Predictive Model for Pitting Corrosion in Buried Oil and Gas Pipelines

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

CORROSION ENGINEERING SECTION

Predictive Model for Pitting Corrosion in Buried Oil


and Gas Pipelines

J.C. Velázquez,* F. Caleyo,‡,* A. Valor,** and J.M. Hallen*

ABSTRACT INTRODUCTION
A predictive model for pitting corrosion in buried pipelines Pitting is recognized as the most severe type of corro-
is proposed. The model takes into consideration the chemi- sion because of the high rate at which pits can grow
cal and physical properties of the soil and pipe to predict in pressure vessels and pipelines. Statistical stud-
the time dependence of pitting depth and rate. Maximum pit
ies have shown that failures in underground oil and
depths were collected together with soil and pipe data at more
gas pipelines in Mexico are most likely due to leak-
than 250 excavation sites over a three-year period. The time
age caused by external pitting corrosion.1 In practice,
dependence of the maximum pit depth was modeled as dmax
= κ(t − t0)ν, where t is the exposure time, t0 is the pit initiation this failure mechanism is extremely difficult to model
time, and κ and ν are the pitting proportionality and exponent on the basis of electrochemical reactions. Therefore,
parameters, respectively. A multivariate regression analysis predictive models for pit growth on underground steel
was conducted with dmax as the dependent variable and the pipelines have been developed mainly using statistical
pipeline age, and the soil and pipe properties as the indepen- approaches instead of electrochemical principles.2-9
dent variables. The dependence of κ and ν on the predictor The statistical approaches available in the litera-
variables was found for the three soil textural classes identi- ture for modeling pitting corrosion use maximum pit
fied in this study: clay, clay loam, and sandy clay loam. The depth as the dependent corrosion variable and envi-
proportionality parameter κ was found to be primarily influ-
ronmental factors as the independent variables.3,7-9
enced by the redox potential, pH value, soil resistivity, and the
Thus, the three essential elements for developing a
dissolved ion concentrations. In contrast, the pitting exponent
ν was found to be influenced mainly by the pipe-to-soil poten-
statistical pit growth model are the law governing the
tial, water content, bulk density, and the pipe coating type. A time dependence of maximum pit depth, the data col-
real-life pipeline integrity assessment is used as a case study lected on environmental factors, and the mathemati-
to illustrate the application of the proposed model and to show cal model used to correlate the measured pit depth to
how it can have a positive impact on integrity management these factors.
programs. The generally accepted form of the dependence
of maximum pit depth, dmax, on exposure time, t, was
KEY WORDS: corrosion, modeling, pipelines, pitting, soils
originally developed by Romanoff,10 who, based on
corrosion data for the long-term (17-year) corrosion of
buried ferrous pipes in a range of soils, proposed that:
Submitted for publication July 2008; in revised form, December
2008. ν
d ma x = κt (1)

Corresponding author. E-mail: fcaleyo@gmail.com. max
* Departamento de Ingeniería Metalúrgica, ESIQIE, IPN, UPALM
Edif. 7, Zacatenco, México D.F. 07738.
** Facultad de Física, Universidad de La Habana, San Lázaro y L, where κ and ν are constant regression parameters.
Vedado, 10400, La Habana, Cuba. From this equation, the influence of soil characteris-

ISSN 0010-9312 (print), 1938-159X (online)


332 09/000057/$5.00+$0.50/0 © 2009, NACE International CORROSION—MAY 2009
CORROSION ENGINEERING SECTION

