International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
Fault Detection of Gas Pipelines Using
Mechanical Waves and Intelligent Techniques
Ahmad Akbarifar1, Morteza Mohammadzaheri2, Ali Joudaki1, Ehsan Jamshidi1,
Saeid Khabbaz3, Mohammareza Parmoon4, and Mohammreza Emadi5

equations. Therefore, despite several years of research, no
practical use of this method has been reported. In opposite,
ultrasound waves have been successfully used to detect leaks
of gas pipelines. The main shortcomings of this method is its
limited range of operations (tens of meters) and high expense
of generators and detectors of ultrasound waves.
This paper proposes a new approach: the use of mechanical
waves, which are much less expensive to generate/detect than
ultrasound waves and can progress much further along the
pipeline due to their lower frequency. However, extraction of
useful data for health monitoring of pipelines from mechanical
waves is not as straight forward as from ultrasound waves.
Finite element modelling (FEM) , statistical and intelligent
techniques were employed in this research to tackle this
complexity. All the employed techniques are detailed in
quadruple subsection of Section II, followed by results,
conclusion and references.
Abstract—This paper proposes a new approach to fault detection
of gas pipes using mechanical waves which are much cheaper and
easier to generate and measure compared to currently employed
ultrasound waves. The proposed method includes finite element
modelling (FEM), transferring the acceleration data to frequency
domain and use of statistical and artificial intelligence techniques.
Besides a fault detection algorithm, the applicable length and
bandwidth of the method is also investigated and found.
Keywords—Gas pipe, radial basis function networks, fault
detection, variance.
I. INTRODUCTION
G
AS pipelines‟ fault diagnosis plays a critical role in gas
transportation both in terms of safety and economy. Two
main categories of approaches are employed to tackle this
class of fault diagnosis problems.
In the first category, the whole pipe should be examined,
that is, the fault detection device needs to be moved or
installed along the entire pipe. Instances are the use of optical
or acoustic sensors to find leaks [1]. Other examples are the
injection of flammable chemicals and use of a flame detector
along the pipeline [2] and simultaneous use of electromagnetic
sources and sensors [3].This latter method is performed
manually or by special robots namely “pigs” [4] in which its
use requires the pipeline to be out of service. Another example
is the installation of optical fibres along the pipe [5].All these
methods are time-consuming and/or expensive.
The second category includes the methods that need
measurement at a limited number of points along the pipeline.
There are two methods in this category, fault diagnosis based
on monitoring the change of fluid characteristics (i.e. flow rate
and pressure) [6, 7] and the use of ultrasound waves [8]. In the
first method, a set of nonlinear equations which describe the
flow dynamics are solved (e.g. through linearization [9] or
discretization [10]) and used to predict the flow rate or
pressure in the presence/absence of faults. This method still
suffers from inaccuracies inherited by complex dynamics of
natural gas and uncertainties in parameters of governing
II. METHODOLOGY
After completion, the proposed fault diagnosis method
suggests to install mechanical force actuators (e.g.
piezoelectric patches) along the pipelines with a defined
distance from each other (e.g. 2000 metres). In addition,
accelerometers (or equivalent devices) are installed in between
force actuators. Each force actuator generates a force impulse
and neighbouring sensors measure the corresponding
accelerations. A time series of accelerations will be the input
to an algorithm which diagnose probable faults. Development
of this algorithm is the focus of this paper. This research can
be summarised in the following stages:
1- Making a FEM model of the gas pipeline, supported by
experimental results.
2- Initial Analysis: Finding the maximum applicable
excitation force, and the effective length and
bandwidth that the acceleration can be measured in.
3- Collection of acceleration data for different faults.
4- Secondary Analysis: Transferring the recorded
acceleration to frequency domain, and their
partitioning and intelligent analysis.
