Reliability Engineering and System Safety 203 (2020) 107089

Contents lists available at ScienceDirect

Reliability Engineering and System Safety

journal homepage: www.elsevier.com/locate/ress

Bayesian network model for buried gas pipeline failure analysis caused by T
corrosion and external interference
⁎
Y. Zhanga,b, W.G. Wenga,b,
a
Institute of Public Safety Research, Department of Engineering Physics, Tsinghua University, Beijing 100084, PR China
b
Beijing Key Laboratory of City Integrated Emergency Response Science, Beijing 100084, PR China

A R T I C LE I N FO A B S T R A C T

Keywords: The unintentional release of urban buried gas pipeline may cause crucial consequences to the economy, society
Buried gas pipeline and environment. Corrosion and external interference are primary causes of pipeline failure incidents. Due to the
Corrosion complexity and unpredictability of outside influence on the buried gas pipeline, this paper presents an approach
External interference to analyze pipeline failure frequency and leakage size caused by corrosion and external interference based on
Failure frequency
pipeline characteristics. Bayesian network method is used to construct a knowledge model. Pipeline char-
Leakage size
Bayesian network method
acteristics statistics and failure data are collected to build the relationships among variables in the model and
verify the applicability of the model. Results show that the proposed model can estimate buried gas pipeline
failure frequency and leakage size caused by corrosion and external interference. It is also capable of high-
lighting the critical parameters to pipeline failure. Practical application of the model is demonstrated on the
underground gas pipeline in the City of H, China. Results indicate that proposed model can explicitly quantify
uncertainties and then put forward practical measures for buried gas pipeline parameter design, laying plan and
operating maintenance.

1. Introduction They combined Analytic Hierarchy Process (AHP) method and Entropy
Method (EM) to determine risk factors weights. Dundulis et al. [7] used
The use of buried pipelines for the transport of large quantities of structural integrity analysis and statistical failure data to estimate
natural gas to city is a common mode of transport of energy. failure probability of gas pipeline. Shan et al. [8] proposed an assess-
Unintentional gas release caused by pipeline damage can easily provoke ment model of gas pipelines failure probability based on historical
fire, explosion and the spread of poisonous gas which may cause crucial failure-related data and modification factors which were calculated by
consequences on society, economy and environment. A better knowl- fuzzy method. In view of the specific reasons for reduced buried pipe-
edge of operating status and reliability of buried gas pipeline may urge line reliability, there are some researches focused on pipeline corrosion.
pipeline managers to implement appropriate maintenance and pre- Caleyo et al. [9] investigated the probability distributions of external-
ventive measures [1]. corrosion pit depth and growth rate in underground pipeline to esti-
The reliability of urban buried gas pipelines is a major concern in mate the reliability evolution of corroded pipelines. The means, var-
the field of safety operation and maintenance. Dong et al. [2,3] pro- iances and shape parameters of these distributions differ notably be-
posed a method which combined fuzzy set theories with fault tree tween soil types. Chen et al. [10] proposed an assessment procedure for
analysis to estimate failure probability of gas pipelines. In their re- predicting failure pressure of high strength pipeline with corrosions.
search, probabilities of the basic events were evaluated by expert eli- Miran et al. [11] proposed depth and length growth models of corroded
citation with fuzzy set theory, which could not reflect real situation buried pipelines and analyzed small leak, large leak and rupture failure
because of ambiguity and imprecision. Brito et al. [4] surveyed gas modes of pipelines based on inspection data. They found that the im-
pipeline hazard scenarios and presented the reasons for a risk assess- pact of the statistical uncertainty in the model is significant. Jana et al.
ment model based on a multi-attribute approach. Wang et al. [5] made [12] developed a fuzzy logic approach to estimate corrosion failure
a statistical analysis of likelihood, causes and consequences of leakage likelihood of gas pipelines. The application of this model depended on
and rupture of U.S. natural gas pipelines. Guo et al. [6] used Fuzzy Petri the accuracy of the values of influence factors which were given by
Net to build a risk evaluation method for long-distance gas pipelines. engineers and experts. For the comprehensive reliability analysis of

⁎
Corresponding author at: Institute of Public Safety Research, Department of Engineering Physics, Tsinghua University, Beijing 100084, PR China.
E-mail address: [email protected] (W.G. Weng).

