Académique Documents
Professionnel Documents
Culture Documents
www.elsevier.com/locate/jcsr
Abstract
The loss of metal in a pipeline due to corrosion usually results in localized pits with various depths
and irregular shapes on its external and internal surfaces. The effect of external corrosion defects was
studied via a series of small-scale experiments and through a nonlinear numerical model based on
the finite element method. After calibration was conducted, based on the experimental results, the
model was used to determine the burst pressure as a function of material and geometric parameters
of different pipes and defects. This paper briefly summarizes these results, which are subsequently
used to develop a simple procedure for estimating the burst pressure of corroded pipes.
© 2005 Elsevier Ltd. All rights reserved.
Keywords: Pipeline; Corrosion; Burst pressure
1. Introduction
∗ Corresponding author. Tel.: +55 21 2562 7792; fax: +55 21 2562 7794.
E-mail address: segen@lts.coppe.ufrj.br (S.F. Estefen).
0143-974X/$ - see front matter © 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.jcsr.2005.02.010
1186 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
Nomenclature
are useful for readily checking the residual strength of pipelines without the need for
sophisticated analyses.
Among the existing criteria for evaluating the residual strength of corroded pipelines,
the ASME B31G code [1], originally developed many years ago, is still the most widely
used criterion. Kiefner and Vieth [2,3] recognized later that the corrosion assessment
methods in the B31G code could be over-conservative for some kinds of defects found
in practice. They modified the code to develop what is known as the 0.85dL method.
Like for the original version, the length of the defect l and its depth d are the only
parameters needed to define the defect. Single defect equations like these form the basis
for predicting the failure loads of more complex problems such as combined internal
pressure and compressive longitudinal stresses, the interaction of single defects, and the
actual complex shape of a corrosion defect. More recently, DNV [4,5] has published
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1187
recommended practices for assessing corroded pipelines under combined internal pressure
and longitudinal compressive stress. Based on both experimental tests and numerical
calculations, the proposed empirical formulae include single and interacting defects, and
complex shaped defects.
Structural analysis software has enabled the modeling of realistic defect shapes and
nonlinear material behavior. Studies using both plane strain and three-dimensional models
have been published recently [6,7]. These more realistic models tend to yield better
predictions. Nevertheless, in most cases the numerical tools used are not simple enough
or readily available to the field engineer.
In this work, the factors governing the behavior of corroded pipelines subject to
internal pressure are investigated through combined experimental and numerical efforts.
An evaluation of the residual strength of pipelines with single longitudinal corrosion
defects was initially made through a series of small-scale experiments. In parallel, a three-
dimensional nonlinear finite element model was developed to predict the burst pressure
of intact and corroded pipes. The model was first validated by reproducing numerically
the physical experiments performed and was subsequently used to carry out an extensive
parametric study. The data set is then reduced to a very simple curve that relates the main
geometric parameters of the pipe and defect to its residual pressure capacity. Finally, codes
currently used in practice are compared to both experimental and numerical results.
2. Experiments
Seven small-scale steel specimens were fabricated; one was intact and six included
induced defects. A schematic view of the specimens with induced defects is shown in
Fig. 1. The actual outer mean diameter (D) and thickness (t) of each tube, and the main
geometric parameters of the defects (d, l, c) are given in Table 1. The overall length of the
specimens was kept around 420 mm (10D). Defects were produced on the pipe through
a spark erosion process, with customized tools for each size of defect. They were made
in circular shapes in both longitudinal and hoop directions in order to obtain maximum
1188 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
Table 1
Geometric properties and burst pressures for the tested tubes
depths (d) of approximately 0.6t, 0.7t, and 0.8t, maximum lengths (l) of 0.5D and D,
and a maximum width (c) equal to 0.31D. Fig. 2 depicts tube T3D and its induced defect
before the experimental test.
