State-of-the-Art Review of Water Leak Detection and Localization Methods through Hydrophone Technology

Abstract

Acoustic technologies are popular for detection of leak detriments in water pipelines. However, problems of false alarms, detection of weak or difficult leaks, accurate leak pinpointing, and the high cost of long-term monitoring remain prevalent. These issues demand a more sophisticated testing approach suitable for real-world application. In particular, hydrophone technology has strong promise for long-range leak detection in high-attenuation conditions. However, existing review studies only cover the methods of leak detection holistically, with limited insight into the practical implementation of sensing technologies for water leak detection. In particular, the problem of detecting and localizing leaks using hydroacoustic data has not yet been extensively studied. The current study, therefore, presents a state-of-the-art review of the extant literature on water leak detection and localization taking hydrophones as a good example of hydroacoustic water leak detection. The study compares hydrophones with other popular sensing technologies such as accelerometers and guides on its better application for detecting water leaks. Current research directions, gaps, and future work foci are also identified to enable further development of a hydrophone-based water leak detection system. Review shows that existing experiments are limited to controlled conditions where impacts of surrounding strata, ambient noise, and difficult pipe geometries cannot be studied. Future studies can apply the technology to real-life cases, developing faster analytical methods and hybrid solutions using a multisensing approach. This can help water leak experts enormously in cost-effective, efficient detection of leaks.

Full text

State-of-the-Art Review
Review of Water Leak Detection and Localization
Methods through Hydrophone Technology
Beenish Bakhtawar1; and Tarek Zayed, F.ASCE2
Abstract: Acoustic technologies are popular for detection of leak detriments in water pipelines. However, problems of false alarms, de-
tection of weak or difficult leaks, accurate leak pinpointing, and the high cost of long-term monitoring remain prevalent. These issues demand
a more sophisticated testing approach suitable for real-world application. In particular, hydrophone technology has strong promise for long-
range leak detection in high-attenuation conditions. However, existing review studies only cover the methods of leak detection holistically,
with limited insight into the practical implementation of sensing technologies for water leak detection. In particular, the problem of detecting
and localizing leaks using hydroacoustic data has not yet been extensively studied. The current study, therefore, presents a state-of-the-art
review of the extant literature on water leak detection and localization taking hydrophones as a good example of hydroacoustic water leak
detection. The study compares hydrophones with other popular sensing technologies such as accelerometers and guides on its better ap-
plication for detecting water leaks. Current research directions, gaps, and future work foci are also identified to enable further development of
a hydrophone-based water leak detection system. Review shows that existing experiments are limited to controlled conditions where impacts
of surrounding strata, ambient noise, and difficult pipe geometries cannot be studied. Future studies can apply the technology to real-life
cases, developing faster analytical methods and hybrid solutions using a multisensing approach. This can help water leak experts enormously
in cost-effective, efficient detection of leaks. DOI: 10.1061/(ASCE)PS.1949-1204.0000574.© 2021 American Society of Civil Engineers.
Author keywords: Hydrophones; Water leaks; Detection; Localization; Water distribution network (WDN).
Introduction
Water leakage is a plaguing issue in the current throes of global
water crises. It is the biggest constituent of the 126 billion cubic
meters volume per year of nonrevenue water (NRW) estimates
(Liemberger and Wyatt 2019). Discontinued water supply for an
extended time during leak rectifications can cause a further nui-
sance to users. Additionally, false alarms and wrong assessment of
leak location can cause huge repair costs. Thus, early and accurate
identification of leaks and timely rectifications can significantly
improve water distribution and supply efficiency. In this regard,
acoustic techniques have the potential for both short- and long-term
leak monitoring and control (Hunaidi and Chu 1999;Khulief and
Khalifa 2013;Xu et al. 2019). Among these techniques, hydro-
phones have invoked interest for their capability of capturing the
in-pipe acoustic signature of leak signals (Khulief et al. 2012). This
is significantly different from out-of-pipe technologies such as ac-
celerometers (Marmarokopos et al. 2018).
Generally, pipe leaks can generate high-frequency noise if there
is unsteady flow separation at the leak location, or low-frequency
noise if hydrodynamic cavitation occurs at the leak location (Gao
et al. 2004a). During cavitation, pressure drop below the vapor pres-
sure can generate shock waves, which, upon falling on the pipewall,
create sound (Khulief and Khalifa 2013;Khulief et al. 2012). In this
situation, leak orifices act as high-pass sound filters, expelling high-
frequency sound energy outside the pipe system and reflecting low-
frequency sound signals into the pipe (Brennan et al. 2019). The
high-frequency sound wave is attenuated significantly inside the
pipe, leaving low-frequency sound signals to be the primary source
of data for leak diagnostic studies. As hydrophones are efficient in
detecting low-frequency noise signals, they can be the most suitable
instrument for leak detection in such conditions (Hamilton and
Charalambous 2020). Plastic pipes with a large diameter having a
high attenuation of sound waves make hydrophones successful in
producing high-resolution correlation in this case as compared to
other technologies (Gao et al. 2017;Hunaidi and Chu 1999). Various
experimental studies demonstrate hydrophones to produce similar or
higher accuracy results as compared to accelerometers, pressure sen-
sors, or ground microphones (Gao et al. 2005). Furthermore, Cody
et al. (2020a) suggest that acoustic methods deliver high-resolution
data for the detection of small leaks. However, coverage of the entire
network through hydrophones is argued to be time-consuming,
costly, and ineffective as compared to other hardware technologies
(Li et al. 2015).
This implies a lack of clarity in the effective implementation of
hydrophone sensor technology for leak detection and localization/
pinpointing. A detailed knowledge base seems to be missing in this
regard to provide practical guidance. Generally, the review studies
available in this domain of knowledge mostly aim to compare all
available techniques for water leak detection. For example, Puust
et al. (2010) discussed the leakage assessment methods, the major
technologies in use for leak detection and pinpointing, and the
hydraulic transient models for managing leaks. Li et al. (2015) cat-
egorized the leak detection methods into hardware- and software-
based methods. Furthermore, Datta and Sarkar (2016) presented a
detailed taxonomy for fault detection methods, distinguishing be-
tween blockage and leakage detection methods. Similarly, El-Zahab
and Zayed (2019) overviewed all popular leak detection methods.
1Research Assistant, Dept. of Building and Real Estate, Hong Kong
Polytechnic Univ., Hung Hom, Kowloon, Hong Kong (corresponding
author). ORCID: https://orcid.org/0000-0003-4994-2253. Email: beenish
.bakhtawar@polyu.edu.hk
2Professor, Dept. of Building and Real Estate, Hong Kong Polytechnic
Univ., Hung Hom, Kowloon, Hong Kong. Email: tarek.zayed@polyu
.edu.hk
Note. This manuscript was published online on July 23, 2021. Dis-
cussion period open until December 23, 2021; separate discussions
must be submitted for individual papers. This paper is part of the
Journal of Pipeline Systems Engineering and Practice,©ASCE,
ISSN 1949-1190.
© ASCE 03121002-1 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