https://doi.org/10.1016/j.ress.2020.107089
Received 16 March 2019; Received in revised form 9 February 2020; Accepted 18 June 2020
Available online 22 June 2020
0951-8320/ © 2020 Elsevier Ltd. All rights reserved.
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

buried pipeline, due to inadequate and unreliable recording of actual frequency per 1000 km per year. The failure frequency is calculated by
failure times and inaccurate measurements of related variables, data dividing the number of incidents by the pipeline exposure. The pipeline
quality and model accuracy become major concerns. In reliability exposure is the length of a pipeline multiplied by its working duration.
analysis of corroded pipeline, the mechanism of pipeline corrosion is Three types of leak size are considered in this paper: pinhole/crack,
not clear and it is difficult to accurately quantify the effect of soil en- hole and rupture. Pinhole/crack represents the effective diameter of the
vironment on pipeline corrosion. Urban buried gas pipelines stretch for leaky hole is smaller than or equal to 2 cm. If the effective diameter of
thousands of kilometers and are usually covered in various soil en- the leak is larger than 2 cm and smaller than or equal to the diameter of
vironments. Most of the studies care less about the leakage caused by the pipe, it is a hole. If the effective diameter of the leak is larger than
pipeline damage and only concentrate on the failure frequency. the pipeline diameter, it belongs to rupture [22]. We constructed two
Bayesian method is widely used by researchers to deal with risk and networks respectively for the occurrence and leakage size estimation of
reliability-related issues on buried infrastructures. Cagno et al. [13] pipeline failure due to corrosion. In this paper, the constructed model is
proposed a robust Bayesian approach to determine the prior distribu- developed using the GeNIe software [25]. It is a tool for artificial in-
tions on gas pipeline failures. The prior probabilities were calculated by telligence modeling and machine learning with Bayesian networks and
AHP using historical data and expert opinion. Francis et al. [14] used a other types of graphical probabilistic models.
knowledge-based BBN to model drinking water distribution system pipe
break prediction. Kabir et al. [15] developed a BBN based data fusion
model for failure rate prediction of water mains. They took soil re- 2.1.1. Pipeline failure estimation model due to corrosion
sistivity and corrosivity into account. Chateauneuf et al. [16] proposed Bayesian network includes a Directed Acyclic Graph (DAG) and
a Bayesian methodology to assess the reliability of systems composed Conditional Probability Tables (CPTs). Directed Acyclic Graph com-
by dependent pipelines subject to different environments conditions. prises of nodes representing variables and directed arcs representing
Shabarchin et al. [17] presented a Bayesian belief network model to probabilistic conditional relationships between nodes. Each variable of
estimate pipeline corrosion defects and probability of failure. Elmasry interests or node is classified in a finite set of state values accompanied
et al. [18] proposed a BBN defect deterioration model to determine the with a probability. Conditional Probability Tables denote those quan-
condition ratings for sewer pipelines. Wang et al. [19-20] used Fault titative probabilistic conditional relationships between child nodes with
Tree method combined with Bayesian Belief Network to analyze failure parent nodes. Based on the EGIG report, 6 variable nodes including 4
probabilities of the urban buried gas pipeline. The estimation data they nodes (X1, X2, X3 and X4) for buried gas pipeline characteristics and 2
used were derived from historical data or expertise. Dahire et al. [21] nodes (X5 and X6) for failure occurrence of pipeline caused by corrosion
proposed an approach for reliability assessment of aging pipeline with a are identified. The state, granularity and scale of these 6 nodes are
focus on the pipeline strength prediction. There are data scarcity and shown in Table 1. The characteristics considered in corrosion failure
relation uncertainty problems in most researches. The calculation of occurrence analysis are year of construction, pipeline material, wall
failure probabilities is difficult and inaccurate when the network is thickness and type of coating. Corrosion is a time dependent phenom-
complex with a number of multi-state variables and the causal re- enon of deterioration of the pipelines [22]. Corrosion condition of aging
lationships between variables are uncertain. pipelines is generally more serious. The characteristics of the pipeline
Based on the survey, corrosion and external interference induced materials also have an effect on the corrosion of the pipeline [24]. Pi-
failures are the primary causes of pipeline damage incidents [17, 19, peline anticorrosion coating is one of the main protective measures
22-24]. Incidents caused by external interference are characterized by which is undertaken to prevent leakage due to corrosion. Meanwhile,
potentially severe consequences. Corrosion has nearly the same fre- the thinner the corroded pipeline wall, the sooner the pipelines fails.
quency rate as external interference, although consequences are much The Directed Acyclic Graph for corrosion failure occurrence Bayesian
less severe [22]. This paper contributes to the failure frequency esti- network is shown in Fig. 1.
mation of buried gas pipeline caused by corrosion and external inter- The conditional probability is defined as the probability of the child
ference based on pipeline property parameters. The proposed method node in the case of its parent node. Take the child node X1a as an ex-
reduces the complexity of failure frequency model and minimize the ample, the ratio of failure frequency to operating pipeline exposure of
impact of subjective factors in analysis. The rest of this paper is orga- the pipelines constructed before 1974 is used as the conditional prob-
nized as follows. A Bayesian network model for estimating buried gas ability of the child node X1a. A set of conditional probabilities for each
pipeline failure frequency caused by corrosion and external interference child node (X1a-X6d) given its parent node (X5a) in the network which is
is presented in Section 2. Section 3 illustrates the validation of the attained based on the statistical data from EGIG [22] is given in Table 2.
model. An application example is given in Section 4 followed by con- These conditional probabilities represents the quantitative relationships
clusions in Section 5. between corrosion failure occurrence and pipeline characteristics. The
probability distributions for each child node (X1a-X4d) given its parent
2. Failure analysis model node (X5b) represent the actual proportion of pipelines with specific
characteristic related to pipeline corrosion accounted for all running
2.1. Corrosion failure analysis of buried gas pipeline pipelines in every year from 1997 to 2016, with details shown in Table
A.1 in the supplementary materials.
Corrosion is one of the primary causes of buried gas pipeline leakage
incidents. Corrosion of buried pipelines is affected by operating con-
ditions, soil environment, and anti-corrosion measures and so on. It is
Table 1
difficult to do quantitative analysis of pipeline corrosion construct Nodes of corrosion failure occurrence Bayesian network.
evolution model. Therefore, the failure analysis of pipeline due to
corrosion in this paper is carried out based on pipeline characteristics Node Variable State (a/b/c/d)

