Sie sind auf Seite 1von 12

See discussions, stats, and author profiles for this publication at: https://www.researchgate.


Pitting Degradation Modelling of Ocean Steel Structures using Bayesian


Article  in  Journal of Offshore Mechanics and Arctic Engineering · April 2017

DOI: 10.1115/1.4036832


4 261

5 authors, including:

Jyoti Bhandari Faisal Khan

Australian Maritime College Memorial University of Newfoundland


Rouzbeh Abbassi Vikram Garaniya

Macquarie University University of Tasmania


Some of the authors of this publication are also working on these related projects:

Pitting Corrosion Degradation Modelling for Marine and Offshore Steel Structures. View project

Retrospective Rosk Analysis and Control for Semabal Gran Storage Hybrid Mixture Explosion View project

All content following this page was uploaded by Jyoti Bhandari on 12 June 2017.

The user has requested enhancement of the downloaded file.

Jyoti Bhandari
Australian Maritime College,
University of Tasmania,
Launceston TAS 7250, Australia
Pitting Degradation Modeling
Faisal Khan1 of Ocean Steel Structures Using
Centre for Risk, Integrity and
Safety Engineering (C-RISE),
Faculty of Engineering and
Bayesian Network
Applied Science,
Memorial University of Newfoundland, Modeling depth of long-term pitting corrosion is of interest for engineers in predicting
St. John’s, NF A1B 3X5, Canada the structural longevity of ocean infrastructures. Conventional models demonstrate poor
e-mail: quality in predicting the long-term pitting corrosion depth. Recently developed phenome-
nological models provide a strong understanding of the pitting process; however, they
Rouzbeh Abbassi have limited engineering applications. In this study, a novel probabilistic model is devel-
Australian Maritime College, oped for predicting the long-term pitting corrosion depth of steel structures in marine
University of Tasmania, environment using Bayesian network (BN). The proposed BN model combines an under-
Launceston TAS 7250, Australia standing of corrosion phenomenological model and empirical model calibrated using
real-world data. A case study, which exemplifies the application of methodology to pre-
dict the pit depth of structural steel in long-term marine environment, is presented. The
Vikram Garaniya result shows that the proposed methodology succeeds in predicting the time-dependent,
Australian Maritime College,
long-term anaerobic pitting corrosion depth of structural steel in different environmental
University of Tasmania,
and operational conditions. [DOI: 10.1115/1.4036832]
Launceston TAS 7250, Australia
Keywords: offshore structures, pitting corrosion, pit depth, Bayesian network,
Roberto Ojeda phenomenological model
Australian Maritime College,
University of Tasmania,
Launceston TAS 7250, Australia

1 Introduction studied with plenty of interest for some time [9–11]. There are
many potential factors that influence pitting corrosion when sub-
Pitting corrosion is a significant potential threat to existing
jected to marine exposure. The influence of environmental factors
marine and offshore infrastructures. It affects the service life of
such as temperature, bacterial community, oxygen concentration,
process equipment and pipelines, and it may result in structural
pH, and velocity for pitting corrosion growth has been previously
failure, leakage, production loss, environmental pollution, and the
studied by several researchers [10,12–17]. Recently, Bhandari
loss of life [1]. Pitting corrosion is regarded as one of the most
et al. [10] reported a wide spectrum of factors that control the pit-
hazardous forms of corrosion in marine and offshore structures
ting corrosion growth and their effects were reviewed.
because its attack is widespread. It generates severe localized
The pitting process has previously been described as random,
damage and is difficult to detect, thereby causing failures. The
sporadic, and stochastic phenomena. In addition, the prediction of
total loss of the materials might be very small but the local rate of
time and location of the pit remains particularly difficult to esti-
attack can be extensive and can lead to early and catastrophic fail-
mate [18]. Conventionally, researchers have simply relied on
ure [2,3]. The oil and gas industry is expanding with a large
mathematical models to estimate the depth of pitting corrosion
increase in offshore platforms, supply vessels, pipelines, under-
and to predict the likely future deterioration of offshore structures
water and floating storages, and shore facilities. However, due to
over long periods of time [19,20]. Katano et al. [21] proposed a
the “build and forget” mentality of the industrial society, the long-
predictive model for pit growth in underground pipes. They pre-
term structural reliability of these structures is still not fully
scribe to the theory that the pitting corrosion rate for metal
understood [4,5].
depends on environmental factors and pitting depth. This relation-
Pitting corrosion in offshore steel structures has a particular
ship was explored through regression analysis, with pitting depth
importance where containment is critical, such as for pressure ves-
as a dependent variable and the environmental factors as inde-
sels, boilers, turbine blades, and metallic containers handling toxic
pendent variables. However, the effect of exposure period on pit-
materials. It also greatly affects the localized structural strength of
ting depth was not considered even though the pitting corrosion
piping, tanks, and general load-bearing structures within ships and
depth is directly influenced by the exposure time. Similarly,
offshore structures [6,7]. Consequently, the knowledge of pitting
Guedes Soares et al. [20] studied the effects of different environ-
corrosion is crucial. This is not only critical in the design phase
mental factors on pitting corrosion behavior of steel plates totally
but is also equally so for assessing any deterioration in the
immersed in salt water conditions. They proposed a corrosion
ongoing strength of offshore structures.
wastage model based on a nonlinear time-dependent function. The
A quantitative understanding is necessary of how the corrosion
model developed in their study accounts for the effect of various
occurs as a function of exposure time and under various environ-
environmental factors including salinity, temperature, dissolved
mental influences. It is also essential to predict the likely depth of
oxygen, pH, and flow velocity [20]. Nevertheless, these models do
corrosion for future defined conditions [8]. The effects of pitting
not incorporate the corrosion phenomenological model presented
corrosion on steel structures in marine environment have been
in Fig. 1. Furthermore, these models are only effective for a spe-
cific site and they are not capable of being updated when new
Corresponding author. observations are available. Another approach for pit depth uncer-
Contributed by the Ocean, Offshore, and Arctic Engineering Division of ASME
tainty estimation and modeling the maximum depth of pits is
Manuscript received February 9, 2016; final manuscript received April 12, 2017; extreme value analysis (EVA). However, the extreme value analy-
published online June 9, 2017. Assoc. Editor: Lance Manuel. sis is not based on the corrosion phenomenological model; it is

Journal of Offshore Mechanics and Arctic Engineering OCTOBER 2017, Vol. 139 / 051402-1
C 2017 by ASME
Copyright V