tics on pit growth can be taken into consideration by pipe coating and pipe-to-soil potential (natural or
using different model parameters from one soil envi- imposed by the cathodic protection) on this damage
ronment to another. Values of ν ranging from 0.33 to mechanism. The possible influences of the soil and
0.70 have been reported elsewhere.4-5,10 pipe characteristics on the pitting initiation time and
Based on Romanoff’s data, Mughabghab and on the pitting exponent also have not been investi-
Sullivan3 investigated the dependence of κ and ν on gated.
the soil properties for carbon steels. A linear correla- The motivation for the present study is the need
tion analysis of κ as a function of soil pH and resistiv- for a more complete and accurate predictive model for
ity revealed that this parameter is influenced primarily pit growth on buried oil and gas pipelines. The goal is
by the pH value of the soil. Also, a linear correlation to propose an empirical prediction formula of maxi-
analysis revealed that the exponent parameter is mum pitting depth using soil and pipe variables. More
determined primarily by the moisture and the clay than 250 maximum pit depths have been measured,
fraction of the soil. The authors concluded that κ and together with the respective soil and pipe data, at
ν are essentially identical for carbon steel, wrought exposed pipeline segments in southern Mexico. Equa-
iron, and open-hearth steel, and that soil environment tion (2) defines the maximum pit depth model used in
determines the amount of corrosion. The roles of other this work:
soil and pipe characteristics that signifycantly impact
ν
pit growth, such as soil resistivity, were found to be d ma
maxx = κ(t – t 0 ) (2)
relatively unimportant. Some factors were not consid-
ered, specifically dissolved chloride concentration, This model differs from that of Equation (1) in the
redox potential, and pipe coating. use of the pit initiation time, t0, which is treated as an
Soil texture(1) is one of the first factors to be con- unknown parameter to be determined as part of the
sidered during corrosion surveys because it deter- analysis. A multivariate regression analysis was per-
mines the degree of aeration and the permeability of formed with dmax as the dependent variable and the
the soil.11 In this regard, Oguzie, et al.,9 recently have exposure time, and the soil and pipe characteristics
analyzed how soil texture influences pitting corrosion as the independent variables. In contrast to previous
of buried steel coupons in a range of soil textures. studies, the prediction model obtained in this work
Using weight-loss analysis through corrosion prod- reflects the particular influence that these predictor
uct count techniques, these authors concluded that variables exert on both the pitting proportionality (κ)
the order of corrosiveness is clay > loam > sand. Their and exponent (ν) parameters. Also, besides the com-
work, however, does not provide any empirical model mon prediction formula derived for all the collected
for relating soil texture to pit growth. samples, a predictive model was obtained for each
Katano, et al.,8 performed a multivariate regres- one of the three soil textural classes identified in this
sion analysis of pitting depth on soil characteristics study: clay, clay loam, and sandy clay loam.
for underground cast iron pipes. Approximately 880 A real-life integrity assessment is used as a case
samples of maximum pit depth and environmental study to show that the proposed model predicts pit
factors were collected at exposed cast iron pipe sec- growth more accurately than the linear growth model
tions from five areas in Japan, and an extensive spec- commonly used in pipeline integrity analyses. This
trum of soil variables was considered (see Table 1 in improvement is expected to have a positive impact
Reference 8). An empirical prediction formula of pit- on integrity management plans addressing the threat
ting depth was proposed based on the power law posed by external pitting corrosion in buried pipelines.
model of Equation (1). The parameter κ was expressed
as a function of the soil characteristics, while ν was EXPERIMENTAL PROCEDURES
assumed to be a constant regression parameter. The
predictor variables for maximum pitting depth were A total of 259 samples of the soil and pipeline
proposed to be the soil type, land condition, soil resis- variables shown in Table 1 were obtained at dig sites
tivity, pH, redox potential, sulfide content, and expo- over a three-year period, for onshore buried pipelines
sure time. operating in southern Mexico. Any corrosion-caused
Despite the significant contribution by Katano metal loss with a diameter equal to or less than two
and coworkers, some issues relevant to pitting cor- times the pipe wall thickness was assumed to be pit-
rosion of buried pipelines still remain and should be ting. The maximum pit depth in each exposed pipeline
properly addressed. Of these, one of the most impor- segment was measured using a micrometer pit depth
tant unaddressed questions is the influence of the gauge attached to a bridging bar. Soil properties like
redox potential, pH, pipe-to-soil potential, and soil
(1)
Throughout this work, the term “soil texture” is used to indicate resistivity were obtained by in situ measurements.
the mineral content of the soil based on a weight ratio of the three Soil samples taken from the pipe ditch were brought
soil separates: sand, silt, and clay.11 The terms “soil type,” “soil
category,” and “soil class” are indistinctly used to refer to soil tex- to the laboratory to be analyzed for texture, water
tural class. content, bulk density, and dissolved chloride, bicar-

CORROSION—Vol. 65, No. 5 333


CORROSION ENGINEERING SECTION

table 1
Investigated Soil and Pipe Characteristics
Measured Variable Symbol Unit        Instrument or Method
Maximum pit depth dmax mm Micrometer pit gauge
Exposure time t years Operating records
Redox potential rp mV ORP meter
pH ph — pH meter
Pipe-to-soil potential pp mV Cu/CuSO4 electrode
Soil resistivity re Ω·m Wenner method (ASTM G57-06)12
Water content wc % Thermogravimetric balance (ASTM D2216-05)13
Soil bulk density bd g/mL ASTM 512G Bouyoucos hydrometer (ASTM D422-63[2007])14
Chloride content cc ppm Titration (ASTM D512-04)15
Bicarbonate content bc ppm Titration (ASTM D1126-02[2007]e1)16
Sulfate content sc ppm Spectrophotometry (ASTM D516-07)17
Coating type ct — Direct observation

Table 2
Non-Parametric Description of the Measured Variables
All Data Clay Sandy Clay Loam Clay Loam
Variable Mean Cov.(A) Min. Max. Mean Cov. Min. Max. Mean Cov. Min. Max. Mean Cov. Min. Max.
dmax 2.02 1.02 0.41 13.4 2.34 0.88 0.41 8.56 1.25 0.80 0.41 5.87 2.03 1.01 0.41 10.4
t 22.9 0.40 5.0 50.0 24.4 0.35 6.0 50.0 18.9 0.36 5.0 39.0 24.6 0.43 5.0 50.0
rp 167 0.51 2.14 348 177 0.50 2.14 348 169 0.56 20 339 158 0.44 19 301
pH 6.13 0.15 4.14 9.88 5.93 0.16 4.14 9.88 6.24 0.13 4.48 7.80 6.34 0.14 4.85 9.36
pp –0.88 0.27 –1.97 –0.42 –0.86 0.28 –1.95 –0.42 –0.95 0.23 –1.97 –0.65 –0.82 0.25 –1.53 –0.47
re 50.2 1.11 1.91 399 61.1 1.07 3.81 399 49.2 0.99 4.26 243 28.2 0.83 1.91 113
wc 23.9 0.27 8.7 66.0 24.0 0.28 8.7 57.5 22.4 0.25 11.3 38.5 24.7 0.21 15.2 37.2
bd 1.30 0.07 1.10 1.56 1.23 0.04 1.11 1.36 1.40 0.04 1.30 1.55 1.32 0.02 1.25 1.38
cc 47.7 1.57 0.0 672 53.1 1.28 6.6 351 21.8 1.08 3.2 185 44.6 1.20 3.5 290
bc 19.6 1.29 0.0 195 19.2 1.31 3.91 156 13.7 0.44 5.1 30.0 22.8 1.03 0.0 104
sc 153 1.10 0.0 1,370 129 0.87 0.0 698 143 0.69 14.9 452 205 1.23 11.3 1,370
ct 0.77 0.17 0.30 1.00 0.77 1.13 0.30 1.00 0.77 0.21 0.30 1.00 0.77 0.17 0.30 1.00
(A)
Absolute value of the covariance.