data which are easy to obtain, such as year of construction, pipe ma- X1 Year of construction before 1974; 1974–1994; after 1994
terial, diameter and so on. The variables which describes pipeline X2 Pipeline material Steel; Cast iron; Polyethylene
characteristics are selected from the 10th Report of European Gas X3 Pipeline wall thickness less than 5 mm; 5–10 mm; greater than 10
Pipeline Incident Data Group (EGIG) [22]. The probabilistic conditional mm
X4 Type of coating Coal tar; Bitumen; Polyethylene; Epoxy
relationships are attained from historical data on about 14,000 km of
X5 Corrosion failure incident Occurrence; Non-occurrence
pipelines every year from 1970 to 2016 recorded in the EGIG database. X6 Corrosion appearance Pitting; Cracking; General; Other
The performance of the pipeline is evaluated through assessing failure

2
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Table 3
Nodes of corrosion failure leakage size Bayesian network.
Node Variable State (a/b/c/d)
X1 Year of construction before 1974; 1974–1994; after 1994
X2 Pipeline wall thickness less than 5 mm; 5–10 mm; greater than 10
mm
X3 Pipeline diameter less than 300 mm; 300–600 mm;
600–900 mm;
greater than 900 mm
X4 Type of coating Coal tar; Bitumen; Polyethylene; Epoxy
X5 Corrosion failure incident Occurrence; Non-occurrence
X6 Corrosion leakage size Pinhole/Crack; Hole; Rupture; Non

Fig. 1. The DAG for corrosion failure occurrence Bayesian network.

2.1.2. Pipeline leakage size estimation model due to corrosion

There are 6 variable nodes including 4 nodes (X1, X2, X3, and X4) for
buried gas pipeline characteristics and 2 nodes (X5 and X6) for failure
leakage size analysis of pipeline caused by corrosion, with details
shown in Table 3. The correlation between pipeline material and the
type of leakage size is not considered in this paper. Thus, the variable of
pipeline material is not included. Smaller leakage size (pinhole/crack,
hole) mainly occurred at pipelines with bigger diameters and bigger
size (rupture) mainly occurred at pipelines with smaller diameters [22].
It is necessary to consider the influence of pipeline diameter on leakage
size. The Directed Acyclic Graph for corrosion failure leakage size Fig. 2. The DAG for corrosion failure leakage size Bayesian network.
Bayesian network is shown in Fig. 2.
Two sets of conditional probabilities for each child node (X1a-X4d; Table 4
X6a-X6c) given its parent node (X6a- X6c; X5a) in the network which are The CPT for corrosion failure leakage size Bayesian network.
attained based on the statistical data from EGIG [22] are given in
a)
Table 4. These conditional probabilities represents the quantitative X1a X1b X1c X2a X2b X2c X3a
relationships among corrosion failure occurrence, failure leakage size X6a 0.0137 0.1931 0.7932 0.7388 0.2612 0 0.6244
and pipeline characteristics. The probability distributions for each child X6b 0 0 1 1 0 0 0.5848
node (X1a-X4d) given its parent node (X6d) represent the actual pro- X6c 0 0 1 1 0 0 1
X3b X3c X3d X4a X4b X4c X4d
portion of pipelines with specific characteristic related to corrosion
X6a 0.1628 0.1828 0.0300 0.2810 0.3326 0.0405 0.3459
failure leakage size accounted for all running pipelines in every year X6b 0.2975 0.1195 0 1 0 0 0
from 1997 to 2016, with details shown in Table A.2 in the supple- X6c 0 0 0 1 0 0 0
mentary materials. b)
X6a X6b X6c
X5a 0.9806 0.0194 0