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

represent the overall pitting corrosion growth for aluminum and
steel alloys. However, Melchers later reported that the power law
model is not consistent with the overall growth of pitting corro-
sion depth; it is only applicable for the first (kinetic or initiation)
phase and the fourth (long-term anaerobic) phase of the pitting
corrosion process shown in Fig. 1. In this study, the power law
model is applied to represent the pitting corrosion depth in long-
term exposures shown in Fig. 1. Parameters for the power law
model were determined from long-term experimental field data.
The long-term phase is believed to be controlled by the amount of
bacterial activity, and the available nutrients for them to grow.
The models discussed so far provide a strong understanding of
the corrosion process; however, they are often hard to apply to
real-world engineering applications. Although these models are
useful for specific site or operating conditions, they still carry
high uncertainty associated with certain data because some data
points may be incorrect owing to errors in observation or data
Fig. 1 Schematic diagram of the theoretical nonlinear corro- processing. Hence, it is essential to use Bayesian statistics to
sion model (Adapted from Ref. [24]) develop a corrosion loss/pitting depth model which considers
“prior” knowledge of the corrosion process involved, together
with data used, in order to calibrate it to a real-world situation.
simply a means to determine the best probability density function Melchers [30] also emphasized the importance of developing
(PDF) to fit the empirical data. The EVA approaches are generally probabilistic models such as using Bayesian statistics in future
used for a specific point in time at which the data are available. models for predicting corrosion loss/pitting depth. This paper
There is no obvious link between the EVA and the corrosion phe- presents a novel methodology that has a high degree of pitting
nomenological model [22,23]. depth prediction capability based on the theoretical corrosion loss
Melchers and Jeffrey [24] describe a more recent progress in model, the empirical pit depth model, and using Bayesian network
the development of a phenomenological model for corrosion loss (BN). It focuses on the development of a predictive model
and for maximum pit depth. This model is based on corrosion sci- adequate for long-term anaerobic conditions in marine environ-
ence that includes the critical role of microbiological activity ments. The proposed methodology is tested and validated using
[16,24,25]. Figure 1 shows the developed phenomenological the data obtained from field testing accompanied by ASTM world-
model based on corrosion principles including a number of wide corrosion study [36].
sequential phases 1–4, i.e., kinetic, diffusion, transition, and
anaerobic phases. In this phenomenological model, stage 2 indi- 2 Bayesian Network
cates the theoretical point at which anaerobic conditions are Bayesian network is a graphical model that connects the causes
reached. Anaerobic conditions are favorable for the activation and (independent variables) and consequences (dependent variables)
growth of sulfate-reducing bacteria (SRB) [19]. The phenomeno- through direct arrows pointing from the causes to the consequen-
logical model in Fig. 1 is widely acknowledged and accepted in ces [37–39]. The BN offers a formal method for encoding joint
the field of corrosion science and engineering. Additionally, each probability distributions using a set of statistical variables in
of these phases is believed to control the corrosion processes in which the information about independencies is explicitly sepa-
both the short- and long-term [5,10,26–29]. rated from numerical quantities [40]. Figure 2(a) represents a
Melchers [30] reported a mathematical and probabilistic model schematic structure of BN; the arrow in the figure represents
for corrosion loss/pitting depth of the structures exposed in sea causal relationship between the variables (A and B) through the
water. In their probabilistic model, the corrosion loss c(t, E) was probability distributions function illustrated in Fig. 2(b). BN
represented by a probabilistic function allows estimating likelihood of rare failure events of complex
structures in an efficient way. It also assists in updating the predic-
cðt; EÞ ¼ bðt; EÞ  f ðt; EÞ þ eðt; EÞ (1) tion by using new information obtained through measurements,
monitoring, and inspection [41].
where t is the exposure time to be contained in the vector E of In BN, “probability inference of an event is conditional on the
environmental and materials parameters, f(t, E) is the mean-value observed evidence.” BN not only implements forward or predic-
function for corrosion loss, e(t, E) is a zero-mean error function, tive analysis, but it also performs backward or diagnostic analysis.
and b(t, E) is a bias function, all as functions of t and E [19,31]. Considering the conditional dependencies of variables, BN repre-
The introduction of a probabilistic framework eliminates the sents the joint probability distribution P(U) of variables U ¼
uncertainty associated with the data when calibrating it to the fA1 ::::::An g as
empirical model [30]. The probabilistic approach has been suc-
cessfully used in many engineering areas including reliability and Yn
PðUÞ ¼ PðA1 jPaðAi ÞÞ (3)
integrity management of complex systems, complex load systems, i¼1
and resistance modeling. Several researchers [20,29,32–35]
adopted the most widely applied “power law” model to represent where Pa(Ai) is the parent set of Ai in the BN, and P(U) reflects
the growth of the pit with respect to the exposure time (Eq. (2)). the properties of the BN [42]. BN takes advantages of Bayes’
The growth of the pit as a function of time is a representative rela- theorem to update the prior occurrence (or failure) given that the
tionship where dðtÞ is a characteristic dimension such as pit depth, observation of another set of variables evidence E. The posterior
and t is the exposure time, while A and B are empirical constants probability distribution of a particular variable can be computed
calibrated using experimental data using different classes of inference algorithms, such as the junc-
tion tree or variable elimination, based on Bayes’ theorem
dðtÞ ¼ AtB (2) (Eq. (4)) [43–45]

Equation (2) represents the time-dependent pitting corrosion depth PðUjEÞ ¼ ¼X (4)
for long-term corrosion (quasi-steady-state situation) phase. This Pð EÞ PðU; EÞ
power law model had been previously applied by Melchers to U

051402-2 / Vol. 139, OCTOBER 2017 Transactions of the ASME

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

Fig. 2 Structure of BN model (the arrow in the network represents the relationship between
the nodes through the probability distributions function) [28]