The availability of X-grade steel pipes in small diameters is limited and, as a
consequence, the specimens were machined from the same long tube made of AISI 1020
mild steel. Initially, three axial test coupons were cut from the tube used in the manufacture
of the specimens and tested under uniaxial tension. In addition, the stress–strain curve
in the hoop direction was determined by internally pressurizing a 210 mm long piece
of tube machined so as to eliminate initial imperfections such as ovality and thickness
eccentricity. During pressurization, the edges were free to move in the axial direction,
i.e., only hoop stress was applied to the specimen. Negligible material anisotropy was
observed on comparing the stress–strain curves in the hoop and axial directions. The
average engineering stress–strain curve calculated from all test results is shown in Fig. 3.
The following material parameters were obtained from the curve:
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1189
Fig. 4. Instrumented tube for testing. (a) Test apparatus; (b) detailed view.
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1191
3. Numerical model
In parallel with the experimental program, a finite element model was developed to
simulate the failure of intact and externally damaged pipes under internal pressure. The
model was developed within the framework of the nonlinear finite element code ABAQUS
(version 6.1).
1192 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
Fig. 6. Typical data recorded in the experiment. (a) Pressure–time history; (b) strain gage signals.
The tubes, including the defect regions, were idealized with three-dimensional, 27-
node, quadratic brick elements (C3D27). The number of integration points per element
was 27, i.e., a full integration scheme was used. The active degrees of freedom were three
displacement components at each node. A typical mesh used in the analyses of “corroded”
tubes is illustrated in Fig. 8. The deformation of the cross-sections of the tubes is assumed
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1193
Fig. 7. Test specimens after the burst. (a) T1I (intact); (b) T2D; (c) T7D.
1194 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
to be symmetric with respect to plane 1–2. Furthermore, plane 2–3, located at the mid-axial
position of the tube (and defect), is also assumed to be a plane of symmetry. The problem
is then reduced to a quarter of its original size. The tubes were represented by twenty
elements in the axial direction, fifteen in the circumferential direction and three through
the thickness. The mesh for the intact tubes was made uniform in each direction. For the
“corroded” tubes, the mesh distribution and refinement varied depending on the geometry
of the defects, though roughly keeping the same total number of elements in all analyses.
Initially, for the sake of simplicity, no attempt was made to reproduce the exact shape of
the defect regions. These were considered as sectors of length l and width c, with constant
thickness equal to t −d, Fig. 8. The numerical model was later improved so as to reproduce
the exact shape of the defects, as shown in Fig. 9 (these two models are henceforth called
the simplified defect shape model – SDS – and the exact defect shape model – EDS).
The material of the tubes was assumed to be J2 -type, elasto-plastic, finitely deforming
solid with isotropic hardening.
Both models (SDS and EDS) were first used to simulate each of the physical
experiments conducted. The actual geometric and material parameters of the tubes tested
were used in the analyses. Fig. 10 shows the calculated P–δυ response for test specimens
T1I and T2D (υo is the initial internal volume of the tube and δυ is the absolute value
of the change of volume evaluated for each pressure level). The curve for the latter case
was obtained considering the defects as idealized sectors of reduced constant thickness
(SDS model). The pressure peaks marked on the curves represent the onset of failure (for
the intact tube, a slight variation in the thickness of the cross-section at x 1 = 0, θ = 0◦
was simulated to provoke failure initiation at this point). Results from experimental tests
(P ), and corresponding analyses using the SDS model ( P ) and the EDS model ( Pˆ ) are
b b b
summarized in Table 2.
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1195
Fig. 10. Calculated internal pressure change in volume response for tubes T1I and T2D.