2.2. External interference failure analysis of buried gas pipeline

2.2.1. Pipeline failure estimation model due to external interference
External interference which means the external activities having Based on the EGIG report, 4 variable nodes including 3 nodes (X1,
caused the incident, such as digging, piling, ground works and so on, is X2, and X3) for buried gas pipeline characteristics and 1 node (X4) for
also one of the main causes of the buried gas pipeline failure. The un- failure occurrence of pipeline caused by external interference are
predictability and uncertainty of pipeline failure caused by external identified. The state, granularity and scale of the nodes are shown in
interference makes it difficult to conduct quantitative analysis from the Table 5. Pipeline diameter, wall thickness and depth of cover are con-
perspective of external causes. In this paper, the ability of pipeline to sidered in this process. Smaller diameter pipelines may be more easily
resist external interference is analyzed based on pipeline character- damaged under external interference. Smaller pipelines tend to have
istics. We construct the pipeline external interference failure analysis thinner wall thickness. The shallower the pipe is laid, the more likely it
model and calculated the probabilistic conditional relationships in the is to be disturbed by external activities. The Directed Acyclic Graph for
same way as mentioned in corrosion analysis process. Two networks external interference failure occurrence Bayesian network is shown in
respectively for the occurrence and leakage size of pipeline external Fig. 3.
interference failure are constructed. A set of conditional probabilities for each child node (X1a-X3c) given

Table 2
The CPT for corrosion failure occurrence Bayesian network.
node X1a X1b X1c X2a X2b X2c X3a X3b X3c
X5a 0.0143 0.1864 0.7993 0.6666 0.3333 0.0001 0.7042 0.2899 0.0059
X4a X4b X4c X4d X6a X6b X6c X6d
X5a 0.3188 0.4348 0.0435 0.2029 0.3834 0.2866 0.1338 0.1962

3
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Table 5
Nodes of external interference failure occurrence Bayesian network.
Node Variable State (a/b/c/d)
X1 Pipeline diameter less than 300 mm; 300–600 mm; 600–900 mm;
greater than 900 mm
X2 Pipeline wall thickness less than 5 mm; 5–10 mm; greater than 10 mm
X3 Depth of cover less than 80 cm; 80–100 cm; greater than 100 cm
X4 External interference failure incident Occurrence; Non-occurrence

Table 7
Nodes of external interference failure leakage size Bayesian network.
Node Variable State (a/b/c/d)
X1 Pipeline diameter less than 300 mm; 300–600 mm;
600–900 mm;
greater than 900 mm
X2 Pipeline wall thickness less than 5 mm; 5–10 mm; greater than
10 mm
X3 Depth of cover less than 80 cm; 80–100 cm;
greater than 100 cm
X4 External interference failure Occurrence; Non-occurrence
incident
X5 Corrosion leakage size Pinhole/Crack; Hole; Rupture; Non

Fig. 3. The DAG for external interference failure occurrence Bayesian network.

its parent node (X4a) in the network which is attained based on the
statistical data from EGIG [22] is given in Table 6. These conditional
probabilities represents the quantitative relationships between external
interference failure occurrence and pipeline characteristics. The prob-
ability distributions for each child node (X1a-X3c) given its parent node
(X4b) represent the actual proportion of pipelines with specific char-
acteristic related to external interference accounted for all running pi-
pelines in each year from 1997 to 2016, with details shown in Table A.3
in the supplementary materials.

2.2.2. Pipeline leakage size estimation model due to external interference

There are 5 variable nodes including 3 nodes (X1, X2 and X3) for
buried gas pipeline characteristics and 2 nodes (X4 and X5) for failure Fig. 4. The DAG for external interference failure leakage size Bayesian network.
leakage analysis of pipeline caused by external interference, with de-
tails shown in Table 7. The pipe characteristics considered are same as
corrosion and external interference from 1997 to 2016 of EGIG are
those in the external interference failure occurrence model. The Di-
performed. The validation flowchart of the Bayesian networks proposed
rected Acyclic Graph for external interference failure leakage size
above is shown in Fig. 5. There are two parts of validation in the
Bayesian network is shown in Fig. 4.
flowchart in Fig. 5, one part is to verify the failure frequency estimation
Two sets of conditional probabilities for each child node (X1a-X3c;
model and the other is to verify the leak size estimation model. Pipeline
X5a- X5c) given its parent node (X5a- X5c; X4a) in the network which is
failure frequency and leak size from 1997 to 2016 are estimated
attained based on the statistical data from EGIG [22] are given in
through the models and compared with the actual values collected from
Table 8. These conditional probabilities represents the quantitative
the EGIG database to validate the model. Evidence data (pipeline
relationships among external interference failure occurrence, failure
parameter statistics) and prior probability (historical failure frequency)
leakage size and pipeline characteristics. The probability distributions
collected from the EGIG database are the input data of the proposed
for each child node (X1a-X3d) given its parent node (X5d) represent the
Bayesian network model. The posterior probability obtained by Baye-
actual proportion of pipelines with specific characteristic related to
sian inference is the predicted value and compared with the actual
external interference failure leakage size accounted for all running pi-
value. Meanwhile, the predicted pipeline failure frequency is the input
pelines in every year from 1997 to 2016, with details shown in Table
data of the leakage size Bayesian network model as evidence informa-
A.3 in the supplementary materials.
tion to obtain the predicted value of the leak size. Bayesian network
inference plays a critical role in validating the proposed model. Three
3. Model validation
validation objects are considered: pipeline corrosion failure frequency,
pipeline external interference failure frequency and leakage size pre-
To validate the proposed model, a set of quantitative validation
dictions for every year from 1997 to 2016. Pipeline characteristics of
approaches for pipeline failure frequency and leakage size caused by

