External corrosion modeling for an underground natural gas pipeline

using COMSOL Multiphysics

Wadie Chalgham*1†, Keo-Yuan Wu*2, Ali Mosleh1
1. Mechanical and Aerospace Engineering Department, University of California, Los Angeles (UCLA)
2. Materials Science and Engineering Department, University of California, Los Angeles (UCLA)
* These authors are both first authors and contributed equally to this work
† Contact author: wadie.chalgham@ucla.edu

Abstract: This paper studies external corrosion of an composed of three layers, namely corrosion,
underground transmission pipeline carrying natural mechanical, and reliability models. The corrosion
gas. Pitting corrosion and Stress Corrosion Cracking models include pitting corrosion and stress corrosion
(SCC) are two most common external corrosion types cracking. The output of this model will be the
for buried pipelines and they usually happen when the corrosion depth based on the corrosion rate results of
mitigation methods such as coating or impressed this paper. The mechanical model calculates the
current cathodic protection (ICCP) fails or remaining strength of the pipe given the dimension of
deteriorates. Pitting corrosion is a localized corrosion the defects and will be used in the reliability model for
which occurs at sites with no coating or cathodic the probability of the pipeline failure calculation.
protection as a result of electrochemical reaction
between the pipe material and a corrosive Keywords: External Corrosion, Natural Gas Pipeline,
environment, while SCC is a joint action of a corrosive Pitting Corrosion, SCC, Bayesian Network, Structural
environment and tensile stress from the soil Health Monitoring
movement. A finite element model was developed in
1. Introduction
COMSOL Multiphysics to study the relationship
Operational and management complexity of pipelines
between corrosion potential, current density, and von
has significantly increased due to the rapid growth of
Mises stress along the corrosion defect as a result of
oil and gas industry. Extreme operating conditions and
near-neutral pH SCC using the Corrosion and
adverse environment often lead to corrosion problems
Structural Mechanics Modules. On the other hand,
and tremendous consequences on the pipelines [1].
pitting corrosion is simulated by the Caleyo et al.’s
Therefore, more efforts have been dedicated to
model, which calculates pit growth rate based on
ensuring the integrity of the infrastructure. With the
several operating inputs. Multiple sensitivity analyses
aid of IoT technology, in-situ pipeline health
were performed to find the relationship between the
monitoring facilitates reliable risk assessment which
defect depth, defect length, and the soil stress with
provides optimal mitigation and maintenance actions
respect to anodic current density, cathodic current
to take based on the pipeline health condition. To meet
density, and corrosion potential in order to obtain
that demand, the authors have been developing a
corrosion rate for SCC. In addition, a demonstration of
pipeline health monitoring and management web
the pitting corrosion was performed by predicting
application that can integrate the data, methods, and
pitting corrosion rate and depth given soil and pipe
technologies into a dynamic pipeline heath monitoring
characteristics such as the soil resistivity, bicarbonate
system supported by multiple probabilistic predictive
ions concentration ([HCO3-]), pH level of the soil,
models [2]. These models include dynamic hybrid
chloride ions concentration ([Cl-]), water content of
causal logic, corrosion prognosis, and sensor
the soil, sulphate ions concentration ([SO42-]),
placement optimization models. Corrosion prognosis
pipe/soil potential, bulk density of the soil, and redox
function requires the development of corrosion
potential. The SCC results show that the maximum
predictive models. A natural gas pipeline can suffer
corrosion potential decreases with the increasing
both internal and external corrosion as shown in
defect depth, but increases with increasing defect
Figure 1. One of the major threat to the gas pipelines
length, and that the maximum von Mises stress
is internal corrosion, which can take place in the form
increases with increasing defect depth but decreases
of uniform corrosion [3], pitting corrosion [4], erosion
with increasing defect length, which indicates that the
corrosion [5], microbiologically-influenced corrosion
corrosion is most severe at the center of the defect. The
[6], and corrosion fatigue [7] depending on the
results of SCC and pitting corrosion models will be the
operating and environmental conditions. Corrosive
input to a Bayesian Network model for external
environment as a result of the presence of CO2, H2S,
corrosion which calculates the probability of pipeline
Cl-, solids, and bacterial activities in the pipelines is
failure in terms of external corrosion. This model is
the main cause of internal corrosion.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 Soil

Soil

Gas

Internal Corrosion External Corrosion

Figure 1. Internal and external corrosion can affect steel Corrosion

Defect Location
pipelines and create corrosion defects.

On the other hand, although the external pipeline Steel Pipeline

surface is protected by coatings and cathodic
protection most of the time, it is possible that coatings
Figure 2. The model under study consists of a steel pipeline
will degrade and cathodic protection will be broken with a corrosion defect and surrounding soil.
due to many factors, leading to external corrosion.
Pitting corrosion and Stress Corrosion Cracking (SCC) The pipeline is made of high strength alloy steel and
are regarded as two of common types of external its material parameters are presented in Table A1 in
corrosion, which take place due to the corrosive the Appendix. The pipeline length is 1 m long with a
environment in soils and the longitudinal strain from wall thickness of 19.1 mm. The corrosion defect on the
the soil movement. exterior side of the pipeline has an elliptical shape with
a variable length and depth as shown in Figure 3.
This paper aims at developing an external corrosion
predictive model to fulfill the corrosion prognosis The defect length is defined as the defect propagation
function for the developing pipeline health monitoring at the surface of the pipe while the defect depth is
and management web application software. Pitting defined as the distance from the pipe surface to deepest
corrosion and SCC are both considered in the model. point the corrosion defect has reached in the vertical
The model is a BN (Bayesian Network)-based model direction.
that enables the graphical demonstration of the whole
system, the use of cause-effect relationships within the Initially, the defect length and depth are set to 1 mm
system, the consideration of uncertainties in data, the and will be varied later on in order to perform a
update with available monitoring data, and the sensitivity analysis. The parameters and variables’
integration of physics-based model, field data, and values used in the simulations can be found in Tables
expert knowledge. A2 and A3 in the Appendix.