The denominator in Eq. (4) is called the probability of observation The performance of the Gaussian KDE depends significantly on
and is the sum of all the conditional probabilities of E given the value of smoothing parameters, which is known as bandwidth
events, and U multiplied by the probabilities of U. The prior and [47]. The Gaussian kernel is assumed, and the smoothing parame-
posterior probabilities can also be considered as “causes” and ters’ “bandwidth” is assumed using asymptotic approximation of
“consequences” of a process. The term consequences is a relation- mean integrated squared error (AMISE). It is performed by apply-
ship between two events in a process—one leading to the other. ing the asymptotic approximation to the random sample. The
An example would be the presence of sulfate-reducing bacteria computational software MATLAB is used in converting the one-
and other environmental factors leading to pitting corrosion of directional data to nonparametric Gaussian Kernel PDF. To
steel structures in a marine environment. validate that Gaussian KDE is the better approach to develop PDF
for the obtained data set, Normality test was also conducted using
Anderson– Darling test considering Gaussian KDE.
3 Statistical Data Analysis Using Gaussian Kernel Figure 3 is the illustrative comparison of different PDFs with
parametric and nonparametric density functions for the data
Density Estimator adapted from the ASTM world-wide corrosion test [36]. The
The application of probability density estimation is the informal Gaussian KDE is tested between normal and logistic PDF. The X-
investigation of the properties of a given set of data [46]. A Kernel axis in Fig. 3 represents the calibrated data for power law constant
density estimator (KDE) is a nonparametric technique used to esti- B. As illustrated in Fig. 3, the nonparametric KDE function best
mate the PDF of the random variable. Nonparametric density esti- fits the data.
mation is particularly valuable for exploratory data analysis and in Figure 4 shows the normality test conducted using Anderson–
situations where available information is insufficient to specify a Darling test considering normal distributions for environmental
parametric model. These methods are “hypersensitive” and no parameter “salinity.” Based on this analysis, it is clear that the
assumptions are needed beyond the smoothing of the functions to data do not follow normal distributions, and the p-value is less
be estimated. Hence, the advantage of using Kernel distributions than or equal to the significance level; hence, the decision is to
is that it produces a nonparametric PDF that adapts itself to the reject the null hypothesis and conclude that the salinity data do
data rather than selecting a density with a particular parametric not follow normal distribution. Also, it is visually clear from the
form and estimating the parameters [47]. It must also be noted normality test that the data points do not follow the fitted distribu-
that the PDF, estimated using the nonparametric approach such as tion line and the data are skewed from a curved line.
KDE, requires less assumptions to be made about the distributions
of the observed data.
Lehmann [48] stated that KDE is an alternative to the paramet- 4 Development of Methodology-Pit Depth Modeling
ric approach in which one specifies a model up to a small number
of parameters and then estimates the parameters via the likelihood Long-term pitting corrosion depth prediction and identification
principle. The advantage of the nonparametric approach is that it in marine and offshore structures is a complex problem for a num-
offers a far greater flexibility in modeling a given data set and, ber of reasons [49]. The pit takes place on a very small scale with
unlike the classical approach, it is not affected by specification passive film in nanometers of thickness and with initiation sites of
bias. A nonparametric density estimator such as KDE can also be similar sizes. Immediately after initiation, the pit growth rate can
used for the summarization of Bayesian posteriors, classification, be extremely high [3]. Due to the complex nature of pitting corro-
and discriminant analysis. In addition to this, the application of sion, no specific methodology exists for predicting the precise pit-
nonparametric density estimators such as KDE is reported to be ting depth under long-term anaerobic conditions. Researchers
useful in Monte Carlo computational methods, such as the have performed several lab and field experiments in order to iden-
smoothed bootstrap method and the particle filter method [47]. tify pitting corrosion loss; however, most of these attempts were
In this study, the Gaussian kernel density estimator was used to in order to find the relationship between pitting depth and the
estimate the PDF for one-dimensional data. Given N independent environmental factors [3,20]. These assessments have not been
realizations XN ¼ ðX1 ; X2 ::::; XN Þ from an unknown continuous successful in developing a model that predicts the future pitting
PDF f on X, the Gaussian kernel density estimator is defined as depth in long-term anaerobic conditions.
In this study, an advanced methodology that integrates a proba-
bilistic approach with the phenomenological and empirical models
1X N
f^ð x; tÞ ¼ /ð x; Xi ; tÞ (5) is presented to predict the pitting depth. The BN is used as a prob-
N i¼1 abilistic data-modeling tool to model complex statistical relation-
pffiffiffiffiffiffiffi ðxX Þ2 =ð2tÞ ships between inputs such as environmental factors and materials
where /ðx; Xi ; tÞ ¼ ð1= p2pt
ffi Þe
is a Gaussian PDF with properties and pitting depth. The overall steps of the proposed
location Xi and scale t. The scale is usually referred to as methodology, as presented in Fig. 5, are divided into two main
bandwidth. parts: experimental analysis and probabilistic modeling. These

Journal of Offshore Mechanics and Arctic Engineering OCTOBER 2017, Vol. 139 / 051402-3

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

Fig. 3 Comparison of different available probability distributions

carefully considered. In the corrosion-related literature, there is a

strong history of empirical model development for long-term
exposure. Apart from the correlation studies conducted by several
researchers [14,20,22,50–53], the most frequent models are based
on a power law (Eq. (2)) which is applied for atmospheric corro-
sion loss/pitting depth.
The development of the proposed pit depth prediction method-
ology starts with the probabilistic approach, which is then inte-
grated with the experimental approach as demonstrated in Fig. 5.
In the first step of the probabilistic approach, the influencing fac-
tors that control the long-term pitting corrosion in marine environ-
ment are identified. The most influencing factors are selected
based on the technical review conducted by Bhandari et al. [10]
and by studying theoretical corrosion phenomenon model. Tem-
perature, salinity, pH, bacterial activity, velocity, exposure period,
and alloy effects are considered to be the most influencing factors
Fig. 4 A probability plot based on Anderson–Darling approach governing the long-term anaerobic conditions. Among these fac-
to identify the best fit distributions for the environmental tors, temperature and bacterial activities play an important role for
parameters such as salinity pit growth in both short- and long-term corrosion. Melchers [16]
reported that seawater temperature is considered to be the major
influencing factor in pitting corrosion loss. He also nominated
two parts are then combined together to predict the time- microbiological activity as another very important factor that con-
dependent pitting corrosion depth of a structural steel in long-term trols long-term pitting corrosion loss in marine environmental
marine environment. conditions. Microbiological growth is a function of available
In pitting corrosion depth modeling, it is crucial to grasp the energy sources and available nutrients, which enables microbio-
significance of the phenomenological model because it provides a logical metabolism [16].
strong understanding of the corrosion process and the contributing In the next step, the probability distributions are developed to
factors that control the pit growth. Likewise, an empirical corro- ascertain the influencing factors that control the pitting corrosion.
sion model that represents the pit depth growth should be These probability distributions are also called the priors because

Fig. 5 Development of a methodology for predicting the long-term time-dependent pit-