Table 6
The CPT for external interference failure occurrence Bayesian network.
X1a X1b X1c X1d X2a X2b X2c X3a X3b X3c
X4a 0.7888 0.1683 0.0389 0.0040 0.7730 0.2014 0.0256 0.5637 0.2615 0.1748

4
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Table 8
The CPT for external interference failure leakage size Bayesian network.
a)
X1a X1b X1c X1d X2a X2b X2c X3a X3b X3c
X5a 0.7596 0.1646 0.0632 0.0126 0.7762 0.1997 0.0241 0.549 0.2636 0.1874
X5b 0.8001 0.1796 0.0203 0 0.7491 0.2150 0.0359 0.5738 0.2654 0.1608
X5c 0.8072 0.1465 0.0463 0 0.8328 0.1672 0 0.5564 0.2483 0.1953
b)
X5a X5b X5c
X4a 0.3962 0.4654 0.1384

Fig. 5. Flowchart of the proposed Bayesian network model validation.

every year are used as evidence, with details shown in Table A.1–3 in question is used as evidence. The posterior probabilities of corrosion
the supplementary materials. The posterior probabilities for failure failure frequency for every year from 1997 to 2016 are attained through
frequency, with details shown in Table A.4 in the supplementary ma- Bayesian network updating. Fig. 6 shows the predicted curve (red one)
terials, and characteristics of damaged pipeline obtained through the is in good agreement with the actual statistical curve (blue one) at-
inference of the failure occurrence Bayesian network, with details tained from the EGIG database. The same process is used to verify the
shown in Table A.5–8 in the supplementary materials, are used as external interference occurrence Bayesian network. Validation result is
evidence for leakage size Bayesian network. The validation data is used shown in Fig. 7. In Fig. 7, the predicted curve (red one) is also in good
to compare with the actual statistical data attained from the report of agreement with actual statistical curve (blue one) attained from the
EGIG [22]. EGIG database. Thus, the proposed model can be used for any urban
Validation result of the corrosion failure occurrence Bayesian net- buried gas pipeline system to predict the failure frequency caused by
work is shown in Fig. 6. Corrosion failure frequency data over a time corrosion and external interference. The required data are: pipeline
span of 1970 to the year in question from the EGIG database is used as material, year of construction, wall thickness, diameter, type of coating,
prior probability. Pipeline characteristics statistics of the year in depth of cover and historical failure frequency.

5
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Fig. 6. Corrosion failure frequency validation result.

Validation results of the leakage size Bayesian network are shown in caused by corrosion and external interference. In Fig. 8, the predicted
Fig. 8. Three types of leakage size are considered: pinhole/crack, hole value only represents failure frequency caused by corrosion and ex-
and rupture. Corrosion always lead to pinhole/crack leakage. Holes and ternal interference and the actual value represents pipeline failure
ruptures are mainly caused by external interference. The red curves in frequency for all causes. The evidence used is the total predicted pi-
Fig. 8 are the posterior probabilities of three types of leakage size peline failure frequency. The accuracy of the results can be improved by
caused by corrosion and external interference failure for every year using the predicted failure frequency for each leakage size. Moreover,
from 1997 to 2016. The blue curves are the actual statistical failure the EGIG does not give the statistical information on the corrosion
frequency per leak size, attained from the EGIG database. In Fig. 8, the failure frequency in relation to pipeline diameter which has influence
predicted value curve (red one) and the actual value curve (blue one) on leakage size. Thus, the posterior probability for diameter of damaged
are not completely consistent due to following factors. According to pipeline cannot be obtained and included in evidence information for
EGIG statistics, there are five initial causes of the pipeline incident: leakage size prediction.
external interference, corrosion, construction defect(material failure),
hot tap made by error and ground movement. Corrosion incidents tend 4. Model application
to have smaller leak sizes(pinhole/crack and hole). Holes are mainly
caused by external interference. For ruptures, external interference and 4.1. Data collection
ground movement both remains the mian causes [22]. This paper only
estimate the failure frequency and leak size of buried gas pipeline The City of H is located in eastern China. City H is one of the sub-

Fig. 7. External interference failure frequency validation result.

