You are on page 1of 11

ARTICLE IN PRESS

International Journal of Pressure Vessels and Piping 84 (2007) 697707


www.elsevier.com/locate/ijpvp

Reliability-based assessment of polyethylene pipe creep lifetime


Rabia Khelifa,b, Alaa Chateauneufc,, Kamel Chaouib
a

LaMI-UBP & IFMA, Campus de Clermont-Fd, Les Cezeaux, BP 265, 63175 Aubie`re Cedex, France
LR3MI, Departement de Genie Mecanique, Universite Badji Mokhtar, BP 12, Annaba 23000, Algerie
c
LGCUniversity Blaise Pascal, Campus des Cezeaux, BP 206, 63174 Aubie`re Cedex, France

Received 2 March 2007; received in revised form 9 July 2007; accepted 1 August 2007

Abstract
Lifetime management of underground pipelines is mandatory for safe hydrocarbon transmission and distribution systems. The use of
high-density polyethylene tubes subjected to internal pressure, external loading and environmental variations requires a reliability study
in order to dene the service limits and the optimal operating conditions. In service, the time-dependent phenomena, especially creep,
take place during the pipe lifetime, leading to signicant strength reduction. In this work, the reliability-based assessment of pipe lifetime
models is carried out, in order to propose a probabilistic methodology for lifetime model selection and to determine the pipe safety levels
as well as the most important parameters for pipeline reliability. This study is enhanced by parametric analysis on pipe conguration, gas
pressure and operating temperature.
r 2007 Elsevier Ltd. All rights reserved.
Keywords: Reliability analysis; Polyethylene pipelines; Lifetime models; Creep

1. Introduction
The distribution networks for natural gas and water
supplies in cities are basically made of plastic pipes with
diameters reaching 250 mm and more depending on the
pressure rate. Newly installed piping gas systems in the
world are exclusively made of polyethylene (PE) because of
its ease of installation, relatively low cost and long-term
resistance to environmental aggressive agents. These
polymers are still the subject of many studies that highlight
various behavioral aspects in terms of service lifetime [1],
mechanical characterization and structure relationship [2],
loading modes [3], residual stresses [4], failure mechanisms
[5] and environmental effects [6]. The design of thermoplastic pipes is achieved through the Rate Process Method
for Projecting Performance of Polyethylene Piping Components, which is standardized in ASTM D-2837 and
D-2513. The calculation is based on a three-parameter
Corresponding author. Tel.: +33 473407526; fax:+33 473407494.

E-mail addresses: rabia.khelif@ifma.fr (R. Khelif),


alaa.chateauneuf@polytech.univ-bpclermont.fr (A. Chateauneuf),
chaoui@univ-annaba.org (K. Chaoui).
0308-0161/$ - see front matter r 2007 Elsevier Ltd. All rights reserved.
doi:10.1016/j.ijpvp.2007.08.006

equation linking the time-to-failure, the hoop stress and the


operating temperature [79].
Property deterioration of PE pipes under aging effects,
related to mechanical, environmental and chemical mechanisms, leads to the reduction of the useful pipe lifetime.
Under continuous loading and relatively high temperature,
the lifetime is governed by the creep strain. Various models
dealing with creep failure are available in the literature
[10,11]. The creep fracture by slow crack growth has been
studied in a medium-density PE at 60 and 80 1C [12]. Tests
were conducted for noteched tensile specimens and for
cracked gas pipes under constant creep strain rate [13]. The
experimental data indicate two regimes of creep deformation related to back stress effect and material cracking. The
stress versus time-to-failure curve may be applied for an
incubation time approach, which is for creep initiation.
After this stage, the process zone of the structure is no
longer subjected to uniaxial loading. The defect within the
material induces stress concentration and stress triaxiality
ratio. The predictive model in [13] clearly shows its
superiority and effectiveness over models that take into
account only one inelastic viscoplastic deformation under
uniaxial conditions. From another point of view, the

ARTICLE IN PRESS
698

R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

extrapolation method requires at least to check the


mechanisms that would operate at the extrapolated time
to rupture (low stress level for engineering components).
This approach has not been considered in the present
paper.
Some comparative studies have been carried out on the
basis of a deterministic framework. However, the uncertainties represent an intrinsic part of the material
behavior, and hence should be appropriately taken into
account.
Reliability analysis is recognized as a powerful decisionmaking tool for risk-based design and maintenance. It
allows us to understand how the uncertainties are
propagated within the system, and hence it gives complementary information for deterministic analysis. For this
reason, several reliability studies have been performed
for steel pipelines [1416]. To the authors knowledge,
there is yet no reliability study on PE pipes in the literature,
especially for lifetime model comparison and qualication.
The aim of this study is the probabilistic characterization
of the lifetime of high-density polyethylene (HDPE) pipes,
in order to assess their reliability levels under operating
conditions. The study is carried out through three steps: the
rst one compares the probabilistic distributions of two
lifetime models, in order to show the insufciency of
deterministic approaches; the second step considers the
time-based model uncertainties to show the high sensitivity
to experimental data, as well as to gas pressure and
temperature; and the third step concerns the pipe reliability
assessment under various operating conditions.

