The 1st International Conference on Business, Management and Information Systems 2019

Abstract ID: ICBMIS-2019-065

A Review on Pipeline Condition Prediction Methods

Firdaus. Basheer1, Mohamed Saleem Haja Nazmudeen2
1
Universiti Teknologi Brunei
[email protected]
2
Universiti Teknologi Brunei
[email protected]

Abstract

Predictive maintenance is considered as cost-efficient and acts as an enabler for businesses to

be more competitive. Maintenance of pipelines can be complex and costly especially when it
is located underground or subsea. Based on the literature survey, many methods have been
developed to predict pipeline conditions. To date, however, none have focused on the eco-
nomic comparison of the various prediction methods. The paper reviews several common
methods of pipeline prediction in order to facilitate future assessment of the economic costs
of each method. These include Split System Approach, Artificial Neural Network, Reliability
Analysis, Monte Carlo Simulation, Standard Data Structure Approach, Bayesian Network,
Fuzzy Logic method and other probabilistic assessments that have produced promising re-
sults. The different types of machine algorithms are compared in terms of accuracy. Artificial
Neural Network is seen as the most popular choice of methods and produced better accuracy
compared to the rest of the machine learning algorithm.

Keywords: pipeline reliability; Machine Learning Algorithm; Artificial Neural Network

1 Introduction

The world is currently facing industrial revolution 4.0 where digitalization plays an important
part in our daily lives. Businesses are racing to start using digitalization to their benefit as it is
seen to improve performance and efficiency, leading to cost optimization, which in turn gen-
erates more revenues for the company in order to thrive in today’s competitive environment.
One area of interest is the application of digitalization for efficient maintenance of equip-
ment. The maintenance of equipment is crucial for businesses as it eliminates or reduces the
number of failures that may occur during production which may disrupt the supply chain.

Maintenance and diagnosis play an important role in promoting machine uptime, as well as
the potentially impacting operating costs. Some data suggest that equipment maintenance can
exceed 30 % of total operating costs, or between 60 and 75 % of equipment lifecycle cost
(Dhillon, 2006). Machinery that is not functioning as per its design specification may nega-
tively impact production yield (e.g. scrapped product). Equipment maintenance can have an
overall positive impact on capital costs. The lifetime of machinery may be enhanced by limit-
ing the number of times it enters a state of disrepair, while ongoing costs may be reduced by
using predictive and preventive maintenance strategies to optimize the scheduling of mainte-
nance activities (O’Donovan, Leahy, Bruton, & O’Sullivan, 2015). Predictive maintenance is

532
F. Basheer and, H. Saleem

favored due to its ability to predict failure and decrease hands-on tool time required to per-
form maintenance work in the field, making it an economical and cost-efficient approach.

Several methods for pipeline condition prediction have shown promising outcomes. Howev-
er, studies conducted on these methods focus primarily on accuracy- either on improving the
accuracy of the method in question or comparing the accuracy of various types of pipeline
prediction methods. To date, none of the work involved cost analysis to evaluate the most
economical and cost-efficient pipeline predictive maintenance method. This paper examines
existing studies on methods of pipeline condition prediction and compares their areas of fo-
cus, objective of prediction, the types of data collected and the outcomes. This analysis can
be used to conduct a future study on the economic cost of these methods. The methods in
question are split system approach, reliability analysis, standard data structure approach, ma-
chine learning algorithms such as Artificial Neural Network, regression analysis, Decision
Tree techniques, Monte Carlo simulation, Markov modeling, Stochastic model, Bayesian
network and Fuzzy based model.

This paper first justifies the case for predicting pipeline conditions and then outlines the dif-
ferent types of methods that have been developed. Research work focusing on the same pre-
diction methods are grouped together to demonstrate the adaptability of the method in solving
different types of problem. Section 3 presents a comparison of the accuracy of the machine
learning algorithm across different methods. The analysis of existing review of the methods is
tabulated in section 4 and this paper concludes with an assessment of the most viable meth-
ods to be further analysed for economic cost assessment.

2 Cases for Pipeline Prediction and Methods Used.

2.1 Split System Approach.

The problem areas widely focused in the literature are the influences of corrosion and creep
on the pipeline failures (Sun, Ma, & Morris, 2009). Sun et al, however, considers the impact
of preventive maintenance (PM) on the system reliability of pipelines. In their research work,
they highlight the need for meeting an effective model that can accurately improve the overall
reliability of the entire pipeline and cost-effective preventive maintenance strategy. They em-
phasize that pipelines are a complex system and imperfect repairs would need to be consid-
ered since most of the assets are assumed to be subjected to PM actions periodically, based
on time-based preventive maintenance method (TBPM). TBPM cannot model the effect of
different PM actions on the system reliability nor does it consider system configuration.
When modeling imperfect maintenance, it uses a method similar to the improvement factor
method. To make an optimal decision, the reliability changes of the entire pipeline with dif-
ferent PM strategies have to be predicted accurately. However, the existing mod-
els/methodologies cannot meet industrial need for the explicit prediction of the reliability im-
provement and the consideration of multiple imperfect repairs because only partial pipeline
will be repaired during a PM action. Therefore, to address this issue, their research aims to
predict the reliability of pipelines with multiple imperfect PM actions using split system ap-
proach (SSA) by considering both TBPM and RBPM. They developed a model for the effects
of PM activities from pipeline repair history and inspection results using reliability function
(Sun et al., 2009).