As expected, the predicted burst pressures for the damaged tubes using the SDS model
were always lower than the experimental counterparts; i.e., making the depth of the
1196 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
Table 2
Comparison between experiments and numerical simulations
ˆ
Pb Pb Pb
Tube Pbi (experiments) Pbi (SDS model) Pbi (EDS model)
Table 3
Material and geometric parameters analyzed
D (mm) 406.4
t (mm) 6.35, 12.7, 19.05, 25.4
d/t 0.1, 0.2, 0.4, 0.6, 0.7, 0.8
l/D 0.5, 1., 1.5, 2.
c/D 0.0785, 0.1047, 0.1571
defect constant in the numerical model leads to a higher detrimental effect on the burst
pressure. This is simply because more material is removed from the tube in the analyses,
as compared to the actual defects induced in the test specimens. The shallower and shorter
the defect, the smaller the difference between experiment and prediction. This trend can
be observed by comparing the results obtained from the analyses of tubes T7D, T5D, T3D
and T1I (the latter showing a very good correlation). On the other hand, the EDS model
slightly overestimates the burst pressure, yet shows in general a better correlation with the
experimental results (with an average deviation of 10%).
Thus, considering its advantageous simplicity, the results indicate that the SDS
numerical model can be reasonably employed to obtain lower bound estimates of the burst
pressure of pipes with defects which in size and shape are similar to those addressed in the
present study.
4. Parametric study
The SDS numerical model was then used to conduct an extensive parametric study on
the influence of geometric parameters of the pipe and defects on the burst pressure of actual
pipelines made of API X-grade steel. The material and ranges of geometric parameters used
in the analyses are given in Table 3.
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1197
Table 4
Tvergaard fit parameters
The uniaxial stress–strain curves of the three materials were represented in the analyses
by the Tvergaard fit:
σp 1 σ n 1
ε= − +1 (1)
E n σp n
for strains less than 10%, and the fit parameters given in Table 4.
The parameters marked in bold in Table 3 were adopted as the base case. For example,
in one series of analyses, D, t, d/t, and c/D were assigned the bold values while changing
l/D from 0.5 to 2.0. This was done for all three materials. Similarly for another series, l,
D, t, l/D, and c/D were kept constant, and d/t was varied from 0.1 to 0.8. These two
series of results are shown in Figs. 11 and 12. When the burst pressures are normalized by
the burst pressures of the intact tubes (36.35, 45.08, and 53.94 MPa for X-52, X-65, and
X-77, respectively), the results for the three materials coalesce.
Again, one can see that the depth of the defect has the strongest detrimental effect
on the burst pressure, but with varied severity depending on the d/t range. For d/t less
1198 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
than 0.2, the loss in the burst capacity is fairly small (within 5%). As the defect grows
deeper, though, this effect gets much more pronounced (e.g., 22%, 43%, and 75% for d/t
equal to 0.4, 0.6, and 0.8, respectively). Inversely, the rate of decay gets smaller as the
length is increased; see Fig. 11. When the length is increased to 0.5D, the burst capacity
is decreased by 13%, with a diminishing effect towards 1.5D. Eventually, a near plateau
is formed between 1.5D and 2.D (1.5% change only), showing that the influence of the
length of the defect is negligible for l/D greater than 1.5.
The results also showed that an increase in the circumferential length of the defect,
for c/D ≥ 0.0785, has very little influence on the burst pressure for the material and
range of geometric parameters analyzed. In the base case, for instance, P̂b / P̂bi is nearly
constant, equal to 0.77, when varying c/D between 0.0785 and 0.1571. Thus, the data set
indicates that it is possible to estimate two geometric bounds beyond which a small or very
small effect is observed in the burst pressure: c/D = 0.0785 and l/D = 1.5. The same
conclusion was obtained from extra series of calculations with different D/t and d/t, but
these results are not shown here for expediency.
In addition to the burst pressure simulation through radial pressure only, another
numerical model considered internal pressure with an end cap effect. However the burst
pressure for the different induced defects presented similar results to those for the case of
radial pressure only and clamped edges.