Detailed descriptions of both pitting corrosion and

SCC models were provided. In this paper, COMSOL
Multiphysics was used to develop a finite element
model of SCC, and the results will be integrated into
the external corrosion BN model. Finally, a case study Defect
length
was done to demonstrate the capability of the model.

2. Numerical Model Defect

depth

In this section, first the model of a pipeline is built in

a modeling finite element software, COMSOL
Multiphysics 5.3.
Figure 2 shows the overall geometry implemented for
the numerical study. It shows that the model consists
of a gas pipeline surrounded by soil. The electrolyte Figure 3. The corrosion defect has an elliptical shape with a
conductivity of soil domain is assumed to be 0.096 variable length and depth.
S/m.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

3. Theory and Governing Equations In order to model the iron dissolution reaction, an
anodic Tafel expression is used as follows:
3.1 SCC
𝜼𝒂
𝒊𝐚 = 𝒊𝟎,𝐚 𝟏𝟎𝑨𝒂 (Eq. 2)
The model presented in this paper is used to study the
behavior of SCC for underground gas pipelines subject where 𝒊𝐚 is the local anodic current density, 𝒊𝟎,𝐚 is the
to longitudinal strain caused by soil movement. Two
exchange anodic current density (2.353·10−3 A/m2),
electrochemical reactions, namely, steel oxidation for
Aa is the Tafel slope (0.118 V) and 𝜼𝒂 is the
anodic reaction and hydrogen evolution for cathodic
overpotential for the anodic reaction. 𝜼𝒂 is found
reaction, respectively are assumed to happen in a near-
using Eq. 3 as follows:
neutral pH environment.
𝜼𝒂 = ∅𝐒 − ∅𝐥 − 𝑬𝐞𝐪,𝐚 (Eq. 3)
3.1.1 Elastoplastic Stress
In this paper, elastoplastic stress is simulated over the
steel pipeline based on the small strain plasticity where 𝑬𝐞𝐪,𝐚 is the equilibrium potential for the anodic
model [8]. In this model, the isotropic hardening reaction. 𝑬𝐞𝐪,𝐚 is found using Eq. 4 as follows:
model is defined as:
∆𝑷𝒎 𝑽𝒎 𝐓𝐑 𝒗𝜶
𝑬𝐞𝐪,𝐚 = 𝑬𝐞𝐪𝟎,𝐚 − 𝐳𝐅
− 𝐳𝐅
𝐥𝐧 (𝐍 𝜺𝒑 + 𝟏) (Eq. 4)
𝝈𝐞 𝟎
𝝈𝐲𝐡𝐚𝐫𝐝 = 𝝈𝐞𝐱𝐩 ,𝜺𝒑 + 1 − 𝝈𝐲𝐬 (Eq. 1)
𝐄
where 𝑬𝐞𝐪𝟎,𝐚 is the standard equilibrium potential for
where 𝝈𝐲𝐡𝐚𝐫𝐝 is the hardening function, 𝝈𝐞𝐱𝐩 is the the anodic reaction (−0.859 V), ΔPm is the excess
experimental stress-strain curve, 𝜺𝒑 is the plastic pressure to elastic deformation (2.687·108 Pa), Vm is
deformation, 𝝈𝐞 is the von Mises stress, E is the the molar volume of steel (7.13·10−6 m3/mol), z is the
Young’s modulus (207·109 Pa), and 𝝈𝐲𝐬 is the yield charge number for steel (2), F is the Faraday’s
strength of high strength alloy steel (806·106 Pa). constant, T is the absolute temperature (298.15 K), R
The experimental stress-strain curve used in the model is the ideal gas constant, ν is an orientation dependent
is shown in Figure 4 and is prescribed in terms of a factor (0.45), α is a coefficient (1.67·1015 m-2) and N0
piecewise cubic interpolation function and is taken is the initial dislocation density (1·1012 m-2).
from [9].
In addition, in order to model the iron dissolution
3.1.2 Electrochemical Reactions reaction, a cathodic Tafel expression is used as
At the corrosion defect surface of pipelines, the two follows:
electrochemical reactions that could occur are: (1) the 𝜼𝒄
anodic (iron dissolution) reactions, and (2) the 𝒊𝐜 = 𝒊𝟎,𝐜 𝟏𝟎𝑨𝒄 (Eq. 5)
cathodic (hydrogen evolution) reactions. In this paper,
it is assumed that the rest of pipeline surfaces are where 𝒊𝐜 is the local cathodic current density, 𝒊𝟎,𝐜 is
electrochemically inactive. the exchange cathodic current density, Ac is the Tafel
slope (−0.207 V) and 𝜼𝒄 is the overpotential for the
cathodic reaction. 𝜼𝒄 is found using Eq. 6 as follows:

𝜼𝒄 = ∅𝐒 − ∅𝐥 − 𝑬𝐞𝐪,𝐜 (Eq. 6)

where 𝑬𝐞𝐪,𝐜 is the standard equilibrium potential for

the cathodic reaction (−0.644 V). 𝒊𝟎,𝐜 is found using
Eq. 7 as follows:
𝝈𝐞 𝑽𝒎
𝒊𝟎,𝐜 = 𝒊𝟎,𝐜,𝐫𝐞𝐟 𝟏𝟎𝟔𝐅(R𝑨𝒄) (Eq. 7)

where 𝒊𝟎,𝐜,𝐫𝐞𝐟 is the reference exchange current density

for the cathodic reaction in the absence of external
stress/strain (1.457×10−2 A/m2).
Figure 4. The experimental stress-strain curve.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

3.2 Pitting corrosion Table 2. Pit initiation time and Regression coefficients for
the maximum pitting depth model
3.2.1 Maximum pitting depth model Parameter (variable, symbol) Value
In order to simulate the pitting corrosion in buried gas 𝛋𝟎 6.08 × 10-1
pipelines, an empirical formula proposed by
𝛂𝟎 8.96 × 10-1
Velázquez et al. [10] is used. This model justifies its
𝛋𝟏 (redox potential, rp) -1.80 × 10-4
physical background by relating pit growth to soil and
pipe characteristics with the multivariate regression 𝛋𝟐 (pH, ph) -6.54 × 10-2
analysis on observed corrosion data. Maximum pit 𝛋𝟑 (resistivity, re) -2.60 × 10-4
depth as a function of time is described by a power law 𝛋𝟒 (chloride, cc) 8.74 × 10-4
model shown as follows: 𝛋𝟓 (bicarbonate, bc) -6.39 × 10-4
𝛋𝟔 (sulphate, sc) -1.22 × 10-4
𝐝𝐦𝐚𝐱 (𝐭) = 𝛋(𝐭 − 𝐭 𝟎 )𝛂 (Eq. 8) 𝛂𝟏 (pipe/soil potential, pp) 5.19 × 10-1
𝛂𝟐 (water content, wc) 4.65 × 10-4
where 𝛋 is the pitting proportionality; 𝛂 is the 𝛂𝟑 (bulk density, bd) -9.90 × 10-2
exponent factor; 𝐭 𝟎 is the pitting initiation time. 𝛂𝟒 (coating type, ct) 4.31 × 10-1
The correlations between the pitting proportionality,
the exponent factors, the soil, and pipe variables are 3.2.1 Pitting rate model
shown as follows: The observed pitting corrosion depth is the result of
accumulation effect by the pitting corrosion rate as a
𝛋 = 𝛋𝟎 + 𝛋𝟏 𝐫𝐩 + 𝛋𝟐 𝐩𝐡 + 𝛋𝟑 𝐫𝐞 + 𝛋𝟒 𝐜𝐜 + (Eq. 9) function of time [11]. Therefore, taking time
𝛋𝟓 𝐛𝐜 + 𝛋𝟔 𝐬𝐜 derivative of the maximum pitting depth model leads
to the pitting rate model shown as follows:
𝛂 = 𝛂𝟎 + 𝛂𝟏 𝐩𝐩 + 𝛂𝟐 𝐰𝐜 + 𝛂𝟑 𝐛𝐝 + 𝛂𝟒 𝐜𝐭 (Eq. 10)
a
𝛎𝐦 (𝐭) = 𝛋` (𝐭 − 𝐭 𝟎 )𝛂 (Eq. 11)
where 𝛋𝐢 and 𝛂𝐢 are regression coefficients for the
corresponding soil and pipe variables. Table 1 shows where 𝛂` = 𝛂 − 𝟏 < 𝟏. 𝟎 and 𝛋` = 𝛋𝛂.
the soil and pipe variables, and Table 2 shows the
regression coefficients of this model, respectively. 3.3 Bayesian Belief Network

It should be noted that since the soil environments Bayesian Network (BN) provides a graphical
have large impact on the corrosion prediction, representation of causal dependencies of a chain of
predicted model from one soil class is unreliable for variables in a probabilistic framework. For example, a
the prediction in other soil classes. Therefore, the chain of operating conditions leading to corrosion and
multivariate regression model provided here considers finally to failure is a BN in this paper. It allows the
only the data from three different common soil classes, calculation of the conditional probability of numerous
namely, clay, clay loam, and sandy clay loam. interconnected parameters based on the Bayes
theorem as shown in Eq. 12:
Table 1. Soil and pipe characteristics
Variable Symbol Unit 𝐏e𝐄h𝐇𝐣 i𝐏(𝐇𝐣 )
𝐏e𝐇𝐣 h𝐄i = (Eq. 12)
Redox potential rp mV ∑𝐧𝐢l𝟏 𝐏(𝐄|𝐇𝐢 )𝐏(𝐇𝐢 )
pH ph -
Pipe-to-soil potential pp mV where 𝐏e𝐇𝐣 h𝐄i is the posterior probability of event 𝐇𝐣
Soil resistivity re Ω·m given the observation E; 𝐏(𝐇𝐣 ) is the prior probability
Water content wc %
of the event 𝐇𝐣 before the observation E; 𝐏e𝐄h𝐇𝐣 i is
Soil bulk density bd g/mL
Chloride content cc ppm the likelihood function and is the probability of the
Bicarbonate content bc ppm observation E given that event 𝐇𝐣 occurred;
Sulfate content sc ppm ∑𝐧𝐢l𝟏 𝐏(𝐄|𝐇𝐢 )𝐏(𝐇𝐢 ) is the sum of all the conditional
Coating type ct - probabilities of E given events 𝐇𝐢 multiplied by the
probabilities of 𝐇𝐢 .