533
Literature Review on Pipeline Condition Prediction Method

2.2 Reliability Analysis Method

Ahammed (1998) addresses the problem of estimating the remaining strength of corroded
pipelines based on deterministic approaches that are free of any uncertainty. In his research,
he carried out reliability analysis which can be used to assess the safety and integrity, and al-
so to predict the remaining life of a corroded pipeline with further corrosion growth. The re-
sulting information can be used to predict safe operating pressure at any time and to prepare
an effective and economic inspection, repair and replacement schedules (Ahammed, 1998).

Hallen et al proposed structural reliability analysis in their research work to evaluate the in-
tegrity of in-service corroded pipelines in Mexico using high-resolution magnetic flux leak-
age (MFL) or ultrasonic technology-based (UT) in line inspection tool (Hallen, Caleyo,
González, & Fernández-Lagos, 2003).

Pandey deals with uncertainties in the decision-making regarding pipeline maintenance, the
frequency of future inspections, and the estimation of associated costs involved with the as-
sessment of condition. The primary objective was to formulate a quantitative probabilistic
approach with the aim of determining the optimal inspection interval and the repair strategy
that would maintain adequate reliability over the remaining service life of the pipeline
(Pandey, 1998).

2.3 Bayesian Network Method

Reza et al focus on dynamic modeling for predicting pipeline performance. This research was
driven by the fact that a majority of studies rely on static models to analyze the failure predic-
tion based on corrosion rates, historical data and/or expert judgment. The main objective of
their research work was to propose a preliminary dynamic reliability model for aging pipeline
performance prediction, by analyzing the pipeline residual life based on internal and external
corrosion rates and several other variables that influence the performance of the system. The
initial model of their research was to utilize qualitative inputs such as experts’ judgment and
assumptions in the model to fill the data gap. As such, the Bayesian Network method was the
preferred choice to build dynamic probabilistic models for its ability to manage and over-
come uncertainties. The network diagram developed in the study was based on the causality
of each variable, from the cause of the corrosion to the effect of the failures. It takes into con-
sideration that corrosion failure is a time-dependent process as the analysis utilizes inspection
history data from several years (Aulia, Tan, & Sriramula, 2019).

Amit et al address the challenges of locating the internal corrosion damage in oil and gas
pipelines that are hundreds of miles long due to the presence of large uncertainties in flow
characteristics, pre-existing conditions, corrosion resistance, elevation data and test meas-
urement despite recent advances in inline inspection technology In their research, they have
considered probabilistic methodology based on internal corrosion direct assessment (ICDA).
Bayesian Network was used to predict the most probable corrosion damage location along the
pipelines using physical models for flow, corrosion rate, and inspection information as well
as uncertainties in elevation data, pipeline geometry and flow characteristics (Kale, Thacker,
Sridhar, & Waldhart, 2004).

2.4 Stochastic Method

534
F. Basheer and, H. Saleem

Mahmoodian and Li conducted research on a reliability-based methodology for assessment of

corroded steel pipes. A stochastic model for strength loss was developed which is related to
key factors that affect the residual strength of a corroded pipe. A pipeline system fails when
its residual strength falls below its operating pressure. An analytical time-variant method is
employed to quantify the probability of failure due to corrosion, so as to determine with con-
fidence the time in which the pipeline may fail and hence require repairs. To deal with the as-
sessment of pipelines with have more than one corrosion pit, system reliability analysis
method is employed. The proposed methodology provides a rational and consistent approach
to conduct a quantitative assessment of pipeline failures. The methodology proposed in this
study can also be extended to other aging structures and infrastructures subjected to localized
deterioration. Monte Carlo simulation technique was applied to verify the results of the ana-
lytical method (Mahmoodian & Li, 2017).

2.5 Monte Carlo Simulation

Shuai et al conducted a study to define an alternative to the conservative model used to pre-
dict burst capacity which is costly in terms of maintenance. In their research, Shuai et al de-
veloped a new burst prediction model that has a good precision for corroded pipelines Based
on this model, a probabilistic method is presented to predict the failure probability of corrod-
ed pipeline by using the Monte Carlo (MC) simulation method. The parameter sensitivity and
model sensitivity of uncertainties are analyzed to find the parameters which have the most in-
fluence on the safety of corroded pipelines (Shuai, Shuai, & Xu, 2017).

