ; : : : ; X m Þ ¼ 0 ð3Þ
pipelines studied had been in service for up to 50 years. The
measured soil variables included resistivity, pH, water content, in which X ¼ ðX 1 ; X 2 ; : : : ; X m Þ = vector of random variables
redox potential, bulk density, and dissolved chloride, bicarbonate, (resistance, corrosion power law, and load), and G = a function that
and sulfate ion concentrations. The pipe-to-soil potential, pipe depicts “failure” when GðXÞ < 0. The limit state functions for pipe
coating type, and pipeline age were also included within each burst failure mode and leakage failure mode are further defined in
dataset. Based on the relative proportions of sand, silt, and clay,
the soil samples were grouped into four classes: clay (C), clay
Table 3. Dependence of Corrosion Parameters on Soil Conditions
loam (CL), sandy clay loam (SCL), and a generic class contain-
ing all collected samples (All). For each variable considered, the Soil category
parameters of the probability distribution fitting the observed data Clay Sandy Average
were determined using the maximum likelihood estimation Parameter Clay loam clay loam type
(MLE) method, and the best probabilistic model was selected ka Mean 0.0092 0.0136 0.0059 0.0087
based on the value of the Kolmogorov-Smirnov (K-S) test STD 0.0035 0.0059 0.0024 0.0033
statistic for the probability distributions tested. Corrosion growth
coefficients k and α in Eq. (2) both depend on the aforemen- α Mean 0.788 0.633 0.631 0.662
tioned soil parameters. MC simulations conducted in the current STD 0.112 0.147 0.106 0.118
study using statistics reported in Caleyo et al. (2009) produced a
k is given in in:=yra ¼ 25.4 mm=yra .

© ASCE 04013012-5 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract. 2014.5.

 Eqs. (6) and (7), respectively, in the subsequent sections. The fail-
ure probability becomes
Z Ps = the pressure load demand, which is a random variable that
Pf ¼ f X ðx1 ; x2 ; : : : ; xm Þdx1 ; dx2 ; : : : ; dxm ð4Þ
Ω
in which f X ðxÞ = joint probability density function of X and the
domain of integration, Ω, is that region of X where GðXÞ < 0.
For well-behaved limit state functions (i.e., those not involving bi-
furcation of equilibrium or large material nonlinearities), the limit
state probability, Pf , can be approximated by Pf ¼ Φð−βÞ, in
which Φð·Þ = standard normal probability distribution (Melchers
1999), and β is the reliability index that is often used as an equiv-
alent measure of structural safety. Because of the complexity of the
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.
where Pf = pipe failure probability, P½·
= the probability of the
event in the brackets, Pbi = burst pressure of the uncorroded pipe,
describes the extreme pressure loads a pipe is likely to experience
during its service lifetime, f R ¼ Pb =Pbi is the burst strength reten-
tion ratio of the corroded pipe expressed by Eq. (1). The corrosion
model expressed in Eq. (2) is used to determine the growth of
the defect depth and length over the time. The statistics in Tables 3
and 4 are used in the MC simulation to estimate pipe reliability.
Analyses were performed for sixteen 610 mm (24 in.) diameter
pipes with different combinations of wall thickness ranging from
2.5 to 13 mm (0.1 to 0.5 in.) and l=D ratio ranging from 0.1 to 0.8,

with a soil environment corresponds to the average soil type (All

limit states considered herein, Monte Carlo (MC) simulation was
used to determine Pf (or β).
The statistical properties of the random variables related to the
corrosion model as discussed in “Corrosion Model for Under-
ground Pipes” are summarized in Table 3, and those of the other
random variables used in the reliability analysis related to load,
material property, and dimension are collected from the literature
(Ahammed and Melchers 1996; Ahammed 1998) and are summa-
rized in Table 4. Parameter B is the modeling error associated with
Eq. (1), and its statistics are obtained from Figs. 4(a and b), depend-
ing on pipe diameter.
The MC simulation revealed that when random variables k and
α as listed in Table 3 are incorporated in Eq. (2), both the mean and
the standard deviation of the defect depth increases with time; how-
ever, the standard deviation increases much faster than the mean,
causing the coefficient of variation (COV) of the defect depth to
increase with time as well. A suitable probabilistic model for defect
depth is found to be the generalized extreme value distribution
(Leadbetter et al. 1983):