These studies only use a broad scope to advance the taxonomical and
epistemological construct for water leak detection methods. Generic
discussion on all methods, however, does not help practitioners gain
a working insight of any hardware-based acoustic methods such as
hydrophones for practical implementation in the field. In this regard,
some of the relevant empirical studies have revealed its merit to be
used for real-time monitoring for leak detection (Lechgar et al. 2016;
Xu et al. 2019). However, practical implementation requires the iden-
tification of constraints for its use for efficient leak detection, accu-
rate localization, and long-term monitoring of real networks. The
current study addresses this knowledge gap by a systematic review
to explore the analytical methods applicable to hydroacoustic data.
Advancement in the application is explored through the identifi-
cation of current research directions, gaps in knowledge, and pos-
sible future directions for the use of hydrophones for water leak
detection.
Research Methodology
The study conducts a review on leak detection in water distribution
systems using hydrophones based on scientometric techniques and
qualitative content analysis. The research themes are first identified
by analyzing the citation data of articles. The identified themes of
research are then discussed in depth to highlight the current prac-
tices, research gaps, and future directions of work. The search de-
sign for articles involved both database search from Web of Science
(WoS) and backward snowballing. First, the literary works were
identified through a focused search on the WoS database. The
WoS search was designed using relevant keywords: hydrophones,
leak detection, pinpointing, localization, burst detection, and water
leaks. In total, 39 relevant literary sources were identified using the
process highlighted in Fig. 1. To find more relevant articles, the
method of backward snowballing was utilized (Mourão et al. 2020).
For the backward snowballing, cited references from the initially
identified 39 works were included in the search. Through this hybrid
approach, a total of 252 more relevant articles were identified. In
total, 291 articles were collected and screened for duplicates and
irrelevant articles based on a Title-Abstract-Keyword search. After
screening, 80 articles were found eligible for further analysis. Full-
text analysis of these 80 articles was then conducted to finalize
72 articles for review. The PRISMA flow diagram showing the de-
tailed shortlisting process and exclusion criteria are presented in
Fig. 1. The shortlisted studies are first classified based on study type,
hydrophone type, and leak detection phases. Furthermore, biblio-
graphic coupling using VOSviewer version 1.6.9 was used to iden-
tify prominent research directions and research topics. Bibliographic
coupling is a similarity-based network analysis method forming a
network of publications based on the number of citations common
between them (Patrício and Ferreira 2020). It does this by forming
clusters using two measures: the number of citations and the total
link strength (TLS). The number of citations shows the individual
influence of the publication itself, and the TLS indicates the strength
of correlation between publications. After categorizing the dominant
research themes, detailed content analysis was carried out to identify
the topics of interest, after which particular research gaps and future
directions in the research were suggested using qualitative full-text
analyses of the articles.
Research Trends
The yearwise analysis of the shortlisted studies is presented in
Fig. 2. Results reveal very limited studies on the subject until
the year 2000, with more than 50% of articles published in the last
five years. A growing interest implies promise for use of hydro-
phone devices for leak detection. Therefore, the current study
presents a timely review examining the scholarly space to identify
useful and pragmatic outlooks for its diverse application in different
aspects of the water leak detection problem.
Classification of Scholarly Works
Types of Research Studies
The type of studies conducted until now in the domain is presented
in Fig. 3. The research in the domain of leak detection is generally
guided by empirical evidence. This is supported by the high num-
ber of experimental studies on the application of hydrophones for
leak detection (63% of total). There are 23% of studies in the data
which are marked both theoretical and experimental. Such studies
offer a conceptual construct as the basis of their experimental work.
Fig. 1. Research methodology. Fig. 2. Yearwise analysis.
© ASCE 03121002-2 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

For example, Gao and Liu (2017) developed a theoretical model for
the relationship between internal pressure and wall displacements
and validated the model using lab experiments. Moreover, only
scattered efforts are made toward studying the physical concepts
and mechanisms of wave propagation inside the pipe. Only 7%
of real-life cases and theoretical studies could be found on the topic,
which implies a big gap in research. For example, Lechgar et al.
(2016) explored hydrophone implementation for leak detection in
Casablanca. Overall, theoretical studies including a review of pre-
vious work on the current research area are also negligible. Find-
ings overall reveal three-scale experimentation using hydrophones:
lab, test bed, and real pipe network. Most studies for leak detection
are based on lab or test-bed experimentation as shown in Fig. 3
(Gao et al. 2009;Hunaidi and Chu 1999). Although effective for
understanding fundamental physical principles, controlled condi-
tions are difficult to replicate in the field. This is due to the presence
of various uncertainties, attenuation effects, and high background
noise in real conditions. Thus, extending the lab experiments to real
networks is an important step for demonstrating the feasibility of
the technique (Guo et al. 2019).
Leak Detection Phases
Generally, the leak detection process can be classified into three
phases: detection, localization, and pinpointing (Hamilton and
Charalambous 2020). It can be seen that in almost all situations,
a leak is suspected in the system if there is a distinct peak in
the sound signal. To determine whether or not the suspected leak
is real, preprocessing is required to remove the ambient noise
effects, as there can be a false alarm in the system. This process
to confirm the presence of a leak in the system is known as leak
detection or leak identification (Cody et al. 2020b). El-Zahab and
Zayed (2019) included differentiating leaks from false alarms as an
essential step of leak identification. Once the presence of a leak is
confirmed, the location of the leak is determined. This process can
be termed as localization, location, or pinpointing, depending upon
the accuracy with which we can determine the distance of the leak
from the sensor (Sun et al. 2020). Many authors differentiate be-
tween these terms for the particular focus of their research or the
purpose of brevity. For example, El-Zahab and Zayed (2019) con-
sider narrowing down the location of the leak to a particular seg-
ment of the water network or a specific district metered area (DMA)
as leak localization, determining the location of the leak with an ac-
curacy of 30 cm as leak location, and determining the leak location
with an accuracy of 20 cm as pinpointing. On the contrary, Zaman
et al. (2020) define leak localization as pinpointing the location of the
leak. In practical terms, these three terms refer to the same process of
estimating the location of the leak (Datta and Sarkar 2016;El-Zahab
and Zayed 2019;Ma et al. 2019). On the basis of the generic three-
phase leak categorization, it was observed through content analysis
that 57 studies discuss leak detection, 41 discuss location/localization,
and only six studies focus on pinpointing the exact location of leak
through hydrophones as shown in Fig. 4. Very few articles focus on
two or three leak detection phases simultaneously.
Significant Research Directions
Analyzing the past research developments in any field helps in
building the theoretical background for more advanced research.
The current study does this using the results of bibliographic cou-
pling. The current study uses the metric to track research develop-
ments because of its retrospect perspective of analysis (Ferreira
2018). Fig. 5shows the developments using the network visuali-
zation in VOSviewer. In total, four main clusters of research foci
are visible in Fig. 5, formed through bibliographic coupling ex-
plained in the section “Research Methodology.”As 60 out of the
72 shortlisted studies were connected, the network formed was
visualized using 60 publications. The resulting clusters have been
assigned names following the dominant research direction of the
articles in the cluster and discussed in the same order in the sub-
sections: Cluster 1 (Section “Leak Detection and Localization Using
Hydrophones and Comparison with Accelerometers”), Cluster 2
(Section “Innovation for Long-Term Leak Detection and Localiza-
tion”), Cluster 3 (Section “Implementation of Hydrophones in Real
Networks”), and Cluster 4 (Section “Hydrophone Measurements
and Pipe Flow Dynamics”) (Fig. 5).
Leak Detection and Localization Using Hydrophones
and Comparison with Accelerometers
Cluster 1 has 21 articles, including the five most cited works in the
field, presented in Tables 1and 2. Mainly the studies in Cluster 1
involve the comparison of hydrophones with other sensors for selec-
tion of suitable acoustic methods (Gao et al. 2005), modeling the
acoustic properties for leak noise and efficient time-delay estimation
(Hunaidi and Chu 1999), and improving the signal analysis and sig-
nal quality of cross-correlation function (Brennan et al. 2019;Gao
et al. 2017). Seminal work from Hunaidi and Chu (1999)empirically
compared the performance of hydrophones and accelerometers
for water leak detection through experimentation. Their results
demonstrated the promising use of hydrophones, especially for
low-frequency leak signal propagation (<50 Hz) in PVC pipes
whose material has high signal attenuation properties. In the ex-
perimentation of their related work, Hunaidi et al. (2000)further
Fig. 3. Study types and scale of experiment for leak detection through
hydrophones. Fig. 4. Study focus with respect to leak detection phases addressed in
the publications.
© ASCE 03121002-3 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