Shu Xin li et al used the cumulative distribution function (CDF) to estimate cumulative prob-
able failure for a pipeline, calculated using the Monte Carlo simulation technique (Li, Yu,
Zeng, Li, & Liang, 2009).

Ossai et al examine the problem of having limited information on failure probability estima-
tion of corroded pipelines. The work aims to predict the failure probability and estimate the
reliability of corroded pipelines using pipeline corrosivity index (PCI) in consideration of the
retained pipe-wall thickness of the corroded pipeline at a given time. Monte Carlo simulation
was used along with Markov modeling to predict the survival probability of corroded pipe-
lines at a given time for different corrosion wastage rates. Weibull probability function was
used to calculate the time-lapse for a pipeline leakage. Markov modelling and Monte Carlo
simulation were chosen because they were used for estimating the failure probability of cor-
rosion defect growth of corroded pipelines and other structures by numerous
researchers(Ahammed, 1998; Breton, Sanchez-Gheno, Alamilla, & Alvarez-Ramirez, 2010;
Keshtegar & Miri, 2014; Pandey, 1998; Zhang et al., 2012). The expectation from the model
developed in this research is that it would be a viable tool for managing the integrity of cor-
roded aging pipeline (Ossai, Boswell, & Davies, 2016).

2.6 Standard Data Structure Approach

Alison et al research mainly focus on developing an understanding of the failure modes and
mechanisms of buried pipe infrastructure systems that are used in a municipal water system.
Their research also considers the pipe material's advantages and disadvantages as well as to
understand all the parameters affecting water pipe infrastructure systems (St. Clair & Sinha,
2014). In order to develop this understanding, they established a standard data structure for
predicting the remaining physical life and consequence of failure of water pipes through

535
Literature Review on Pipeline Condition Prediction Method

physical or structural parameters, operational or functional parameters and environmental pa-

rameters. Correlating the various pipe material types with their pipe life cycle failure modes
and mechanisms are crucial in defining the various parameters that affect the water pipelines
(St. Clair & Sinha, 2014).

2.7 Artificial Neural Network

Najafi et al in their research address the challenge of completing more than 304 miles of in-
spection on 2200 miles of local sewers for cracks, collapses and blockages as part of their
sewer system evaluation survey (SSES) effort. They developed a pipeline condition predic-
tion model based on the neural network algorithm which can identify pipelines at risk of fail-
ure so that inspections can be prioritized (Najafi, M., and Kulandaivel, 2005).

Kexi Liao et al came up with an alternative approach for internal corrosion direct assessment
of wet gas pipeline and developed a model using ANN (Artificial Neural Network) based on
116 groups of data from wet gas gathering pipelines (Liao, Yao, Wu, & Jia, 2012).

El Abbasy et al conducted several studies on the prediction model for pipeline using the Arti-
ficial Neural Network method. In 2014, they studied condition assessment and prediction
models in order to develop expected deterioration curves for offshore oil and gas pipelines
based on several factors besides corrosion (El-Abbasy, Senouci, Zayed, Mirahadi, &
Parvizsedghy, 2014a). In 2016, they developed forecasting models using ANN, regression
analysis and decision tree techniques to predict the different types of pipeline failure by stud-
ying the critical factors affecting the condition of oil and gas pipelines and the design condi-
tion forecasting models for unpiggable oil and gas pipelines. They also developed expected
deterioration curves for the pipelines considered in their research exercise. The models were
tested, validated and compared with their previously developed models for piggable offshore
pipeline (El-Abbasy, Senouci, Zayed, Parvizsedghy, & Mirahadi, 2016).

Ren et al considered the use of back propagation (BP) neural network to predict the corrosion
rate in long distance natural gas pipeline. The input parameter was based on natural gas pipe-
line mileage, elevation difference, pipe inclination, pressure, and Reynolds number, whereas
the maximum average corrosion rate of the pipeline was used as an output parameter, to pre-
dict natural gas pipeline internal corrosion rate (Ren, Qiao, & Tian, 2012).

Achim et al explored the possibility of using neural networks with the aim of improving fail-
ure predictions for CWW (City West Water Ltd), replacing regression analysis on past failure
histories. The models were based on age only and are considered unsatisfactory, due to rela-
tively low correlations between predicted and actual failure rates (Righetti 2001). The neural
network developed was compared against previous modeling STPM (shifted time power
model) and STEM (shifted time exponential model) (Achim, Ghotb, & McManus, 2007).