bonate, and sulfate ion concentrations. In Table 1, the eters κ and ν were treated as a linear combination of
instruments or methods used to measure the soil and the soil and pipe characteristics under the constraint
pipe variables are listed. The pipeline age and coating that each predictor variable could only be associated
type were also recorded. A scoring model was used to with one of the parameters. The best regression model
account for the influence of the pipe coating type on was chosen on the basis of the value of the (multiple
pit growth. regression) correlation coefficient, R,18 taking care to
The collected soil samples were categorized first keep a consistent model for all soil categories.
into the textural classes used by the U.S. Department
of Agriculture (USDA, Washington, DC). The soil tex- RESULTS
ture was defined on the basis of the weight ratio of
the three soil separates: sand, silt, and clay.11 The Data Analysis
mineral content of the soil was determined using the Soil Categorization — The soil categories identified
Bouyoucos hydrometer method (ASTM 152H hydrom- from the collected samples were:
eter).14 All subsequent analyses were conducted for —clay (110 samples)
each soil category and for a generic category that —sandy clay loam (79 samples)
included all collected samples. Standard exploratory —clay loam (61 samples)
statistical methods were used to analyze the distribu- —silty clay loam (6 samples)
tion of the measured variables, determine the pres- —silty clay (2 samples)
ence of outliers, and examine the correlation between —silt loam (1 sample)
the variables. Multivariate regression analyses were A non-parametric description of the considered vari-
conducted based on the model of Equation (2), using ables is shown in Table 2 for all observed samples (All)
the computer statistical software Statistical Pack- and for the clay (C), sandy clay loam (SCL), and clay
age for Social Sciences (SPSS†).18 The pitting param- loam (CL) samples. These were the only categories
with a large enough number of samples to ensure sta-

Trade name. tistical significance.

334 CORROSION—MAY 2009


CORROSION ENGINEERING SECTION

From the mean values of the variables given


in Table 2, it can be inferred that the investigated
pipeline population has suffered from long-term
(~23 years) exposure to moderately corrosive envi-
ronments. Although the soils under investigation are
types of clay, the moderated potential for corrosive-
ness can be assumed based on the fact that, on aver-
age, the population has been exposed to weakly acidic
soil environments with a relatively high resistivity
(~5,000 Ω·cm), a redox potential above 100 mV, and a
pipe-to-soil potential (natural or imposed) close to the
optimum level of cathodic protection.11,19
It is important to note that the uncertainty in the
measurements is large, with the exception of mea-
surements of bulk density. Among the dependent vari-
ables, soil resistivity and dissolved ion concentrations
show the largest uncertainties. Besides the unavoid-
able experimental errors, the seasonal fluctuations FIgURe 1. Experimental and fitted distributions for all measured
of the soil characteristics might play a role in deter- maximum pit depths.
mining the large covariance associated with field and
laboratory measurements. In this regard, it is worth
noting that the potential variations in soil properties TAble 3
from one season to the next have been incorporated, Coating Type Frequency by Soil Category
to some degree, into the analysis as data collection Soil Category
was carried over a span of three years. Coating (ct) C Cl SCl All
Maximum Pit Depth Distribution — Figure 1 shows
Bare pipe 0.102 0.119 0.194 0.118
the distribution of all measured maximum pit depths. Asphalt enamel 0.019 0.017 0.039 0.024
The distinctive right-skewed, leptokurtic distribution Wrap tape 0.421 0.288 0.429 0.382
commonly observed for maximum pitting corrosion Coal tar 0.449 0.542 0.273 0.443
penetration was also observed in this study. As mea- FBE 0.009 0.034 0.065 0.033
sured by the Kolmogorov-Smirnov statistic (D), the
best fitting distribution (D = 0.071) was the general-
ized extreme value distribution GEVD(ζ, σ, µ): made in a wide variety of soil environments and expo-
sure periods. Thus, these results only suggest that
 –1  the maximum depths are sampled from a population
   χ – µ ξ
GEVD( x ) = exp
exp  – 1 + ξ    (3) in which, for a sufficiently long exposure, the deep-
   σ   
 est pits are the most likely to grow.20-21 This makes the
tail at large depths more extensive and leads to the
where ζ, σ, and µ are the shape, scale (mm–1), and distinguishing right-skewed, leptokurtic-type distribu-
location (mm) parameters, respectively. The value of tion observed in Figure 1. Because the maximum pit
ζ (0.422) indicates that the Fréchet distribution is depth is recorded at each exposed pipeline segment, it
the most likely distribution for describing the maxi- is not surprising that, instead of a lognormal distribu-
mum pit depth data. For the sake of comparison, tion, one of the three extreme value distributions for
the lognormal distribution, LogND(σ, µ), fitted to the maxima arises as the most likely distribution in Fig-
observed data is also presented. In this case, σ and µ ure 1.
represent the mean and standard deviation of the nat- Scoring Model for Pipeline Coating — Field obser-
ural logarithm of the variable depth, respectively. vations confirmed that most of the exposed pipeline
The lognormal distribution has been reported by segments were protected with anticorrosive coating.
Katano, et al.,8 as the best option in describing the Coatings were categorized into five major categories
distribution of maximum pit depth data of cast iron according to their type: fusion-bonded epoxy (FBE),
pipes in soil. According to Figure 1 and the corre- coal-tar, single- or double-wrapped polyolefin tape,
sponding goodness of fit test (D = 0.071), the lognor- asphalt enamel, and non-coated or bare pipe. Table
mal distribution can also provide a good description 3 shows the frequency with which each coating type
of the maximum pit depths measured in this study. was observed in each soil category.
However, it is worth mentioning that in both works, Because coating disbondment is a precursor to
Katano’s and the present one, the statistical fitting pit initiation and growth, pipeline coating significantly
of all measured pit depths has little physical mean- influences the pitting process.11,19,22 Recently, Race,
ing because the observations include measurements et al.,23 have developed a scoring predictive model for

