Accepted Manuscript

Leak detection of water distribution pipeline subject to failure of socket joint

based on acoustic emission and pattern recognition

Suzhen Li, Yanjue Song, Gongqi Zhou

PII: S0263-2241(17)30649-8
DOI: https://doi.org/10.1016/j.measurement.2017.10.021
Reference: MEASUR 5023

To appear in: Measurement

Received Date: 23 May 2017

Revised Date: 27 July 2017
Accepted Date: 11 October 2017

Please cite this article as: S. Li, Y. Song, G. Zhou, Leak detection of water distribution pipeline subject to failure
of socket joint based on acoustic emission and pattern recognition, Measurement (2017), doi: https://doi.org/
10.1016/j.measurement.2017.10.021

This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers
we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and
review of the resulting proof before it is published in its final form. Please note that during the production process
errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.
Leak detection of water distribution pipeline subject to failure of
socket joint based on acoustic emission and pattern recognition
Suzhen Li a,b, Yanjue Songb and Gongqi Zhoub
a
Tongji Univeristy, State Key Laboratory of Disaster Reduction in Civil Engineering, Siping 1239, Shanghai,
China, 200092
b
Tongji Univeristy, College of Civil Engineering, Siping 1239, Shanghai, China, 200092

Abstract. Early leak detection is of great importance for life-cycle maintenance and management of municipal
pipeline system. Due to economic and technical efficiency, ductile iron pipe segments and socket joints are
widely used in practice to construct water distribution systems. The ductile configuration of the socket joint
allowing for large deformation constitutes the most common cause for water leakage. Using acoustic emission
(AE) techniques, this paper presents an experimental study on leak detection of a water distribution system
subject to failure of socket joint. The acoustic characteristics of leak signals in the socket and spigot pipe
segments are investigated. After feature extraction and selection, a classifier based on artificial neural network
(ANN) is established. It has been validated that the dominant frequencies of the AE leak signals due to the
failure of the socket joint concentrate on 0~10kHz. The proposed ANN-based method can achieve good
estimation accuracy of 97.2% and 96.9% by using the feature set {Peak, Mean, Peak Frequency, Kurtosis} and
{Mean, Peak Frequency}.

Keywords: water leak detection, socket joint, acoustic emission, pattern recognition, artificial neural network

Address all correspondence to: Suzhen Li, Tongji University, College of Civil Engineering and State Key
Laboratory of Disaster Reduction in Civil Engineering, Siping 1239, Shanghai, China, 200092; Tel: 86-21-
65981505; E-mail: Lszh@tongji.edu.cn