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 Figure 5. BBN model of external corrosion for a natural gas pipeline.
External corrosion for a natural gas pipeline is The quantification of the conditional probability tables
simulated by integrating two most common corrosion was done based on the corrosion models, expert
types, namely, pitting and SCC into a BN model. The knowledge, and literature data. The details of each
BN of external corrosion is shown in Figure 5. Note discretized node of the BN model are described in the
that every node is linked to the next by cause- authors previous work [2]. In this paper, the
consequence relationships. application of this model is demonstrated on an
underground natural gas pipeline as a case study.

(a) (c)

(b) (d)

Figure 6. von Mises stress on the pipe under various longitudinal tensile strains: (a) 1 mm, (b) 2 mm, (c) 3 mm, and (d) 4 mm.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

4. Simulation Results In other words, the purpose of the simulation results
analysis is to find an understanding of the effect of soil
4.1 Stress Corrosion Cracking (SCC) model strain and defect size on the corrosion rate of the
pipeline. The detailed results of sensitivity analysis of
This section provides the results of the SCC SCC model are shown in the Appendix Table A5.
simulations described in the previous sections. The
results aim at showing the effect of different strains 4.1.1 Von Mises Stress
and different defect depths and lengths on von Mises Figure 6 shows the von Mises stress distribution in
stress and corrosion potential. MPa on the pipe under various longitudinal tensile
6 2
strains. The strain is simulated by inducing different
soil displacements. It can be seen that the stress is
larger at the defect location compared to the rest of the
pipeline domain.

In addition, stress increases significantly when the soil

strain applied increases as well as the propagation
length. For instance, at a location of 1.05 m in the x
direction, the stress felt is negligible when the soil
displacement is 1 mm, is around 400 MPa when the
displacement is 2 mm, is around 600 MPa when the
displacement is 3 mm, and is around 700 MPa when
the displacement is 4 mm. These results show that the
effect of the soil strain on the defect location as well
(a) as on the rest on the pipe is not negligible.
6 6

Figure 7 shows the von Mises stress under different

defect depths and at a fixed defect length of 6 mm.
This sensitivity analysis aims at understanding the
effect of the stress on the pipe under different defect
depths. It can be seen that the increase in the corrosion
defect depth increases the stress at the bottom edge of
the defect but decreases in the surface of the pipe. The
stress seems to concentrate in the defect location and
have a more significant effect on the inside of the pipe
rather than the surface area.

4.1.2 Corrosion Potential

Figure 8 shows the corrosion potential under different
(b) defect depths and at a fixed defect length of 6 mm.
6 10
This sensitivity analysis aims at understanding the
corrosion potential variation under different defect
depths. It can be seen that the increase in the corrosion
defect depth decreases the corrosion potential at the
bottom edge of the defect.

(c)
Figure 7. von Mises stress under different defect depths:
(a) 2 mm, (b) 6 mm, and (c) 10 mm.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 6 2

(a) Figure 9. Linear distribution of corrosion potential along

the length of the corrosion defect (length and depth of 5 mm)
6 6
under various longitudinal tensile strains: soil displacements
of 1, 2, 3, and 4 mm.

Figure 9 shows the distribution of the corrosion

potential along the length of the corrosion defect under
various longitudinal tensile strains: soil displacements
of 1, 2, 3, and 4 mm. For this simulation, the defect
length and depth were fixed at 10 mm. It is shown that
for the smaller tensile strains of 1 and 2 mm, the
variation in the corrosion potential is almost uniform
along the length of the corrosion defect. However, for
higher tensile strains of 3 and 4 mm, the variation in
the corrosion potential is clearly nonuniform with the
more negative corrosion potential at the center of the
(b) corrosion defect than that at both the sides of the
corrosion defect.
6 10
Figure 10 shows the linear distribution of anodic
current density (iron dissolution) along the length of
the corrosion defect under various longitudinal tensile
strains: soil displacements of 1, 2, 3, and 4 mm. For
lower soil displacements of 1 mm and 2 mm, the
anodic current density is almost uniform along the
length of the corrosion defect, with an increase around
the defect center. However, for higher soil
displacements of 3 mm and 4 mm, the anodic current
density varies and is significantly nonuniform,
especially at the center of the corrosion defect. The
increase in the anodic current density for higher strains
could be explained by the plastic deformation
observed at the center of the corrosion defect.
(c)
Figure 8. Corrosion potential under different defect depths:
(a) 2 mm, (b) 6 mm, and (c) 10 mm.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 6, 2
6, 4
6, 6
6, 8
6, 10

Figure 10. Linear distribution of anodic current density Figure 12. Linear distribution of corrosion potential along
along the length of the corrosion defect under various the bottom of corrosion defect with a fixed length of 3 mm
longitudinal tensile strains: soil displacements of 1, 2, 3, and and with various depths.
4 mm.

2, 6
4, 6
6, 6
8, 6
10, 6

Figure 11. Linear distribution of cathodic current density Figure 13. Linear distribution of corrosion potential along
along the length of the corrosion defect under various the bottom of corrosion defect with a fixed depth of 3 mm
longitudinal tensile strains: soil displacements of 1, 2, 3, and and with various lengths.
4 mm.