ting corrosion depth

051402-4 / Vol. 139, OCTOBER 2017 Transactions of the ASME

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

they assign a basic probability to the factors that lead to the partic- order to access generic data, to statistically evaluate data, and to
ular consequences. The prior probabilities are crucial in modeling develop incident distributions using the data [55].
the causes-consequences relationship using BN. The prior proba- Finally, the BN result and experimental results are integrated
bilities of the influencing factors are estimated from their corre- by PDFs of constants A and B. Using the maximum probability
sponding probability distributions, which are developed based on from BN, the actual numerical value of constants A and B is
available observational data. For example, Fig. 6 illustrates the obtained from their posterior PDFs developed using experimental
prior probability distribution for temperature. analysis. The obtained numerical value for constants A and B is
Subsequently, a BN model is developed to estimate the maxi- then applied to the power law model to predict the time-
mum occurrence probability of constants A and B. In the BN dependent pitting corrosion depth.
model, inputs such as the environmental factors and the material
properties are required in order to exercise the model. A condi-
tional probability table (CPT) is developed to estimate the proba-
bilities of each state of the variables and considers each 5 Application of the Methodology: A Case Study
combination of the parent states. The conditional probabilities for A case study is used to illustrate the application of the proposed
the child nodes were created by studying the relationship between methodology to predict the time-dependent pit depth of structural
different factors that control pitting corrosion depth in long-term steel in long-term marine exposure. The data were obtained from
exposure. Most of the information was derived from the field data. the world-wide test program undertaken by ASTM Task Group
In addition to field data, the theoretical corrosion loss model and [36], which assessed the relative corrosivity of seawater at 14 var-
additional expert knowledge were used to develop the conditional ious test sites. These field tests were accomplished by exposing
probability table. aluminum alloy, copper–nickel alloy, and mild steel specimens at
To exercise the experimental analysis, the first step requires the 14 distinctive test sites. However, this study is only focused on the
collection of long-term exposure pitting corrosion data (>2.5 pitting corrosion observations of mild steel specimen exposed for
years) along with the specific operating conditions. The power 0.5, 1, 3, and 5 years in marine environments. Regrettably, suita-
law model is then applied to represent the pit depth as the function ble data sources for the calibration of this model are very limited
of time. The model is then calibrated using a thorough re- and relatively few data sources provide sufficient information
examination of literature data and the extensive use of specially about operating (environmental) conditions. Therefore, in this
commissioned field tests. The parameters of the power law model case study, the ASTM worldwide corrosion tests provide the sole
(A&B) for each site are derived using linear stepwise regression. data observed for pitting corrosion depth as well as the operating
First, the hypothetical value is assigned to empirical constants A conditions. Their data were exclusively adopted to conduct this
and B to calculate the pitting corrosion depth c(t). Later, the pre- study and to test the applicability of the proposed methodology.
dicted pitting corrosion depth c(t) was compared with the experi- The pitting corrosion data and operating conditions adapted from
mental pitting corrosion depth. The standard deviation was ASTM worldwide test can be found in Appendices A and B,
calculated to find the amount of variation in a power law constant respectively [36].
values. Based on the ASTM field tests, the contributing factors of long-
Following this, the PDF for constants A and B is developed term pitting corrosion in marine environment are identified as
based on the numerical value obtained from fitting the model to temperature, salinity, pH, oxygen concentration, exposure period,
the observed data. The PDFs are determined empirically (that is, and velocity. Some of these factors, such as the existence of SRB
based entirely on available observational data). Furthermore, the and the alloying effect, were not calibrated when ASTM tests
assumed probability distributions are verified in the light of avail- were conducted. For this study, the authors have included the
able data, using the normality test as explained in Sec. 4. This test effects of SRB and alloy composition based on the phenomeno-
is implemented by considering several PDFs, and by carefully logical model and from several published works including the
selecting the one that best fits the calibrated data. This approach is recent review by Bhandari et al. [10]. SRB is the most active con-
generally used to eliminate the uncertainty associated with the tributor to pitting in long-time exposure of carbon steel materials
estimation of the probability of occurrence of undesirable events in marine environment because the metabolic activities result in
and their consequences [54]. As discussed in Sec. 4, quantitative sulfate ion reduction to hydrogen and sulfide. The sulfide ion
data analysis is also preferred as a part of the methodology in attacks the steel electrochemically causing more pitting corrosion
due to an increase in anodic/cathodic reactions caused by sulfate
reduction. The effect of oxygen concentration was omitted in this
study because it does not affect the long-term pitting corrosion
process (phase 4 in Fig. 1).
The prior probabilities for environmental parameters and mate-
rial properties are obtained through the distribution developed for
individual factors. The distribution for each particular factor is
established by considering the operating conditions of all 14 test
sites. When considering a PDF, the prior probability of different
states (such as high, average, and low value) is estimated for each
factor as demonstrated in Table 1. The high and low states repre-
sent the higher and lower bound in the PDFs.
After establishing the prior probabilities for the environmental
factors and material properties, a causes-consequences relation-
ship model is developed using BN. To predict the probability of
the constants (A and B), the contributing factors and their assigned
prior probabilities are applied to BN as listed in Table 1.
Figure 7 demonstrates the developed BN model, which esti-
mates the maximum probability of constants A and B, respec-
tively. The probability of these constants is a complex function of
environmental variables, such as temperature, exposure period,
and bacterial activities, as shown in Table 1. These variables
Fig. 6 Prior probability distribution of a influencing factor influence the occurrences probability of constants A and B, which
“temperature” indirectly controls the growth of the pitting corrosion depth.

Journal of Offshore Mechanics and Arctic Engineering OCTOBER 2017, Vol. 139 / 051402-5

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

Table 1 Parameter for predicting long-term pitting corrosion depth

Node State Operating conditions Probability Intermediate event State Probability

Temperature Minimum 0–18 0.15 Sulfate-reducing bacteria Present 0.909

Average 18–22 0.2 Absent 0.091
Maximum 22–30 0.65
Salinity Present Yes 0.67
Absent No 0.33 Alloying effect Yes 0.868
Acidic 4.0–5.5 0.3 No 0.132
pH Neutral 5.5–7 0.1
Basic 7.0–9.0 0.6
Steel composition Present >0.5% 0.6
Absent <0.5% 0.4
Carbon content Present Yes 0.67
Absent No 0.33
Exposure period Maximum >3.5 years 0.65
Average 2.5–3.5 years 0.3
Minimum <2.5 years 0.05
Velocity High >2.20 m/s 0.64
Low 0.220 m/s 0.36

Fig. 7 Developed BN model to predict pitting depth constants (A&B)

Therefore, the developed model should take into account the rela- was reported at Wrightsville Beach, NC and the lowest tempera-
tionship between these variables to predict the maximum proba- ture was reported in Sjaelland, Denmark [36]. For the short-term
bility of the constants. The final probability of constants A and B exposure, the pH of seawater is found to be consistent for all sites;
depends on the conditional probability matrix. As previously however, the effect of carbon dioxide can affect the range of pH
stated, the conditional probabilities table in this study is developed during long-term corrosion exposure [56]. The pH is used in three
based on the perception of theoretical corrosion model, from the different states in this study and, as presented in Table 1, the
field test data, and through expert judgments. acidic, neutral, and basic states are found to be within the range of
In the BN model, inputs such as environmental factors and 4–9. The exposure period is crucial for predicting long-term pit-
materials properties are used to exercise the model. As mentioned ting depth. Only long-term corrosion (more than 0.5 years) is con-
earlier, the effect of temperature has long been recognized as an sidered in this study. Melchers and Jeffrey [57] also considered
important influencing factor. This is confirmed in the BN model the data for 0.5 to more than 4 years of exposure in their study.
based on both theoretical corrosion principles and field observa- However, they stated that there is usually a considerable increase
tions [13,52]. In this study, the operating temperature is consid- in the instantaneous corrosion rates for both weight loss and for
ered to be within the range of 0–30  C. The highest temperature pitting depth after 2 years exposure.