6
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Fig. 8. Leakage size of pipeline failure validation results: a. pinhole/crack; b. hole; c. rupture.

7
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Fig. 9. Summary statistics of the buried gas pipeline.

center of the Yangtze River Delta Urban Agglomerations. The 2018 4.2.2. Failure frequency and leakage size prediction
census reported 8.08 million people living in the City of H. The EGIG report [22] illustrates the decreasing pipeline failure
The low-pressure pipeline (maximum operating pressure lower than frequencies over the years and the decrease may be explained by
0.01 MPa) is not included in this study. The buried gas pipeline system technological developments in inspection and maintenance which are
composed of 82,163 mains with a total length of 2122.168 km is ana- not considered in this paper. However, the proposed Bayesian network
lyzed in this paper. Most of the pipeline is laid after 1994. Steel, cast model can perform probabilistic inferences or belief updating based on
iron and polyethylene are common pipe materials. The pipeline coating new data or observations. Thus, to take into account the impact of
is mostly made of polyethylene. The diameter of the thickest pipe is less technological progress, the five year moving average failure frequency
than 900 mm. The summary of the used pipeline characteristics data is of 2016 from the EGIG database is used as the prior probability. The
shown in Fig. 9. five year moving average for the year in question means that the cal-
culations have been performed over the five previous year. The pos-
terior probability of corrosion failure frequency for the City of H un-
4.2. Results and discussions derground gas pipeline system (minimum operating pressure higher
than 0.01 MPa) is 0.0009 per 1000 km•yr. The posterior probability of
4.2.1. Sensitivity analysis external interference failure frequency is 0.0201 per 1000 km•yr. The
To identify critical buried gas pipeline characteristics parameters for posterior probabilities of corrosion failure frequency per leak size
the buried gas pipeline protection against corrosion and external in- (pinhole/crack, hole and rupture) are 0.0004, 0.001 and 0 per 1000
terference, the sensitivity analysis is performed using the GeNIe soft- km•yr. The posterior probabilities of external interference failure fre-
ware in this paper. The results are shown in Fig. 10, and the darker the quency per leak size (pinhole/crack, hole and rupture) are 0.0227,
color is, the more critical the variable is. 0.0278 and 0.009 per 1000 km•yr. There are uncertainties in the pre-
As can been seen in Fig. 10, the critical pipeline characteristics diction of leakage size due to the following factors. The EGIG does not
parameters in corrosion failure incident for the city of H underground give the statistical information on the corrosion failure frequency in
gas pipelines are year of construction, pipeline material, type of coating relation to the diameter of pipeline which has influence on leakage size.
and wall thickness. The critical pipeline characteristics parameters in Failure frequency and leakage size statistics of H buried gas pipeline are
corrosion failure leakage size are year of construction and pipeline not available, and are replaced by the statistical data of 2016 from the
diameter. The critical pipeline characteristics parameters in external EGIG report [22] as the prior probabilities. Data on the depth of cover
interference failure incident are pipeline diameter, depth of cover and for H underground gas pipeline is not available, and is replaced by the
wall thickness. These three parameters all have obvious influence on data from the EGIG report [22]. The lack and inaccuracy of data results
the external interference failure leakage size. These critical factors in uncertainty problems in leakage size estimation.
should be given priority in buried gas pipeline parameter design, laying The pipeline corrosion failure frequency is extremely low, and the
plan and operating maintenance. Pipeline materials can significantly overall corrosion failure estimation has little guiding significance for
affect the corrosion resistance. Priority should be given to polyethylene pipeline operating maintenance. Thus, the corrosion resistance of pi-
pipelines with thick wall in severe corrosion environment. In daily pelines with different specific characteristics is analyzed. There are 26
maintenance, more attention should be paid to the aging pipeline as the types of underground pipelines (illustration shown in Table A.5 in the
year of construction is one of the leading indicators for the failure in- supplementary materials) with different characteristics parameters
cidents. The larger the diameter, the thicker the wall thickness should (year of construction, material, wall thickness and type of coating),
be as such a combination can effectively reduce the failure frequency.