CORROSION—Vol. 65, No. 5 335


CORROSION ENGINEERING SECTION

table 4 knowledge on the susceptibility of pipeline coatings to


Scoring Model for Pipe Coating Type in-service failure.11,19,22-23
Coating (ct) Variable Score Correlation Analysis — The plot of maximum pit
depth against exposure time is shown in Figure 2 for
Bare pipe 1.0
Alkyd enamel 0.9
all measured depths. Although an apparent positive
Wrap tape 0.8 correlation is observed between maximum pit depth
Coal tar 0.7 and pipeline age, it is impossible to postulate the
FBE 0.3 existence of a statistically significant linear correla-
tion between these two variables. For each exposure
time, a significant scatter is observed in the measured
depths. This scatter is undoubtedly associated with
the variability of the soil and pipe characteristics from
one exposed pipeline segment to another. This obser-
vation supports the need for realistic nonlinear pit
growth models capable of overcoming the oversimpli-
fied assumption of constant pit growth rate made in
pipeline integrity and reliability assessments.
The results of the correlation analysis between
the soil and pipe variables and the maximum pitting
depth are shown in Table 5. The first observation that
can be gleaned from this table is the relative degree
of influence exerted by each predictor variable on the
observed maximum pit depth. Besides the exposure
time, dissolved chloride ion concentration, bulk den-
sity, soil pH, and pipe-to-soil potential are the most
influential variables among those considered. This is
in agreement with the general body of experimental
Figure 2. Measured maximum pit depth as a function of exposure
time (all data). knowledge about the effect of soil characteristics on
pitting corrosion.11,19,21 It is also worth noting that the
direction (sign) of the correlation of each independent
table 5 variable agrees with its documented influence on pit-
Correlation Between Soil and Pipe Variables ting corrosion penetration. That is, dmax increases with
and dmax for All Data decreasing pH value (negative correlation) and with
increased chloride content (positive correlation).
Variable Correlation with dmax
The correlation matrix for the soil and pipe vari-
t   0.420 ables is shown in Table 6. The use of redundant
rp –0.065
information in the pursued prediction model can
ph –0.369
pp   0.334 be avoided by detecting a high degree of correlation
re –0.140 between the soil and pipe variables. The results pre-
wc   0.304 sented in Table 7 indicate that only a weak-to-mod-
bd –0.365 erate correlation exists between these variables.
cc   0.406
The strongest, though moderate, correlation occurs
bc –0.266
sc –0.122 between soil resistivity and water content. Based
ct –0.118 on this result, the variables shown in Table 6 were
all incorporated into the nonlinear pit growth model
developed in this work.
pipeline external corrosion rates. In this model, coat- Detection of Outlier Observations — An outlier is
ing type is a key variable because it determines the an observation that lies outside the overall pattern
susceptibility of the coating to failure. In contrast to of the data distribution. Four different techniques
the work of Katano, the influence of pipeline coating were used to detect outliers: the boxplot, standard-
was modeled in the present study through an inde- ized residuals, Cook’s distance, and the hat matrix.18
pendent variable (coating type [ct] in Table 1) whose Data found to be outliers in three of the four tests
value was assigned by means of the scoring model were considered spurious. Nine outliers were detected
shown in Table 4. The scores proposed in Table 4 and the corresponding data sets were removed. As
were selected based on the criteria given elsewhere.22 an example, Figure 3 shows the boxplot analysis for
Higher scores are given to coatings that provide lower each dependent variable in the clay data after nor-
protection against corrosion. It is intended that Table malization. The outlier observations (marked with an
4 will summarize and quantify the general body of asterisk in Figure 3) identified in this stage of model

336 CORROSION—MAY 2009


CORROSION ENGINEERING SECTION

TAble 6
Correlation Matrix for the Independent (Random) Variables for All Data
ph pp re wc bd cc bc sc ct
rp –0.268 –0.248 0.339 –0.291 0.050 –0.267 –0.124 –0.133 0.145
ph 0.048 –0.320 0.165 0.206 –0.051 0.440 0.129 –0.227
pp –0.351 0.274 –0.124 0.302 0.038 0.158 –0.226
re –0.532 –0.092 –0.130 –0.199 –0.188 0.180
wc –0.155 0.196 0.117 0.193 –0.149
bd –0.229 0.166 0.159 0.179
cc 0.102 0.204 –0.052
bc 0.268 0.029
sc 0.082