051402-6 / Vol. 139, OCTOBER 2017 Transactions of the ASME

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

The effect of seawater salinity is conventionally considered to Indeed, there is a need of model calibration to extensive data-
be a very important factor with regard to pitting corrosion. It was derived new field observations. Data for four different sites are
found that the amount of salinity present in the seawater was rela- plotted in Fig. 8, where the line represents the power function
tively inconsistent for all the considered exposure sites. However, regression curves and the dot represents the actual data. This
it should be noted that the variations of salinity for different sites graph shows fairly acceptable agreements between the experimen-
were fairly minimal. Two binary states of “present or absent” are tal corrosion data and the power function regression curves
considered for the occurrences of seawater salinity in this study. obtained using Eq. (2). In Fig. 8, some of the graph does not
Similarly, the effect of alloy composition acts on the tendency for appear to have a perfect fit; this could be due to the lack of enough
an alloy to initiate pit and also affects the rate of corrosion [58]. observed data points from the field test.
The effect of the alloy composition may have a favorable effect in The parameters of the power law model (A&B) for each site
corrosion resistance ability of the materials under certain exposure are derived using linear stepwise regression. Then the PDF for
conditions such as seawater immersion and atmospheric condi- constants A and B is developed based on the numerical value
tions. In this study, the significant effect of alloy composition in obtained from fitting the model to the observed data. Figure 9
predicting the probability of constants (A and B) is observed as shows the posterior PDF for constants A and B developed from
demonstrated in the BN. The effect of water velocity on long-term the value obtained by fitting Eq. (2) to the calibrated data from all
pitting corrosion is also found to be significant. This was found to the test sites. In order to estimate the depth of the pit as a function
be in line with Melchers’ [7] report, which states that the water of time, the result obtained from BN is used to read the corre-
velocity increases the rate of pitting corrosion nonlinearly. Melch- sponding numerical value of A and B when considering PDFs.
ers [7] indicated that when corrosive products and/or marine Considering the maximum probability of 78.4% (obtained from
growth is disrupted or removed (as through erosion or abrasive BN), the numerical value of A and B is found to be 1.5 and 0.8,
action), the effect of water velocity on pitting corrosion can respectively.
become more severe. This study indicates that the effect of micro- In predicting the value of constants A and B from the empirical
biological activities is significant in predicting pitting corrosion pit depth model, comparable outcomes are reported in several pre-
depth. From the developed BN, it can be seen that the sulfate- viously published studies. Sowinski and Sprowls [60] considered
reducing bacteria increase the maximum values of constants A the power law model for various alloys exposed to the marine
and B which, in turn, increases the depth of pit as a function of environment for up to 30 years in U.S.; they found the value of B
time consistently. Southwell et al. [59] and Melchers and Jeffrey to range between 0.33 to more than 1. Melchers [61] considered
[24] have also discussed the influence of the microbiological the power law in his previous studies and indicated that B should
activities on both the short- and long-term corrosion of steels. be 0.5 for pure Fickian diffusion and homogenous rusts; however,
The BN model illustrated in Fig. 7 estimates the maximum pos- his calibration to field data invariably showed that B varies
sible probability of constants A and B. The value estimated repre- between 0.3 and 0.8. For short-term laboratory studies of the pit
sents the occurrence probability of constants A and B to a depth, this value may differ. Aziz and Godard [62] reported that
maximum probability when all the operating conditions and their for short-term laboratory tests on pitting depth, the value of B is in
conditional probabilities are satisfied. The final probability of con- the range of 0.33 and 0.55. Based on the previous literature, the
stants A and B is calculated as 0.784 (refer Fig. 6). It can be seen value for A has not been reported as comparable to B. However,
that the probability of A and B is remarkably similar; however, the Hou and Liang [26] established the multicorrelation coefficient
relationship between the variables and the conditional probability value of A and B as 0.82 and 0.9 after analyzing 16 years exposure
table is different. This will lead to a separate numerical value for data of steel. Their value of A shows a slight discrepancy from the
the individual constants (A and B) when considering the similar value reported in this study. The reason could be attributed to the
probability received from the BN. uncertainty associated with each data point, or to errors in their
In the experimental analysis, the corrosion loss/pitting depth experimental data and/or data processing.
versus the exposure time observation (for each of 14 sites) is plot- Southwell et al. [59] reported a data of 8-years exposure tests of
ted to examine whether the data agree with the phenomenological structural steel exposed to seashore and inland environments in
corrosion model trends as shown in Fig. 1. As illustrated in Fig. 8, the Panama Canal Zone. Their data were applied to the proposed
the data are considerably at variance with the pit depth growth methodology to estimate the corrosion loss model using power
law considered in this study. It clearly shows that the idealized law considering the operational conditions. Figure 10 shows the
model of the type shown in Fig. 1 is suitable for pitting corrosion. comparison of power law model calibrated to the actual data for a

Fig. 8 Localized corrosion data for mild steel exposed to surface seawater conditions at
four different sites (the solid lines represent power law model and dotted line represents
actual data)

Journal of Offshore Mechanics and Arctic Engineering OCTOBER 2017, Vol. 139 / 051402-7

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

Fig. 9 Posterior probability distribution for constants (A and B)

individual factors affecting the pitting depth. The BN model cal-

culates the percentage contributions by comparing each case with
the base case when all factors are active. The result shows that the
temperature and the exposure period are the most influencing fac-
tors affecting the pitting depth. Comparable investigations are
reported by several researchers [3,6,32,34,63]. Melchers [52]
reported that the pitting corrosion rate should be doubled for every
10  C increase in temperature. He further stated that the microbio-
logical reaction process for corrosion becomes faster with increase
in temperature after the initial stage; this suggests that the corro-
sion rate increases with the rise in seawater temperature. Melchers
[52] also confirmed that both temperature and the exposure time
play a vital role toward increasing the biological activities in