    defect shape factor alone. Similar analyses were performed for
x − μ −1=ξ
Fðx; u; ξÞ ¼ exp − 1 þ ξ ð5Þ
u iceability checking. The yield limit state occurs when the operating
where x = state variable, ξ = shape factor, μ = location factor, and
u = scale factor. These parameters can be determined through MLE
using samples of defect depth obtained from MC simulation, and
their values are dependent on the time as well as on the soil expo-
sure conditions.
in Table 3). For each pipe studied, reliabilities are computed for
a 50-year service period. Fig. 7 shows the effect of pipe wall
thickness on time-dependent pipe reliability index, β. Clearly, pipe
wall thickness has a strong effect on pipe reliability against burst,
and pipes with thin walls are especially susceptible to corrosion-
induced rupture failure. Fig. 8 shows the impact of defect length,
l=D, on pipe reliability when other parameters are held constant.
The pipe reliability index decreases as the ratio l=D increase from
0.1 to 0.8. A comparison of Figs. 7 and 8 reveals that, although both
the pipe wall thickness and the l=D ratio have an obvious impact
on the pipe reliability, the impact of the pipe wall thickness is much
stronger.
To further investigate the combined effect of (d=t) and (l=D),
which are the two basic components of the defect shape factor,
the time-dependent reliability indices for all 16 pipes analyzed are
plotted against the defect shape factor in Fig. 9. The dashed line is a
polynomial best-fit curve for all cases considered, indicating that
the pipe reliability can be nicely expressed as a function of the
other soil environments as well as for the yield limit state for serv-
pressure causes the onset of yielding in the pipe in the defect
region. The resulting reliability indices, β, are plotted in Fig. 10(a)

Burst and Yield

Both strength (burst) and service (yield) limit states are considered
in this study. The burst limit state has already been defined in the
section “Parametric Study.” The failure probability of a pipe by
burst can be estimated using the following equation:

Pf ¼ P½B · Pbi fR < Ps
ð6Þ

Table 4. Random Variables and Their Parameters

Random variable Distribution characteristics
Symbol Description Type Bias COV
D Pipe diameter Normal 1.0 0.03
l Defect length Normal 1.0 0.10
t Pipe wall thickness Normal 1.0 0.05
Ps Maximum operating pressure Normal 0.9 0.10
σy Steel yield stress lognormal 1.10 0.08
k Proportionality factor Normal Table 3 Table 3
α Exponent factor Normal Table 3 Table 3 Fig. 7. Effect of pipe wall thickness on the time-dependent burst
B Modeling error of the Eq. (1); Normal 1.10 0.05 failure probability of corroded pipes (D ¼ 24 in:; l=D ¼ 0.2).
obtained from Fig. 5 (1 in: ¼ 25.4 mm)

© ASCE 04013012-6 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract. 2014.5.


Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.
Fig. 8. Effect of the ratio of defect longitudinal length to pipe diameter
on the time-dependent burst failure probability of defected pipes
(D ¼ 24 in:, t ¼ 0.3 in:) (1 in: ¼ 25.4 mm)
as functions of defect shape factor for both yielding and burst limit
states, for small pipes (D < 610 mm) and all four soil conditions
considered in this study. Similar results are obtained for large pipes
(D > 610 mm), as shown in Fig. 10(b).
Figs. 10(a and b) can be used to evaluate pipe reliability against
burst or yielding based on the estimated defect shape factor. As
indicated in these figures, a corrosion-free pipe designed by ASME
Standard B31 (2009) has a reliability index of about 3.3 ∼ 3.5
against yielding; for the burst limit state, which is associated with
more severe failure consequences, the initial reliability index is
Fig. 9. Pipe reliability index as a function of calculated defect shape
factor (1 in: ¼ 25.4 mm)
© ASCE 04013012-7
J. Pipeline Syst. Eng. Pract. 2014.5.
greater than 4. However, as soon as corrosion propagates, pipe
reliabilities start to decrease rapidly.
As reflected in Figs. 10(a and b), for a prescribed target reliabil-
ity level, β T , with respect to either the burst or the yielding limit
state, the defect shape factor for a given pipe has to be within cer-
tain limiting values to ensure that the safety requirement is met.
Steel pipe design is currently based on the allowable stress method,
and the target reliability index is not stipulated in the design codes
or their supporting documents. Typically, target reliability index
values in the 2.5–4.5 range have been used in reliability-based
design and evaluation of building and bridge structures with service
lives on the order of 50 years (Ellingwood 2000). Accordingly,
a target reliability index of 2.5 was adopted for existing pipes
for both burst and yielding limit states to show how practical
reliability-based evaluation criteria can be developed. To ensure the
safety requirement imposed by the target reliability of 2.5, the de-
fect shape factor must be within a certain range and subsequently
the two components of the defect shape factor, (d=t) and (l=D),
must be consistent with that range; this enables the space of (d=t)
and (l=D) to be mapped into safe and unsafe regions. Fig. 11 shows
the safe and unsafe regions divided by the values of (d/t) and (l=D)
for small pipes with respect to the design level pressure load for
a target reliability index of 2.5 for both yielding and burst limit
states, corresponding to a failure probability of 1.25 × 10−4 =yr
over a 50-year service life. As indicated in the Fig. 11 will not
print in color. Please make any necessary adjustments. Fig. 11,
when d=t is small, the target reliability is satisfied for any reason-
able values of l=D, but as d=t increases as a result of continuing
corrosion, the range of tolerable l=D values decreases rapidly and
approaches zero when d=t reaches about 0.4.
Figs. 10 and 11 provide useful tools for inspection engineers
to evaluate pipe safety without conducting a full probabilistic
analysis. The underlying target reliability index of 2.5 in Fig. 11
can actually be any safety benchmark expressed in terms of reli-
ability index stipulated by either code writer or pipe owner. If pipe
owners have higher expectations on pipe performance than the
minimum requirements stipulated in the code, these expectations
can be conveniently incorporated in these pipe fitness-for-service
evaluation tools.
Leakage
Leakage is another corrosion-induced limit state that is commonly
encountered during service life of buried steel pipes. Assume that
leakage will initiate if pipe wall thickness is reduced to 10% of its
original value due to corrosion; then the limit state probability can
be estimated as
Pf ¼ P½d > 0.9t
ð7Þ
where d = corrosion defect depth and t = pipe wall thickness. Using
the statistics presented in Tables 3 and 4, the reliability of the pipe
against leakage can be computed. Fig. 12 shows the reliability in-
dex against pipe leakage as a function of defect depth for pipes
buried in the average soil environment with wall thickness ranging
from 3 to 12 mm (0.1 to 0.5 in.). Clearly, a pipe with a thicker wall
is always less vulnerable to leakage failure. Assume further that
defect depth measured at the time T 1 is d1 ; then the remaining ser-
vice life of the pipe, ΔT R , following T 1 , can be estimated using the
corrosion model introduced in Eq (2) as
 1=α  1=α
dF d
ΔT R ¼ ðΔT R þ T 1 − T 0 Þ − ðT 1 − T 0 Þ ¼ − 1
k k
ð8Þ
J. Pipeline Syst. Eng. Pract.
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.