demonstrated that hydrophones can detect small leaks even at
6L=min (1.6 gpm).
These studies have inspired many researchers to design similar
experiments and improve the leak detection process and cross-
correlation accuracy. For example, Gao et al. (2004a) developed
a model for cross-correlation of leak signals in plastic pipes, explor-
ing the effect of antialiasing filters on the removal of noise from
low-frequency noise data. By combing the cross-correlation with
the concepts of wave propagation in fluid-filled pipes, a cut-off fil-
ter range of 10–50 Hz was estimated for leak detection through
hydrophones. Gao et al. (2004b,2005) further compared the find-
ings of hydrophone data with accelerometers. It was found that
acoustic pressure data from hydrophones is useful for correlation
for leak cases with small signal-to-noise (SNR) ratio. Furthermore,
Gao et al. (2006) explained the method of cross-correlation for leak
localization in depth and compared the different time delay estima-
tors used for the purpose.
Overall, there are two basic methods of cross-correlation: basic
cross-correlation (BCC) and generalized cross-correlation (GCC).
The difference between the two methods is that in the GCC method,
the signals are prefiltered through a prewhitening process for re-
moving the background noise and sharpening the peak of the
cross-correlation function. For BCC, acoustic signals are mea-
sured at two different access points at both sides of suspected
leaks through hydrophones. These hydrophones can be either at-
tached to the hydrants or inserted in the pipe through valve open-
ings (Gao et al. 2005,2009;Khulief and Khalifa 2013;Xu et al.
2019). The presence of a leak is confirmed if there is a distinct
peak in the cross-correlation of the two measured signals, s1ðtÞ
and s2ðtÞ. The location of the leak can then be calculated concern-
ing either of the sensor locations, l1or l2, using the time delay
(tpeak) between the arrival times of signals to the sensor locations,
given as
l1¼l−ctpeak
2ð1Þ
where l= distance between access points and the propagation wave
speed cin the buried pipe. For any two random signals, s1ðtÞand
s2ðtÞ, the theoretical BCC function is given as
Rs1s2ðtdÞ¼E½s1ðtÞs2ðtþtdÞ ð2Þ
Fig. 5. Bibliographic coupling of shortlisted publications.
Table 1. Most significant articles with respect to number of citations in
Cluster 1
Rank Most significant articles No. of citations
1 Hunaidi and Chu (1999) 149
2 Hunaidi et al. (2000) 124
3 Gao et al. (2004a)95
4 Gao et al. (2005)91
5 Gao et al. (2006)78
Table 2. Most significant articles with respect to TLS in Cluster 1
Rank Publications Total link strength
1 Almeida et al. (2015) 184
2 Gao et al. (2006) 172
3 Gao et al. (2009) 161
4 Brennan et al. (2019) 157
5 Gao et al. (2017) 149
© ASCE 03121002-4 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

where td= time lag; and E= operator for maximization of td. The
largest value of tdcan be considered the time delay estimation
(TDE), tpeak. For real data, the BCC function can be calculated by
the inverse Fourier transform of the signal. The cross-correlation
function is preferred to be expressed in the normalized form, on
a−1to þ1scale, known as the correlation coefficient (Almeida
et al. 2014). The highest value of the correlation coefficient will
be gained when the two sensors are equidistant from the leak
source; however, practically one sensor is always nearer to the leak
source than the other (Gao et al. 2004b). Thus, only the relative
ratio between the two distance values, l1=l2, is important. Gao et al.
(2005) compared hydrophones with accelerometers and geophones
and found that hydrophones showed the lowest sensitivity toward
changes in the distance ratio as compared to others. Specifically,
for good correlation, the ratio of distance should satisfy 1=10 ≤
ðl1=l2Þ≤10.
For the GCC method, prefiltering is done on the input signals
either in time or frequency domains (Gao et al. 2009,2017;Hunaidi
et al. 2000). For the time domain, signals can be filtered before delay
calculation, and for the frequency domain, window or weighting
functions can be applied to the cross-spectral density (CSD) function
before the application of inverse Fourier transform. The weighting
functions used for the purpose are Roth impulse response, smoothed
coherence transform (SCOT), the Wiener, the phase transform
(PHAT), and the maximum likelihood (ML) estimators and help in
increasing the resolution of the cross-correlation function. The
GCC function Rg
s1s2ðtÞbetween two random signals, s1ðtÞand
s2ðtÞ, is given by
Rg
s1s2ðtÞ¼F−1½φgðωÞCs1s2ðωÞ ¼ 1
2πZ−∞
þ∞
φgðωÞCs1s2ðωÞeiωtlω
ð3Þ
where F−1½ = inverse Fourier transform; Cs1s2ðωÞ= cross-spectral
density; and φgðωÞ= frequency weighting function. For φgðωÞ¼1,
the GCC ¼BCC. Further details of the suitability of these methods
for different conditions can be seen in Gao et al. (2006,2009)
which establish the effect of pipe dynamics and reflective proper-
ties of the pipe material on the cross-correlation peaks. It was found
that the PHAT estimator gives the best results and can be termed as
an improved GCC method or the GCC-PHAT method. It prewhitens
the modulus of cross-spectrum and leaves the phase spectrum in-
formation only, from which the time delay can then be efficiently
calculated. However, the method does not take coherence between
two signals into account and assigns equal weights to all frequen-
cies irrespective of signal strength. Instead of improving the GCC
function, an alternate method for the TDE is the generalized phase
spectrum method (GPS).
The GPS method defines the best time-delay estimate as the one
for which the mean square error between the measured and esti-
mated phase of the CSD is minimized over a predefined frequency
bandwidth. Brennan et al. (2007) compared the GPS method with
the BCC and GCC-PHAT methods and found the time and fre-
quency domain analysis equivalent. Both hydrophones and accel-
erometers were used for demonstrating the GPS method. Most
of the signals detected from the hydrophone ranged from 10 to
120 Hz. The coherence between signals was generally better than
accelerometers. However, the phase spectrum revealed phase shifts
at 60 and 80 Hz. Such phase shifts can cause inaccurate time delays
and were attributed to hydrophone mounting resonances. Thus, the
frequency bandwidth was limited to 10–50 Hz for the hydrophone.
Almeida et al. (2014) presented the TDE using phase spectrum as
per Eq. (4)
tpeak ¼Pi
j¼1½Ds1s2ðωjÞθðωjÞωj
Pi
j¼1½Ds1s2ðωjÞω2
j
ð4Þ
Among the recent studies, Almeida et al. (2014) used this
method to explore the choice of acoustic sensors by experimenta-
tion on leak noise data from a test rig. It was found that in the case
of their experiment, all sensors proved efficient in detecting strong
leaks (high SNR). However, hydrophones were unable to detect
weak leaks (low SNR) because of their invasiveness and presence
of high background. This is in contrast to the findings of Gao et al.
(2005), who particularly recommended the use of a hydrophone sen-
sor for low SNR signal detection, as they are least affected by attenu-
ation. This indicates the limitation of the lack of standardization in
experiments. Furthermore, Almeida et al. (2015) highlighted that for
Eq. (1), estimation of the speed of wave propagation cin the pipe
needs to be accurate to find the location of the leak from time-delay
estimation. The study proposed an in situ measurement of cto con-
trol error in leak localization. Furthermore, to control the resonance
effects and extra peaks due to reflections, and achieve a good coher-
ence, the PHAT correlation estimator has been demonstrated (Gao
et al. 2017). Additionally, Brennan et al. (2019) examined the effect
of instrumentation issues such as clipping and quantization through
the application of signum function and random telegraph theory. For
the hydrophones, it was observed that clipping effects cause severe
distortions on amplitude, coherence, and phase angle above frequen-
cies >50 Hz. However, the clipping effect has a negligible effect on
the normalized cross-correlation and thus does not affect the TDE.
The studies give fundamental knowledge for hydrophone use in the
leak detection field. However, they only consider basic modeling and
signal processing for the improvement of time-delay estimation
and involve manual computations which are laborious and time-
consuming.
Innovation for Long-Term Leak Detection
and Localization
In Cluster 2, there are 22 articles, 17 of which are published in the
last three years. Most cited articles in Cluster 2, given in Tables 3
and 4, are the top articles with respect to TLS. This implies that
these publications can help identify the current state of research
directions in the area. After an in-depth review of articles, it is
Table 3. Most significant articles with respect to number of citations in
Cluster 2
Rank Publications Citations
1 Khulief et al. (2012)57
2 Casillas et al. (2013)46
3 Li et al. (2018)22
4 Butterfield et al. (2017)11
4 Brunner and Barbezat (2006)11
5 Gao et al. (2018)9
Table 4. Most significant articles with respect to TLS in Cluster 2
Rank Most significant articles Total link strength
1 Gao et al. (2018) 220
2 Li et al. (2018) 161
3 Cody et al. (2020b) 150
4 Butterfield et al. (2017) 146
4 Butterfield et al. (2018a) 107
5 Ma et al. (2019) 107
© ASCE 03121002-5 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