Fig. 10 The pitting corrosion loss model developed from 6 Conclusions

Ref. [59] data, which is applied to proposed methodology
A BN-based methodology is proposed for predicting pitting
corrosion depth of structural steel exposed to seawater in long-
term anaerobic conditions. Emphasis is placed on modeling the
pitting depth by combining the multiphase phenomenological and
variety of metals in the Panama Canal Zone. The compassion empirical models. The proposed model is critically important for
shows the remarkable agreement between actual data (lines) and the management of older offshore structures as well as for the
the power law model calibrated to the field data (dots). design of new infrastructure to ensure safety and serviceable long
Furthermore, Hou and Liang [26] used the regression analysis life. Application of this proposed methodology to a case study
approach that does not take account of the conditional dependency demonstrates that the use of the power law model is appropriate
among variables, whereas the BN is a powerful technique, which for estimating the long-term pitting corrosion depth. The calcu-
predicts the probabilities of constants. This study used the prior lated values of constants (A and B) using BN are compared with
knowledge of pitting corrosion phenomena, together with the the previously conducted studies and relatively comparable results
observed data. BN is also capable of establishing the probabilistic are achieved. Hence, the proposed model is efficient for predicting
relationship between variables to deal with the uncertainty associ- the pitting depth of a steel structure exposed to seawater for a lon-
ated with the data set. ger term. Moreover, the proposed BN-based methodology is also
This analysis confirms that the proposed methodology is suc- capable of providing the quantitative relationship between pitting
cessful in predicting the long-term anaerobic pitting corrosion corrosion and environmental factors. The model is capable of
depth of structural steel as a function of time in different environ- being updated whenever new information becomes available
mental and operational conditions. The developed methodology in through measurements, monitoring, and inspection. Seawater tem-
this study has advantages for estimating the pitting depth as com- perature and the longer exposure periods are found to be the most
pared to the previously used conventional techniques. This BN- influencing factors that increase the microbiological activities,
based methodology can be practiced to predict the future deterio- which subsequently increase the pitting corrosion depth in anaero-
ration of newly built or older offshore infrastructures using their bic conditions.
specific operating conditions. This novel concept for modeling pit
depth focuses on interpreting data in terms of a priori model using
Bayesian statistics rather than using data alone to indicate a Acknowledgment
model, such as typical correlation studies do. Using this methodol- Authors thankfully acknowledge support provided by National
ogy, the result illustrates that this approach is successful in provid- Centre for Maritime Engineering and Hydrodynamic (NCMEH)
ing new interpretations and better understanding of empirical and Australian Maritime College (AMC). The grant received from
“field” observations. University of Tasmania (2016) under the Research Enhancement
Furthermore, the sensitivity analysis is performed to investigate Grants Scheme (REGS) is highly appreciated. The authors greatly
the most influencing factors as well as their contributions to the appreciate the editors and the reviewers for their insightful com-
pitting depth. It is conducted by assigning a zero probability to the ments and suggestions.

051402-8 / Vol. 139, OCTOBER 2017 Transactions of the ASME

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

Appendix A: Pitting Corrosion Depth Data Adapted From ASTM [36]

Ocean City, NJ Wrightsville Beach, NC Key West, FL

Exposure period Corrosion rate Corrosion loss Exposure period Corrosion rate Corrosion loss Exposure period Corrosion rate Corrosion loss
(years) (lm/year) (pit depth) (mm) (years) (lm /year) (pit depth) (mm) (years) (lm /year) (pit depth) (mm)

0.5 192 0.36 0.5 161 0.31 0.5 136 0.16

0.5 213 0.33 0.5 165 0.5 0.5 146 0.36
1 209 1 1 123 0.68 1 107 0.81
1 322 1.31 1 213 1.28 1 119 0.39
3 100 1.57 3 87 2.44 3 124 2.17
3 107 1.87 3 214 4.44 3 84 1.19
5 92 2.24 5 68 1.39 5 84 1.23
5 100 1.46 5 224 4.11 5 75 1.40
Freeport, TX Port Huenema, CA Talara, Peru

0.5 120 0.27 0.5 550 0.95 0.5 196 0.48

0.5 105 0.37 0.5 245 0.44 0.5 180 0.35
1 105 0.53 2 113 0.8 1 147 0.73
1 91 2.58 2 121 0.83 1 137 0.84
3 99 2.77 2.9 125 3.5 3 129 3.4
3 137 6.07 2.9 173 1.88 3 126 2.52
5 135 6.07 5 207 2.12 5 131 2.98
5 86 2.59 5 136 2.59
KeAhole, Kona, HI Innisfail, Queensland, Australia Sakata Harbor, Japan

0.5 240 0.84 1 207 3.63 0.5 86 0.18

0.5 235 0.44 1 233 2.23 0.5 89 0.27
1 165 0.73 3 174 1.62 1 168 0.22
1 536 6.07 3 185 1.59 1 171 0.27
3.2 204 2.11 5 162 2.2 3.1 68 1.32
3.2 79 1.27 5 146 1.59 3.1 62 1.21
5 56 1.09 5 94 1.98
5 97 1.41 5 68 2.61
Genoa, Italy Kyndby Isefjord, Denmark Studsvik, Sweden

0.5 145 0.54 0.5 82 0.25 0.5 108 0.25

0.5 194 0.32 0.5 106 0.22 0.5 103 0.22
1 158 0.79 1.5 95 0.41 1 128 0.6
1 214 0.75 1.5 122 0.7 1 131 0.69
3 215 4.42 3.1 91 1.69 3.2 69 1
3 224 6.07 3.1 90 1.72 3.2 62 0.92
5 115 3.88 5 100 1 5 73 1.58
5 123 1.86 5 79 1.85 5 72 1.4
Bohus-Malmon, Sweden Isle of Wight, UK

0.5 103 0.18 0.5 226 0.36

0.5 95 0.21 0.5 156 0.31
1 86 0.35 1 148 0.44
1 94 0.43 1 355 1.06
3.1 99 2 3.1 84 1.07
3.1 119 1.83 3.1 104 1.42
5 101 1.88 5 87 1.19
5 79 2.25 5 79 1.58

Journal of Offshore Mechanics and Arctic Engineering OCTOBER 2017, Vol. 139 / 051402-9

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

Appendix B: Parameters for Predicting the Long-Term Pitting Corrosion Depth Adapted From ASTM [36]

No. Locations Temperature (  C) pH Oxygen concentration (ppm) Salinity

1 Ocean City, NJ 1.0–29.0 7.5–8.2 5.2–11.7 31–34

2 Wrightsville Beach, NC 7.0–30.0 7.9–8.2 5.0–9.6 31.8–37.6
3 Key West, FL 16–31 8.0–8.2 4.0–8.0 33–39
4 Freeport, TX 15–17 8.0–8.2 1.5–6.0 11.7–19.4
5 Port Huenema, CA 14–21 7.9–8.1 3.6–5.3 33
6 Talara, Peru 18–22 8.2 5.0–6.0 19.8
7 KeAhole, Kona, HI 18–22 8–8.3 6.0–14.0 34.6–35
8 Innisfail, QLD, Australia 23–30 8–8.5 5.1–6.5 31.7–37.2
9 Sakata Harbor, Japan 2.0–28.0 8.4 7.1–13 16.8–18.3
10 Genoa Harbor, Italy 11.0–25.0 8 4.5–6.0 35
11 Sjaelland, Denmark 0–18 7.5–8.0 4.6–5.2 18–28
12 Baltic Sea, Sweden 2.0–20.0 7.4–7.6 6.0–10.0 7.8–8.1
13 North Sea, Sweden 2.0–20.0 8.0–8.2 6.0–10.0 21–28
14 Isle of Wight, UK 5.0–22.0 7.8–8.4 8.8–11.8 34–34.6