8
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Fig. 10. Sensitivity analysis results.

with details shown in Table A.6 in the supplementary materials, in the thickness), with details shown in Table A.8 in the supplementary ma-
City of H. The corrosion resistance of these 26 types of pipelines is terials, in the City of H. Thus, the external interference resistance of
analyzed through the Bayesian network approach, shown in Fig. 11. each type pipeline with different depth interval of cover is analyzed,
Most of the pipelines coated with a polyethylene coating have a sig- shown in Fig. 12. Increasing the buried depth of pipelines in state 1, 4
nificant lower failure frequency. Older pipelines, with tar or bitumen, and 7 with lower resistance to external interference will greatly im-
have higher failure frequencies which should be paid more attention. prove its resistance which should be given priority, especially the pi-
The results are in accordance with EGIG report. The pipelines in state pelines in state 1. These types of pipelines should be avoided to lay in
19–26 should be avoided to lay in strong corrosive areas and should be urban areas where external activity is generally higher. In consideration
paid more attention in daily maintenance subject to corrosion. of economic factor, there is no need to significantly increase the buried
As mentioned above, data on the depth of cover for City H under- depth of pipelines in state 3, 8 and 9.
ground gas pipeline is not available. There are 9 kinds of underground
pipelines (illustration shown in Table A.7 in the supplementary mate-
rials) with different characteristic parameters (diameter and wall

Fig. 11. Corrosion resistance of 26 types pipelines.

9
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

Fig. 12. External interference resistance of 9 types pipelines with different depth of cover.

5. Conclusions resistance of pipes with different characteristics (diameter and wall

thickness) to external interference in different depth range of cover is
In this paper, a Bayesian network model for buried gas pipeline analyzed. Meanwhile, the critical pipeline characteristics parameters
failure analysis caused by corrosion and external interference has been for the buried gas pipeline protection against corrosion and external
proposed to connect the relationship from pipeline characteristics to interference are identified. In the pipeline corrosion failure prevention,
failure incident and potential leakage size. The advantage of the pro- year of construction and material should be paid more attention. Year
posed model is that it can perform probabilistic inferences or belief of construction and pipeline diameter are the most important factors
updating to predict pipeline failure frequency and potential leakage size influencing the leakage size of corrosion failure incident. In the pipeline
based on the latest data or observations. Based on the results, useful external interference failure prevention, pipeline diameter and depth of
suggestions can be provided for buried gas pipeline parameter design cover should be paid more attention. Pipeline diameter, wall thickness
specification and daily maintenance. and depth of cover all have a significant effect on the leakage size of
The statistical data of pipeline characteristics parameters, failure external interference failure incident. According to conclusion, specific
frequency and leakage size caused by corrosion or external interference recommendations to daily maintenance and overhaul of pipeline can be
from 1970 to 2016 were collected from the EGIG database to construct provided to improve pipeline safety management.
the Bayesian network model. The data from 1997 to 2016 were used to
validate the proposed model. Applying Bayesian inference method to Declaration of Competing Interest
model validation is the innovation of this paper. In order to verify the
failure frequency prediction model, the pipeline characteristics statis- None.
tics are used as evidence. The failure frequency over the entire period
before is used as prior probability. The failure frequency predictions are Acknowledgement
in good agreement with the actual statistical data from the EGIG da-
tabase. Thus, based on readily available statistics on pipeline char- This work was supported by National Science Fund for
acteristics data, the failure frequency and potential leak size of the Distinguished Young Scholars of China (Grant No 71725006). The
buried gas pipelines caused by corrosion and external interference can support is gratefully acknowledged.
be estimated. A quantitative evaluation method is provided to evaluate
the unintentional leakage of pipelines under the influence of external Supplementary materials
interferences. In order to verify the leakage size prediction model, the
pipeline characteristics predictions of damaged pipeline, obtained Supplementary material associated with this article can be found, in
through the failure frequency prediction model, are used as evidence. the online version, at doi:10.1016/j.ress.2020.107089.
Posterior probability for failure frequency which is also obtained
through the failure frequency prediction model is used as prior prob- References
ability. But there are uncertainties in the prediction of leakage size due
to lack of failure frequency statistics in related to pipeline diameter [1] Kabir G, Balek NBC, Tesfamariam S. Consequence-based framework for buried in-
data. In conclusion, this model can be used to predict the failure fre- frastructure systems: a Bayesian belief network model. Reliab Eng Syst Saf
quency and potential leakage size based on the statistical data on pi- 2018;180:290–301.
[2] Yuhua D, Datao Y. Estimation of failure probability of oil and gas transmission pi-
peline characteristics parameters for two main causes of buried gas pelines by fuzzy fault tree analysis. J Loss Prev Process Ind 2005;18(2):83–8.
pipeline failure: corrosion and external interference. [3] Yuhua D, Datao Y. Estimation of failure probability of oil and gas transmission pi-
To demonstrate the efficacy and applicability of the proposed pelines by fuzzy fault tree analysis. J Loss Prev Process Ind 2005;18(2):83–8.
[4] Brito AJ, de Almeida AT. Multi-attribute risk assessment for risk ranking of natural
model, the underground gas pipelines of the City of H, China is in- gas pipelines. Reliab Eng Syst Saf 2009;94(2):187–98.
vestigated. Through this model, critical pipelines vulnerable to unin- [5] Wang H, Duncan IJ. Likelihood, causes, and consequences of focused leakage and
tentional gas release due to corrosion are identified, in related to year of rupture of US natural gas transmission pipelines. J Loss Prev Process Ind
2014;30:177–87.
construction, material, wall thickness and type of coating. The