2.8 Fuzzy Based Model

Senouci et al developed a fuzzy-based model for oil and gas pipeline prediction in extension
to their previous research activity on the same subject that involved regression and artificial
neural network (ANN). The model took into consideration the prediction of failure types such
as corrosion, mechanical, third-party, natural hazard, and operational. The model was built
based on historical data collected from a report prepared by CONCAWE (Davis et al. 2010).

536
 F. Basheer and, H. Saleem

However, the developed models using regression analysis and ANN suffered from certain
shortcomings, such as the precision incorrectly predicting the natural hazard type of failure.
The data set was also subjective in nature, which led to many assumptions. They believed a
fuzzy-based model will tackle the problems of uncertainties and subjectivity in the available
variables. The main objectives of the study conducted were to identify and study the critical
failure causes of oil pipelines, and to design a failure type prediction model for such pipelines
(Senouci, El-Abbasy, & Zaye, 2014).

Maneesh and Tore considered the fuzzy logic method to come up with risk-based inspection
planning. They realized that an important feature of plant operation is the availability of a
considerable amount of information regarding the corrosion of pipelines, which are qualita-
tive and imprecise knowledge. This knowledge cannot be easily applied using traditional
mathematics such as differential and algebraic equations. Hence, the incorporation of the sub-
jective knowledge along with the objective information is crucial in order to develop a practi-
cal model that is simple to use and flexible enough to be modified according to the require-
ment of plant section and field data. The fuzzy logic approach is an attractive option when (a)
the available data is not precise enough to allow conventional methods of computing; (b)
there is a significant tolerance to the imprecision, allowing the development of a simple and
robust model; (c) the available information is too incomplete to allow the development of a
proper model; and (d) the model is too difficult to allow easy computation [1–5]. The intent
of their research was to develop an optimal inspection plan considering various sources of in-
formation like recommended models, standards and their own plant experience (Singh &
Markeset, 2009).

3 Machine Learning Algorithm Methods Comparison

The literature review indicates that most pipeline condition prediction methods utilize ma-
chine learning algorithms. The algorithm is widely used, not only for predictive maintenance
but also for data mining in other areas such as cloud computing, medical and banking sectors.
The table below compares the accuracy of the methods as per gathered from the literature
survey.

Accuracy Rate
Tahyudin Caruana Kalanaki Mollaza Ahmed Correa
Machine Learning (Tahyudin, Utami, (Caruana & (Kalanaki & de et al Senouci (Correa,
Algorithm Methods & Amborowati, Niculescu- Soltani, 2013) (Mollaza (Senouci et Bielza, &
2013) Mizil, 2006) de, Omid, & al., 2014) Pamies-
Arefi, 2012) Teixeira, 2009)
Artificial Neural 100% Average 99.92% 96.33% 92% 96.35%
Network
Regression Analysis 100% Poor - 90%

Support Vector Ma- 100% Excellent 100% 95.67%

chine
Bayesian Network 75.16% Poor 94.33% 94.84%

Decision Trees 80.01% Poor 94.67%

Neuro Fuzzy 99.80% 93%

Table 1 Comparison of machine learning algorithm (El-Abbasy et al., 2014a)

537
Literature Review on Pipeline Condition Prediction Method

From the table below, it can be summarized that ANN is widely used in most research activi-
ties, and has a good accuracy rate. Although Support Vector Machine (SVM) is not discussed
in this literature survey, from the table above it can be seen that SVM has an excellent out-
come and outperforms ANN in terms of accuracy.

4 Summary of Literature Review

This literature survey paper reviews 19 articles, forming a data bank or database of infor-
mation on various prediction methods of pipeline condition assessment. The information is
summarized and tabulated below (Table 2).

5 Conclusion

Based on the literature review, there are many methods that can be used for the prediction of
pipeline condition and each method has its own limitations. Artificial Neural Network (ANN)
is widely used for pipeline prediction and the extent of the prediction can go beyond the in-
ternal corrosion based on pipeline failure history. The fuzzy model is also seen as the next
promising method after ANN. Fuzzy logic can outperform ANN, but it can be inaccurate if
the models are not trained enough. For probabilistic analysis, Monte Carlo simulation is gen-
erally used to calculate or analyze the reliability of the pipeline.

Most of the methods above are using machine learning algorithm to process, train and ana-
lyze data to predict pipeline failure. In this literature review, the effort is also extended to de-
termine the accuracy of various machine learning algorithms as used in other applications.
Based on the outcome, ANN is frequently used and has a good accuracy rate in general. Alt-
hough Support Vector Machine (SVM) outperformed ANN in some of the studies conducted,
not many pipeline predictions utilize the SVM method.

The common methods that have been identified here would be used for future works on eco-
nomics assessment to determine the most viable method technically and economically. This
would help the pipeline operator in selecting the best methods for their pipeline predictive
maintenance program in the long run.