1. INTRODUCTION

As a major municipal infrastructure to deliver water with appropriate quality, quantity and pressure,
water distribution pipeline plays an important role in modern society. A serious issue existing in the
daily operation of the water supply system is leakage, which may cause considerable cost difference
between production and sale of water and even pose a threat to public safety. Early detection and
precise location of leakage is of great importance for life-cycle maintenance and management of
widely-distributed pipeline system.
Many methods have been proposed for leak detection in water pipes, including visual inspection,
electromagnetic methods, acoustic methods, ultrasound methods, radiographic methods, and
thermography methods (Liu & Kleiner, 2013). In the past few years, acoustic emission (AE)
techniques have demonstrated to be an excellent tool for on-line leak detection given the fact that the
leakage can release elastic energy in form of transient stress waves and generate the signals
representative of the abnormal AE events (Dipen, 2005; ASTM, 2011). Many efforts have been made
to investigate acoustic features of leakage source, propagation characteristics of acoustic waves along
pipelines, and the relation of AE signals with different parameters like leakage rate, propagating
distance, material and geometric properties of pipelines, operating conditions etc. (Gao et al., 2004,
2006; Brunner and Barbezat, 2006; Yang et al., 2008; Tang et al., 2009; Khulief et al., 2012; Juliano et
al., 2013). Compared with the leak detection methods using hydrophones or accelerometers for
measurements of fluid-borne waves or vibrations (Puust et al., 2010; Yazdekhasti et al., 2016), AE
techniques provides the signals which are very sensitive to the leakage (Gao et al., 2005). Some
successful detections have been reported with considerable detecting distances (Anastasopoulos et al.,
2009; Lim, 2015) and adaptability to various pipe materials (Hunaidi et al., 2000; Martini et al., 2016).
Leak detection using AE signals has gone through several stages of development. Parameter
analysis is a common method for the early AE-based leak detection. Some characteristic indices are
defined or constructed from the raw AE data to discriminate leak signals from environmental noise,
including peak amplitude, mean value, root-mean-square (RMS) value, variance, kurtosis, etc. in time
domain (Grosse & Ohtsu, 2008) as well as dominant frequency, peak frequency, frequency centroid,
energies at various sub-bands etc. in frequency domain (Meng et al., 2012). However, the poor
performance of AE sensors and instruments at the early stage, such as narrow bandwidth and limited
acquisition power, restricts the effectiveness and accuracy of AE techniques (Hunaidi & Chu, 1999).
With the enormous progress in hardware, AE sensors are able to achieve records with high efficiency
and reliability. Meanwhile, more sophisticated methods are employed as well to improve the accuracy
of AE signal processing, among which empirical mode decomposition and wavelet transform are
widely used for de-noising and feature extraction (Ahadi & Bakhtiar, 2010; Ding et al., 2004; Ferrante
& Brunone, 2003; Guo et al., 2016; Meng et al., 2012; Romano et al., 2011; Sikorska & Mba, 2008;
Yuan et al., 2015).
An important issue existing in the AE based leak detection system is to distinguish the leak signal
without giving false alarms. Considering that the acoustic wave obtained by the data acquisition
system is digital signal in essence, modern pattern recognition methods for data processing have been
adopted to improve the accuracy of the two-class classification problem in leak detection (Valizadeh et
al., 2009; Zadkarami et al., 2016). Among them, logistic regression (LR), artificial neural network
(ANN), and support vector machine (SVM) are the most popular algorithms. It is commonly accepted
that there is not a so-called “best method” and the efforts should be devoted to a good solution for a
specific engineering problem. In general, ANN and SVM classifiers present better performance than
LR for the non-linear, non-stationary data processing like AE leak signals (Dreiseitl & Ohno-Machado,
2002).
Although the pattern recognition strategy has made great advances in AE-based leak detection, the
most important thing from the view of engineering is to determine the “pattern”, which is highly case-
dependent on the specific application. Due to economic and technical efficiency, ductile iron pipe
segments and socket joints are widely used in practice to construct water distribution systems.
However, the ductile configuration of the socket joint allowing for large deformation constitutes the
most common cause for water leakage. Although many work have been reported regarding the leak
detection of water pipeline using acoustic signals, most of them are focused on the leak either due to
the opening of valves and faucets (Khulief et al., 2012), or by manipulatively introducing a crack, a
hole or other damages to the pipe segments (Ahadi & Bakhtiar, 2010; Butterfield et al., 2017; Osama
Hunaidi et al., 2000). The theoretical and experimental investigation on AE-based leak detection of a
full-scale pipeline model subject to failure of socket joint are rarely found.
This paper presents a recent effort on AE leak tests of a water distribution pipeline system with two
ductile iron pipe segments connected by a socket joint. A specific experimental apparatus is set up to
simulate the leakage of the water pipeline under operating condition due to the joint failure. The
acoustic characteristics of leak signals in the socket and spigot pipe segments are investigated. After
feature extraction and selection, a classifier based on artificial neural network (ANN) is established
and used for leak detection of water distribution pipeline.