stress due to overlaying soil can be expressed by [17]

2. Long-term behavior

where tf is the time to failure (h), T is the temperature (K),


sc is the hoop stress (MPa) and A, B, C and D are
parameters to be determined from experimental results.
Experience shows that the rate process method (RPM)
provides the best correlation between long-term performance projections and known eld data for several PE
piping materials [20]. Testing of pipes has to be carried out
according to ASTM D 1598: Standard test method for
time-to-failure of plastic pipe under constant pressure.

sc sp ss st ,

(1)

where sp is the stress due to internal pressure, ss is due


to soil loading and st is the trafc bending stress. For
creep analysis, only permanent load is considered and so
trafc stress is not considered for long-term design (it is
noted that this stress is neglected compared with pressure
stress). For thin-wall tubes, the pressure stress is given
by [17]
pr
sp ,
(2)
h
where p is the internal pressure, r is the internal pipe radius
and h is the pipe wall thickness. The circumferential bending

(3)

2.2. Lifetime model


In 1962, the Plastic Pipe Institute (PPI) selected stress
rupture testing as the most suitable test method for rating
plastic piping materials. The design of the pipes is achieved
through the Rate Process Method for Projecting Performance of Polyethylene Piping Components, which is also
described in ISO 9080. It is well established that the PE
creep rupture curve may be divided into three regions and
the same testing procedure allows us to distinguish between
various PE resins and manufacturing processes. Fig. 1
shows the typical test results for lifetime for different PE
lots under various hoop stress levels [18,19].
The long-term hydrostatic strength (LTHS) is dened as
the hoop stress that, when applied continuously, will cause
the failure of the pipe at 105 h (11.4 years). This strength
determines the design lifetime for thermoplastic pipes. For
design purposes, the standard extrapolation method (SEM)
can be applied according to the ISO/TR 9080 model:


 
B
C
D log sc ,
log tf A
(4)

T
T

100
PE2306-IA, lot: ANL0283
PE2306-IA, lot: ANL482
PE2306-IA, lot: BCL374
PE2306-IIC, lot: ANL031683
PE2306-IIC, lot: BCL1075

10

50 years

In general, thin-wall underground pipelines are mainly


subjected to radial, longitudinal and circumferential
stresses. The radial and longitudinal stresses are not
considered in this study because they have no signicant
inuence on the creep of buried plastic pipes. The lifetime
models are dened in terms of hoop stress sc, which can
be determined by the superposition of three principal
stresses:

6km C d gB2 Ehr


,
Eh3 24kd pr3

where B is the ditch width at the pipe top level, Cd is the


coefcient of earth pressure, E is the modulus of elasticity, km
is the bending coefcient depending on load and soil reaction,
kd is the deection coefcient and g is the soil density.

Applied Stress, MPa

2.1. Mechanical stress

ss

0.1

10

100

1000

10000

100000 1000000

Time to Failure, hr

Fig. 1. Stress versus time-to-failure for 50 mm OD pipes [18,19].

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

Using the slit failure data, the three-coefcient RPM


equation is given by putting D 0 in Eq. (4), leading to
log tf A

B C log sc

.
T
T

(5)

For the available test data at 60 and 80 1C (Fig. 2), the


constants are tted to give A 16.241, B 9342.2 and
C 1120.4. As it can be seen in Fig. 2, the data show a
large scatter in lifetime results leading to large uncertainties
on the model parameters. For design purposes, these
uncertainties are traditionally covered by the applied safety
factor. However, a more realistic approach can be
developed by the use of the probabilistic design theory, in
order to better balance the cost/safety requirements.
It is to be noted that the model parameters are very
sensitive to the type of HDPE; for example, Trankner
et al. [9] have found A 12.931, B 5904.042 and
C 996.957, for HDPE with a density of 953 kg/m3 and
a nominal yield point of 23.7 MPa; the comparison
between experimental results and tted equation also
showed a signicant test data scatter.
Beside these uncertainties, the choice of the failure
criterion is still a key point and the consequences on
lifetime predictions are far from being negligible. Farshad
[21] has performed an interesting study on lifetime
predictive models at 20 1C, using the parameters: a
A B=T and b C=T. He compared two new criteria for
predicting the long-term (creep rupture) behavior of plastic
pipes under hydrostatic pressure. One of these is the
ultimate strain extrapolation method (USEM) and the
other is called the distortion energy extrapolation method
(DEEM). The three models employed were the stressbased, the strain-based and the energy-based regression
analyses. These models have been compared: linear
regression (equivalent to SEM) leading to a 40.75
and b 25, ultimate strain extrapolation leading to
a 45.5 and b 25, and distortion energy extrapolation leading to a 7.714 and b 14.286. For these
three methods, the 50-year failure stress is, respectively,