538
539
 The 1st International Conference on Business, Management and Information Systems 2019

Table 2 Summary of Literature Survey

Method Paper Problem addressed Objective Data Collected Outcome
Split System Approach (Sun et al., Assets are assumed to have PM ac- To predict the reliability of Pipeline repair his- The prediction model can deal with multiple im-
2009) tions periodically based on time-based pipelines with multiple im- tory perfect repairs more effectively and more accurate-
preventive maintenance (TBPM) perfect preventive mainte- ly.
method. This method cannot model the nance actions.
effects of different PM actions on sys-
tem reliability.
Bayesian Network (Kale et al., Locating internal corrosion damage in To predict the most proba- Pipeline Inspection The developed model was able to focus on loca-
2004) gas pipelines is challenging due to the ble corrosion damage loca- data tions with a high probability of damage.
presence of large uncertainties in flow tion along the pipelines, and
characteristics, pre-existing conditions, then update this prediction
corrosion resistance, elevation data, using inspection data.
and test measurement.
Reliability Analysis (Ahammed, For a pipeline containing live corro- To quantify the reduction in Depth of metal loss This methodology allows calculation of the relia-
1998) sion defects. safety and hence the re- and axial length bility index, failure probability, relative contribu-
maining life for deteriorat- tion of the random variables and also allows the
ing corroded steel pressur- study of the effect of variation of the variance of
ized pipelines at any point the random variables on the overall pipeline relia-
in time. bility.
Structural Reliability (Hallen et al., Significant effort required to assess the To formulate operating and Inspection data The method produced cost-effective maintenance
Analysis 2003) corroded pipeline using data obtained maintenance strategies for strategies and target reliability levels are used as
from high-resolution magnetic flux the safe operation of this reference to assess the condition of the pipeline as
leakage (MFL) or ultrasonic technolo- pipeline over its targeted measured by its failure probability.
gy-based (UT) ILI tools also known as service life.
‘smart pigs’
Monte Carlo Simulation (Li et al., The traditional design code is unable To employ Cumulative Dis- Pipeline data CDF is more appropriate to characterize the pipe-
2009) to predict the failure probability of tribution Function (CDF) in line failure probability compared to Probability
corroded pipelines at a given time. a probabilistic model to Density Function (PDF) and reliability index since
predict pipeline failure for the cumulative effect will be more obvious as ex-
the underground pipeline. posure period of pipeline increases. Further effort
is required for more precise predictions.

540
 F. Basheer and, H. Saleem

Monte Carlo Simulation (Shuai et al., Various models were used to predict To develop a new burst pre- Pipeline data The comparison results indicated that the proposed
2017) the burst capacity of corroded pipe- diction model that has a model shows good precision when compared with
lines that are conservative and may good precision for corroded commonly used codes.
cause many defects that need to be re- pipelines using the Monte
paired which lead to more costs. Carlo (MC) simulation
method.
Probabilistic method (Pandey, The assessment of the extent of corro- To determine the optimal Inspection data The probabilistic analysis results are able to deter-
1998) sion for underground pipelines carry- inspection interval and the mine the optimal inspection interval and the repair
ing oil and gas that are inaccessible for repair strategy that would strategy that would maintain adequate reliability
direct inspection. maintain adequate reliability throughout the service life. The method was vali-
over the remaining service dated using Monte Carlo simulation. The proposed
life of the pipeline. approach can minimize the overall cost of inspec-
tion and repair of existing pipelines.
Dynamic modeling using (Aulia et al., Pipeline corrosion - pipeline reliability To propose a preliminary Operational data Bayesian networks were seen effective for analyz-
Bayesian Networks 2019) studies rely on static models to analyze dynamic reliability model and inspection re- ing the corroded pipeline performance. The dy-
the failure prediction based on corro- for aging pipeline perfor- sults namic model produced a realistic result of the pipe-
sion rates, historical data and/or expert mance prediction. line residual life, which lead to an efficient
judgments. estimate of the future condition of the pipeline
based on the current and historical data.
Reliability deterministic (Noor, The inherent uncertainties embedded To manipulate the polyno- Metal Loss Infor- The method has improved the capability of predict-
prediction Ozman, & within metal-loss data play significant mial equation of safety fac- mation; ing the future growth of corrosion defects deter-
Yahaya, roles in reducing the accuracy of pipe- tors in the DNV RP-F101 to ministically. The new equation is derived so that
2011) line future assessment. make the capacity equation the partial safety factors can be recalculated to
capable of predicting the fu- consider the growth of uncertainties as corrosion
ture growth of defects by progresses.
deriving a time-function
standard deviation equation
of the inspection tool.
Markov Modelling and (Ossai et al., Limited information on failure proba- To predict the survival Inline inspection The probability of failure of internally corroded
Monte Carlo Simulation 2016) bility estimation of corroded pipelines probability of corroded data of X52 grade pipelines that are expected to fail by small leakage,
using the Pipeline Corrosivity Index pipelines at a given time for onshore transmis- burst and rupture are considered.
(PCI). different corrosion wastage sion pipeline was
rates whilst using Weibull used.
probability function to cal-
culate the time-lapse for
pipeline leakage.