2. EXPERIMENTAL SETUP

To simulate the leakage due to failure of socket joint in water distribution pipeline system, a specific
setup is elaborately designed, as shown in Figure 1. Two ductile iron pipe segments of 200mm inside
diameter and 3m length are employed and connected by using a socket joint. A blind flange is fit at
each end of the segment. The socket joint adopts a newly developed configuration, capable of water
leakage prevention. Owing to the filled-in gasket made of water swelling rubber, the space between
socket and spigot can be decreased with the progressive expansion of the rubber, which prevents the
aggravation of water leakage.
 Socket joint Displacement gauge
Water pressure meter d2 d1 Pressure regulator
Blind flange
Spigot segment Socket segment
Outlet AE sensor (A2) Measuring cup AE sensor (A1) Inlet

Hose
Water tank Water pump

Hose
Water pump Socket joint Pipe segment

Water tank
Measuring cup

Figure 1. Experimental setup

A water cycling system is constructed to simulate the operational state of water pipeline, which
consists of a water pump, a hose and a water tank. Water in the tank is driven by the pump, via the
hose runs through the pipeline and then flows back to the tank. Two pressure regulators and water
pressure meters are respectively installed onto the ends of the pipe to adjust and measure the pressure
of the water inlet and outlet. A displacement gauge is fixed at the socket joint to measure the relative
displacement of the joint. The AE sensing system adopted in the experiment is composed of an 8-
channel PXI-5105 data acquisition card with the maximum sampling rate of 50 MS/s, two PAC R15α
sensors, two PXPAIV pre-amplifiers and a personal computer (PC). The two AE sensors are attached
on the top of the pipeline using adhesive types: A1 is on the socket segment with d1 away from the
joint; and A2 is on the spigot segment with d2 away from the joint.
One end of the tested specimen is fixed to a load device and the other end is placed against a
reaction wall. Relying on this setup, pulling load can be applied to the pipeline by controlling the joint
displacement. The water dropping down from the joint is collected by a measuring cup, based on
which the leakage rate can be estimated.
The internal pressure is first stabilized at 0.2MPa by adjusting the pressure regulators at the inlet
and outlet, whereby the signal involving environmental noise is recorded by the AE sensors. The
tensile load is progressively applied to the pipeline until the water drip-drop at the socket joint can be
visibly observed. By now, the measured joint displacement is 67mm and the leakage rate is 140mL/s.
Keep the load constant and the internal pressure being 0.2MPa. With the gradual expansion of the
water swelling rubber, the severity of water leakage is mitigated as time goes on. After an hour, the
leakage rate is finally stabilized at 20mL/s.
In the progress of the water leakage rate decreasing from 140 mL/s to 20mL/s, the AE sensors are
placed at the different distance (d1, d2) away from the joint to record the AE signals in succession. The
sampling rate is 1MHz and the duration of each acquisition is 1s. The testing cases under different
working conditions are summarized in Table 1.

Table 1 Testing cases

Pump on Pump off
Leak Dura Dura Leak Dura Dura
d1 Cases d2 Cases d1 Cases d2 Cases
rate tion tion rate tion tion
(m) No. (m) No. (m) No. (m) No.
(mL/s) (s) (s) (mL/s) (s) (s)
0 0.5 C0-1 6 1 C0-2 6 0.5 Z1-1 7 0.5 Z1-2 6
[15,25)
[15,25) 0.5 C1-1 2 1 C1-2 7 1 Z2-1 6 1 Z2-2 7
 1.5 C2-1 9 2 C2-2 13 1.5 Z3-1 6 2 Z3-2 9
2.5 C3-1 9 2 Z4-1 9
2.5 Z5-1 11
1.5 C4-1 8 1 C3-2 8 0.5 Z6-1 8 1 Z4-2 1
[25,45)
2.5 C5-1 7 2 C4-2 7 [25,45) 1.5 Z7-1 4 2 Z5-2 1
≥100 1.5 C6-1 2 1 C5-2 2 2.5 Z8-1 4

3. PROCEDURE FOR SIGNAL-BASED LEAK DETECTION

A standard procedure for signal-based leak detection is given in Figure 2. Using the AE signals
recorded in the above experimental cases, the procedure is mainly followed by the three steps: (1)
feature extraction; (2) feature selection; (3) leak detection based on pattern recognition. The methods
are introduced in details as follows.

Figure 2. Leak detection procedure

3.1. Feature extraction

Since the sampling rate of AE signals is usually high, it is impossible to directly use the raw data
for pattern recognition algorithm. Data de-noising and compression should be first conducted by
extracting features as representation of the original data for the decision model (Hao et al., 2014). In
general, the features can be constructed in time domain or frequency domain. The most commonly
used features of AE signals (Li et al., 2016) are listed in Table 2, where X   x1 , x2 , , xN  denotes an
AE signal in time domain, and f   f1 , f 2 , , f n  in frequency domain with xf   xf1 , xf 2 , , xf n  .