699

25.37, 11.97 and 18.59 MPa; note that for the strain-based
criterion, the failure strain is dened as 0.9%, which is a
pessimistic value, compared with a short-term failure strain
of 1.5%.
2.3. Long-term creep model
An alternative formulation of pipe lifetime can be given
in terms of long-term deformations. On the basis of
experimental observations, Lai et al. [22] proposed
an expression to predict long-term creep deformation,
given by
"


9
X
te
t sg D0
Di 1  exp 
1

mti as
i1
"
#!!#
1m
t

1
1
,
6
te
where s is the applied stress, as and g correspond,
respectively, to horizontal and vertical shift factors,
estimated by as 100:4s2 and g 100:04375s2 for
so10 MPa, m is a factor estimated by 0.69 for so6 MPa, ti
are the characteristic retardation times chosen as ti 10i ,
te is the elapsed time for creep compliance measurement
(taken as 4 h), and D0 and Di are tabulated coefcients
obtained by tting the experimental data [22].
The equivalence between the time-based and the
deformation-based criteria are given by specifying an
appropriate failure creep strain. As shown by Farshad
[21], setting a low failure creep strain leads to strongly
under-estimate the long-term failure stress, by a factor
close to 2. The consistency condition implies that coherent
limits have to be dened for strain, stress and material
modulus, in order to prevent arbitrary choice of lifetime
predictive models. This adjustment implies some difculties
that are related not only to average values but also to data
and model scatter; this will be discussed in the following
sections.
3. Reliability model

7.00

Hoop stress (MPa)

In this section, the probabilistic model, to be used in


pipeline reliability analyses, is presented. The strength and
loading variables are represented by a set of random
variables, described by the distribution type and parameters (generally, the mean and the standard deviation).
Specic algorithms are then applied for searching the most
probable failure conguration.

Tests at 80C
Tests at 60C
RPM for 80C
RPM for 60C

6.00
5.00
4.00
3.00
2.00

3.1. Design limit states

1.00
0.00

1000

2000

3000

Time-to-failure (hours)

Fig. 2. Experimental results and tted RPM curves.

4000

In this work, reliability analysis is applied to an


underground pipeline made of HDPE tubes, which is used
to convey gas under a working pressure of 0.4 MPa. The
failure function G(x) corresponds to the lifetime safety
margin dened by the difference between the failure age

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

700

and the required service lifetime tservice, that is


Gxi txi  tservice ,

3.2. Pipe and loading uncertainties


(7)

where xi are the random variables in the system. The


condition G(xi)40 indicates safety and G(xj)p0 corresponds to conventional failure. In this expression, the
failure age depends on the hoop stress and the model
parameters reecting the material properties (i.e. aging
resistance). To evaluate the failure probability, one can
apply Monte Carlo techniques to produce a random
sample of pipe lifetime distribution. This procedure is
convenient for the evaluation of the distribution parameters, but it requires a very large number of simulations
for the evaluation of low failure probabilities, which is
generally the case in engineering design. In order to reduce
the computation time, iterative algorithms [23] are conveniently applied to deal with nonlinear limit state
functions.
For the failure scenario (7), the reliability index b is
dened as the minimum distance between the median point
and the failure domain in the equivalent Gaussian space.
This index is evaluated by solving the constrained
optimization problem:

r
X
b minimize
T i xj 2 subjected to Gxj p0,
(8)
i

where Ti(xj) is any suitable probabilistic transformation.


The solution of this optimization problem can be obtained
by standard optimization algorithms. In our case, specic
reliability algorithms have been used and combined with a
line search procedure. The solution of problem (8) is
usually referred to as the design point, noted P*. In rstorder reliability methods (FORM) [24,25], the failure
probability Pf is simply calculated by
Pf PrGp0 Fb,

(9)

where Pr[ ] is the probability operator and F is the


cumulative Gaussian probability function.

The pipe uncertainties are related to geometry, loading,


manufacturing and service conditions. On the basis of data
from the literature [16], Table 1 indicates the statistical
parameters for the selected random variables. Due to lack
of information, correlation between variables is assumed to
be neglected. While the uncertainty associated with the
coefcient Cd is rather large, the coefcients kd and km
have moderate uncertainties since they need to be selected
for a given situation on the basis of imperfect information.
The geometrical parameters such as ditch width B, pipe
radius r and wall thickness h contain uncertainties highly
dependent on workmanship and quality control after the
ditch construction and pipe production process. The
pressure parameters are dened by operating conditions.
For the considered HDPE pipe with diameter 200 mm,
the gas pressure of 0.4 MPa produces a hoop stress,
sp 2.93 MPa, which is much higher than the stress
ss 0.01 MPa due to soil loading (for information, the
bending stress due to trafc wheels is st 0.06 MPa); that
is why soil and trafc stresses can be neglected.
3.3. Model uncertainties
As described in Section 2.2, the RPM has been calibrated
to give the best tting with experimental data. It is thus
necessary to take account for model uncertainties in
predicting the real lifetime of the pipeline. In the present
work, two uncertain parameters d1 and d2 are introduced in
order to reect the scatter observed during pipe testing.