apparent that the research directions focus on: (1) improvements in
TDE methods, (2) application of advanced data analytics for auto-
mated leak detection and localization, (3) exploring innovative
technologies for long-term monitoring and early detection of a leak,
and (4) factors of complexity in the adopted experimental approach.
In Cluster 2, some studies have focused on presenting alternative
methods for the prefiltering requirement in the GCC method. The
most significant work in the cluster in terms of link strength given
in Table 5, Gao et al. (2018) introduced the differentiation process
(DIF) as an improved version of the GCC. Instead of adopting the
prewhitening methods of the GCC, the DIF method modifies the
pipe system characteristics by applying a higher-order frequency
weighting function φfðωÞ¼ωnbefore the cross-correlation. This
makes the pipe system act as a high-band pass filter, making avail-
able more information than what could be achieved through prewhit-
ening GCC methods and BCC. This reduces the effect of resonance
and ambient noise effect at low frequencies and allows for a more
reliable cross-correlation peak. The method works well with hydro-
phones but has limited application for accelerometer-based noise
correlators due to the diminishing effect on SNR. It is previously
established that TDE can be done using the phase spectrum
(Almeida et al. 2014). In related work, Ma et al. (2019) further de-
veloped a novel method for TDE. They developed a new frequency
response function (FRF) using only the phase information of the leak
Table 5. Methods for leak detection and localization/pinpointing application for hydrophones
Type Method Advantages Limitations References
TDE BCC 1. Computationally simplistic
approach
2. Less sensitivity for system
resonance giving more robust
results for time delay
Fundamental approach and pipe
conditions such as discontinuities
and background noise can lead to
errors; thus, quiet conditions are
required
Almeida et al. (2018) and
Gao et al. (2009)
GCC 1. Applicable to multiple leak
detection technologies
2. Prefiltering enhances the signal
resolution, suppresses
background noise, and sharpens
the correlation function to make
accurate detection of a leak
1. Manual preprocessing is tedious
2. Continuous data from two
sources is required for
ong-term monitoring, which
proves expensive
3. Variables and conditions need
to be known a priori
Gao et al. (2017)
GCC-PHAT
methods
The differentiation method can
control resonance effects at low
frequency for a more reliable TDE
in real networks
May reduce the SNR limiting
application to leak detection
through hydrophones
Gao et al. (2018)
GPS 1. No resolution problems like
in GCC
2. Only phase spectrum information
is required
3. Independent of sensor used for
detection
4. No prefiltering or prewhitening
efforts required in comparison
to GCC
5. The adaptive PHAT method
based uses the LMS algorithm
also effective for low SNR
Only considers the phase
information of signals so can ignore
the dispersion during leak noise
propagation in a pipe
Brennan et al. (2007) and
Ma et al. (2019)
Novelty detection
methods
Basic statistical
feature analysis
Automated leak detection enabled
for small-diameter HDPE service
pipes; in-pipe measurements can
also be made for pinpointing using
a swimming hydrophone
1. Reliance on basis statistical
features increases sensitivity for
baseline conditions
2. Applicability of identified
features may not apply to real
conditions
3. Large scale application difficult
due to tedious feature analysis
for each location
Khulief et al. (2012) and
Martini et al. (2017b)
Parametric 1. Potential for autonomous
long-term leak detection and
localization
2. Applicable for complex pipe
geometries
3. Can even detect small leaks
1. Multiple leaks and pipe backfill
conditions cause significant
deviations during real
application
2. Depends on the minimum
threshold selection
3. Historical data is required for
training sets
Cody et al. (2020b)
Nonparametric 1. Completely data-driven
2. Computationally efficient
3. Early leak detection for small
and difficult leaks
1. Limited to leak identification
only
2. Can only detect new leaks
Cody et al. (2020a)
© ASCE 03121002-6 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

signals and proposed an adaptive phase transform (ADPHAT) algo-
rithm based on it for TDE. Time delay can be estimated by taking the
inverse Fourier transform of the FRF, given by Eq. (5)as
hðnÞ¼F−1ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
H12ðωÞH21 ðωÞ
pð5Þ
It is worth mentioning that the method is an improved version of
the least mean square (LMS) method and can estimate time delay
even without information on spectral characteristics of the signals.
Currently, novelty detection methods have become appealing
for the application of long-term leak detection and localization.
These methods identify novel or anomalous events from a normal
set of data in any machine learning system using either statistical
approaches or neural network approaches (Markou and Singh
2003a,b). These methods are data-driven and only require the sig-
nal data for leak diagnostics. For fully supervised training, histori-
cal data for both leaks and the no-leak condition is required.
However, for semisupervised training, only the normal state data
is required to detect anomalies (Cody et al. 2018). As the historical
data sets for leak data might not be available, semisupervised ap-
proaches seem more attractive to leak detection practitioners (Cody
et al. 2020a). In this regard, Cody et al. (2018) used single-
spectrum analysis (SSA) and a one-class support vector machine
(OCSVM) for leak detection. This is a nonparametric approach
to effectively decompose signals into components showing prom-
ising ability to detect leaks in strong background noise. However,
the method does not apply to leak localization. Similarly, Har-
mouche and Narasimhan (2019) employed association rules
(AR) mining, a nonparametric unsupervised learning approach
for leak detection. In this study, a mean-shift clustering was used
for finding the ARs, based on which a leak index matrix is then
established. This leak index then serves as a reference model for
detecting any new leak in the system with high sensitivity. On the
contrary, Cody et al. (2020b) uses a linear prediction (LP) method
for leak detection and localization. This is a parametric signal
processing method assuming a Gaussian mixture model for nov-
elty detection from a baseline. The detected leak is localized using
cross-correlation.
One commonality between these studies is the use of feature
engineering during the preprocessing of hydroacoustic data. In fea-
ture extraction, domain knowledge is applied to extract relevant
features from a data set acting as an input for the algorithm or
analysis being used for leak detection (Zheng and Casari 2018).
Though effective for lab experiments, the variability of the features
over a real water distribution network (WDN) is not established.
Additionally, repeating this step for each new location can be tedi-
ous and complex. Cody et al. (2020a) avoided the use of feature
engineering through a novel approach applying autoencoders based
on deep learning. The technique uses spectrograms of hydroacous-
tic data for novelty detection. In the study, a convolution neural
network (CNN) is used with a variational autoencoder (VAN), with
which the spectrograms are used as training sets for normal data.
The approach requires neither prior training of data sets nor any
feature extraction for leak detection. This is the only study for
hydroacoustic data applying the neural network–based novel de-
tection. A 97% classification accuracy was achieved during test-
bed experiments under realistic noisy conditions. A unique fea-
ture of the experiment design of the study is the use of only a
single hydrophone installed at the base of the hydrant for data
collection.
Based on a detailed review, it is evident that various factors
affect signal propagation of leaks, determining the accuracy of
cross-correlation. These factors are either internal or external.
Internal factors are related to internal flow and pipe system
characteristics, including flow rate, line pressure, wave propaga-
tion speed, pipe size, leak size, pipe material, pipe geometry, pipe
condition (old or new), unknown discontinuities in the pipe
network, number of leaks in the pipe, and whether the leak is
old or new (Butterfield et al. 2018a;Gao et al. 2005,2009;
Martini et al. 2017a). External factors include the surrounding
factors such as soil pressure due to backfill in buried pipes, plac-
ing of sensors on the pipe, type of sensor, season, time of sound-
ing, traffic and other background noise, the experience of
engineers, and the conditions during experimentation (Butterfield
et al. 2018b;Hunaidi and Chu 1999;Zhao et al. 2020). There
are, however, very few existing studies that have explored such
attenuation factors or incorporated these effects into their calcu-
lations. In Cluster 2, various studies have focused their attention
on expanding the complexity of experimentation to include
such factors. For example, Khalifa et al. (2012) established some
important factors affecting acoustic wave propagation in pipe-
lines, such as the effect of sensor placing from the leak source,
the effect of changes in flow conditions such as line pressure,
and flow rate.
One impending challenge is the scale and scope of leak experi-
ments, as very few real network-based studies can be found in the
shortlisted publications (Bracken and Cain 2012;Butterfield et al.
2018a;Lechgar et al. 2016;Ma et al. 2019). In lab conditions,
standardization of experiments to study the impact of any single
factor is difficult. In real networks, there can be multiple leaks
in the system at the same time, and the pipe geometry can involve
bends, valves, fittings, discontinuities, varying diameters, and cor-
responding line pressure, depending on pipe functions and differ-
ences in pipe material within the network. In this regard, Butterfield
et al. (2018a) investigated the effect of different pipe materials on
the efficacy of sensors for cross-correlation of leak localization.
The experiments were done on a real WDN using artificially cre-
ated leaks and showed poor cross-correlation results for plastic
pipes using vibration. However, the use of hydrophones has been
encouraged to get better results.
In Cluster 2, in addition to such methodical and analytical im-
provements, technological innovations are also discussed. These
innovations are briefly explained in the section “Research Trends”
as a case of technological innovation for overcoming instrument
capability issues, where Khulief et al. (2012) designed a novel ex-
periment for leak detection using a free-swimming hydrophone.
Although the use of a free-swimming sensor allows the long-
distance survey of pipe network and greater reach to less approach-
able pipeline sections, the applicability has some limitations as
well. Past experiment scenarios based their findings on a compari-
son of the leak and no-leak states. So, in this case, data will need to
be collected in a leak-free pipe section as well, or correlation using
a multisensor system will be needed. Additionally, uninterrupted
access to GPS and continuous recording of sound data are depen-
dent on battery life. Design and access limitations also exist in
terms of navigating around sharp bends and narrow pipe diameters.
Apart from innovation in the method of deployment of hydro-
phones, alternatives to piezoelectric materials for fabrication are
also being explored for better precision, low cost, and smaller size
e.g., microelectromechanical system (MEMS)-based and fiber-
optic hydrophones. The application of hydrophones for metal pipes
and long-term monitoring is also a new endeavor due to its capabil-
ity of long-range detection. Recently, Guo et al. (2019) demon-
strated a long-term monitoring system for leakage in water pipes
by deploying four fiber-optic-based hydrophones on the pipe walls
of a real network of ductile cast iron pipes. The hydrophones mon-
itor the pipe condition in real time and trigger an alarm upon any
leak. The positioning accuracy of leak localization was 99.829%.
© ASCE 03121002-7 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