References for Environmental Factors. Discussion,” Soc. Nav. Archit. Mar. Eng., Trans.,
[1] Chaves, I., and Melchers, R., 2013, “Long Term Localised Corrosion of Marine 113, pp. 306–329.
Steel Piling Welds,” Corros. Eng., Sci. Technol., 48(6), pp. 469–474. [21] Katano, Y., Miyata, K., Shimizu, H., and Isogai, T., 2003, “Predictive Model
[2] Saville, G., Richardson, S., and Barker, P., 2004, “Leakage in Ethylene Pipe- for Pit Growth on Underground Pipes,” Corrosion, 59(2), pp. 155–161.
lines,” Process Saf. Environ. Prot., 82(1), pp. 61–68. [22] Aziz, P., 1956, “Application of the Statistical Theory of Extreme Values to the
[3] Bhandari, J., Lau, S., Abbassi, R., Garaniya, V., Ojeda, R., Lisson, D., and Analysis of Maximum Pit Depth Data for Aluminum,” Corrosion, 12(10), pp.
Khan, F., 2017, “Accelerated Pitting Corrosion Test of 304 Stainless Steel 35–46.
Using ASTM G48; Experimental Investigation and Concomitant Challenges,” [23] Melchers, R. E., 2008, “Extreme Value Statistics and Long-Term Marine Pit-
J. Loss Prev. Process Ind., 47, pp. 10–21. ting Corrosion of Steel,” Probab. Eng. Mech., 23(4), pp. 482–488.
[4] Stewart, M. G., and Al-Harthy, A., 2008, “Pitting Corrosion and Structural [24] Melchers, R., and Jeffrey, R., 2008, “The Critical Involvement of Anaerobic
Reliability of Corroding RC Structures: Experimental Data and Probabilistic Bacterial Activity in Modelling the Corrosion Behaviour of Mild Steel in
Analysis,” Reliab. Eng. Syst. Saf., 93(3), pp. 373–382. Marine Environments,” Electrochim. Acta, 54(1), pp. 80–85.
[5] Melchers, R. E., 2014, “Microbiological and Abiotic Processes in Modelling [25] Melchers, R. E., 2015, “Using Models to Interpret Data for Monitoring and
Longer-Term Marine Corrosion of Steel,” Bioelectrochemistry, 97, pp. 89–96. Life Prediction of Deteriorating Infrastructure Systems,” Struct. Infrastruct.
[6] Abood, T. H., 2008, “The Influence of Various Parameters on Pitting Corrosion Eng., 11(1), pp. 63–72.
of 316L and 202 Stainless Steel,” Master thesis, University of Technology [26] Hou, W., and Liang, C., 2004, “Atmospheric Corrosion Prediction of Steels,”
(UOT), Baghdad, Iraq. Corrosion, 60(3), pp. 313–322.
[7] Melchers, R., 2004, “Pitting Corrosion of Mild Steel in Marine Immersion [27] Chaves, I. A., and Melchers, R. E., 2014, “Extreme Value Analysis for Assess-
Environment—Part 2: Variability of Maximum Pit Depth,” Corrosion, 60(10), ing Structural Reliability of Welded Offshore Steel Structures,” Struct. Saf.,
pp. 937–944. 50, pp. 9–15.
[8] Davydov, A., 2008, “Analysis of Pitting Corrosion Rate,” Russ. J. Electro- [28] Jain, S., Beavers, J. A., Ayello, F., and Sridhar, N., 2013, “Probabilistic Model
chem., 44(7), pp. 835–839. for Stress Corrosion Cracking of Underground Pipelines Using Bayesian
[9] Yevtushenko, O., Bettge, D., Bohraus, S., B€aßler, R., Pfennig, A., and Kranz- Networks,” CORROSION, Orlando, FL, Mar. 17–21, SPE Paper No. NACE-
mann, A., 2014, “Corrosion Behavior of Steels for CO2 Injection,” Process Saf. 2013-2616.
Environ. Prot., 92(1), pp. 108–118. [29] Melchers, R., 2005, “Statistical Characterization of Pitting Corrosion—Part 1:
[10] Bhandari, J., Khan, F., Abbassi, R., Garaniya, V., and Ojeda, R., 2015, Data Analysis,” Corrosion, 61(7), pp. 655–664.
“Modelling of Pitting Corrosion in Marine and Offshore Steel Structures: A [30] Melchers, R. E., 2012, “Modeling and Prediction of Long-Term Corrosion of
Technical Review,” J. Loss Prev. Process Ind., 37, pp. 39–62. Steel in Marine Environments,” Int. J. Offshore Polar Eng., 22(4), pp. 257–263.
[11] Bhandari, J., Khan, F., Abbassi, R., Garaniya, V., and Ojeda, R., 2016, [31] Melchers, R., 2003, “Modeling of Marine Immersion Corrosion for Mild and
“Reliability Assessment of Offshore Asset Under Pitting Corrosion Using Low-Alloy Steels—Part 1: Phenomenological Model,” Corrosion, 59(4), pp.
Bayesian Network,” CORROSION, Vancouver, BC, Canada, Mar. 6–10, SPE 319–334.
Paper No. NACE-2016-7070. [32] Melchers, R. E., 2006, “Advances in Mathematical Probabilistic Modelling of
[12] Melchers, R. E., and Jeffrey, R., 2011, “Bacteria Have Transient Influences on the Atmospheric Corrosion of Structural Steels in Ocean Environments,” Third
Marine Corrosion of Steel,” The University of Newcastle, NSW, Australia. International ASRANet Colloquium, Glasgow, UK, June 10–12, pp. 1–12.
[13] Guedes Soares, C., Garbatov, Y., and Zayed, A., 2011, “Effect of Environmen- [33] Caines, S., Khan, F., and Shirokoff, J., 2013, “Analysis of Pitting Corrosion on
tal Factors on Steel Plate Corrosion Under Marine Immersion Conditions,” Steel Under Insulation in Marine Environments,” J. Loss Prev. Process Ind.,
Corros. Eng., Sci. Technol., 46(4), pp. 524–541. 26(6), pp. 1466–1483.
[14] Ha, H.-Y., and Kwon, H.-S., 2012, “Effects of pH Levels on the Surface [34] Melchers, R., 2006, “Examples of Mathematical Modelling of Long Term Gen-
Charge and Pitting Corrosion Resistance of Fe,” J. Electrochem. Soc., 159(9), eral Corrosion of Structural Steels in Sea Water,” Corros. Eng., Sci. Technol.,
pp. C416–C421. 41(1), pp. 38–44.
[15] Nesić, S., 2007, “Key Issues Related to Modelling of Internal Corrosion of Oil [35] Melchers, R. E., and Jeffrey, R., 2008, “Probabilistic Models for Steel Corro-
and Gas Pipelines–A Review,” Corros. Sci., 49(12), pp. 4308–4338. sion Loss and Pitting of Marine Infrastructure,” Reliab. Eng. Syst. Saf., 93(3),
[16] Melchers, R., 2014, “Modelling Long Term Corrosion of Steel Infrastructure in pp. 423–432.
Natural Marine Environments,” Understanding Biocorrosion: Fundamentals [36] Phull, B. S., Pikul, S. J., and Kain, R. M., 1997, Seawater Corrosivity Around
and Applications, Vol. 66, The University of Newcastle, Callaghan, Australia, the World: Results From Five Years of Testing, Vol. 1300, ASTM Interna-
p. 213. tional, West Conshohocken, PA, pp. 34–73.
[17] Szklarska-Smialowska, Z., 1986, “Pitting Corrosion of Metals,” National Asso- [37] Nielsen, T. D., and Jensen, F. V., 2009, Bayesian Networks and Decision
ciation of Corrosion Engineers (NACE), Houston, TX. Graphs, Springer Science & Business Media, Aalborg, Denmark.
[18] Ryan, M. P., Williams, D. E., Chater, R. J., Hutton, B. M., and McPhail, D. S., [38] Neapolitan, R. E., 2004, Learning Bayesian Networks, Pearson Prentice Hall,
2002, “Why Stainless Steel Corrodes,” Nature, 415(6873), pp. 770–774. Upper Saddle River, NJ.
[19] Melchers, R. E., and Wells, T., 2006, “Models for the Anaerobic Phases of [39] Pearl, J., and Russell, S., 1998, “Bayesian Networks,” University of California,
Marine Immersion Corrosion,” Corros. Sci., 48(7), pp. 1791–1811. Los Angeles, CA.
[20] Guedes Soares, C., Garbatov, Y., Zayed, A., Wang, G., Melchers, R., and Paik, [40] Yang, M., Khan, F., and Amyotte, P., 2015, “Operational Risk Assessment: A
J., 2005, “Non-Linear Corrosion Model for Immersed Steel Plates Accounting Case of the Bhopal Disaster,” Process Saf. Environ. Prot., 97, pp. 70–79.