Table 2 Typical features of AE signals

Features in time Features in
Expressions Expressions
domain frequency domain
3
N  fi   f 
max abs( xi )
1
Peak Skewness
N
 
i 1  f


4
1 N N  fi   f 
abs( x )
1
Mean
N
i Kurtosis
N
 
i 1  f

i 1

N
1
 x   
Standard deviation 2
xfi | fi maxabs( f )
Peak frequency
(STD) N
i
i 1
 1 N xf  f T
 x 
Root-mean-square 2
Frequency centroid

n
(RMS) N
i
f
i 1 i 1 i

max abs( xi )
Crest factor 1
  xi 
N 2
i 1
N
1 N
  xi 
2
Energy
N i 1

3.2. Feature selection

After feature extraction, totally 10 features as shown in Table 1 are obtained from the original AE
signals. To find which features are most suitable for pattern recognition, feature selection plays an
important role before training a robust classifier. In general, if the separating capacity of the features is
weak, the classifier would have poor performance; if the features contains a fair amount of redundant
information, the efficiency and accuracy of the classifier would be affected. Feature selection is the
process of detecting the relevant features and discarding the irrelevant ones.
For this reason, a selection algorithm based on cross entropy is employed here to rank the features
from best to worst and then, a set of top listed features is selected for classification. Cross entropy,
which is also known as Kullback-Leibler (KL) distance (d), is defined as (Theodoridis and
Koutroumbas, 2008):
d  D12  D21 (1)
p  x w1 
D12   p  x w1  log 2
p  x w2 
(1-1)

p  x w2 
D21   p  x w2  log 2
p  x w1 
(1-2)

where w1 , w2 are two classes representative of the cases with leak or no leak; x   x1 , x2 , xn  is a
T

sort of features for the n samples. Actually, cross entropy measures the ‘‘distance’’ between the
probability functions p  x w1  and p  x w2  .In case of two completely overlapped classes, d=0. The
values of d increase as the two distributions move far apart from each other. In the other words, the
better features correspond to the higher values of d.

3.3. Artificial neural network (ANN) based leak detection

After feature extraction and selection, a classifier is finally constructed using the features to
determine whether a signal is leak or not. Many algorithms have been developed for such two-class
classification problems. Among them, artificial neural network (ANN) is one of the most common
classifiers in pattern recognition and has a recent resurgence as the state-of-the-art technique for many
applications (Paliwal & Kumar, 2009).
For leak detection using acoustic signals in this work, ANN with a back-propagation (BP)
algorithm is employed. The neural network is consisted of three layers, including one input layer, one
hidden layer and one output layer. The nodes in the input layer are the selected features as given in
Section 3.2. Two nodes are adopted in the output layer, where the value “1” stands for leak and “0” for
the case without leak.
Sigmoid function is chosen as the transfer function between the two adjacent layers. Taking z as
the input of a neuron, the transfer function is given by
 1
g ( z)  (2)
1  e z
Suppose there are m training samples denoted as X   X1 , X 2  X i  X m  , in which the column vector
T

X i   xi1 , xi 2 ,, xin  is the n features of the i input sample; ai  denotes the i node in the (j+1)
T th j 1 th th

layer;  jk is the weight of the direct edge from the j node in the i layer to the k node in the (i+1)th
i th th th

layer. Define a1  X i for the input layer, and the nodes in the rest layers can be determined layer by
layer according to:

ai j 1  g  j  a j
T
 (3)
th th
Thus,
z（k j +1）, the output of the k node in the (j+1) layer, is
m
z（k j +1） ikj a j (4)
i 1

So far, the procedure called forward propagation is accomplished.