The error in estimation increases with the distance of the sensor
from the leak. Xu et al. (2019) used MEMS hydrophones for leak
detection and pinpointing. The sensors were installed both inside
and outside the pipes to study attenuation effects on correlation. It
was found that inside installation had a lower chance of false alarms
and the setup could detect both old and new leaks for flow rates as
low as 30 L=min. Although the feasibility of MEMS and fibre optic
hydrophone (FOH) hydrophones is demonstrated for leak detection
and pinpointing, the technologies are still in the testing stage.
Use of Hydrophones in Real Networks
In Cluster 3, nine articles deal with (1) leak detection and locali-
zation in service pipes; and (2) the implementation of hydrophones
in real networks. Among these, Martini et al. (2017b) have a TLS of
193 with 21 citations. The study has explored longitudinal leak de-
velopment in high-density polyethylene (HDPE) service pipes. The
leaks in this pipe are difficult to detect due to the low flow rate,
making the service leaks one of the significant components of pipe
losses. There are very few studies that deal with this pipe type. The
analytical approach adopted in the study is a statistics-based nov-
elty detection method. But the difference between this study and the
studies in Cluster 2 is that this study only adopts the basic statistical
feature analysis for leak detection. The parameter used for detection
is the standard deviation of raw signal for various condition
changes. It was observed that hydrophones and accelerometers per-
formed similarly on these pipes, with hydrophones being able to
detect leaks farther away than accelerometers. In a similar work,
Martini et al. (2015) presented a leak monitoring index (MI) and
compared the MI efficiency of both hydrophones and accelerometers
for leaks induced in small-diameter polyethylene pipes. It was ob-
served that hydrophones showed much better performance than ac-
celerometers for long-distance leaks. However, accelerometers are
still commonly much cheaper than hydrophones. In another related
work, Martini et al. (2018) further demonstrated the successful use of
the autocorrelation method for detecting and pinpointing water leaks
in service pipes using vibroacoustic measurements. Meanwhile,
Muntakim et al. (2017) and Lechgar et al. (2016)bothperformed
experiments on hydroacoustic data on real network cases in Canada
and Casablanca, respectively. Muntakim et al. (2017) focused on
leak identification using the coherence of two acoustic signatures
from sensors installed on the fire hydrants as part of a setup called
LeakFinder-ST. The leak is further localized using advanced com-
mercial correlation software. Lechgar et al. (2016), on the other hand,
presented a unique case study for demonstrating artificial intelligence
as a potential way for the optimal placing of hydrophones for fast and
efficient leak detection in real networks. Otherwise, full coverage of
the network using hydrophones is not financially viable as compared
to cheaper accelerometers. The method used the greedy algorithm
and SLOTS algorithms as inputs for the genetic algorithm for leak
detection. Such case studies serve as good examples for future dem-
onstration of developed methods on real networks.
Hydrophone Measurements and Pipe Flow Dynamics
Overall, there are very few studies that have focused on studying
the attenuation characteristics of a submerged plastic pipe. Thus,
Cluster 4 studies are unique in their theoretical contribution to
understanding the pipe and water interaction in buried plastic pipes.
In this regard, Muggleton and Brennan (2004) developed a theo-
retical model for wave attenuation for plastic pipes and investigated
its validity via laboratory experiments. As leak-generated acoustic
energy is buried, water pipes propagate at a low frequency, so it is
useful to study the pipe flow dynamics at frequencies less than
200 Hz (Muggleton et al. 2002). At lower frequencies than the ring
frequency, acoustic energy dissipates in different wave types. Out
of these, most of the leak energy is concentrated in the axisymmet-
ric (n¼0), fluid-borne wave (s¼1), the behavior of which is
studied in relevant studies in the cluster (Pinnington and Briscoe
1994). The axisymmetric, fluid-borne wave is denoted by the wave-
number n¼0,s¼1in the literature (Gao and Liu 2017). Using an
assembly of three hydrophones suspended along a centerline inside
an MDPE pipe, the attenuation effects were experimentally com-
pared with the theory. It was found that for low frequencies, the
wave propagation speed and attenuation pattern are almost the
same whether the pipe is suspended in air or submerged in water
or soil.
However, it should be noted that the soil type has an impact on
the attenuation effects. Sandy soil may have air pockets, creating
similarly high attenuation as occurring in in-air pipe. These relation-
ships can further judge the attenuation impacts on hydrophone-
collected leak signals and improve process efficiency. For example,
in Cluster 4, Gao and Liu (2017) have the highest TLS of 228 with
four citations. The study has a theoretically developed relationship
between internal pressure and wave propagation and demonstrated
using two accelerometers and one hydrophone. The relationship be-
tween internal pressure and radial displacement in the pipe wall for
the s¼1wave is given in Eq. (6)as
Ps1¼ω2ρϝ
kr
s1Joðkr
s1aÞWs1ð6Þ
where Ps1and Ws1= acoustic pressure and radial pipe displacement,
respectively; kr
s1=radialwavenumberofs¼1wave; Jo=Bessel
function of order zero; and a= radius of the pipe.
For any noise-free case, the ratio between the distances of two
sensors (l1=l2) from the leak should be less than about 10 (or
greater than 1/10) for pressure responses and less than 3 or greater
than 1/3 for acceleration responses. In this regard, Gao and Liu
(2017) practically demonstrated that the pressure signals from the
hydrophone can be detected at two locations between 10 and 93 m
from the leak source. For accelerometers, the range was 26–77 m
from the leak source. This implies that instead of the absolute
distances of the sensors from the leak source, the ratio of the
two distances is more meaningful while deploying sensors for data
collection. Further, Li et al. (2019) developed a theoretical model
for the effect of pipe wall thickness and radius on wave attenuation.
For a higher radius/thickness ratio, the radial vibration of the pipe
wall increases, creating a strong power dissipation to the surrounding
medium, causing higher wave attenuation. Further, using the find-
ings on asymmetric wave propagation, Sun et al. (2020)compared
new polyvinylidene fluoride (PVDF) wire sensors as a nonintrusive,
inexpensive, and easily deployable alternative to commercially avail-
able piezoelectric hydrophones for leak detection and localization.
Highlights, Research Gaps, and Future Directions
Although the scholarly works have been discussed in depth in the
section “Classification of Scholarly Works,”it is pertinent to high-
light some notable features of the research to guide practitioners
looking to use hydrophones as their sensor choice for leak detec-
tion. Such features are presented in this section and are shown in
Fig. 6, identifying the gaps and future direction of the research.
Method of Analysis
Many models have been identified in literature which can be
combined with other analytical methods and signal processing
© ASCE 03121002-8 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