In order to study the influence of material shape and degree of strain hardening in
the plastic range on the burst pressure, a nominal steel API grade X-65 was used as a
reference. This curve was used for varying the ratio σ0 /σu and with the extent of the
Ludder’s band plateau region (εLD ) modified. εLD values equal to 80% and 120% of
the respective reference value were assumed for the two curves; see Fig. 13. The burst
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1199
Fig. 13. Stress–strain curves in the plastic range with different Ludder’s band plateau sizes.
pressures obtained, 33.21 and 32.81 MPa, differed by only 0.6% from the burst pressure
obtained for the reference curve (33.02 MPa). Another two curves with yield stress values
equal to 80% and 110% of the reference value were also adopted, Fig. 14, with respective
burst pressures of 26.51 and 36.08 MPa. These values differed by −19.7% and 9.27% with
respect to the burst pressure using the reference curve. In all four modified curves, nominal
values have been maintained from plastic strains beyond approximately 0.05 (Figs. 13 and
14). On the basis of these results, it can be concluded that small variations of the Ludder’s
band plateau have a negligible influence on the burst pressure (less than 1%). On the other
hand the ratio σo /σr has a more significant influence.
5. A simple assessment
Full numerical calculations like the ones described here or by other authors [8] are
viable but, in many cases, they may not be the most practical option. Alternatively, there are
several codes and recommended practices available in the literature [1,4] that can be used
to assess the residual strength of corroded pipes under internal pressure. They usually take
the form of formulae that relate the pipe geometry, material and geometric characteristics
of the defect to the burst pressure of the damaged pipeline. Although this could definitely
be done here, it is proposed to go one step further in simplifying the problem. The idea is
to establish a procedure that can enable the surveyor to instantly evaluate the detrimental
effect of a single defect. The operator could then later decide on which type of analyses
and/or actions would be more suitable in each case.
Such a procedure is developed using the experimental data and the numerical results
from the parametric study. A similar methodology was proposed in the past [9,10] for
1200 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
Fig. 14. Stress–strain curves in the plastic range with different yield stresses.
different problems. Here, we assume that the burst pressure is dependent on the major
problem parameters, i.e.,
Pb = f (D, t, d, l, c, Pbi ). (2)
From Buckingham’s Π theorem, this can be reduced to a relationship between
nondimensional variables:
Pb t l d c
=F , , , , (3)
Pbi D D D D
which is simplified to the power series below:
∞ α1 α2 α3
n
Pb t l d c α4
= An . (4)
Pbi n=0
D D D D
We further simplify the expression by neglecting higher order terms (n > 1). On setting
Ao = 1, Eq. (4) condenses to
α1 α2 α3
Pb t l d c α4
≈ 1 + A1 . (5)
Pbi D D D D
The numerical data were reduced through an iterative scheme based on the least-squares
error method. Since the parametric study indicated little influence of the parameter c/D
on Pb for c/D > 0.0785, the exponent α4 was assigned as zero. The result of this process
produced the following equation:
1.6 0.4
Pb d l
= 1 − 0.9435 , (6)
Pbi t D
for c/D ≥ 0.0785, 0.1 ≤ d/t ≤ 0.8, and l/D ≤ 1.5.
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1201
Fig. 15. Burst pressure versus the empirical function of geometric parameters and the best linear fit.
Fig. 16. Comparison between experiments, the linear fit, and predictions from DNV [5] and B31G [1] codes.
Fig. 15 shows all numerical data plotted versus the parameter given on the RHS of Eq.
(6), and the corresponding best linear fit. The results nicely coalesce when plotted in this
fashion, with a very good correlation with the linear fit.
1202 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
The experimental results are also plotted against the same parameter in Fig. 16. The
predictions obtained using the equations from B31G [1] and DNV [4] codes are shown
in the same figure. The B31G and DNV equations for predicting the burst pressure are,
respectively,
2t 1 − 2
3
d
t
PB31G = (1.1σ y )
(7)
D 1 − 2 d M −1 3 t T
where
L2
MT = 1 + 0.8
Dt
and
∗
2tSMTS 1 − γd dt
PDNV = γm
∗ (8)
γd d
t
(D − t) 1 − Q
where
2
l
Q= 1 + 0.31 √
Dt
and
∗
d d d
= + εd StD .
t t med t
In the above equations, the following values where adopted in order to allow a direct
comparison between experiments, analyses, and the predictions from both codes:
γd = 1; γm = 0.77
and SMTS was taken as the ultimate stress obtained from the uniaxial stress–strain tests,
i.e. SMTS = 392 MPa.