Figure 11 shows the linear distribution of cathodic that the increase of the defect depth decreases the
current density (hydrogen evolution) along the length corrosion potential along the defect location. In
of the corrosion defect under various longitudinal addition, it is shown that at low defect depths of 2 and
tensile strains: soil displacements of 1, 2, 3, and 4 mm. 4 mm, the corrosion potential distribution is almost
This figure shows that the increase in the tensile strain uniform. However, when the defect depth increases to
decreases the cathodic current density, especially at 6, 8, and 10 mm, the corrosion potential is not uniform
the center of the corrosion defect where it is the most anymore with the highest corrosion potential at the
negative. Similar to the anodic current density center of the defect.
distribution results, the nonuniformity in the cathodic
current density increases when the tensile strain Figure 13 shows the linear distribution of corrosion
increases. potential along the bottom of corrosion defect with a
fixed depth of 6 mm and with various lengths where a
Figure 12 shows the linear distribution of corrosion 1 mm longitudinal tensile strain is applied. This figure
potential along the bottom of corrosion defect with a shows that the increase of the defect length increases
length of 6 mm and with various depths where a 1 mm the corrosion potential along the defect location.
longitudinal tensile strain is applied. This figure shows

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

4.2 Pitting Corrosion model

This section provides the results of the pitting

corrosion simulations described in the previous
sections. The purpose of this section is to demonstrate
the capability of this model by showing the predicted
pitting corrosion rate and depth given soil and pipe
characteristics. A set of field data of underground gas
pipelines collected across southern Mexico including
soil and pipe data were used for demonstration (partly
from [11]). Table 3 lists the data in a Distribution Type
(mean, variance) format for every variable except for
the coating type which is a constant. Monte Carlo
(MC) simulation method was used to estimate the Figure 14. Distribution of pitting rate for underground gas
corrosion predictions in terms of rate and depth. pipelines after 20 years of operation.

Table 3. Soil and pipe data of underground gas pipelines

collected across southern Mexico

Variable, symbol (units) Probability function

Resistivity, re (Ω-m) Lognormal (50, 2931)
Sulphate, sc (ppm) Lognormal (154, 25328)
Bicarbonate, bc (ppm) Lognormal (19, 436)
Chloride, cc (ppm) Lognormal (41, 3135)
Water content, wc(%) Normal (24, 38)
pH, ph Gumbel (6.13, 0.84)
Pipe/soil potential, pp (V)1 Normal (-0.86, 0.04)
Bulk density, bd (g/ml) Normal (1.30, 0.007)
Figure 15. Distribution of maximum pitting depth for
Redox potential, rp (mV)2 Uniform (2.14, 348) underground gas pipelines after 20 years of operation.
Deterministic function
Coating type, ct Constant (0.7651) 4.3 External Corrosion BN model
Operating time, t Constant (20)
1
Relative to a Cu/CuSO4 reference electrode.
2 A cause-consequence relationship external corrosion
Relative to the standard hydrogen electrode.
model was developed to predict external corrosion
4.2.1 Pitting Rate considering both SCC and pitting corrosion as shown
Figure 14 displays the distribution of pitting rate for in Figure 5. A small portion of the external corrosion
underground gas pipelines after 20 years of operation. BN model was highlighted and shown in Figure 16 to
The distribution has a long tail in high pitting rate illustrate how the model works. The conditional
region in a reflection of the uncertainties of soil and probability table was developed by the models for
SCC and pitting corrosion described in this paper and
pipe parameters. Although the mean pitting rate is
expert knowledge. The comprehensive details of each
roughly 0.03 mm/y, the maximum pitting rate can be
discretized node in the BN model were listed in the
as high as 0.27 mm/y.
authors’ previous work for reader’s reference [2].
4.2.2 Pitting Depth An example of the Bayesian calculations shown in
Figure 15 displays the distribution of maximum pitting Figure 16 are displayed in Table 4. All nodes in this
depth for underground gas pipelines after 20 years of BN model are discrete. Also, the states for each node
operation. As pitting depth is the accumulative are uniformly distributed. The results shows that when
consequence of pitting rate over time, the distribution the coating type is more likely to be coal-tar with
also has a long tail in high maximum pitting depth 44.3% probability; pipe-soil potential is more likely to
region. The mean maximum pitting depth is 0.96 mm, between -1.0 to -0.5 V with 72.1% probability; bulk
but the maximum pitting depth is as high as 7.08 mm. density is more likely to between 1 to 1.5 g/ml with
99.1% probability; water content is more likely to be
Although the maximum pitting depth is less likely to
between 0 to 25% with 56.9% probability; the
happen, it should still be treated seriously because the
exponent of the pitting model is likely to be between
location with the highest pitting is usually where the
0.5 to 0.75 with 73.5% probability.
failure happens.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 are used. The pipeline is 1000 km in length and 10 mm
in thickness. The type of the pipe material is X100
pipeline steel. Only coating is applied to the pipeline
as a protection against external corrosion, and it has
been operating for 20 years. Due to the lack of
information regarding corrosion initiation time, it was
assumed that the corrosion defect started to grow after
the pipeline was operated. In addition, it should be
noted that the SCC model developed in COMSOL
Multiphysics can be used to predict the corrosion rate
but not the corrosion depth. Therefore, the corrosion
Figure 16. A small portion of the external corrosion BN defect as a result of SCC was assumed to grow linearly
model. with respect to time for simplicity. For external
corrosion, leakage is considered a most common
Table 4. Results of Bayesian calculation for Figure 16
failure; therefore, the probability of failure due to
leakage was calculated assuming the leakage will
Parameters Value Certainty (%)
happen when 80% of the wall thickness is lost due to
Bare pipe 11.8 corrosion. The inputs for the external corrosion BN
Asphalt enamel 2.4 model are summarized in Table 5.
Coating type Wrap-tape 38.2
Coal-tar 44.3 Table 5. Soil and pipe data of a demonstrative underground
FBE 3.3 gas pipeline subject to external corrosion
(-2.0)-(-1.5) 0.1
Pipe-to-soil
(-1.5)-(-1.0) 23.9 Certainty
potential Parameters Value
(-1.0)-(-0.5) 72.1 (%)
(V)
(-0.5)-0 3.9 Bare pipe 11.8
0-0.5 0 Asphalt
2.4
Bulk density 0.5-1 0.1 enamel
Coating type
(g/ml) 1-1.5 99.1 Wrap-tape 38.2
1.5-2 0.8 Coal-tar 44.3
0-25 56.9 FBE 3.3
Water content 25-50 43.1 (-2.0)-(-1.5) 0.1
(%) 50-75 0 Pipe-to-soil potential (-1.5)-(-1.0) 23.9
75-100 0 (V) (-1.0)-(-0.5) 72.1
0-0.25 0 (-0.5)-0 3.9
0.25-0.5 6.5 0-0.5 0
Exponent Bulk density 0.5-1 0.1
0.5-0.75 73.5
0.75-1 20 (g/ml) 1-1.5 99.1
1.5-2 0.8
The same treatment was performed for the other nodes 0-25 56.9
in a cause-consequence relationship. The overall Water content 25-50 43.1
model tracks much more nodes than the demonstrated (%) 50-75 0
one; each of which has direct or indirect effects on 75-100 0
external corrosion. The comprehensive details of each 1-250 99.94
node in the BN model were listed in the authors’ Soil resistivity 250-500 0.04
previous work for reader’s reference [2]. (Ω-m) 500-750 0.02
750-1000 0
5. Case Study 0-10 0.44
Sulphate content 10-100 39.28
For the sake of model demonstration on an operating (ppm) 100-1000 58.52
underground gas pipeline subject to external 1000-2500 1.76
corrosion, the developed external corrosion BN model 0-10 29.12
was used to calculate the corrosion rate and depth of Bicarbonate content 10-100 69.58
an underground gas pipeline via a case study. The (ppm) 100-1000 1.30
experimental data available in the references [8, 11]
1000-4000 0