development were detected only on the basis of the TAble 7


distributional properties of the predictor variables Result of the Regression Analysis by Soil Category
using the boxplot analysis. Further detection of out- Soil Category/R2
lier observations was conducted in the latest stages Coefficient C/0.90 Cl/0.88 SCl/0.71 All/0.87
of the model development based on the prediction for-
t0 3.05 3.06 2.57 2.88
mula obtained for each soil category. k0 5.51E–01 9.84E–01 5.99E–01 6.08E–01
n0 8.85E–01 2.82E–01 9.65E–01 8.96E–01
Regression Analysis and Pit Growth Model k1 (rp) –8.98E–05 –1.06E–04 –1.82E–04 –1.80E–04
Prediction Model — After the preliminary data k2 (ph) –5.90E–02 –1.15E–01 –6.42E–02 –6.54E–02
analysis, a multivariate regression of maximum pit k3 (re) –2.15E–04 –2.99E–04 –2.12E–04 –2.60E–04
k4 (cc) 8.38E–04 1.80E–03 8.62E–04 8.74E–04
depth on the exposure time and soil and pipe vari- k5 (bc) –1.28E–03 –4.88E–04 –6.78E–04 –6.39E–04
ables was conducted for each soil category. To estab- k6 (sc) –5.33E–05 –2.09E–04 –1.13E–04 –1.22E–04
lish the influence of the predictor variables on the n1 (pp) 4.93E–01 4.61E–01 5.12E–01 5.19E–01
parameters of model (2), κ and ν were expressed as n2 (wc) 3.72E–03 1.69E–02 4.50E–03 4.65E–03
n3 (bd) –1.01E–01 –9.87E–02 –1.58E–01 –9.90E–02
linear combinations of the soil and pipe variables.
n4 (ct) 4.67E–01 5.67E–01 4.34E–01 4.31E–01
Therefore, the regression formula for the estimated
^
maximum pit depth (dmax) was proposed as:

m
n0 + ∑ n j x j
ˆ  n 
d max
max =  k 0 + ∑ k i x i  (t – t 0 ) j=1
(4)
 i =1 

where xi represents the i-th predictor variable (Tables


1 and 2) and ki and nj are the regression coefficients
for this predictor.
For each soil category, a regression analysis was
conducted for each one of the 1,024 possible com-
binations for the distribution of predictor variables
among κ and ν. The best model obtained for the data
from the clay, sandy clay loam, and clay loam soil
types was the one for which:

κ = k 0 + k1rp + k 2 ph
ph + k 3 re + k 4cc
cc + k 5 bc + k 6 s
sc (5)

ν = n 0 + n1pp + n 2 wc
wc + n 3 bd + n 4c
ct (6) FIgURe 3. Boxplot analysis of the clay data. Extreme outliers,
indicated by asterisks, are defined as points beyond three times the
According to these formulas, the proportional- interquantile range from the edge of the box.
ity parameter κ is most likely a function of the redox
potential, pH, resistivity, and dissolved ion concentra-
tions. In contrast, the pitting exponent ν is a function all soils, the values of the multiple regression correla-
of the pipe-to-soil potential, water content, bulk den- tion coefficients are high enough to support the con-
sity, and the coating type. sistency of the prediction formula for maximum pit
The numerical results of the regression analy- depth. Also, the sign of each regression coefficient
sis are shown in Table 7. It can be observed that, in reflects correctly the influence of the correspond-

CORROSION—Vol. 65, No. 5 337


CORROSION ENGINEERING SECTION

table 8 each soil category, the time evolution of the mean of


Model Parameters for Pit Growth in Typical Conditions the simulated maximum pit depth distributions was
Soil Category fit to model (2) using the corresponding value of t0
Parameter C CL SCL All given in Table 7. The model parameters obtained by
this method are shown in Table 8, where they are
κ(A) 0.178 0.163 0.144 0.164
ν 0.829 0.793 0.734 0.780 assumed to be unbiased estimates of κ and ν for typi-
(A)
cal conditions in each soil category. Thus, they can be
Expressed in (mm/y1/ν) units. used, together with the respective value of t0, to pro-
duce an empirical formula for the time evolution of
the average maximum pit depth in the soil of interest.
Figure 4 shows the time evolution of the average
maximum pit depth, by soil category, predicted from
the Monte Carlo simulations. From the results shown
in Table 8 and Figure 4, the order of corrosiveness
predicted by the proposed model is clay > clay loam >
sandy clay loam. This order agrees with the findings
of previous works on pitting corrosion of low-carbon
steel in soils.6,11
Importantly, if the differences between soil envi-
ronments were not considered in the analysis, a com-
mon prediction formula with the pitting parameters
for the “All” category would be obtained. The resulting
estimations would be close to those obtained for pit-
ting in clay loam soils but would be very unreliable if
the damage has to be assessed in the other two soil
classes; this is especially true for long (>15 years)
Figure 4. Pit growth predicted for typical average conditions. exposure times. This comes to confirm that, even for
average soil conditions, considering the actual values
of the environmental factors (soil and pipe-related pre-
ing variable on pit growth. Further discussion on the dictors) will significantly improve the accuracy of any
physical meaning of the obtained prediction model is predictive model for pit depth growth in underground
provided in the next (Model Discussion) section. pipelines.
The maximum pit depth in a buried pipeline seg- Sensitivity Analysis — Corrosion modeling is use-
ment at a given time point can be estimated by eval- ful in determining which variables have a greater
uating Expression (4) for the soil and pipe variables influence on pit growth. Figure 5 shows a sensitiv-
measured at that particular segment. The regression ity graph for the model developed in this work. In the
coefficients to be used in the estimations are those graph, greater values of the magnitude of the slope
given in Table 7 for the soil textural class of interest. of a line correspond to greater influence of the corre-
The time evolution of the average maximum pit sponding predictor variables on maximum pit depth.
depth in a soil category can be predicted from Monte The sign of the slope indicates whether the influence
Carlo simulations or from the evaluation of the pro- exerted by the respective predictor variable on pit
posed prediction formula using the respective average growth is positive or inverse; a positive slope is related
values in Table 2. The latter approach has the disad- to a positive influence, whereas a negative slope
vantage that using the mean value of the predictor denotes an inverse influence.
variables in the model may produce biased estimates In agreement with the current knowledge on pit-
of the average maximum pit depth because the model ting corrosion, the present pit growth model is more
is nonlinear. Therefore, the approach based on Monte sensitive to the pH value, pipe-to-soil potential, pipe
Carlo simulations was adopted to predict how the coating type, bulk density, water content, and the dis-
average of the maximum pit depth evolves with time. solved chloride concentration, in that order. The rest
The distributions of the independent variables of the measured variables are less influential and
were used as inputs into Monte Carlo simulations. have not been included in Figure 5. The model sen-
Random values drawn from the distributions were sitivity information is of paramount importance for
used to evaluate Equations (4), (5), and (6) 5,000 developing simplified pit growth models based on the
times for different exposure times. The regression most influential soil and pipe properties. Importantly,
coefficients used in the simulations were also those the experience gained during this work indicates that
given in Table 7. Following this procedure, the distri- such models will succeed in predicting external cor-
butions of the maximum pit depths in each soil class rosion pit growth only if they account for soil texture
were obtained for different exposure times. Then, for categorization.