techniques to identify and locate the leak. As presented in Table 5,
two main kinds of methods can be found: TDE and novelty detec-
tion methods. For TDE, various signal processing models have
been developed following the correlation principle, and very few
studies use the phase spectrum information to locate the leak.
Moreover, the adoption of an automated GPS approach called the
ADPHAT method is limited. But automated approaches with less
computational and preprocessing requirements can increase the
feasibility of hydrophones for long-term leak detection and locali-
zation. Conversely, novelty detection methods detect anomalies in a
normal system through fully supervised or semisupervised machine
learning approaches. But very few studies utilize statistical model-
ing and machine learning–based algorithms to analyze hydroacous-
tic data. Table 5indicates that such models are innovative and
require further work for efficient leak detection and precision pin-
pointing. Future studies can further improve the limitations of the
discussed methods as presented in Table 5.
Application to Real Networks
There are studies available that demonstrate the leak detection
methods on real networks (Cody et al. 2020a;Martini et al. 2018).
However, the authors observe two limitations: (1) most studies
demonstrate their developed methods on a lab or test-bed scale
and do not discuss the performance of their analysis methods on
real networks, and (2) the experiments presented for real networks
provide limited insight into the implementation issues and exper-
imental designs they followed for other practitioners to follow. In
real WDNs, the results might not be the same (Hamilton and
Charalambous 2020). In real networks, there can be multiple leaks
in the system at the same time, and the pipe geometry can involve
bends, valves, fittings, discontinuities, varying diameters, and corre-
sponding line pressure depending on pipe functions and differences
in pipe material within the network. In this regard, Butterfield et al.
(2018a) investigated the effect of different pipe materials of a real
WDN on the efficacy of sensors for cross-correlation of leak locali-
zation, and the results showed poor cross-correlation results for plas-
tic pipes using vibration, unlike expectation. Therefore, it is important
to apply the developed model to the WDN network.
The selection of leak detection techniques for real network test-
ing is not straightforward. Various factors such as life-cycle cost,
efficiency, and time duration required for leak detection are in-
volved in this decision-making. Additionally, every technology
has associated advantages and limitations. For example, leak noise
correlators have high accuracy but have high investment costs (Lai
et al. 2016). Hydrophones can locate very difficult leaks, but their
installation in the field is difficult due to access issues (Khulief et al.
2012). Vibroacoustic sensors such as accelerometers result in very
clear correlation results; however, they get affected by obstructions
and background noise (Brennan et al. 2019). Leak noise loggers
prove to very effective but report the problem of false alarms
(El-Abbasy et al. 2016). Hydrophones generally have a difficult
deployment method and are considered invasive if they are directly
inserted in water. On the contrary, accelerometers can be connected
to the pipelines using duct tape or magnets and are thus considered
noninvasive and convenient to use in the field. Accelerometers are
not considered very efficient for leak detection in plastic pipes or
conditions of low SNR, whereas hydrophones perform well for
detecting small leaks and leaks in plastic pipes.
To develop a best practice or optimized approach, hybrid solu-
tions for leak detection may be of interest. Such hybrid approaches
can be created in two ways: by mounting multiple sensors on the
same device or by combining different sensors in the system during
the data collection. Either way, compatible methods of analysis need
to be developed. For example, from Eq. (15), it should be noted that
as the pressure and wall displacement are directly proportional,
acceleration signals can be obtained by double-differentiating the
pressure signals from hydrophones (Gao et al. 2018). Thus, there is
potential to combine the accelerometers with the hydrophones for
leak detection and localization. Similarly, more comparative studies
are encouraged to develop a best-practices framework. Such a frame-
work can serve as a guiding tool for the selection of technology as
per conditions. For example, Hamilton and Charalambous (2020)
have developed a guide for practitioners for the selection of technol-
ogy based on the pipe diameter, flow conditions, and pipe material.
Based on such a framework, hybrid mechanisms for efficient long-
term leak detection monitoring can be enabled in real time.
Practical Implications for the Use of Hydrophones
As limited real-time testing has been reported, the situations where
hydrophones will be the best choice cannot be recommended with
Fig. 6. Summary of main findings.
© ASCE 03121002-9 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

certainty. Still, some reasonable deductions regarding their use can
be made through literature as follows.
Frequency Ranges, Hydrophone Type, and Pipe Material
Three hydrophone technologies have been reported in the literature:
piezoelectric, fiber optics, and MEMS. In earlier studies, mostly
piezoelectric hydrophones were used, which worked well for low-
frequency signal ranges within the threshold range of 5–50 Hz.
This made them effective for use in high-attenuation conditions
such as that of plastic pipes or large diameter mains (Gao et al.
2005;Hunaidi and Chu 1999). This can serve as a good rule of
thumb; however, Guo et al. (2019) recently demonstrated their po-
tential application for detecting high-frequency water leaks in metal
pipes through fabrication technology improvements. They used a
new custom fiber-optic hydrophone in their experiment. As fiber
optics is a new technology, it is relatively expensive compared to
other commercially available hydrophones. As an alternative, Xu
et al. (2019) and Phua et al. (2020) have recommended the use
of a MEMS-based hydrophone which is low-cost, low-power,
and smaller in size. It is easier to deploy than commercial piezo-
electric hydrophones and is compatible with IoT technology. This
implies that with certain innovations, the testing difficulties for the
use of hydrophones in long-term leak detection systems may be
overcome.
Detection, Localization, or Pinpointing
As shown in Fig. 4, most studies related to the use of hydrophones
deal with either leak detection or localization, or both. There are
very few studies dealing with the use of hydrophones for pinpoint-
ing a leak. The studies using a hydrophone for pinpointing use a
free-swimming hydrophone. Also, some of the methods presented
in Table 5are limited to application for leak detection only.
Further work on such machine learning–based methods for real
networks should be demonstrated for the detection of both old and
new leaks.
Experimentation Complexity and Variations
One of the main drawbacks of studies conducted on the subject is
the lack of standardization of experiments. Many designs for ex-
periments are available offering different flow and line pressure
conditions, pipe geometries, and hydrophone types that exist in
the literature. However, the consideration for pipe bends and dis-
continuities is rarely considered. The number of sensors that
should be used for specific conditions is also vague. From the lit-
erature, it seems that for locating small or difficult leaks, a single
swimming hydrophone can be used. For step surveys, the use of a
typical two-sensor assembly looks plausible. However, for real
scenarios such as villages or small towns where two hydrants are
not spaced within 100 m of each other, what strategy should be
adopted?
Additionally, whether the hydrophone can be fitted onto the
hydrant/valve or be installed in the pipe with help of a tether will
also affect the design. Such scenarios have not been considered
while designing the experiments. Moreover, there is no guideline
to choose the best design of the experiment for testing the use of
hydrophones for leak detection. For this purpose, further review of
experiments needs to be carried out to select appropriate conditions
for testing. It is observed that none of the studies related their re-
sults with the sensor specifications, sensitivity, or directional abil-
ities. To optimize the design, the impact of model selection needs to
be considered. Moreover, most of the experiments use simulated
leaks in the test rig for testing the hydrophones. This is beneficial,
but results may differ for real leaks. A summary of the main find-
ings of the sections “Classification of Scholarly Works”and
“Significant Research Directions”is presented in Fig. 6.
Comparison of Hydrophones with Other Sensors
As elaborated on in the section “Leak Detection and Localization
Using Hydrophones and Comparison with Accelerometers,”hydro-
phones have been currently compared with accelerometers, geo-
phones, pressure transducers, and PVDF wire sensors (Almeida
et al. 2014;Gao et al. 2005;Khalifa et al. 2010;Sun et al. 2020).
They have been reported to show either better or similar perfor-
mance to these sensors. Hydrophones offer a significant advantage
in being less prone to ambient noise and high attenuation due to
in-pipe measurement producing high-quality data, reducing the
chance of false alarms. Accelerometers and PVDF wire sensors,
although noninvasive, can face serious issues in the real environ-
ment due to slippage, pipe bulging, and ambient noise. Hydro-
phones are not prone to such issues and offer high sensitivity to
low-frequency data as compared to accelerometers and geophones.
Accurate prediction of leak location, however, relies on wave
speed. Wave speed estimate changes with temperature variations,
changing Young’s modulus of pipe material and making the season
during data collection relevant. The data collection for vibroacoustic
sensing devices is affected by ambient noise. This is a frequently
identified drawback and is overcome by taking data in low-usage
hours during the night or early morning. The process can be further
automated using a data logger and cloud system with the hydrophone
for efficient data transfer.
Conclusion
The current study conducts an in-depth review of hydrophone ap-
plications for water leak detection. Main research directions, gaps
in the literature, and future directions of work were identified to aid
practical implementation. Presently, the acoustical characteristics,
pipe flow dynamics, and attenuation properties for various pipe
materials and flow conditions for water leak detection have been
studied for hydrophones. Hydrophones perform well for high-
attenuation conditions and have a long range of leak detection.
Additionally, they collect data directly from water, thus showing less
sensitivity to background noise and reduced chances of false alarms.
Hydrophones, however, face limitations for sensing high-frequency
noise or detecting leaks in metal pipes, and should be combined with
other approaches to enhance efficiency. Future work requires the de-
velopment of hybrid approaches for multiple-sensing te chn ologies.
Further, the sophistication of experimental designs, real-world
case studies, and new fabrication technologies such as MEMS
and fiber optics need to be tested for performance improvement.
Data Availability Statement
All data supporting the findings of this study are available from the
corresponding author upon reasonable request.
Acknowledgments
The authors gratefully acknowledge the support from the Innova-
tion and Technology Fund [Innovation and Technology Support
Programme (ITSP)] under Grant No. ITS/067/19FP and the Water
Supplies Department of the Government of Hong Kong.
References
Almeida, F., M. Brennan, P. Joseph, S. Whitfield, S. Dray, and A.
Paschoalini. 2014. “On the acoustic filtering of the pipe and sensor
© ASCE 03121002-10 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