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 0-100 89.88 requires hundreds to thousands of simulations, once it
100-1000 10.12 is generated no more simulation is needed. There are
Chloride content several advantages including that if evidence of a node
1000-10000 0
(ppm) is available (certainty is 100%), the predictive result
10000-
0 can then be updated. On the other hand, even though
100000
4-5 0 the information of a certain node is unavailable, BN
5-6 4.74 model can simply take the uncertainty into account by
pH 6-7 45.34 assigning all values of that node with equal
7-8 34.86 probabilities. In addition, the cause-consequence
8-9 15.06 relationship feature allows it to integrate many kinds
1-100 28.40 of data, namely, physics-based model, field data,
expert knowledge, and literature data, among which
100-200 35.58
Redox potential the physics-based model is more reliable but also rare
200-300 35.66
and that is the motivation that this paper uses
300-400 0.36
COMSOL Multiphysics to generate physics-based
(-10-4) – (- simulation data of SCC.
10-3)
0
(-10-3) – (- Table 6. Results of Bayesian calculation for a demonstrative
Cathodic i01 50
10-2) underground gas pipeline subject to external corrosion
(A/m2) 50
(-10-2) – (-
0
10-1) Parameters Value Certainty (%)
(-10-1) – (-1) 0-0.01 0.25
10-4 – 10-3 50 External 0.01-0.1 99.34
Anodic i01 10-3 – 10-2 50 Corrosion Rate 0.1-1 0.41
(A/m2) 10-2 – 10-1 0 (mm/y) 1-5 0
10-1 – 1 0 5-10 0
0-0.1 0 0-0.1 0.
Defect Depth 0.1-1 55 External 0.1-1 0
(mm) 1-10 45 Corrosion Depth 1-5 99
10-20 0 (mm) 5-10 1
0-0.1 0 10-20 0
Defect Length 0.1-1 55 0-20 99
(mm) 1-10 45 Probability of 20-40 0.51
10-20 0 Failure due to 40-60 0.49
10-2 – 10-1 0 leakage (%) 60-80 0
Strain/Displacement2
10-1 – 1 50 80-100 0
(mm)
1 – 10 50
1
Cathodic i0 and Anodic i0 are exchange cathodic current
density and exchange anodic current density respectively,
6. Conclusions
which are related to operating conditions such as In conclusion, this paper presented an external
temperature, pH, etc. corrosion modeling for an underground natural gas
2
Strain/Displacement is the longitudinal strain of the pipe pipeline subject to Stress Corrosion Cracking (SCC)
due to soil movements. and pitting corrosion. For the SCC, the model is
implemented using the Solid Mechanics and the
The results of the demonstrative gas pipeline are Secondary Current Distribution interfaces in
shown in Table 6. The corrosion rate is likely to fall COMSOL Multiphysics. This model studied the
into the range between 0.01 to 0.1 mm/y, while the impact of elastoplastic deformations on
corrosion depth is likely to fall into the range between electrochemical reactions. It was shown that the
1 to 5 mm after 20 years of operation. The pipe is not maximum corrosion potential along the defect length
in urgent demand for repair at this time as there are decreases with the increasing defect depth, but
99% chance that the probability of failure due to increases with increasing defect length, and that the
leakage is under 20%. maximum von Mises stress increases with increasing
The developed BN model can be easily applied to defect depth but decreases with increasing defect
other similar gas pipelines for external corrosion length, which indicates that the corrosion is most
prediction because although the development of severe at the center of the defect. That trend can also
conditional probability tables of the BN model be seen by quantifying the corrosion level into