338 CORROSION—MAY 2009


CORROSION ENGINEERING SECTION

MODEL DISCUSSION
Discussion of the physical implications of the
proposed prediction model is focused firstly on the
results shown in Table 7 and Figure 4 for the aver-
age time for pit initiation. In contrast to previous
models, the present model is capable of accounting
for the fact that pits do not initiate immediately after
the pipeline has been constructed or commissioned.
The parameter t0 can be related unambiguously to
the total elapsed time from pipeline commissioning
to coating damage plus the time period in which the
cathodic protection is effective in preventing or atten-
uating external pitting corrosion after coating dam-
age. It is important to emphasize that, on average, the
numerical values predicted for t0 in the present study
are in good agreement with those reported in the pipe- Figure 5. Model sensitivity analysis.
line corrosion literature.19 Unfortunately, uncertain-
ties in the estimates of t0 do not allow for the drawing
of solid conclusions about the influence of soil cat- between resistivity and variables such as water con-
egory on this parameter. From the results shown in tent and dissolved ion concentrations can conceal the
Tables 3 and 7 it can be inferred that the pit initiation role of resistivity on pit growth.
times determined during the regression analyses mar- If the only predictor variable was resistivity, its
ginally reflect the influence that coating type exerts on strong correlation with the predicted pit depth could
t0. The fact that the smallest value of t0 is predicted be observed readily. However, in models such as those
for the sandy clay loam soils might be associated with developed by Oguzie and coworkers and the pres-
the fact that, in the present investigation, this soil ent one, the effect of the soil resistivity on pit growth
category showed the largest proportion of non-coated is most probably masked by the marked effect of
pipelines and asphalt enamel-coated pipelines. How- soil water content on this phenomenon. Based on
ever, the database is dominated by tape wrap and the results presented in Table 6 and Figure 5, it is
coal-tar coatings, which may explain the observation expected that the effect of highly influential predictor
that pitting initiation time does not vary significantly variables may conceal the effect of those variables cor-
from one soil category to another. related to them. As an example, it could be postulated
Another important issue deserving discussion that the effect of bicarbonate ions is masked by the
is the distribution of the predictor variables among significant role that pH plays on pitting.
the two pitting parameters, κ and ν. It is widely rec- The other variables considered in the proportion-
ognized that the exponent ν largely depends on the ality parameter are the dissolved chloride, sulfate, and
aeration of the soil. For example, the results reported bicarbonate ion concentrations. In the basic theory of
by Mughabghab and Sullivan,3 Oguzie, et al.,9 and pitting corrosion, the major role of dissolved ions and
Romanoff10 indicate that poorly aerated soils such as pH in pitting of low-carbon steels is well understood
clay have a greater corrosion exponent. The proposed and documented.20-21,24 The insertion of these factors
model captures this effect correctly in two ways. First, into the set of predictors used to estimate κ can be
the model associates the pitting exponent with the rationalized by recalling that, according to Equation
predictor variables that determine aeration, i.e., bulk (2), the magnitude of the pit penetration is determined
density and moisture. Second, the model produces primarily by κ while ν determines the form of the pen-
estimates of the exponent that indicate appropriately etration rate curve.10 Therefore, at a given time point,
the degree of soil corrosiveness. The estimates of the soil conductivity (1/re), soil acidity (pH), and dissolved
model parameters shown in Table 8 for pit growth in ion concentrations (chloride content [cc], sulfate con-
typical conditions support these statements. tent [sc], and bicarbonate content [bc]) would seem
The fact that the soil pH and resistivity are both to determine the magnitude of pitting penetration.
associated with the proportionality parameter κ has This result is in accordance with the one-dimensional
also been reported by Oguzie, et al.,9 who also found pit model under mass-transport, diffusion-controlled
that resistivity does not significantly influence pit growth.21
depth. This latter result seems to be in disagree- On the other hand, the fact that pipe-related vari-
ment with the general body of practical knowledge on ables, such as coating type and pipe-to-soil potential,
soil corrosion, where resistivity is a key factor in pre- and soil aeration variables, such as bulk density and
dicting soil corrosiveness. This apparent contradic- water content, are considered in the pitting exponent
tion can be explained by noting that the correlation and can be interpreted as evidence that they play a