10
Y. Zhang and W.G. Weng Reliability Engineering and System Safety 203 (2020) 107089

[6] Guo Y, Meng X, Wang D, Meng T, Liu S, He R. Comprehensive risk evaluation of for water mains: bayesian belief network based data fusion. Knowl Based Syst
long-distance oil and gas transportation pipelines using a fuzzy Petri net model. J 2015;85:159–69.
Nat Gas Sci Eng 2016;33:18–29. [16] Chateauneuf A, Laggoune R. Bayesian approach for the reliability assessment of
[7] Dundulis G, Žutautaitė I, Janulionis R, Ušpuras E, Rimkevičius S, Eid M. Integrated corroded interdependent pipe networks. Int J Pressure Vess Pip 2016;148:46–58.
failure probability estimation based on structural integrity analysis and failure data: [17] Shabarchin O, Tesfamariam S. Internal corrosion hazard assessment of oil & gas
natural gas pipeline case. Reliab Eng Syst Saf 2016;156:195–202. pipelines using Bayesian belief network model. J Loss Prev Process Ind
[8] Shan K, Shuai J, Xu K, Zheng W. Failure probability assessment of gas transmission 2016;40:479–95.
pipelines based on historical failure-related data and modification factors. J Nat Gas [18] Elmasry M, Hawari A, Zayed T. Defect based deterioration model for sewer pipe-
Sci Eng 2018;52:356–66. lines using Bayesian belief networks. Can J Civ Eng 2017;44(9):675–90.
[9] Caleyo F, Velázquez JC, Valor A, Hallen JM. Probability distribution of pitting [19] Wang W, Shen K, Wang B, Dong C, Khan F, Wang Q. Failure probability analysis of
corrosion depth and rate in underground pipelines: a Monte Carlo study. Corros Sci the urban buried gas pipelines using Bayesian networks. Process Saf Environ Protect
2009;51(9):1925–34. 2017;111:678–86.
[10] Chen Y, Zhang H, Zhang J, Li X, Zhou J. Failure analysis of high strength pipeline [20] Wang W, Shen K, Wang B, Dong C, Khan F, Wang Q. Failure probability analysis of
with single and multiple corrosions. Mater Des 2015;67:552–7. the urban buried gas pipelines using Bayesian networks. Process Saf Environ Protect
[11] Miran SA, Huang Q, Castaneda H. Time-dependent reliability analysis of corroded 2017;111:678–86.
buried pipelines considering external defects. J Infrastruct Syst [21] Dahire S, Tahir F, Jiao Y, Liu Y. Bayesian Network inference for probabilistic
2016;22(3):04016019. strength estimation of aging pipeline systems. Int J Pressure Vess Pip
[12] Jana DK, Bej B, Wahab MHA, Mukherjee A. Novel type-2 fuzzy logic approach for 2018;162:30–9.
inference of corrosion failure likelihood of oil and gas pipeline industry. Eng Fail [22] European Gas Pipeline Incident Data Group. (2018). 10th Report of the European
Anal 2017;80:299–311. Gas Pipeline Incident Data Group.
[13] Cagno E, Caron F, Mancini M, Ruggeri F. Using AHP in determining the prior dis- [23] Lam C, Zhou W. Statistical analyses of incidents on onshore gas transmission pi-
tributions on gas pipeline failures in a robust Bayesian approach. Reliab Eng Syst pelines based on PHMSA database. Int J Press Vess Pip 2016;145:29–40.
Saf 2000;67(3):275–84. [24] Ossai CI, Boswell B, Davies IJ. Pipeline failures in corrosive environments–A con-
[14] Francis RA, Guikema SD, Henneman L. Bayesian belief networks for predicting ceptual analysis of trends and effects. Eng Fail Anal 2015;53:36–58.
drinking water distribution system pipe breaks. Reliab Eng Syst Saf 2014;130:1–11. [25] GeNIe. Decision systems laboratory, 1998–2017. Available at: 〈https://dslpitt.org/
[15] Kabir G, Demissie G, Sadiq R, Tesfamariam S. Integrating failure prediction models genie/〉.

11