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

corrosion rate. For the pitting corrosion, the applied [5] S.Nešic andJ.Postlethwaite, “A predictive
pitting corrosion model is able to predict the pitting model for localized erosion—corrosion,”
corrosion rate given soil and pipe data at different Corrosion, vol. 47, no. 8, pp. 582–589, 1991.
types of soil conditions. Finally, the proposed BN [6] B. F. M.Pots et al., “Improvements on de
model considering both SCC and pitting corrosion has Waard-Milliams corrosion prediction and
shown its potential in predicting external corrosion applications to corrosion management,” in
rate and depth of underground gas pipelines. The CORROSION 2002, 2002.
advantages such as flexibility of updating with [7] K.Zaman andR. P.Wei, “Probability approach
evidence and ability of taking uncertainty of inputs for prediction of corrosion and corrosion
makes the model more flexible and easier for field fatigue life,” AIAA J., vol. 32, no. 10, pp.
application to aid decision makings. 2073–2079, 1994.
[8] L. Y.Xu andY. F.Cheng, “Development of a
Acknowledgements finite element model for simulation and
prediction of mechanoelectrochemical effect
The authors would like to acknowledge the of pipeline corrosion,” Corros. Sci., vol. 73,
sponsorship of the Petroleum Institute, Khalifa pp. 150–160, 2013.
University of Science and Technology, Abu Dhabi, [9] L.Xu, “Assessment of corrosion defects on
UAE and the University of Maryland (Department of high-strength steel pipelines.” University of
Mechanical Engineering) for the research work Calgary, 2013.
presented in this paper, which is part of the Pipeline [10] J. C.Velázquez, F.Caleyo, A.Valor, andJ.
System Integrity Management research project. In M.Hallen, “Predictive model for pitting
addition, the authors would like to acknowledge the corrosion in buried oil and gas pipelines,”
SEAS-net Computing Facility and UCLA for proving Corrosion, vol. 65, no. 5, pp. 332–342, 2009.
the license to use COMSOL Multiphysics, which [11] F.Caleyo, J. C.Velázquez, A.Valor, andJ.
supported the numerical study. M.Hallen, “Probability distribution of pitting
corrosion depth and rate in underground
References pipelines: A Monte Carlo study,” Corros.
Sci., vol. 51, no. 9, pp. 1925–1934, 2009.
[1] O.Shabarchin andS.Tesfamariam, “Internal
corrosion hazard assessment of oil & gas
pipelines using Bayesian belief network
model,” J. loss Prev. Process Ind., vol. 40,
pp. 479–495, 2016.
[2] W.Chalgham, M.Diaconeasa, K.-Y.Wu,
andA.Mosleh, “A Dynamic Pipeline Network
Health Assessment Software Platform for
Optimal Risk-based Prioritization of
Inspection, Structural Health Monitoring, and
Proactive Management,” Proceedings of the
ASME 2019, International Mchanical
Engineering Congress and Exposition
IMECE2019, November 11-14, 2019, Salt
Lake City, UT, USA.
[3] K. Y.Wu andA.Mosleh, “Effect of Temporal
Variability of Operating Parameters in
Corrosion Modelling for Natural Gas
Pipelines Subjected to Uniform Corrosion,”
J. Nat. Gas Sci. Eng., p. 102930, 2019.
[4] S.Papavinasam, A.Doiron, andR. W.Revie,
“Model to predict internal pitting corrosion of
oil and gas pipelines,” Corrosion, vol. 66, no.
3, pp. 35006–35011, 2010.

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

Appendix

Table A1. High-strength alloy steel material properties

Name Value Unit

Density 7850 kg/m³
Young's modulus 207e9 Pa
Poisson's ratio 0.33
Initial yield stress 806e6 Pa
Relative permeability {{1, 0, 0}, {0, 1, 0}, {0, 0, 1}}
Heat capacity at constant pressure 475 J/(kg*K)]
Thermal conductivity {{44.5, 0, 0}, {0, 44.5, 0}, {0, 0, 44.5}} W/(m*K)
Electrical conductivity {{4.032e6, 0, 0}, {0, 4.032e6, 0}, {0, 0, 4.032e6}} S/m
Relative permittivity {{1, 0, 0}, {0, 1, 0}, {0, 0, 1}}
Coefficient of thermal expansion {{12.3e-6, 0, 0}, {0, 12.3e-6, 0}, {0, 0, 12.3e-6}} 1/K
Lamé parameter λ 1.15e11 Pa
Lamé parameter μ 7.69e10 Pa

Table A2. Parameters used for the simulation

Name Expression Value Description

disp 0.001 [m] 0.001 m Displacement
Eeq0_fe -0.859 [V] −0.859 V Equilibrium potential for iron
dissolution vs SCE in absence of
stress
Eeq0_h -0.644[V] −0.644 V Equilibrium potential for hydrogen
evolution vs SCE in absence of stress
i0a 2.353e-3[A/m^2] 0.002353 A/m² Exchange current density for iron
dissolution
ba 118[mV] 0.118 V Tafel slope for iron dissolution
i0c_h 1.457e-2[A/m^2] 0.01457 A/m² Exchange current density for
hydrogen evolution
bc -207[mV] −0.207 V Tafel slope for hydrogen evolution
deltaPm 806e6[Pa]/3 2.6867E8 Pa Excess pressure to elastic
deformation
Vm 7.13e-6 [m^3/mol] 7.13E−6 m³/mol Molar volume of steel
zm 2 2 Charge number
T 298.15[K] 298.15 K Temperature
nu 0.45 0.45 Orientation dependent factor
alpha 1.67e11[1/cm^2] 1.67E15 1/m² Coefficient
N0 1e8[1/cm^2] 1E12 1/m² Initial dislocation density
deltaEeqae -(deltaPm*Vm/(zm*F_const)) −0.0099269 V Change in equilibrium potential due
elastic deformation
sigmal 0.096 [S/m] 0.096 S/m Soil conductivity
defect_depth 5 [mm] 0.005 m Initial defect depth
defect_length 5 [mm] 0.005 m Initial defect length