CORROSION—Vol. 65, No. 5 339


CORROSION ENGINEERING SECTION

the constant rate model. For each pit, the growth


rapidity, υ, was computed as the ratio of the mea-
sured pit depth, dm, to the time interval elapsed from
the pipeline commissioning time, tC, to the time of
inspection, tI:

dm
υ= (8)
tI – tC

The time-to-failure by leakage (tL) was predicted


^
by equating dcr(t) to the pipe wall thickness in Equa-
tion (7) and solving it for the variable time (t). Dur-
ing the years that follow the in-line inspection, repairs
were scheduled based on the predicted time-to-fail-
ure and many sections of this pipeline were excavated
and, when necessary, rehabilitated. As a part of this
process, the depth of a total of 123 pits detected at
the threatened pipeline sections was measured in the
field to calibrate the prediction model. For these pits,
the time-to-failure by leakage was estimated to be in
the range from 1 year to 8 years. Given that 18 years
FIgURe 6. Pit depth predicted using the linear pit growth model elapsed from the date of the pipeline commissioning
against actual (field-measured) depth. PWT: pipe wall thickness. to the date of the in-line inspection, the exposure time
for the measured pit depth data ranges from 19 years
to 26 years.
role in determining the time dependence of the pitting In Figure 6, the depths estimated using Equa-
penetration rate. However, further statistical studies tions (7) and (8) are plotted against the field-mea-
are warranted to obtain a more complete understand- sured (actual) depths of the 123 pits of interest. As a
ing of the physical meaning of the pit growth predic- reference, the perfect match (1:1) line is also shown.
tion model. Besides the random errors always associated with this
kind of plot, one notices that the predicted depths are
EXAMPLE Of APPLICATION on average greater than the actual depths. The bias
toward greater depth values can be explained by the
The proposed model is applied to a real-life case
fact that the linear trend assumed for pit growth leads
study involving the integrity assessment of an onshore
to more conservative prediction when the estimations
buried pipeline with an outside diameter of 914 mm
are made for long exposure periods.
and a pipe wall thickness of 11 mm. This pipeline was
In contrast to the results presented in Figure
commissioned in 1979 and inspected in 1997 using
6, the application of the prediction model developed
magnetic flux leakage in-line inspection (ILI). The
in this work produced less conservative predictions.
threat posed by the deeper (external) corrosion pits
From model (2), the time evolution of the corrosion
detected and sized during the inspection was evalu-
rate was expressed as:
ated and the remaining life of the pipeline was pre-
dicted. For these pits, the most credible failure cause
υ( t ) = κ
κνν( t – t 0 )ν –1 (9)
was determined to be leakage due to the perforation of
the pipe wall.
To predict pit depth for any future time, the above
For the sake of comparison, in a first analysis,
expression was used with the parameter values given
the growth of the detected pits was assumed to be lin-
in Table 8 for the “All” category (κ = 0.164 mm/y1/0.780
ear with respect to time, i.e., a constant growth rate
and ν = 0.780). In this way, the growth of each pit
was assumed. Due to the lack of reliable information
was modeled for the average conditions found in this
regarding pit initiation time, it was also assumed that
study for all collected samples as:
pits started to grow right after the pipeline was com-
missioned. Based on these assumptions, the time
dependence of pit depth was predicted as: ˆ pm ( t ) = d m + ∫ t L υall ( tt′′ )dt
d )dt ′ (10)
t l

ˆ cr ( t ) = υ( t – t I )
d (7) Here, υall(t) = 0.128(t – 2.88)–0.220 (in mm/y) is the
time-dependent average corrosion rate in all soils; tL is
where tI is the time of in-line inspection and the sub- the predicted time-to-leakage; t′ is a dummy variable;
script “cr” means that the prediction is made using and the subscripts “pm” and “all” mean that the pre-

340 CORROSION—MAY 2009


CORROSION ENGINEERING SECTION

dictions were made using the proposed model for all


data.
The results of the application of this prediction
method are shown in Figure 7. Note that the bias
toward larger predicted depths observed in the lin-
ear growth model has been reduced significantly. The
relatively small underprediction that is still observed
in Figure 7 can be explained by the fact that the envi-
ronmental factors in this particular example likely
differed from those measured for the “All” soil cate-
gory. For example, the use of the model parameters
obtained for the average conditions observed for clay
soils could further improve the predictions because of
the higher average corrosion rates observed for these
soils. Although less evident, the reduction of the scat-
ter of the predicted depth around the perfect corre-
lation line (with respect to that observed in Figure 6)
indicates that a reduction in the uncertainty of the pit
depth estimations can be associated with the applica-
tion of the proposed model.
From the point of view of pipeline integrity, the
increase in accuracy and precision of the pit depth Figure 7. Pit depth predicted using the proposed model against
estimations results in longer remaining life being actual (field-measured) depth. PWT: pipe wall thickness.
predicted and thus fewer repairs being scheduled.
It is interesting to note that, in the present exam-
ple, a positive impact on integrity management was v For the soils considered in this work, the order of
achieved using average conditions. It is reasonable to corrosiveness was found to be: clay > clay loam >
expect that collecting the actual values of the soil and sandy clay loam. Importantly, this order remains un-
pipe characteristics at each exposed pipeline section altered as the pipelines remain in service. The appli-
would further improve the accuracy and precision of cation example developed at the end of this paper has
the maximum pit depth estimations. This is the sub- proven that the use of a realistic pit growth model
ject of ongoing investigations. that takes into account the actual soil and pipe char-
acteristics could greatly improve the quality of current
Conclusions and future integrity management plans.
v Finally, it is worth acknowledging the limitations of
v A predictive model has been proposed for pit- the proposed model. Although microbiologically influ-
ting corrosion in buried oil and gas pipelines. In con- enced corrosion was not identified as a credible cause
trast to previous models, the present model accounts of the observed pitting damage, the role of micro-
for the influence of the soil and pipe characteris- bial corrosion, notably by sulfate-reducing bacteria,
tics on external pitting corrosion in different soil tex- could be incorporated in a more complete predictive
tural classes. It has been shown that multivariate model. Seasonal fluctuations in soil properties and
regression analysis can be used to obtain a physically their influence on pit growth were also not consid-
rational model relating pit growth to soil and pipe ered in this work. However, the fact that samples were
variables. collected within a time span of several season cycles
v The power law model proved to be valid for describ- allows the authors to assume that the observed dis-
ing the functional form of the time dependence of tributions for the soil characteristics contain informa-
maximum pit depth. It has been shown that it is tion about the variations in soil properties from one
possible to establish a physically sound correlation season to another. It is also worth nothing that the
between the pitting model parameters and the soil results shown in Tables 7 and 8 will be applicable to
and pipe variables. The proportionality parameter κ any corroded pipeline in contact with clay, clay loam,
is determined primarily by the soil pH, redox poten- and sandy clay loam soils in tropical regions. Caution
tial, resistivity, and dissolved ion concentrations. The should be exercised when applying the results
pitting exponent is determined primarily by the pipe obtained in the present study to pipelines in contact
coating, pipe-to-soil-potential, soil bulk density, and with these textural classes in non-tropical regions,
water content. Overall, the variables with the greatest for which the probability distribution of the soil char-
influence on pit growth are pH, pipe-to-soil potential, acteristics could be quite different to those given in
coating type, bulk density, water content, and dis- Table 2. Lastly, it also would be desirable to increase
solved chloride concentration, in that order. the number of collected samples and to include other