B
d2 C
log10 sc .
log10 tf d1 A
(10)

T
T
In this expression, the parameters d1 and d2 are assumed
to follow a normally distributed probability function with
mean values equal to one and with coefcients of variation
identied by test results. By applying the resampling
technique, it has been found that the coefcients of
variation of 0.004 and 0.015 are appropriate for d1 and

Table 1
Random variables and corresponding parameters
Variable

Description

Mean value

Coefcient of variation

Geometry
r
h
B

Internal radius of the tube


Wall thickness
Width of the ditch

100 or 62.5 mm
11.4 mm
440 mm

0.02
0.05
0.10

Coefcients
Cd
kd
km
X

Calculation coefcient
Deection coefcient
Bending moment coefcient
Unit weight of soil

1.32
0.108
0.235
1.89  105

0.20
0.15
0.15
0.10

Loading
P
T

Internal pressure
Operating temperature

0.2 to 0.5 MPa


20 1C

0.10
0.10

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

1.4
1.2

2.8
Log() in MPa

Hoop stress (MPa)

701

2.6

1.0
0.8
0.6
0.4

2.4

0.2
2.2

0.5

1.5

2.5

Time-to-failure (log-hours)

0.0

Log(time) in hours

Fig. 3. Scatter of the RPM tting for the two tested PE types.

d2, respectively. For T 80 1C, Fig. 3 shows the scatter of


the time-to-failure and some test results.
4. Pipeline reliability assessment
The reliability assessment is rstly focused on the
comparison of the lifetime prediction models based on
time and on creep strain, in order to identify the model
sensitivity to pipe uncertainties. In the second step, the
sensitivity of the RPM model parameters is analyzed
according to the available data; for instance, this cannot be
performed for creep strain due to lack of sufcient
experimental data. Finally, the reliability assessment of a
pipeline is considered for a practical design situation.
4.1. Probabilistic comparison of lifetime models
In this section, the RPM is compared with the ultimate
deformation model. In this part of the work, the random
variables are related to pipe geometry and applied loading,
as given in Table 1. To allow for fair comparison, the
model parameters are considered deterministic for both
models.
The choice of the target reliability results from engineering practice and society acceptability. The admissible safety
level depends on the failure consequences. According to
natural risks in human activities, the admissible failure
probability ranges from 102 for low failure consequences
to 107 for high failure consequences (e.g. nuclear power
plants). For civil structures, the admissible value of 104
seems to be appropriate as described by the Joint
Committee on Structural Safety (JCSS) [26,27].
4.1.1. RPM model
For pipes with diameters of 200 mm at a constant
temperature of 20 1C, Fig. 4 shows the lifetime distribution
according to the RPM model, obtained by 10,000 Monte
Carlo simulations. Under the service pressure of 0.3 MPa,
the mean lifetime is 115 years while the standard deviation

Fig. 4. Lifetime distribution under system uncertainties.

is 36 years. The distribution is clearly skewed and the


lognormal law allows us to adequately estimate the lifetime
scatter, as shown in Fig. 4. At 20 1C, the probability of
failure before reaching 50 years of service is 4.9  103,
which is signicant, knowing the material strength
uncertainties.
The sensitivity of system reliability is shown in Fig. 5 for
the three signicant parameters: pipe wall thickness, gas
pressure and pipe diameter. The other parameters in Table 1
have very low effects on the pipe safety. The difference
between the importance factors for the two pipes with
diameters 125 and 200 mm is very low for practical use.
For pipe with diameter 125 mm, Fig. 6 shows the
cumulated distribution functions (CDF) of the pipe lifetime
under different internal pressures (while the left-side gure
gives the overall distributions, the right-side one illustrates
these distributions in the range of interest for design). The
effect of gas pressure on the lifetime distribution is

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

702

Diameter
8.8%

Diameter
8.2%
Gas pressure
37.1%

Thickness
54.1%

Gas pressure
40.6%

Thickness
51.2%

Pipe 125

Pipe 200

Fig. 5. Importance factors for pipes with diameters of 125 and 200 mm.