Fig. 10. Pipe reliability index as a function of the calculated defect shape factor for (a) small pipes [D < 24 in: (610 mm)]; (b) large pipes
[D > 24 in: (610 mm)]

Fig. 12. Reliability index against leakage failure for steel pipes
Fig. 11. Safe/unsafe regions with respect to target reliability index of
2.5 for both yielding and bursting limit states [D < 24 in: (610 mm)]

a pressure level comparable to the maximum allowable (or rated)

where d1 = defect depth measured at time T 1 and dF = defect depth
obtained from Fig. 12 that corresponds to the target reliability
index. Fig. 12 together with Eq. (8) can be used to predict the
remaining service life against pipe leakage to assist engineers in
setting priorities in risk-informed pipe management.
Leakage is often regarded as the most frequently observed fail-
ure mode in pipelines. Note that the probability of leakage in a
pipeline is only determined by the corrosion propagation in the pipe
wall; the probability of burst, however, is governed by both the cor-
rosion propagation and the maximum pressure load experienced by
the pipe. Therefore, for many pipes, the probability of leakage is
often higher than that of burst, but for other pipelines that operate at
pressure, the probability of burst could well exceed the probability
of leakage and become the critical failure mode in the service life of
the pipes. Furthermore, since pipe wall thickness is the dominant
design parameter for both the burst and leakage limit states, the
most cost-effective maintenance schedule should depend on the
wall thickness of the pipe.
Conclusions
This paper presents a methodology for evaluating fitness-
forservice of pressurized steel pipes based on structural reliability
principles. The strength of a steel pipe with corrosion defect(s)

© ASCE 04013012-8 J. Pipeline Syst. Eng. Pract.