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

Table A3. Variables used for the simulation

Name Expression Unit Description

hardening max(0,stress_strain_curve(solid.epe+solid.mises/solid Pa Hardening function
.E) - solid.sigmags)
deltaEeqap -T*R_const/(zm*F_const)*log((nu*alpha*solid.epe) V Change in anode
/N0 + 1) equilibrium potential due
plastic deformation
Eeqa Eeq0_fe+deltaEeqae+deltaEeqap V Anode equilibrium potential
including elastic and plastic
deformation terms
ic_h i0c_h*10^(-solid.mises*Vm/(6*F_const*bc)) A/m² Cathode exchange current
density including stress
factor

Table A4. Mesh Properties

Description Value
Minimum element quality 0.6912
Average element quality 0.9424
Triangle 6013
Edge element 697
Vertex element 9
Maximum element size for the soil domain 0.067
Minimum element size for the soil domain 3.0E-4
Curvature factor 0.3
Maximum element growth rate 1.3
Maximum element size for the pipeline domain 0.002
Minimum element size for the pipeline domain 6.0E-4

Table A5. Results of sensitivity analysis of SCC model for fixed cathodic i0 at 0.01457 A/m2 and fixed anodic i0 at
0.002353 A/m2 where the defect length and depth were varied

Anodic Cathodic Maximum

Defect Defect Corrosion
Simulation Displacement current current corrosion
length depth rate
# (mm) density density potential
(mm) (mm) (mm/y)
(A/m2) (A/m2) (V)
1 1 2 2 0.0351 -0.0381 -0.7204 0.0409
2 2 2 2 0.0354 -0.0382 -0.7201 0.0411
3 3 2 2 0.0355 -0.038 -0.7199 0.0412
4 1 2 4 0.0349 -0.0382 -0.7208 0.0405
5 1 2 6 0.0348 -0.0383 -0.7209 0.0404
6 1 2 8 0.0347 -0.0385 -0.7209 0.0404
7 1 2 10 0.0347 -0.0386 -0.7210 0.0404
8 2 2 10 0.0349 -0.0383 -0.7207 0.0406
9 3 2 10 0.0349 -0.0382 -0.7207 0.0405
10 1 2 16 0.0349 -0.0384 -0.720 0.0406
11 1 4 2 0.0353 -0.037 -0.7201 0.0410

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston

 12 1 4 4 0.0351 -0.038 -0.7204 0.0409
13 1 4 6 0.0350 -0.0382 -0.7206 0.0407
14 1 4 8 0.0349 -0.0383 -0.7207 0.0406
15 1 4 10 0.0349 -0.0384 -0.7208 0.0405
16 1 6 2 0.0353 -0.0363 -0.7201 0.0411
17 1 6 4 0.0353 -0.03775 -0.7202 0.0410
18 1 6 6 0.0351 -0.038 -0.7204 0.0409
19 1 6 8 0.0350 -0.003825 -0.7205 0.0408
20 1 6 10 0.0350 -0.0385 -0.7206 0.0407
21 1 8 2 0.0353 -0.036 -0.7201 0.0410
22 1 8 4 0.0353 -0.037 -0.7201 0.0410
23 1 8 6 0.0353 -0.0378 -0.7202 0.0410
24 1 8 8 0.0352 -0.038 -0.7203 0.0409
25 1 8 10 0.0351 -0.03825 -0.7204 0.0408
26 1 10 2 0.0353 -0.03555 -0.7201 0.0410
27 1 10 4 0.0353 -0.0365 -0.7201 0.0411
28 1 10 6 0.0353 -0.0375 -0.7201 0.0410
29 1 10 8 0.0353 -0.038 -0.7202 0.0410
30 1 10 10 0.0352 -0.0382 -0.7203 0.0409
31 1 20 2 0.0352 -0.0354 -0.7203 0.0409
32 2 20 2 0.0360 -0.0364 -0.7191 0.0418
33 3 20 2 0.0368 -0.0371 -0.7181 0.0427

Table A6. Results of sensitivity analysis of SCC model for fixed defect length and depth of 6 mm and fixed
displacement of 1 mm where the exchange anodic and cathodic current densities were varied

Exchange Exchange Anodic Cathodic

Corrosion Corrosion
Simulation anodic current cathodic current current current
Potential rate
# density density density density
(V) (mm/y)
(A/m2) (A/m2) (A/m2) (A/m2)
1 0.001353 0.00457 -0.7402 0.0137 -0.0151 0.0159
2 0.001353 0.01457 -0.7023 0.0288 -0.0311 0.0334
3 0.001353 0.02457 -0.6852 0.0401 -0.0438 0.0466
4 0.002353 0.00457 -0.7582 0.0168 -0.0183 0.0195
5 0.002353 0.01457 -0.7204 0.0351 -0.0384 0.0408
6 0.002353 0.02457 -0.7033 0.0490 -0.0538 0.0569
7 0.003353 0.00457 -0.7698 0.0191 -0.0214 0.0222
8 0.003353 0.01457 -0.7319 0.0399 -0.0432 0.0463
9 0.003353 0.02457 -0.7148 0.0557 -0.0611 0.0647

Excerpt from the Proceedings of the 2019 COMSOL Conference in Boston