1.00
0.90
0.80
0.70
0.60
0.50
0.40
0.30
0.20
0.10
0.00

1.0E-01

Lifetime distribution in terms of mean gas pressure

1.0E-02
1.0E-03

Gas pressure = 3 bars


Gas pressure = 4 bars
Gas pressure = 5 bars

Lifetime CDF

Lifetime CDF

Lifetime distribution in terms of mean gas pressure

1.0E-04
1.0E-05
1.0E-06

Gas pressure = 3 bars

1.0E-07

Gas pressure = 4 bars

1.0E-08
0

250

500

750

1000 1250 1500 1750 2000

Time (years)

1.0E-09

Gas pressure = 5 bars

50

100

150

200

Time (years)

Fig. 6. Pipe lifetime cumulated probability functions under different internal pressures.

naturally very high. The admissible failure rate of 104


leads to a lifetime of 46 years for a gas pressure of 0.5 MPa,
95 years for 0.4 MPa and more than 200 years for 0.3 MPa.
For the three pressures 0.3, 0.4 and 0.5 MPa, the failure
probabilities at 50 years are, respectively, 6.2  1022,
4.2  109 and 1.3  103; it is to be noted that these values
correspond to conditional probabilities, as material uncertainties are not yet taken into account. Given the
prescribed uncertainties, the gas pressure of 0.4 MPa seems
to be convenient for this kind of pipes.
The effect of temperature on the 50-year failure
probability is shown in Fig. 7 for both diameters. The
dominant role of temperature is highly underlined by these
results. Considering the 125 mm pipes under a pressure of
0.4 MPa, while the failure probability is below 108 for
20 1C, it goes above 102 for 25 1C and becomes 37% for
30 1C. The same trend is observed for the pipe of diameter
200 mm. This very high sensitivity could not be expected by
deterministic analysis, because the reduction of the average
lifetime cannot predict the safety losses. To emphasize this

idea, let us consider the pipe of diameter 125 mm, the


deterministic lifetime is reduced by a factor of 0.42 when
temperature increases from 20 to 25 1C, while the failure
probability is multiplied by a factor of 106. The situation is
even worst when the temperature follows random variations during the service lifetime. For this reason, much care
is required in the use of HDPE pipes in warm regions, in
order to ensure the target safety level.
4.1.2. Ultimate deformation criterion
In order to allow for comparison between the two failure
criteria, the ultimate strain is taken as the strain
corresponding to the deterministic lifetime given by the
RPM. So, the deterministic lifetimes are 313 years for the
pipe of diameter 125 mm under a pressure of 0.4 MPa, and
109 years for the pipe of diameter 200 mm under a pressure
of 0.3 MPa. From the long-term strain curves, the strains
are found to be eu125 0.0099 and eu200 0.0128 for these
pipes at 313 and 109 years, respectively. These values
are taken as the strain at failure, leading to the following

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

703

Fig. 7. Pressuretemperature effects on the 50-year failure probability.

Fig. 8. Cumulated distribution functions for RPM and EDC criteria.

limit states:
G 125 u125  t,
G 200 u200  t,

11

where e(t) is the long-term strain determined by the applied


hoop stress (Eq. 6). These limit states are called herein the
equivalent deformation criteria (EDC). From the deterministic point of view, the conditions on the time or on the
strain lead to almost the same expected lifetime, as the
ultimate strain is calibrated by using the mean values for
pipe and loading parameters. To consider the resulting
lifetime scatter, the load and geometry are considered as
random while the model parameters are kept deterministic
for both models. This conguration is chosen in order to
give fair comparison without the effect of the creep
uncertainties, which are unknown for the strain law;
otherwise the scatter of both RPM and EDC models
should be considered for the same material under the same
testing conditions. In other words, our comparison is still
valid independently from the material composition. Fig. 8
shows the cumulated distribution functions for both

criteria (RPM and EDC) for the two pipe diameters 125
and 200 mm, under the pressures 0.4 and 0.3 MPa,
respectively. These curves show that, under the same
geometrical and loading dispersion, the RPM probability
distribution has a much lower standard deviation than for
the EDC model, as the RPM curve points are closer to the
mean value. This can also be shown by looking for the
intersection of the curves with any arbitrary probability
level. While the two distributions intersect at 50% of
failure probability, the lifetime previsions are completely
different for other probabilities. The high scatter of the
EDC is clearly observed by the low slope of the central part
of the probability curve. Under loading uncertainties, the
EDC leads to pessimistic prediction of the pipe safety, as
signicant failure probabilities appear even for low pipe
ages; for example, the 10% failure probability of diameter
200 mm is reached for 75 years in RPM and for only 20
years in EDC. It is important to notice that for design
purposes, only the lower tail of the probabilistic curve is to
be considered, as the admissible failure rate should be
always very small (i.e. much lower than 50%). Therefore,
the herein probabilistic comparison allows us to select

ARTICLE IN PRESS
704

R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

comprehensively the time-based models, such as RPM,


rather than strain-based models.

generally, the trend is to decrease the lower tail of the


lifetime distribution when the dispersion of d2 increases.