Confirming previous works [8], the B31G code proved to be over-conservative in all
cases. Yet more conservative results were obtained using the DNV equation. The linear
fit (Eq. (6)) approximately fell between the predictions from the two codes and the
experimental results, with a closer correlation with the experiments (see Fig. 16).
6. Recommendations
As previously mentioned, the numerical model tends to be more conservative when
longer and deeper defects are analyzed. This tendency is somewhat reflected in Fig. 16,
when comparing the linear fit with the experimental results corresponding to higher values
of the burst pressure. Thus, in general, this fit can be used to get a lower bound estimate
of the burst pressure of pipes as a function of the key variables. In summary, the following
T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204 1203
7. Conclusions
Small-scale steel pipe specimens have been fabricated, one intact and six with different
induced defects, and tested under internal pressure to investigate the residual strength of
the damage pipes. Numerical models based on the finite element method incorporating
nonlinear ties due to both geometrical and material behavior have also simulated the burst
pressure. The numerical and experimental results presented good correlation. A parametric
study was then performed for different steel grades and defect geometries. On the basis of
both experimental and numerical results an analytical formula has been derived for the pipe
burst pressure with a single defect. Using this formula a simple inspection assessment has
been proposed.
Special care should be taken when using the proposed approach in order to avoid
extrapolation of the defect geometry parameters. It is recommended that one use a realistic
representation of the defect geometry as well as the proper material stress–strain curve.
Variations in geometries and material properties could be taken into account assuming the
proposed limit state equation in association with reliability techniques.
Acknowledgments
The authors would like to thank for technical support the Submarine Technology
Laboratory/COPPE and also the sponsors National Petroleum Agency (ANP) and Brazilian
Research Council (CNPq).
References
[1] ASME B31G. Manual for determining the remaining strength of corroded pipelines. A supplement to
ANSI/ASME B31 code for pressure piping. 1991.
[2] Kiefner JF, Vieth PH. Evaluating pipe 1: new method corrects criterion for evaluating corroded pipe. Oil &
Gas Journal 1990;88(32):56–9.
[3] Kiefner JF, Vieth PH. Evaluating pipe conclusion: PC program speeds new criterion for evaluating corroded
pipe. Oil & Gas Journal 1990;88(34):91–3.
1204 T.A. Netto et al. / Journal of Constructional Steel Research 61 (2005) 1185–1204
[4] DNV Recommended Practice, RP-F101. Corroded pipelines, Det Norske Veritas. April, 1999.
[5] Bjornoy OH, Sigurdsson G. Residual strength of corroded pipelines, DNV test results. In: Proc. int’l offshore
and polar engineering conference, vol. II. 2000. p. 189–96.
[6] Kim HO, Maple JA. A plane strain analysis of corroded pipelines. In: Proc. int’l offshore and polar
engineering conference, vol. II. 1996. p. 89–94.
[7] Fu B, Kirkwood MG. Predicting failure of internally corroded linepipe using finite element method. In:
Proc. int’l conference on offshore mechanics and Arctic engineering—OMAE’95, vol. V. 1995. p. 175–84.
[8] Stephens DR, Francini RB. A review and evaluation of remaining strength criteria for corrosion defects in
transmission pipelines. In: Proc. int’l conference on offshore mechanics and arctic engineering—OMAE’00.
2000 [CD Rom].
[9] Netto TA, Estefen SF. Buckle arrestors for deepwater pipelines. Marine Structures 1996;9:873–83.
[10] Kyriakides S, Park T-D, Netto TA. On the design of integral buckle arrestors for offshore pipelines. Applied
Ocean Research 1998;20:95–104.