541
 Literature Review on Pipeline Condition Prediction Method

Stochastic model and (Mahmoodian Metal corrosion is a common threat to To provide a rational and Pipeline data The effect of yield strength of the pipe material on
Monte Carlo Simulation & Li, 2017) the structural integrity of aging oil and consistent approach to make failure probability is not significant, while the ef-
gas pipelines. a quantitative assessment of fect of operating pressure on the service life of the
pipelines failures to be ex- pipe is remarkable. The proposed time-variant reli-
tended to other aging struc- ability method can be used as a rational tool for
tures and infrastructures failure assessment of corrosion affected oil and gas
subjected to localized dete- pipelines with a view to determining the service
rioration. life of the pipeline system.
Artificial Neural (Najafi, Unable to prioritize inspection plan for To develop a pipeline con- Sewer System The neural network is capable of learning the dete-
Network M., and rapidly deteriorating sewer system that dition prediction model to Evaluation Survey rioration trends
Kulandaivel, is 2200 miles long. identify pipelines at risk of
2005) failure so that inspections
can be prioritized.
Artificial Neural (El- Most of the models developed have To design a condition as- Pipeline Inspection Based on the comparison between ANN and the
Network Abbasy et al., used corrosion features alone to assess sessment and prediction data regression model, ANN technique provided better
2014a) the condition of pipelines. models and develop ex- results. The models are expected to help pipeline
pected deterioration curves operators to assess and predict the condition of ex-
for offshore oil and gas isting oil and gas pipelines and hence prioritize
pipelines based on several their inspections and rehabilitation planning.
factors besides corrosion.
Artificial Neural (Liao et The calculated corrosion rate using the To develop an effective Pipeline detail; The The test results show a satisfactory degree of
Network al., 2012) internal corrosion prediction model numerical method to evalu- internal corrosion matching between the prediction corrosion rate and
(ICPM) may deviate from the realistic ate the corrosion rate of wet inspection data the inspection data. This model was selected as the
corrosion rates when the internal envi- gas gathering pipelines. best model for the prediction of corrosion rates.
ronmental parameters of the inspection
segments are not within the scope of
the prediction model.
Artificial Neural (Ren et al., To predict the corrosion rate on long BP neural network is ap- Pipeline data BP neural network model can be used to describe
Network 2012) natural gas pipeline plied to predict the corro- the natural gas pipeline in different corrosive envi-
sion rate of long-distance ronments the maximum corrosion rate.
pipeline based on natural
gas pipeline mileage, eleva-
tion difference, pipe inclina-
tion, pressure, and Reynolds
number.

542
 F. Basheer and, H. Saleem

Artificial Neural (Achim et Regression analysis to predict future To improve failure predic- Historical data The neural network outperforms the statistical
Network al., 2007) failure rates are based on age only and tions by comparing the neu- models. However, some improvement is required
are considered unsatisfactory, due to ral network method against due to databases that are relatively large and noisy.
relatively low correlations between previous modeling STPM
predicted and actual failure rates (Shifted Time Power Mod-
(Righetti 2001). el) and STEM (Shifted
Time Exponential Model).

Regression Analysis (El- Majority of developed condition as- The main objective of the Historical inspec- The model developed is expected to help decision-
Abbasy, sessments or failure prediction only study is to develop expected tion data collected makers assess and predict the condition of existing
Senouci, deal with one failure causes such as condition deterioration from three different oil and gas pipelines and hence prioritize their in-
Zayed, corrosion. No standard condition as- curves for pipelines with re- pipelines in Qatar. spections and rehabilitation planning.
Mirahadi, & sessment scale for oil and gas pipe- spect to age.
Parvizsedghy, lines (condition rating system and its
2014b) associated rehabilitation actions) has
been developed.
Fuzzy Logic (Singh & The traditional modeling approaches – To develop an optimal in- Operating condi- Flexible and easy calibration/modification It is
Markeset, semi-empirical and mechanistic have spection plan considering tion of the pipe, based on a combination of expert knowledge, ob-
2009) been successful in making accurate various sources of infor- wall thickness, and jective measurements and quantifiable objective
predictions only to a limited degree. mation like the recommend- corrosion rate which are more easily available.
ed models, standards and
their own experience of the
plant.
Fuzzy Logic (Senouci et The majority of developed condition The main objectives of the Historical pipeline The results showed that the fuzzy model outper-
al., 2014) assessment and failure prediction present study are: (1) to data formed regression and ANN techniques. However,
models are either subjective (i.e., de- identify and study the criti- as a model limitation, it is concluded that the mod-
pending only on expert opinion con- cal failure causes of oil el is not always capable of accurately predicting
sidering no historical data), or are not pipelines, and (2) to design the natural hazard failure types due to limited de-
comprehensive (i.e., dealing with only a failure type prediction tails in database. As a result, the model was not
one failure cause). Therefore, there is a model for such pipelines. trained enough to predict such type of failure.
need for the development of a more
objective failure prediction model for
oil and gas pipelines to allow pipeline
operators to take the necessary actions
to prevent catastrophic failures