To adjust the weights to get the best fitting model for the training set, a BP algorithm should be
conducted by following these steps:
(1) Initialize the weights close to zero randomly;
(2) Perform the forward propagation process with all m groups of training set X to get the predicting
results Y  Y1 , Y2 Yi Ym  , where Yi   yi1 , yi 2  denotes the two-class classification results for the ith
T T

training sample;
(3) Calculate the difference between the predicting results Y and the expected results
D   D1 , D2  Di  Dm  to construct the cost function with the weights Θ :
T

  yij  dij 
1 m 2
J Θ = (5)
2m i 1 j 1
(4) Use the gradient descent method to optimize the weights:

 jik   jik  J Θ (6)
 jik
To implement gradient descent, the derivatives of the cost function with respect to the weights are
calculated from the output layer to the input layer via a BP process. Here take one training sample for
example:
 zik  
J Θ   k J  Θ   aij  k J  Θ  (7)
 jik  ji zi
k
zi
As the start of the BP algorithm, the errors of the output layer are defined as

J  Θ   yi  di (8)
 jiL 1
Through a recursive computation for the hidden layers, it holds
 aik  aik
J  Θ    J  Θ     jik (9)
zik zik aik zik j

When several samples are taken into consideration, J  Θ  / ai is equal to the sum of the deviation of
k

all the samples.

(5) Repeat step 2 ~ step 4 until the algorithm convergence or the maximum number of iteration are
achieved.

4. LEAK DETECTION FOR THE EXPERIMENT

4.1. Raw data of AE signals

The primary comparison of the time and frequency characteristics of the raw AE signals in the
socket and spigot pipelines without and with leakage is illustrated in Figure 3 and Figure 4. It can be
observed that: (1) the dominant frequency of environmental noise due to the is less than 2kHz; (2) the
dominant frequencies of the AE leak signals in the two pipe segments concentrate on 0~10kHz,
slightly influenced by socket joint; (3) the data in time and frequency domain present significant
difference between AE leak signals and the environmental noise; (4) the AE sensor fixed to the spigot
pipeline displays a bit larger measuring noise than that fixed to the socket pipeline.
Amplitude/V

Amplitude/V
0.1 0.1

0 0

-0.1 -0.1
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
Time/s Time/s
-3 -3
x 10 x 10
Amplitude/V

Amplitude/V
5 5

0 0
0 2 4 6 8 10 0 2 4 6 8 10
Frequency/kHz Frequency/kHz
(1) Socket pipeline (2) Spigot pipeline
Figure 3. Acoustic characteristics of environmental noise
Amplitude/V
Amplitude/V

0.1 0.1

0 0

-0.1 -0.1
0 0.1 0.2 0.3 0.4 0 0.1 0.2 0.3 0.4
Time/s Time/s
-3 -3
x 10 x 10
Amplitude/V
Amplitude/V

5 5

0 0
0 2 4 6 8 10 0 2 4 6 8 10
Frequency/kHz Frequency/kHz
(1) Socket pipeline (2) Spigot pipeline
Figure 4. Acoustic characteristics of leak signals (41mL/s)

4.2. Feature extraction and selection

Since the main components of the AE signals concentrate on 0~10kHz, a low-pass filter below
15kHz is introduced to pre-process the raw data. The de-noised AE signal for each case is first
separated into a series of samples by setting the duration as 0.03s with a 50% overlap between two
adjacent samples. For each sample, the ten features as shown in Table 2 are calculated and used for
feature selection. With respect to each feature, the cross entropy or say the KL distance (d) is
calculated according to Eq.(1). Table 3 lists the ranking of the features in time and frequency domain.
It can be seen that the features in time domain except energy have a big value of d, which verifies the
good capability of these features for separating the leak signals from the environmental noise. As for
those in frequency domain, kurtosis, peak frequency and skewness perform better. Regarding this, four
top ranking features including peak, mean, Kurtosis and peak frequency are employed as the
candidates for ANN-based pattern recognition.