CORROSION—Vol. 65, No. 5 341


CORROSION ENGINEERING SECTION

soil categories, thereby increasing the statistical 9. E.E. Oguzie, I.B. Agochukwu, A.I. Onuchukwu, Mater. Chem.
Phys. 84, 1 (2004): p. 1.
power of the model. 10. M. Romanoff, “Underground Corrosion,” National Bureau of
Standards (NBS) Circular 579 (Washington, DC: National Bureau
of Standards, 1957).
Acknowledgments 11. S. Bradford, Practical Handbook of Corrosion Control in Soils,
CASTI Corrosion Series, vol. 3 (Edmonton, Canada: CASTI Pub-
This work was performed during A. Valor’s stay lishing Inc., 2000).
12. ASTM G57-06, “Standard Test Method for Field Measurement of
at the National Polytechnic Institute (ESIQIE-IPN), Soil Resistivity Using the Wenner Four-Electrode Method (West
Mexico, under research project no. 428817805. The Conshohocken, PA: ASTM International, 2006).
authors are grateful to Petróleos Mexicanos (Pemex) 13. ASTM D2216-05, “Standard Test Methods for Laboratory Deter-
mination of Water (Moisture) Content of Soil and Rock by Mass”
for permission to publish these results. Comments (West Conshohocken, PA: ASTM International, 2005).
provided by the reviewers are also deeply appreciated. 14. ASTM D422-63(2007), “Standard Test Method for Particle-Size
Analysis of Soils” (West Conshohocken, PA: ASTM International,
References 2007).
15. ASTM D512-04, “Standard Test Methods for Chloride Ion in
Water” (West Conshohocken, PA: ASTM International, 2004).
1. F. Caleyo, L. Alfonso, J. Alcantara, J.M. Hallen, F. Fernández, H. 16. ASTM D1126-02(2007)e1, “Standard Test Method for Hardness
Chow, “On the Estimation of Failure Rates of Multiple Pipeline in Water” (West Conshohocken, PA: ASTM International, 2007).
Systems,” Proc. ASME 6th Biennial Int. Pipeline Conf., IPC06, 17. ASTM D516-07, “Standard Test Method for Sulfate Ion in Water”
paper no. IPC06-10526 (New York, NY: ASME International, 2006). (West Conshohocken, PA: ASTM International, 2007).
2. M. Kowaka, H. Tsuge, M. Akashi, K. Masamura, H. Ishimoto, An 18. B. Tabachnick, Using Multivariate Statistics, 5th ed. (New York,
Introduction to the Life Prediction of Plant Materials, Applications NY: Pearson, Harper, and Row Publishers, 2005).
of Extreme Value Statistical Methods for Corrosion Analysis (New 19. A.W. Peabody, Control of Pipeline Corrosion, 2nd ed., ed. R. Bi-
York, NY: Allerton Press, 1994; Tokyo, Japan: Maruzen Pub., anchetti (Houston, TX: NACE International, 2001).
1984). 20. G.T. Burstein, C. Liu, R.M. Souto, S.P. Vines, Corros. Eng. Sci.
3. S.F. Mughabghab, T.M. Sullivan, Waste Manage. 9 (1989): p. Technol. 39 (2004): p. 25.
239. 21. P. Marcus, ed., Corrosion Mechanisms in Theory and Practice
4. A.K. Sheikh, J.K. Boah, D.A. Hansen, Corrosion 46 (1990): p. (New York, NY: Marcel Dekker, Inc., 2002).
190. 22. S. Papavinasam, R.W. Revie, “Pipeline Protective Coating Evalua-
5. P.J. Laycock, R.A. Cottis, P.A. Scarf, J. Corros. Sci. Eng. 137, 1 tion,” CORROSION/2006, paper no. 06047 (Houston, TX: NACE,
(1990): p. 64. 2006).
6. T. Shibata, ISIJ Int. 31 (1991): p. 115. 23. J.M. Race, S.J. Dawson, L.M. Stanley, S. Kariyawasam, J. Pipe-
7. F. Kajiyama, Y. Koyama, Corros. Eng. 53, 2 (1997): p. 156. line Eng. 6 (2007): p. 15.
8. Y. Katano, K. Miyata, H. Shimizu, T. Isogai, Corrosion 59, 2 24. Z. Szklarska-Smialowska, Pitting and Crevice Corrosion (Houston,
(2003): p. 155. TX: NACE, 2005), p. 129.

342 CORROSION—MAY 2009

You might also like