4.2. RPM model uncertainties

4.3. Reliability analysis of HDPE pipes

In this section, the material strength uncertainties are


introduced in the probabilistic model. The scatter of the
RPM pipe lifetime can be dened by random sampling of
the parameters d1 and d2 (for instance, the other variables
are kept constant). The experimental data have shown a
coefcient of variation (COV) of 0.004 for d1 and 0.015 for
d2, considered in Section 3.3. These coefcients of variation
are related to two sources of uncertainties: (1) experimental
procedure and specimen size and (2) intrinsic material
dispersion. While the rst kind can be reduced by more
experiments and better procedures, the second one is due to
material properties and manufacturing process, and cannot
be easily reduced. To check the effect of model uncertainties, 10,000 Monte Carlo samplings have been carried out
for different coefcients of variation and gas pressures, as
illustrated in Fig. 9. It is observed that the scatter of the
time-to-failure can be correctly modeled by lognormal
distributions. The lifetime standard deviation is quite large,
which is already conrmed by tests and models developed
by the PPI [7]; e.g. in a given application of a pipe subjected
to 0.2 MPa at 20 1C, the PPI found the 5% quantile equal
to only 65 years, when the mean is equal to 165 years
(under normality assumption, this corresponds to a COV
of 0.36).
These results show that the COV of the lifetime is more
sensitive to the dispersion of parameter d2 than for d1. By
comparing the two congurations, it can be observed that
the COV is practically independent of the pipe conguration (i.e. diameter and pressure), within the considered
service range. Moreover, Fig. 9 shows that the lifetime
distribution is rather normal for low parameter dispersion,
lognormal for moderate dispersion (which corresponds to
engineering practice) and tends to converge to exponential
when the COV of parameter d2 increases above 0.03. More

Reliability analysis is now carried out for a pipe by


considering all kinds of uncertainties: loading and geometry (Table 1), as well as the RPM uncertainties d1 and d2.
For the 125 mm diameter pipe subjected to 0.4 MPa of gas
pressure, the failure probability is given by Pf 103. If
the COV of the parameter d2 is taken as 0.004, instead of
0.015, the failure probability is reduced to 1.3  104. The
variable importance on the pipe safety is illustrated in
Fig. 10, where temperature is observed as the most
important variable, followed by the lifetime parameter d2
and the pipe thickness; the other parameters in Table 1
have no signicant effect on pipe safety. The reduction of
the dispersion of d2 leads to a signicant increase of the
importance of temperature (from 33% to 48%), thickness
(from 17% to 22.4%) and gas pressure (from 11% to
15%). The same trend is observed for the 200 mm pipe
(Table 2).
Table 3 indicates the 50-year failure probabilities for
different uncertainty considerations, with the probabilistic
data for geometry and loading (soil, gas pressure
and temperature) drawn from Table 1. When the model
and the temperature are deterministic, the failure probability levels are rather low. However, when the model
uncertainties are considered, especially with a COV:
cd2 0.015, the failure probability is multiplied by 10,000
and by 10 for the two pipes of diameter 125 and 200 mm,
respectively. The consideration of temperature uctuations
multiplies again these probabilities by a factor of nearly 10.
While the failure probability is still low for the diameter
125 mm under 0.4 MPa, the pipe of diameter 200 mm under
0.3 MPa indicates a 50-year failure probability close
to 10%, which is extremely high. Naturally, these
results should be considered with special attention as
they strongly depend on the assumed probabilistic models.

Fig. 9. Monte Carlo samplings with various model uncertainties.

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707
1: 6%

705

1: 8%
2: 3%

T : 33%

2 : 30%

T : 48%
p : 15%

: 3.6%

p: 11%

h: 17%

h: 22.4%

: 3%
C.O.V. of 2 =0.015

C.O.V. of 2 =0.004

Fig. 10. Variable importance on the pipe safety in terms of the COV of d2.

Table 2
Lifetime (LT) parameters under various coefcients of variation of model parameter
Case

Diameter 125 mm
Gas pressure 0.4 MPa

COV of d1
COV of d2
Mean of LT (years)
Std. dev. of LT (years)
COV of LT

0.001
0.004
314
29
0.092

0.001
0.015
329
109
0.331

Diameter 200 mm
Gas pressure 0.3 MPa
0.004
0.004
318
53
0.170

0.004
0.010
323
86
0.266

0.004
0.015
333
122
0.366

0.001
0.004
109
11
0.101

0.001
0.015
115
40
0.348

0.004
0.004
111
19
0.171

0.004
0.010
113
30
0.265

0.004
0.015
117
44
0.376

Table 3
Fifty-year failure probabilities for various uncertainty considerations
Case

Geometry
(Table 1)

Loading
(Table 1)