Table 3 Feature ranking

 Features in time domain d Features in frequency domain d
Peak 51.71 Kurtosis 49.30
Mean 51.57 Peak Frequency 49.12
STD 51.50 Skewness 49.08
RMS 51.50 Frequency Centroid 5.70
Crest Factor 51.49
Energy 8.15

To investigate the potential influencing factors on pattern recognition, peak and kurtosis of the
leak signals, which are the top ranking feature in time and frequency domain, are presented in Figure 5
with respect to various situations. Consider the influence of socket joint, different propagating distance
and leak rate on the two features in succession. When one factor is under investigation, all the others
are kept same. It can be found that there is no significant difference in kurtosis with respect to various
factors. Kurtosis, on the other hand, seems more discrete but still cannot distinguish the leak signals in
terms of different cases. By and large, socket joint, propagating distance and leak rate have little
influence on the classification features from the aspect of pattern recognition.

0.25 0.5 0.5m 0.5

Socket [20,25)
1.0m
Spigot 0.4 0.4 [25,45)
0.2 1.5m
≥100
2.0m
0.3 0.3
Peak/V

0.15
Peak/V

2.5m
Peak/V

0.1 0.2 0.2

0.05 0.1 0.1

0 0 0
0 500 1000 0 1000 2000 3000 0 500 1000 1500
Sample No. Sample No. Sample No.
8000
4000 8000 0.5m
[20,25)
Socket [25,45)
1.0m
Spigot 6000 ≥100
3000 6000 1.5m
2.0m
Kurtosis
Kurtosis

Kurtosis

2.5m 4000
2000 4000

2000
1000 2000

0 0 0
0 500 1000 0 1000 2000 3000 0 500 1000 1500
Sample No. Sample No. Sample No.

(1) Socket joint (2) Propagating distance (3) Leak rate (Unit: mL/s)
Figure 5. Features under different situations

4.3. Leak detection

Considering the cases with pump on, 30% of the raw AE data including leak signals and
environmental noise are randomly selected as the training set. The rest 70% are employed as the
testing set 1. Meanwhile, the data acquired in the cases with pump off is adopted as the testing set 2.
Following the procedure as given in Section 3.3, the ANN-based classifier is constructed based on the
training set.
 Using different feature combination by taking one or more features from the candidates including
peak, mean, Kurtosis and peak frequency, the estimation accuracy of the trained ANN model are
presented in Table 4. It can be seen that most cases achieve good accuracy of over 90%. The accuracy
of the cases using the testing set 1 is generally higher than that using the testing set 2 since the training
set is also taken from the cases with pump on. Comparing with the features in time domain, those in
frequency domain present good performance regardless of the effect of water flow. When the top
ranking features {Peak, Mean, Peak Frequency, Kurtosis} are adopted, the estimation accuracy using
all testing tests is 97.2%. The feature set {Mean, Peak Frequency} achieves the accuracy of 96.9%. It
indicates that the combination of time- and frequency-domain features reduces false alarm
significantly. In summary, two features with one from time domain and the other from frequency
domain can work well for water leak detection due to the failure of socket joint.

Table 4 Accuracy of leak detection using different feature sets

Accuracy of testing Accuracy of
set 1 (%) testing set 2 (%) Accuracy of all
Feature set
testing sets (%)
Safe Leak Leak
{Peak, Mean, Peak
96.5 99.3 95.9 97.2
Frequency, Kurtosis}
{Peak, Kurtosis} 92.9 99.2 92.4 94.8
{Mean, Peak Frequency} 97.6 98.1 95.0 96.9
{Mean, Kurtosis} 95.7 98.7 90.6 95.0
{Mean} 57.6 99.0 98.9 85.2
{Peak Frequency} 3.3 100.0 100.0 67.8

5. CONCLUSIONS

This paper presents an experimental investigation on AE-based leak detection of a water

distribution system subject to failure of socket joint. A pattern recognition procedure based on artificial
neural network is proposed. The major work and conclusions are summarized as follows.
(1) With regard to the specifically designed experiments, the dominant frequency of environmental
noise is less than 2kHz and the dominant frequencies of the AE leak signals in the two pipe segments
concentrate on 0~10kHz.
(2) After data preprocessing using a low-pass filter below 15kHz, feature selection based on cross
entropy ranks the ten features extracted from the de-noised AE signals from best to worst. Based on
this, four features including peak, mean, Kurtosis and peak frequency are employed as the candidates
for training and testing of the ANN model.
(3) It is validated that socket joint, propagating distance and leak rate have little influence on the
classification features from the aspect of pattern recognition. The selected features present good
performance regardless of the effect of water flow.
(4) The proposed ANN-based method can achieve good estimation accuracy of 97.2% and 96.9%
using the feature set {Peak, Mean, Peak Frequency, Kurtosis} and {Mean, Peak Frequency}. It
indicates that the combination of time- and frequency-domain features reduces false alarm
significantly.