A. Modal Analysis and FEM
Initially, modal analysis was performed on a sample pipe
(depicted in Fig.1), then an FEM of the same pipe was
constructed in ABAQUS software package. As a result, a
suitable element type and mesh size were chosen so as the
2
Texas
A&M
University
at
Qatar,
email:
[email protected].
Islamic Azad University, 1Semnan and 5Mahdishahr branches, Iran.
3
Research Council of Semnan Provincial Gas Company, Iran
4
Sharif University of Technology, Iran
http://dx.doi.org/10.15242/IIE.E0115012
63
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
TABLE I
COMPARISON OF NATURAL FREQUENCIES OBTAINED FROM MODAL
ANALYSIS AND FEA RESULTS
natural frequencies of the model are close enough (with an
error of less than 2%) to modal analysis results. Based on this
stage of research, mesh size of 1 cm and element type of
explicit shell and meshing technique of structured quad were
used in the rest of FEMs of the paper. The density, elasticity
modulus and poison ration of the pipe are 7861 kg/m3, 207
GPa and 0.3 respectively.
FEA Results[Hz]
Modal Analysis [Hz]
40.101
110.03
214.26
351.05
40.020
110.95
216.55
356.018
After selection of element type and mesh size, the pipe is
extended by 1 km with bearings constraining axial motions of
the pipe and installed every five meters.
B. Initial Analysis
The maximum magnitude of excitation force was determined
300 N. Then an impulse with this magnitude was applied
between the first and the second bearing of the pipeline.
Afterwards, the time series of accelerations recorded in the
distance of 200,400,600,800 and 1000 meters from the
excitation point with a sampling rate of 8 kHz were
transformed to the frequency domain as depicted in Fig.3.
In order to determine the maximum length/bandwidth that
the acceleration signal is measurable, a measurability criterion
is required. Having an acceleration signal with a magnitude of
several times larger than the resolution of a typical
accelerometer (0.001 m/s2) is defined as the measurability
criterion. With this criterion, maximum length and frequency
of 1000 m and 3 kHz were chosen for this analysis.
Fig. 1 Sample pipe
C. Faults
For the purpose of fault diagnosis, a 50 meters long pipe was
investigated to decrease computation expenses. 15 holes with
the consistent diameter of 2mm is placed within this pipe. All
located on 90º position (top) of the cross section. The distance
between two consequent faults is 3.25 m, unless those located
on bearings which have 3.5 meters distant with their previous
faults. A schematic of meshed faults is shown in Fig. 4.
Fig. 2 Modal analysis of the sample pipe
Table 1 lists some of the natural frequencies obtained from
modal analysis and finite element analysis (FEA).
Fig. 3 Acceleration 1km far from the force source
http://dx.doi.org/10.15242/IIE.E0115012
64
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
For faulty pipes, all these numbers are subtracted from the
ones of the faultless pipe. As a result, 100 numbers associated
with 100 frequency ranges are obtained for each fault location.
These numbers are called „signature numbers‟ of each fault.
Then variance of each signature number in all fault locations
is calculated. This shows, how sensitive the signature number
is respect to change of fault location. Then the signature
numbers with highest variance (e.g. >0.05) are selected as
inputs to a radial basis function network (RBFN) [11] with the
output of fault location.
Fig. 4 Fault Mesh
III. RESULTS
The faults are added to the pipe one at a time, and the
resultant pipe was excited and the corresponding acceleration
at the distance of 50m was collected; an example is depicted in
Fig.5. The faultless pipe underwent the same process.
Fig. 6 shows the variance of signature numbers and their
associated frequency range. The trained radial basis function
network was tested with the data of faults locations of 18 and
33 meters from the force source. The data of these two faults
were not used in training. The estimated locations are 23.0501
and 33.0000 metres respectively. Considering the number of
data sets used in training (13 sets), the achieved accuracy is
promising.