Model

1
2
3
4
5

Random
Random
Random
Random
Random

Random
Random
Random
Random
Random

Deterministic
cd1 0:004
cd1 0:004
cd1 0:004
cd1 0:004

The acquisition of long-term model uncertainties allows


us to improve such an estimation, but the lack of practical
information on in-situ long-term behavior still remains the
main difculty. However, these results present a major
importance in the identication of the most sensitive
parameters, in order to orient the data collection procedures and the inspection/maintenance strategies.
Fig. 11 shows the reliability index evolution as a function
of the up-rated pressure at different temperatures. It is clear
that for permanent temperature above 30 1C, the service
lifetime of 50 years cannot be safely ensured. For the safety
level of 3.8 (i.e. failure probability equal to 104), the allowed
gas pressure in 125 mm pipes is given by 3 bar at 25 1C,
2.6 bar at 20 1C and 5 bar at 15 1C. These curves conrm the
high inuence of ambient temperature on pipe reliability.

cd2
cd2
cd2
cd2

0:004
0:015
0:004
0:015

Temperature

+ 125 mm
(0.4 MPa)

+ 200 mm
(0.3 MPa)

Deterministic
Deterministic
Deterministic
cT 0.1
cT 0.1

4.2  109
2.1  107
7.7  105
1.3  104
9.9  104

4.9  103
1.2  102
5.1  102
5.5  102
9.2  102

The evolution of the importance factors with respect to


the mean applied pressure (Fig. 12) shows that parameter
d2 becomes more important with high pressures (which is
not surprising as it represents the slope of the lifetime curve
with respect to stress) and a signicant decrease of the
importance of the gas pressure itself (which is not that
evident to understand in deterministic analysis), as well as
pipe thickness and operating temperature. The importance of the other parameters is unchanged for all pressure
levels.
5. Conclusion
One of the major outcomes of this study lies in the use of
the probabilistic tools for lifetime model qualication and

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707

706

Pipe diameter of 125 mm

Pipe diameter of 125 mm


1.0E+00

0.2

0.25

0.3

0.35

0.4

0.45

0.5

1.0E+00

0.25

0.3

0.35

0.4

0.45

0.5

1.0E-02

1.0E-04
10C
15C
20C
25C
30C

1.0E-06

1.0E-08

Failure probability

1.0E-02
Failure probability

0.2

1.0E-04
10C
15C
20C
25C
30C

1.0E-06

1.0E-08

1.0E-10

1.0E-10

Mean gas pressure (MPa)

Mean gas pressure (MPa)

Fig. 11. Failure probability as a function of up-rated pressure at various temperatures.

45%

Importance of 2

Importance of gas pressure


20%

25%

18%

23%

Importance of operating temperature


60%

Importance of pipe thickness

30%
25%

15%
13%
10%

50%
Importance factor

Importance factor

Importance factor

35%

Importance factor

55%
40%

20%
18%
15%

20%

8%

13%

15%

5%

10%

45%
40%
35%
30%
25%
20%
15%
10%
5%

0.2 0.25 0.3 0.35 0.4 0.45 0.5

0.2 0.25 0.3 0.35 0.4 0.45 0.5

0.2 0.25 0.3 0.35 0.4 0.45 0.5

0.2 0.25 0.3 0.35 0.4 0.45 0.5

Mean gas pressure (MPa)

Mean gas pressure (MPa)

Mean gas pressure (MPa)

Mean gas pressure (MPa)

Fig. 12. Importance factors for model, thickness, pressure and temperature uncertainties.

pipe safety assessment, which is mandatory to supply the


designer and the maintenance service with a realistic image
of the pipeline risk of failure at various lifetime instances
with regard to pipe aging. The choice of the lifetime model
should not be based only on the equivalence between the
deterministic nominal values, but should also consider the
uncertainties related to the input parameters: geometry,
loading and material properties. In the probabilistic sense,
two models are equivalent only when both averages and
dispersions are identical; otherwise, even with identical
mean values, the predicted lifetime could not lead to
equivalent safety predictions. In this study, the time-based
criterion leads to more consistent results than the strainbased criterion.
According to the available data, the model parameter
uncertainties should be considered in pipeline safety
assessment. The full pipe analysis shows that the inuence
of service temperature and model uncertainties on the pipe
safety is much larger than the gas pressure uctuations and
the geometry precision. A practical conclusion can be

formulated by orienting the costsafety balance to the


service and installation conditions, rather than to excessive
pipe thickness controls.
Acknowledgments
The authors wish to strongly acknowledge the FrancoAlgerian cooperation program for nancially supporting
this work. The experimental works have been carried out at
both LR3MI (Annaba, Algeria) and LaMI (UBP-IFMA,
Clermont-Ferrand, France).
References
[1] Chudnovsky A, Sulkhin Y. Application of crack layer theory to
modeling of slow crack growth in polyethylene. Int J Fract 1999;97:
83102.
[2] Hubert L, David L, Seguela R, Vigier G, Coras-Zuccalli C, Germain
Y. Physical and mechanical properties of PE pipes in relation to
molecular architecture: short-term creep of isotropic and drawn
materials. J Appl Polym Sci 2002;84:230817.