051402-10 / Vol. 139, OCTOBER 2017 Transactions of the ASME

Downloaded From: on 06/11/2017 Terms of Use: http://www.asme.or

[41] Abimbola, M., Khan, F., Khakzad, N., and Butt, S., 2015, “Safety and Risk [53] Melchers, R., and Ahammed, M., 1994, Nonlinear Modelling of Corrosion of
Analysis of Managed Pressure Drilling Operation Using Bayesian Network,” Steel in Marine Environments, University of Newcastle, NSW, Australia.
Saf. Sci., 76, pp. 133–144. [54] Khan, F. I., Haddara, M. M., and Bhattacharya, S. K., 2006, “Risk-Based Integ-
[42] Bhandari, J., Abbassi, R., Garaniya, V., and Khan, F., 2015, “Risk Analysis of rity and Inspection Modeling (RBIIM) of Process Components/System,” Risk
Deepwater Drilling Operations Using Bayesian Network,” J. Loss Prev. Process Anal., 26(1), pp. 203–221.
Ind., 38, pp. 11–23. [55] Zhu, W., 2008, “An Investigation Into Reliability Based Methods to Include
[43] Khakzad, N., Khan, F., and Amyotte, P., 2013, “Dynamic Safety Analysis of Risk of Failure in Life Cycle Cost Analysis of Reinforced Concrete Bridge
Process Systems by Mapping Bow-Tie Into Bayesian Network,” Process Saf. Rehabilitation,” Master thesis, RMIT University, Victoria, Australia.
Environ. Prot., 91(1), pp. 46–53. [56] Pardo, A., Otero, E., Merino, M., Lopez, M., Utrilla, M., and Moreno, F., 2000,
[44] Abimbola, M., Khan, F., and Khakzad, N., 2014, “Dynamic Safety Risk Analy- “Influence of pH and Chloride Concentration on the Pitting and Crevice Corro-
sis of Offshore Drilling,” J. Loss Prev. Process Ind., 30, pp. 74–85. sion Behavior of High-Alloy Stainless Steels,” Corrosion, 56(4), pp. 411–418.
[45] Bhandari, J., Arzaghi, E., Abbassi, R., Garaniya, V., and Khan, F., 2016, [57] Melchers, R., and Jeffrey, R., 2008, “Modeling of Long-Term Corrosion Loss
“Dynamic Risk-Based Maintenance for Offshore Processing Facility,” Process and Pitting for Chromium-Bearing and Stainless Steels in Seawater,” Corrosion,
Saf. Prog., 35(4), pp. 399–406. 64(2), pp. 143–154.
[46] Silverman, B. W., 1986, Density Estimation for Statistics and Data Analysis, [58] Melchers, R. E., 2004, “Effect of Small Compositional Changes on Marine
CRC Press, Boca Raton, FL. Immersion Corrosion of Low Alloy Steels,” Corros. Sci., 46(7), pp. 1669–1691.
[47] Botev, Z. I., Grotowski, J. F., and Kroese, D. P., 2010, “Kernel Density Estima- [59] Southwell, C., Forgeson, B., and Alexander, A., 1958, “Corrosion of Metals in
tion Via Diffusion,” Ann. Stat., 38(5), pp. 2916–2957. Tropical Environments—Part 2: Atmospheric Corrosion of Ten Structural
[48] Lehmann, E. L., 1990, “Model Specification: The Views of Fisher and Neyman, Steels,” Corrosion, 14(9), pp. 55–59.
and Later Developments,” Stat. Sci., 5(2), pp. 160–168. [60] Sowinski, G., and Sprowls, D. O., 1982, Atmospheric Corrosion, Wiley, New
[49] Frankel, G., and Sridhar, N., 2008, “Understanding Localized Corrosion,” York, pp. 297–328.
Mater. Today, 11(10), pp. 38–44. [61] Melchers, R. E., 2008, “Development of New Applied Models for Steel Corro-
[50] Abdel-Ghany, R., Saad-Eldeen, S., and Leheta, H., 2008, “The Effect of Pitting sion in Marine Applications Including Shipping,” Ships Offshore Struct., 3(2),
Corrosion on the Strength Capacity of Steel Offshore Structures,” ASME Paper pp. 135–144.
No. OMAE2008-57844. [62] Aziz, P., and Godard, H. P., 1952, “Pitting Corrosion Characteristics of
[51] Al-Fozan, S. A., and Malik, A. U., 2008, “Effect of Seawater Level on Corro- Aluminum-Influence of Magnesium and Manganese,” Ind. Eng. Chem., 44(8),
sion Behavior of Different Alloys,” Desalination, 228(1), pp. 61–67. pp. 1791–1795.
[52] Melchers, R., 2002, “Effect of Temperature on the Marine Immersion Corro- [63] Moayed, M. H., Laycock, N., and Newman, R., 2003, “Dependence of the Critical
sion of Carbon Steels,” Corrosion, 58(9), pp. 768–782. Pitting Temperature on Surface Roughness,” Corros. Sci., 45(6), pp. 1203–1216.

Journal of Offshore Mechanics and Arctic Engineering OCTOBER 2017, Vol. 139 / 051402-11

stats on 06/11/2017 Terms of Use: http://www.asme.or