IV. CONCLUSION
This paper introduced and investigated a new technology for
fault diagnosis of gas pipes. The stages of the proposed
method can be summarised as (i) developing a FEM of the
pipe, (ii) creating simulated faults on different locations of the
pipe FEM, (iii) exciting the pipe with and without faults and
calculating signature numbers, (iv) finding the variance of
signature numbers and selection of signature numbers with
high variance as the inputs to an RBFN to estimate fault
location and (v) designing and training the RBFN.
Fig. 5 Acceleration at the end of a 50m pipe with a fault 12m from
the force source
D. Secondary Analysis
For any of 16 acceleration versus frequency series of data.
The selected bandwidth of 3 kHz in divided into 100 areas of
30 Hz. The average magnitude of acceleration is derived for
any of them. That is, each pipe (with or without faults) is
represented by 100 values of average acceleration.
Fig. 6 Variance of signature numbers in m2/s4
http://dx.doi.org/10.15242/IIE.E0115012
65
International Conference on Artificial Intelligence, Energy and Manufacturing Engineering (ICAEME'2015) Jan. 7-8, 2015 Dubai (UAE)
REFERENCES
[1]
J. Zhang, "Designing a cost-effective and reliable pipeline leakdetection system," Pipes and Pipelines International, vol. 42, pp. 20-26,
1997.
[2] M. Liu, S. Zang, and D. Zhou, "Fast leak detection and location of gas
pipelines based on an adaptive particle filter," International Journal of
Applied Mathematics and Computer Science, vol. 15, p. 541, 2005.
[3] C. He, L. Hang, and B. Wu, "Application of homodyne demodulation
system in fiber optic sensors using phase generated carrier based on
LabVIEW in pipeline leakage detection," in 2nd International
Symposium on Advanced Optical Manufacturing and Testing
Technologies, 2006. http://dx.doi.org/10.1117/12.676892
[4] A. G. Di Lullo, G. Pinarello, and A. Canova, "Apparatus and method for
monitoring the structural integrity of a pipeline ", ed: US Patent
20,140,312,887, 2014.
[5] Y. Zhou, S.-j. Jin, and Z.-g. Qu, "Study on the distributed optical fiber
sensing technology for pipeline leakage protection," in Advanced Laser
Technologies, 2006.
[6] J. Gong, A. C. Zecchin, A. R. Simpson, and M. F. Lambert, "Frequency
Response Diagram for Pipeline Leak Detection: Comparing the Odd and
the Even Harmonics," Journal of Water Resources Planning and
Management, 2013.
[7] P. J. Lee, M. F. Lambert, A. R. Simpson, J. P. Vítkovský, and J. Liggett,
"Experimental verification of the frequency response method for
pipeline leak detection," Journal of Hydraulic research, vol. 44, pp. 693707, 2006. http://dx.doi.org/10.1080/00221686.2006.9521718
[8] S. Wang, M. Lian, S. Zhou, J. Feng, and Y. Rao, "Research and Outlook
on Leak Detection Technology of Natural Gas Pipeline," Bridges, vol.
10, 2014.
[9] X.-J. Wang, M. F. Lambert, A. R. Simpson, J. A. Liggett, and J. P. V ı´
tkovský, "Leak detection in pipelines using the damping of fluid
transients," Journal of Hydraulic Engineering, vol. 128, pp. 697-711,
2002. http://dx.doi.org/10.1061/(ASCE)0733-9429(2002)128:7(697)
[10] K. Fukushima, R. Maeshima, A. Kinoshita, H. Shiraishi, and I.
Koshijima, "Gas pipeline leak detection system using the online
simulation method," Computers & chemical engineering, vol. 24, pp.
453-456, 2000. http://dx.doi.org/10.1016/S0098-1354(00)00442-7
[11] M. Mohammadzaheri, L. Chen, and S. Grainger, "A critical review of
the most popular types of neuro control," Asian Journal of Control, vol.
16, pp. 1-11, 2012 http://dx.doi.org/10.1002/asjc.449
http://dx.doi.org/10.15242/IIE.E0115012
66