in a buried plastic water pipe and its effect on leak detection: An ex-
perimental investigation.”Sensors 14 (3): 5595–5610. https://doi.org/10
.3390/s140305595.
Almeida, F. C., M. J. Brennan, P. F. Joseph, S. Dray, S. Whitfield, and A. T.
Paschoalini. 2015. “Towards an in-situ measurement of wave velocity in
buried plastic water distribution pipes for the purposes of leak location.”
J. Sound Vib. 359 (Dec): 40–55. https://doi.org/10.1016/j.jsv.2015
.06.015.
Almeida, F. C. L., M. J. Brennan, P. F. Joseph, Y. Gao, and A. T.
Paschoalini. 2018. “The effects of resonances on time delay estimation
for water leak detection in plastic pipes.”J. Sound Vib. 420 (Apr):
315–329. https://doi.org/10.1016/j.jsv.2017.06.025.
Bracken, M., and B. Cain. 2012. “Transmission main and plastic pipe leak
detection using advanced correlation technology: Case studies.”In
Proc., Pipelines Conf. 2012, 147–157. Reston, VA: ASCE.
Brennan, M. J., Y. Gao, P. C. Ayala, F. C. L. Almeida, P. F. Joseph, and
A. T. Paschoalini. 2019. “Amplitude distortion of measured leak noise
signals caused by instrumentation: Effects on leak detection in water
pipes using the cross-correlation method.”J. Sound Vib. 461 (Nov):
114905. https://doi.org/10.1016/j.jsv.2019.114905.
Brennan, M. J., Y. Gao, and P. F. Joseph. 2007. “On the relationship be-
tween time and frequency domain methods in time delay estimation for
leak detection in water distribution pipes.”J. Sound Vib. 304 (1–2):
213–223. https://doi.org/10.1016/j.jsv.2007.02.023.
Brunner, A. J., and M. Barbezat. 2006. “Acoustic emission monitoring of
leaks in pipes for transport of liquid and gaseous media: A model ex-
periment.”Adv. Mater. Res. 13–14: 351–356. https://doi.org/10.4028
/www.scientific.net/AMR.13-14.351.
Butterfield, J. D., R. P. Collins, and S. B. Beck. 2018a. “Influence of pipe
material on the transmission of vibroacoustic leak signals in real com-
plex water distribution systems: Case study.”J. Pipeline Syst. Eng.
Pract. 9 (3): 05018003. https://doi.org/10.1061/(ASCE)PS.1949-1204
.0000321.
Butterfield, J. D., A. Krynkin, R. P. Collins, and S. B. M. Beck. 2017.
“Experimental investigation into vibro-acoustic emission signal
processing techniques to quantify leak flow rate in plastic water dis-
tribution pipes.”Appl. Acoust. 119 (Apr): 146–155. https://doi.org/10
.1016/j.apacoust.2017.01.002.
Butterfield, J. D., V. Meruane, R. P. Collins, G. Meyers, and S. B. M. Beck.
2018b. “Prediction of leak flow rate in plastic water distribution pipes
using vibro-acoustic measurements.”Struct. Health Monit. 17 (4):
959–970. https://doi.org/10.1177/1475921717723881.
Casillas, M. V., V. Puig, L. E. Garza-Castan´on, and A. Rosich. 2013.
“Optimal sensor placement for leak location in water distribution net-
works using genetic algorithms.”Sensors 13 (11): 14984–15005.
https://doi.org/10.3390/s131114984.
Cody, R., J. Harmouche, and S. Narasimhan. 2018. “Leak detection in
water distribution pipes using singular spectrum analysis.”Urban Water
J. 15 (7): 636–644. https://doi.org/10.1080/1573062X.2018.1532016.
Cody, R. A., P. Dey, and S. Narasimhan. 2020a. “Linear prediction for leak
detection in water distribution networks.”J. Pipeline Syst. Eng. Pract.
11 (1): 04019043. https://doi.org/10.1061/(ASCE)PS.1949-1204
.0000415.
Cody, R. A., B. A. Tolson, and J. Orchard. 2020b. “Detecting leaks in water
distribution pipes using a deep autoencoder and hydroacoustic spectro-
grams.”J. Comput. Civ. Eng. 34 (2): 04020001. https://doi.org/10.1061
/(ASCE)CP.1943-5487.0000881.
Datta, S., and S. Sarkar. 2016. “A review on different pipeline fault detec-
tion methods.”J. Loss Prev. Process Ind. 41 (May): 97–106. https://doi
.org/10.1016/j.jlp.2016.03.010.
El-Abbasy, M. S., F. Mosleh, A. Senouci, T. Zayed, and H. Al-Derham.
2016. “Locating leaks in water mains using noise loggers.”J. Infra-
struct. Syst. 22 (3): 04016012. https://doi.org/10.1061/(ASCE)IS
.1943-555X.0000305.
El-Zahab, S., and T. Zayed. 2019. “Leak detection in water distribution
networks: An introductory overview.”Smart Water 4(1):1–23. https://doi
.org/10.1186/s40713-019-0017-x.
Ferreira, F. A. 2018. “Mapping the field of arts-based management: Biblio-
graphic coupling and co-citation analyses.”J. Bus. Res. 85 (Apr):
348–357. https://doi.org/10.1016/j.jbusres.2017.03.026.
Gao, Y., M. J. Brennan, and P. Joseph. 2004a. “Detecting leaks in buried
plastic pipes using correlation techniques. Part 1: A model of the cor-
relation function of leak noise.”In Proc., 18th Int. Congress on Acous-
tics. Southampton, Hampshire: Univ. of Southampton.
Gao, Y., M. J. Brennan, and P. Joseph. 2004b. “Detecting leaks in buried
plastic pipes using correlation techniques. Part 2: On the selection of
acoustic/vibration sensors.”In Proc., 18th Int. Congress on Acoustics.
Southampton, Hampshire: Univ. of Southampton.
Gao, Y., M. J. Brennan, and P. F. Joseph. 2006. “A comparison of time
delay estimators for the detection of leak noise signals in plastic water
distribution pipes.”J. Sound Vib. 292 (3–5): 552–570. https://doi.org/10
.1016/j.jsv.2005.08.014.
Gao, Y., M. J. Brennan, and P. F. Joseph. 2009. “On the effects of reflec-
tions on time delay estimation for leak detection in buried plastic water
pipes.”J. Sound Vib. 325 (3): 649–663. https://doi.org/10.1016/j.jsv
.2009.03.037.
Gao, Y., M. J. Brennan, P. F. Joseph, J. M. Muggleton, and O. Hunaidi.
2005. “On the selection of acoustic/vibration sensors for leak detection
in plastic water pipes.”J. Sound Vib. 283 (3–5): 927–941. https://doi
.org/10.1016/j.jsv.2004.05.004.
Gao, Y., M. J. Brennan, Y. Y. Liu, F. C. L. Almeida, and P. F. Joseph. 2017.
“Improving the shape of the cross-correlation function for leak detec-
tion in a plastic water distribution pipe using acoustic signals.”Appl.
Acoust. 127 (Dec): 24–33. https://doi.org/10.1016/j.apacoust.2017
.05.033.
Gao, Y., and Y. Liu. 2017. “Theoretical and experimental investigation into
structural and fluid motions at low frequencies in water distribution
pipes.”Mech. Syst. Sig. Process. 90 (Jun): 126–140. https://doi.org/10
.1016/j.ymssp.2016.12.018.
Gao, Y., Y. Liu, Y. Ma, X. Cheng, and J. Yang. 2018. “Application of the
differentiation process into the correlation-based leak detection in urban
pipeline networks.”Mech. Syst. Sig. Process. 112 (Nov): 251–264.
https://doi.org/10.1016/j.ymssp.2018.04.036.
Guo, C., K. Shi, and X. Chu. 2019. “Experimental study on leakage mon-
itoring of pressurized water pipeline based on fiber optic hydrophone.”
Water Supply 19 (8): 2347–2358. https://doi.org/10.2166/ws.2019.116.
Hamilton, S., and B. Charalambous. 2020. Leak detection: Technology and
implementation. London: IWA Publishing.
Harmouche, J., and S. Narasimhan. 2019. “Long-term monitoring for leaks
in water distribution networks using association rules mining.”IEEE
Trans. Ind. Inf. 16 (1): 258–266. https://doi.org/10.1109/TII.2019
.2911064.
Hunaidi, O., W. Chu, A. Wang, and W. Guan. 2000. “Detecting leaks in
plastic pipes.”J. Am. Water Works Assn. 92 (2): 82–94. https://doi.org
/10.1002/j.1551-8833.2000.tb08819.x.
Hunaidi, O., and W. T. Chu. 1999. “Acoustical characteristics of leak sig-
nals in plastic water distribution pipes.”Appl. Acoust. 58 (3): 235–254.
https://doi.org/10.1016/S0003-682X(99)00013-4.
Khalifa, A. E., R. Ben-Mansour, K. Youcef-Toumi, and C. Choi. 2012.
“Characterization of in-pipe acoustic wave for water leak detection.”
In Proc., ASME International Mechanical Engineering Congress
and Exposition. New York: ASME.
Khalifa, A. E., D. M. Chatzigeorgiou, K. Youcef-Toumi, Y. A. Khulief, and
R. Ben-Mansour. 2010. “Quantifying acoustic and pressure sensing for
in-pipe leak detection.”In Proc., ASME Int. Mechanical Engineering
Congress and Exposition, 489–495. New York: ASME.
Khulief, Y. A., and A. Khalifa. 2013. On the in-pipe measurements of
acoustic signature of leaks in water pipelines. New York: ASME.
Khulief, Y. A., A. Khalifa, R. Ben Mansour, and M. A. Habib. 2012.
“Acoustic detection of leaks in water pipelines using measurements in-
side pipe.”J. Pipeline Syst. Eng. Pract. 3 (2): 47–54. https://doi.org/10
.1061/(ASCE)PS.1949-1204.0000089.
Lai, W. W., R. K. Chang, J. F. Sham, and K. Pang. 2016. “Perturbation
mapping of water leak in buried water pipes via laboratory validation
experiments with high-frequency ground penetrating radar (GPR).”
Tunnelling Underground Space Technol. 52 (Feb): 157–167. https://doi
.org/10.1016/j.tust.2015.10.017.
Lechgar, H., A. Mallouk, M. E. I. Malaainine, T. Nahhal, and H. Rhinane.
2016. “Artificial intelligence (AI) contribution to GIS in optimal posi-
tioning of hydrophone sensors using genetic algorithm (Case study:
© ASCE 03121002-11 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.