543
 The 1st International Conference on Business, Management and Information Systems 2019

References

Journal articles

Achim, D., Ghotb, F., & McManus, K. J. (2007). Prediction of water pipe asset life using
neural networks. Journal of Infrastructure Systems, 13(1), 26–30.
https://doi.org/10.1061/(ASCE)1076-0342(2007)13:1(26)
Ahammed, M. (1998). Probabilistic estimation of remaining life of a pipeline in the presence
of active corrosion defects. International Journal of Pressure Vessels and Piping, 75(4),
321–329. https://doi.org/10.1016/S0308-0161(98)00006-4
Aulia, R., Tan, H., & Sriramula, S. (2019). Prediction of corroded pipeline performance
based on dynamic reliability models. Procedia CIRP, 80, 518–523.
https://doi.org/10.1016/j.procir.2019.01.093
Breton, T., Sanchez-Gheno, J. C., Alamilla, J. L., & Alvarez-Ramirez, J. (2010).
Identification of failure type in corroded pipelines: A Bayesian probabilistic approach.
Journal of Hazardous Materials, 179(1–3), 628–634.
https://doi.org/10.1016/j.jhazmat.2010.03.049
Caruana, R., & Niculescu-Mizil, A. (2006). An empirical comparison of supervised learning
algorithms. ACM International Conference Proceeding Series, 148, 161–168.
https://doi.org/10.1145/1143844.1143865
Correa, M., Bielza, C., & Pamies-Teixeira, J. (2009). Comparison of Bayesian networks and
artificial neural networks for quality detection in a machining process. Expert Systems
with Applications, 36(3 PART 2), 7270–7279.
https://doi.org/10.1016/j.eswa.2008.09.024
Dhillon, B. S. (2006). Maintainability, Maintenance, and Reliability for Engineers. In
Maintainability, Maintenance, and Reliability for Engineers.
https://doi.org/10.1201/9781420006780
El-Abbasy, M. S., Senouci, A., Zayed, T., Mirahadi, F., & Parvizsedghy, L. (2014a).
Artificial neural network models for predicting condition of offshore oil and gas
pipelines. Automation in Construction, 45, 50–65.
https://doi.org/10.1016/j.autcon.2014.05.003
El-Abbasy, M. S., Senouci, A., Zayed, T., Mirahadi, F., & Parvizsedghy, L. (2014b).
Condition prediction models for oil and gas pipelines using regression analysis. Journal
of Construction Engineering and Management, 140(6).
https://doi.org/10.1061/(ASCE)CO.1943-7862.0000838
El-Abbasy, M. S., Senouci, A., Zayed, T., Parvizsedghy, L., & Mirahadi, F. (2016).
Unpiggable oil and gas pipeline condition forecasting models. Journal of Performance
of Constructed Facilities, 30(1). https://doi.org/10.1061/(ASCE)CF.1943-5509.0000716
Hallen, J., Caleyo, F., González, J., & Fernández-Lagos, F. (2003). Probabilistic Condition
Assessment of Corroding Pipelines in Mexico. Aaende.Org.Ar. Retrieved from
http://www.aaende.org.ar/sitio/biblioteca/material/T-126.pdf
Kalanaki, M., & Soltani, J. (2013). Performance assessment among hybrid algorithms in
tuning SVR parameters to predict pipe failure rates. Advances in Computer Science: An
International Journal, 2(5), 40–46.
Kale, A., Thacker, B. H., Sridhar, N., & Waldhart, C. J. (2004). A probabilistic model for
internal corrosion of gas pipelines. Proceedings of the Biennial International Pipeline
Conference, IPC, 3, 2437–2445.