ARTICLE IN PRESS
R. Khelif et al. / International Journal of Pressure Vessels and Piping 84 (2007) 697707
[3] Hamouda HBH, Simoes-betbeder M, Grillon F, Blouet P, Billon N,
Piques R. Creep damage mechanisms in polyethylene gas pipes.
Polymer 2001;42:542537.
[4] Chaoui K, Moet A, Chudnovsky A. Consequences of residual stress
on crack propagation in PE pipes. In: Proceedings of the 10th
international conference on experimental mechanics, Lisbon, Portugal, 1994. p. 8116.
[5] Zhou Y, Lu X, Zhou Z, Brown N. The relative inuence of molecular
structure on brittle fracture by fatigue and under constant loads in
polyethylene. Polym Eng Sci 1996;36(16):21017.
[6] Choi B-H, Zhou Z, Chudnovsky A, Stivala SS, Sehanobish K,
Bosnyak CP. Fracture initiation associated with chemical
degradation: observation and modeling. Int J Solid Struct 2005;42:
68195.
[7] The Plastic Pipe Institute. Nature of hydrostatic stress rupture curves,
Technical note TN-7/2000, p. 11.
[8] Palermo EF, DeBlieu IK. Rate process concepts applied to
hydrostatically rating PE pipes. In: Ninth plastic fuel gas pipe
symposium, American Gas Association, New Orleans, 1985.
p. 21540.
[9] Trankner T, Hedenqvist M, Gedde UW. Structure and crack growth
in gas pipes of medium-density and high-density polyethylene. Polym
Eng Sci 1996;36(16):206976.
[10] Song MS, Hu GX, Hu LJ. Prediction of long-term mechanical
behavior and lifetime of polymeric materials. Polym Test 1998;17(5):
31132.
[11] OConnell PA, Bonner MJ, Duckett RA, Ward IM. The relationship
between slow crack propagation and tensile creep behavior in
polyethylene. Polymer 1995;36(12):235562.
[12] Hamouda HBH, Laiarinandrasana L, Piques R. Fracture mechanics
global approach concepts applied to creep slow crack growth in a
medium density polyethylene (MDPE). Eng Fract Mech 2007;74:
2187204.
[13] Hamouda HBH, Laiarinandrasana L, Piques R. Viscoplastic
behaviour of a medium density polyethylene (MDPE): constitutive

[14]

[15]

[16]

[17]
[18]
[19]
[20]

[21]
[22]
[23]
[24]
[25]

[26]
[27]

707

equations based on double nonlinear deformation model. Int J Plast


2007;23:130727.
Zhou J. Reliability assessment method for pressure piping containing
circumferential defects based on fuzzy probability. Int J Press Vessels
Piping 2005;82(9):66978.
Ko"owrocki K. Asymptotic approach to reliability evaluation of large
multi-state systems with application to piping transportation. Int J
Press Vessels Piping 2003;80(1):5973.
Amirat A, Chateauneuf A, Chaoui K. Reliability assessment of
underground pipelines under the combined effect of active corrosion
and residual stress. Int J Press Vessels Piping 2006;83(2):10717.
Spangler MG, Handy RL. Soil engineering. 4th ed. New York:
Harper & Row; 1982.
Kausch HH. Polymer fracture, 2nd revised and enlarged edition.
Berlin: Spring, 1987.
Young JE, Raphaelian LA, Raske DT. Project review meeting, Gas
Research Institute, Chicago, 1986.
The Plastic Pipe Institute Inc. Rate process method for projecting
performance of polyethylene piping components. Technical note TN16/99, 1999.
Farshad M. Two new criteria for the service life prediction of plastics
pipes. Polym Test 2004;23(8):96772.
Lai J, Bakker A. Analysis of the non-linear creep of high-density
polyethylene. Polymer 1995;36(1):939.
Ditlevsen O, Madsen HO. Structural reliability methods. New York:
Wiley; 1996.
Yan-Gang Z, Tetsuro O. A general procedure for rst/second-order
reliability method (FORM/SORM). Struct Saf 1999;21:95112.
Castillo E, Sarabia JM, Solares C, Gomez P. Uncertainty analyses in
fault trees and Bayesian networks using FORM/SORM methods.
Reliabil Eng Sys Saf 1999;65:2940.
Madsen HO, Krenk S, Lind NC. Methods of structural safety.
New York: Dover Publications; 2006.
Lemaire M, Chateauneuf A, Mitteau JC. Structural reliability. UK:
Iste Publishing Company; 2006.

You might also like