ACKNOWLEDGMENTS
The authors are grateful to the State Key Laboratory of Disaster Reduction in Civil Engineering
(Project: SLDRCE14-B-19) and the National Key Research and Development Program of China
(Project: 2016YFC0802406) for the financial support of this work.

REFERENCES

Ahadi, M., & Bakhtiar, M. S. (2010). Leak detection in water-filled plastic pipes through the application of tuned
wavelet transforms to Acoustic Emission signals. Applied Acoustics, 71(7), 634-639.
doi:10.1016/j.apacoust.2010.02.006
Anastasopoulos, A., Kourousis, D., & Bollas, K. (2009). ACOUSTIC EMISSION LEAK DETECTION OF
LIQUID FILLED BURIED PIPELINE. Journal of Acoustic Emission, 27.
ASTM. (2011). Standard terminology for nondestructive examinations Astm.
Brunner,A.J. and Barbezat, M, (2006). Acoustic Emission Monitoring of Leaks in Pipes for Transport of Liquid
and Gaseous Media: A Model Experiment, Advanced Materials Research, 13-14:351-356
Butterfield, J. D., Krynkin, A., Collins, R. P., & Beck, S. B. M. (2017). Experimental investigation into vibro-
acoustic emission signal processing techniques to quantify leak flow rate in plastic water distribution pipes.
Applied Acoustics, 119, 146-155. Ding, Y., Reuben, R. L., & Steel, J. A. (2004). A new method for waveform
analysis for estimating AE wave arrival times using wavelet decomposition. Ndt & E International, 37(4),
279-290.
Dipen N. Sinha, (2005). Acoustic Sensor for Pipeline Monitoring, Technology Report, Los Alamos National
Laboratory, USA
Dreiseitl, S., & Ohnomachado, L. (2002). Logistic regression and artificial neural network classification models:
a methodology review. Journal of Biomedical Informatics, 35(5–6), 352-359.
Ferrante, M., & Brunone, B. (2003). Pipe system diagnosis and leak detection by unsteady-state tests. 2. Wavelet
analysis. Advances in Water Resources, 26(1), 107-116.
Gao, Y., Brennan, M. J., Joseph, P. F., Muggleton, J. M., & Hunaidi, O. (2005). On the selection of
acoustic/vibration sensors for leak detection in plastic water pipes. Journal of Sound & Vibration, 283(3–5),
927-941.
Grosse, C., & Ohtsu, M. (2008). Acoustic Emission Testing: Springer Berlin Heidelberg.
Guo, C., Wen, Y., Li, P., & Wen, J. (2016). Adaptive noise cancellation based on EMD in water-supply pipeline
leak detection. Measurement, 79, 188-197.
Hao, J., Zhang, L., Wei, L., & Ding, Q. (2014). Integrated leakage detection and localization model for gas
pipelines based on the acoustic wave method. Journal of Loss Prevention in the Process Industries, 27(1), 74-
88.
Hunaidi, O., Chu, W., Wang, A., & Guan, W. (2000). Detecting leaks in plastic pipes. Journal, 92(2), 82-94.
Hunaidi, O., & Chu, W. T. (1999). Acoustical characteristics of leak signals in plastic water distribution pipes.
Applied Acoustics, 58(3), 235-254. doi:Doi 10.1016/S0003-682x(99)00013-4
Juliano, T., Meegoda, J., and Watts, D. (2013). Acoustic Emission Leak Detection on a Metal Pipeline Buried in
Sandy Soil, Journal of Pipeline Systems Engineering and Practice, 4(3):149-155
Khulief, Y. A., Khalifa, A., Ben Mansour, R., & Habib, M. A. (2012). Acoustic Detection of Leaks in Water
Pipelines Using Measurements inside Pipe. Journal of Pipeline Systems Engineering and Practice, 3(2), 47-54.
doi:10.1061/(Asce)Ps.1949-1204.0000089
Li, Z., Zhang, H., Tan, D., Chen, X., & Lei, H. (2016). A novel acoustic emission detection module for leakage