544
F. Basheer and, H. Saleem

Keshtegar, B., & Miri, M. (2014). Reliability analysis of corroded pipes using conjugate HL-
RF algorithm based on average shear stress yield criterion. Engineering Failure
Analysis, 46, 104–117. https://doi.org/10.1016/j.engfailanal.2014.08.005
Li, S. X., Yu, S. R., Zeng, H. L., Li, J. H., & Liang, R. (2009). Predicting corrosion
remaining life of underground pipelines with a mechanically-based probabilistic model.
Journal of Petroleum Science and Engineering, 65(3–4), 162–166.
https://doi.org/10.1016/j.petrol.2008.12.023
Liao, K., Yao, Q., Wu, X., & Jia, W. (2012). A numerical corrosion rate prediction method
for direct assessment of wet gas gathering pipelines internal corrosion. Energies, 5(10),
3892–3907. https://doi.org/10.3390/en5103892
Mahmoodian, M., & Li, C. Q. (2017). Failure assessment and safe life prediction of corroded
oil and gas pipelines. Journal of Petroleum Science and Engineering, 151(November
2016), 434–438. https://doi.org/10.1016/j.petrol.2016.12.029
Mollazade, K., Omid, M., & Arefi, A. (2012). Comparing data mining classifiers for grading
raisins based on visual features. Computers and Electronics in Agriculture, 84, 124–131.
https://doi.org/10.1016/j.compag.2012.03.004
Najafi, M., and Kulandaivel, G. (2005). Copyright ASCE 2005 767 Pipelines 2005 Pipelines
2005. Pipelines 2005: Optimizing Pipeline Design, Operations, and Maintenance in
Today’s Economy, 767–781.
Noor, N. M., Ozman, N. A. N., & Yahaya, N. (2011). Deterministic prediction of corroding
pipeline remaining strength in marine environment using DNV RP -F101 (Part A).
Journal of Sustainability Science and Management, 6(1), 69–78.
O’Donovan, P., Leahy, K., Bruton, K., & O’Sullivan, D. T. J. (2015). An industrial big data
pipeline for data-driven analytics maintenance applications in large-scale smart
manufacturing facilities. Journal of Big Data, 2(1), 1–26.
https://doi.org/10.1186/s40537-015-0034-z
Ossai, C. I., Boswell, B., & Davies, I. J. (2016). Application of Markov modelling and Monte
Carlo simulation technique in failure probability estimation - A consideration of
corrosion defects of internally corroded pipelines. Engineering Failure Analysis, 68,
159–171. https://doi.org/10.1016/j.engfailanal.2016.06.004
Pandey, M. D. (1998). Probabilistic models for condition assessment of oil and gas pipelines.
NDT and E International, 31(5), 349–358. https://doi.org/10.1016/S0963-
8695(98)00003-6
Ren, C. Y., Qiao, W., & Tian, X. (2012). Natural gas pipeline corrosion rate prediction model
based on BP neural network. Advances in Intelligent and Soft Computing, 147 AISC,
449–455. https://doi.org/10.1007/978-3-642-28592-9_47
Senouci, A., El-Abbasy, M. S., & Zaye, T. (2014). Fuzzy-based model for predicting failure
of oil pipelines. Journal of Infrastructure Systems, 20(4), 1–11.
https://doi.org/10.1061/(ASCE)IS.1943-555X.0000181
Shuai, Y., Shuai, J., & Xu, K. (2017). Probabilistic analysis of corroded pipelines based on a
new failure pressure model. Engineering Failure Analysis, 81(February), 216–233.
https://doi.org/10.1016/j.engfailanal.2017.06.050
Singh, M., & Markeset, T. (2009). A methodology for risk-based inspection planning of oil
and gas pipes based on fuzzy logic framework. Engineering Failure Analysis, 16(7),
2098–2113. https://doi.org/10.1016/j.engfailanal.2009.02.003
St. Clair, A. M., & Sinha, S. (2014). Development of a standard data structure for predicting
the remaining physical life and consequence of failure of water pipes. Journal of
Performance of Constructed Facilities, 28(1), 191–203.
https://doi.org/10.1061/(ASCE)CF.1943-5509.0000384
Sun, Y., Ma, L., & Morris, J. (2009). A practical approach for reliability prediction of
pipeline systems. European Journal of Operational Research, 198(1), 210–214.

545
Literature Review on Pipeline Condition Prediction Method

https://doi.org/10.1016/j.ejor.2008.07.040
Tahyudin, I., Utami, E., & Amborowati, A. (2013). Comparing Clasification Algorithm Of
Data Mining to Predict the Graduation Students on Time. Information Systems
International Conference (ISICO), (December), 2–4.
Zhang, G., Luo, J., Zhao, X., Zhang, H., Zhang, L., & Zhang, Y. (2012). Research on
probabilistic assessment method based on the corroded pipeline assessment criteria.
International Journal of Pressure Vessels and Piping, 95, 1–6.
https://doi.org/10.1016/j.ijpvp.2011.12.001

Books

Dhillon, B. S. (2006). Maintainability, Maintenance, and Reliability for Engineers. In

Maintainability, Maintenance, and Reliability for Engineers.
https://doi.org/10.1201/9781420006780

546