J. Pipeline Syst. Eng. Pract. 2014.5.

 subjected to internal pressure has been analyzed using nonlinear
finite-element analyses, leading to an analytical model for estimat-
ing the retained pipe strength as a function of pipe dimensions and
defect geometrical parameters. This model was then used for reli-
ability analysis taking into account the numerous uncertainties
attendant to active corrosion in buried pipes. Both strength and
service limit states were considered. A practical reliability-based
methodology and simple tools to evaluate fitness-for-service and
remaining service life of pressurized steel pipelines has been pre-
sented. These tools can assist pipe owners and engineers in making
risk-informed decisions related to continued safe pipe operation
and optimal pipe inspection intervals and maintenance schedules.
The results presented herein pertain specifically to low-carbon
Downloaded from ascelibrary.org by Western Michigan University on 01/23/15. Copyright ASCE. For personal use only; all rights reserved.
steel pipes buried in soil. However, the reliability framework is
equally applicable to pipelines with different mechanical properties
and exposure conditions, provided that the burst pressure model
pertain to the specific pipeline metal of interest and the statistical
data on corrosion under relevant environmental conditions are
available to determine the time-dependent corrosion mechanics for
the pipe under investigation.
Acknowledgments
The research described in this paper was supported by Palo Verde
Nuclear Generating Station, Arizona Public Service Company.
This support is gratefully acknowledged.
References
Abaqus version 6.10 (2010). [Computer software]. Dassault Systèmes
Simulia Corporation, Providence, RI.
Ahammed, M. (1998). “Probabilistic estimation of remaining service life
of a pipe in the presence of active corrosion.” Int. J. Pressure Vessels
Piping, 75(4), 321–329.
Ahammed, M., and Melchers, R. E. (1996). “Reliability estimation of
pressurized pipelines subject to localized corrosion defects.” Int. J.
Pressure Vessels Piping, 69(3), 267–272.
ASME. (2009). “ASME B31G-2009—Manual for determining the remain-
ing strength of corroded pipe: A supplement to ASME B31 code
for pressure piping.” ANSI/ASME B31G-2009 (Revision of ASME
B31G-1991), New York.
Anghel, C. I. (2009). “Risk assessment for pipelines with active defects
based on artificial intelligence methods.” Int. J. Pressure Vessels
Piping, 86(7), 403–411.
Batte, D., Fu, B., Kirkwood, M. G., and Vu, D. (1997). “Advanced method
for integrity assessment of corroded pipelines.” Pipes Pipelines Int.,
5–11.
© ASCE 04013012-9
J. Pipeline Syst. Eng. Pract. 2014.5.
Caleyo, F., Gonzalez, J. L., and Hallen, J. M. (2002). “A study on reliability
assessment methodology for pipelines with active corrosion defects.”
Int. J. Pressure Vessels Piping, 79(1), 77–86.
Caleyo, F., Velazquez, J. C., Valor, A., and Hallen, J. M. (2009). “Proba-
bility distribution of pitting corrosion depth and rate in underground
pipelines: A Monte Carlo study.” Corrosion Science, 51(9), 1925–1934.
Chauhan, V., Swankie, T. D., Espiner, R., and Wood, I. (2009). “Develop-
ment in methods for assessing the remaining strength of corroded pipe-
lines.” NACE Int. Corrosion Conf. & Expo., Huston, TX.
Cosham, A., and Hopkins, P. (2004). “The assessment of corrosion in
pipelines—Guidance in the pipeline defect assessment manual (PDAM).”
Pipeline Pigging and Integrity Management Conf., Amsterdam,
Netherlands.
Cronin, D. S., and Pick, R. J. (2000). “Experimental database for corroded
pipe: Evaluation of RSTRENG and B31G.” Proc., of the 3rd Int. Pipe-
line Conf. (IPC 2000), Vol. 2, American Society of Mechanical
Engineers, Calgary, AB, Canada, 757–767.
DNV Recommended Practice. (2004). RP-F101, Corroded Pipelines.
Det Norske Veritas.
Ellingwood, B. (2000). “LRFD: Implementing structural reliability in
professional practice.” Eng. Struct., 22(2), 106–115.
Hasan, S., Khan, F., and Kenny, S. (2011). “Probability assessment of burst
limit state due to internal corrosion.” Int. J. Pressure Vessels Piping, 89,
48–58.
Kiefner, J. F., and Vieth, P. H. (1989). “A modified criterion for evaluating
the remaining strength of corroded pipes.” American Gas Association
Catalog, 1989, No. L51609.
Kiefner, J. F., and Vieth, P. H. (1990). “Evaluating Pipe. 1: New Method
Corrects Criterion for Evaluating Corroded Pipe.” Oil Gas J., 6, 56–59.
Leadbetter, M. R., Lindgren, G., and Rootzén, H. (1983). Extremes and
related properties of random sequences and processes, Springer,
New York.
Ling, Y. (1996). “Uniaxial true stress-strain after necking.” AMP J.
Technol., 5, 37–48.
Manjoine, M. J. (1976). “Determination of plastic or creep strains by grids.”
Recent developments in mechanical testing, ASTM STP 608, ASTM,
91–105.
Melchers, R. E. (1999). Structural reliability: Analysis and prediction,
2nd Ed., John Wiley, Chichester, UK.
Netto, T. A., Ferraz, U. S., and Estefen, S. F. (2005). “The effect of
corrosion defects on the burst pressure of pipelines.” J. Constr. Steel
Res., 61(8), 1185–1204.
Romanoff, M. (1957). “Underground corrosion.” NBS Circular 579,
National Bureau of Standards, Washington, DC.
Teixeira, A. P., GuedesSoares, C., Netto, T. A., and Estefen, S. F. (2008).
“Reliability of pipes with corrosion defects.” Int. J. Pressure Vessels
Piping, 85(4), 228–237.
Xu, L. Y., and Cheng, Y. F. (2011). “Reliability and failure pressure pre-
diction of various grades of pipeline steel in the presence of corrosion
defects and pre-strain.” Int. J. Pressure Vessels Piping, 89, 75–84.
Zapp, P. E. (1996). “Pitting growth rate in carbon steel exposed to simulated
radioactive waste.” ASET, report WSRC-TR-96-0024.
J. Pipeline Syst. Eng. Pract.