recognition in a gas pipeline valve. Process Safety & Environmental Protection, 105, 32-40.
Lim, J. (2015). Underground Pipeline Leak Detection Using Acoustic Emission and Crest Factor Technique
Advances in Acoustic Emission Technology (pp. 445-450): Springer.
Liu, Z., & Kleiner, Y. (2013). State of the art review of inspection technologies for condition assessment of water
pipes. Measurement, 46(1), 1-15.
Martini, A., Troncossi, M., & Rivola, A. (2016). Leak Detection in Water-Filled Small-Diameter Polyethylene
Pipes by Means of Acoustic Emission Measurements. Applied Sciences, 7(1), 2.
Meng, L. Y., Li, Y. X., Wang, W. C., & Fu, J. T. (2012). Experimental study on leak detection and location for
gas pipeline based on acoustic method. Journal of Loss Prevention in the Process Industries, 25(1), 90-102.
doi:10.1016/j.jlp.2011.07.001
Paliwal, M., & Kumar, U. A. (2009). Review: Neural networks and statistical techniques: A review of
applications. Expert Systems with Applications, 36(1), 2-17.
Romano, M., Kapelan, Z., & Savić, D. A. (2011). Real-Time Leak Detection in Water Distribution Systems.
Paper presented at the Conference on Water Distribution Systems Analysis.
Puust, R., Kapelan, Z., Savic, D., & Koppel, T. (2010). A review of methods for leakage management in pipe
networks. Urban Water Journal, 7(1), 25-45.
FSikorska, J. Z., & Mba, D. (2008). Challenges and obstacles in the application of acoustic emission to process
machinery. ARCHIVE Proceedings of the Institution of Mechanical Engineers Part E Journal of Process
Mechanical Engineering 1989-1996 (vols, 222(1), 1-19.
Tang, X., Liu, Y., Zheng, L., Ma, C., & Wang, H. (2009). Leak Detection of Water Pipeline Using Wavelet
Transform Method. Paper presented at the International Conference on Environmental Science and
Information Application Technology.
Theodoridis, S., & Koutroumbas, K. (2008). Pattern Recognition (Fourth Edition): Academic Press.
Valizadeh, S., Moshiri, B., & Salahshoor, K. (2009). Leak Detection in Transportation Pipelines Using Feature
Extraction and KNN Classification. Paper presented at the Pipelines Specialty Conference.
Y. Gao, M.J. Brennan, P.F. Joseph, J.M. Muggleton and O. Hunaidi, (2004). A model of the correlation function
of leak noise in buried plastic pipes, Journal of Sound and Vibration, 277(1-2): 133–148
Y. Gao, M. J. Brennan, P. F. Joseph, (2006). A Comparison of Time Delay Estimators for the Detection of Leak
Noise Signals in Plastic Water Distribution Pipes, Journal of Sound and Vibration, 292: 552-570.
Yazdekhasti, S., Piratla, K. R., Atamturktur, S., & Khan, A. A. (2016). Novel vibration-based technique for
detecting water pipeline leakage. Structure & Infrastructure Engineering, 1-12.
Yang, J., Wen, Y., & Li, P. (2008). Leak location using blind system identification in water distribution pipelines.
Journal of Sound & Vibration, 310(1–2), 134-148.
Yuan, F., Zhong, Z., Cai, Y., & Cheng, E. (2015). Leak Detection Research of Water Supply Pipeline Based on
HHT: Springer Berlin Heidelberg.
Zadkarami, M., Shahbazian, M., & Salahshoor, K. (2016). Pipeline leakage detection and isolation: An
integrated approach of statistical and wavelet feature extraction with multi-layer perceptron neural network
(MLPNN). Journal of Loss Prevention in the Process Industries, 43, 479-487.