water network, Casablanca, Morocco).”In Proc., of the Mediterranean
Conf. on Information & Communication Technologies 2015,11–19.
Cham, Switzerland: Springer.
Li, R., H. Huang, K. Xin, and T. Tao. 2015. “A review of methods for burst/
leakage detection and location in water distribution systems.”Water Sci.
Technol. Water Supply 15 (3): 429–441. https://doi.org/10.2166/ws
.2014.131.
Li, S. Z., Y. J. Song, and G. Q. Zhou. 2018. “Leak detection of water dis-
tribution pipeline subject to failure of socket joint based on acoustic
emission and pattern recognition.”Measurement 115 (Feb): 39–44.
https://doi.org/10.1016/j.measurement.2017.10.021.
Li, Z., L. Jing, W. Wang, P. Lee, and R. Murch. 2019. “The influence of
pipeline thickness and radius on guided wave attenuation in water-filled
steel pipelines: Theoretical analysis and experimental measurement.”
J.Acoust.Soc.Am.145 (1): 361–371. https://doi.org/10.1121/1
.5087703.
Liemberger, R., and A. Wyatt. 2019. “Quantifying the global non-revenue
water problem.”Water Supply 19 (3): 831–837. https://doi.org/10.2166
/ws.2018.129.
Ma, Y., Y. Gao, X. Cui, M. J. Brennan, F. C. Almeida, and J. Yang. 2019.
“Adaptive phase transform method for pipeline leakage detection.”Sen-
sors 19 (2): 310. https://doi.org/10.3390/s19020310.
Markou, M., and S. Singh. 2003a. “Novelty detection: A review—Part 1:
Statistical approaches.”Signal Process. 83 (12): 2481–2497. https://doi
.org/10.1016/j.sigpro.2003.07.018.
Markou, M., and S. Singh. 2003b. “Novelty detection: A review—Part 2:
Neural network based approaches.”Signal Process. 83 (12): 2499–
2521. https://doi.org/10.1016/j.sigpro.2003.07.019.
Marmarokopos, K., D. Doukakis, G. Frantziskonis, and M. Avlonitis. 2018.
“Leak detection in plastic water supply pipes with a high signal-to-noise
ratio accelerometer.”Meas. Control 51 (1–2): 27–37. https://doi.org/10
.1177/0020294018758526.
Martini, A., A. Rivola, and M. Troncossi. 2018. “Autocorrelation analysis
of vibro-acoustic signals measured in a test field for water leak detection.”
Appl. Sci.-Basel 8 (12): 2450. https://doi.org/10.3390/app8122450.
Martini, A., M. Troncossi, and A. Rivola. 2015. “Automatic leak detection
in buried plastic pipes of water supply networks by means of vibration
measurements.”Shock Vib. 2015. https://doi.org/10.1155/2015/165304.
Martini, A., M. Troncossi, and A. Rivola. 2017a. “Leak detection in water-
filled small-diameter polyethylene pipes by means of acoustic emission
measurements.”Appl. Sci.-Basel 7 (1): 2. https://doi.org/10.3390
/app7010002.
Martini, A., M. Troncossi, and A. Rivola. 2017b. “Vibroacoustic measure-
ments for detecting water leaks in buried small-diameter plastic pipes.”
J. Pipeline Syst. Eng. Pract. 8 (4): 04017022. https://doi.org/10.1061
/(ASCE)PS.1949-1204.0000287.
Mourão, E., J. F. Pimentel, L. Murta, M. Kalinowski, E. Mendes, and
C. Wohlin. 2020. “On the performance of hybrid search strategies for
systematic literature reviews in software engineering.”Inf. Software
Technol. 123 (Jul): 106294. https://doi.org/10.1016/j.infsof.2020
.106294.
Muggleton, J., and M. Brennan. 2004. “Leak noise propagation and attenu-
ation in submerged plastic water pipes.”J. Sound Vib. 278 (3): 527–537.
https://doi.org/10.1016/j.jsv.2003.10.052.
Muggleton, J. M., M. J. Brennan, and R. J. Pinnington. 2002. “Wavenum-
ber prediction of waves in buried pipes for water leak detection.”
J. Sound Vib. 249 (5): 939–954. https://doi.org/10.1006/jsvi.2001.3881.
Muntakim, A. H., A. S. Dhar, and R. Dey. 2017. “Interpretation of acoustic
field data for leak detection in ductile iron and copper water-distribution
pipes.”J. Pipeline Syst. Eng. Pract. 8 (3): 05017001. https://doi.org/10
.1061/(ASCE)PS.1949-1204.0000257.
Patrício, L. D., and J. J. Ferreira. 2020. “Blockchain security research:
Theorizing through bibliographic-coupling analysis.”J. Adv. Manage.
Res. 18 (1): 1–35. https://doi.org/10.1108/JAMR-04-2020-0051.
Phua, W. K., S. M. Rabeek, B. Han, E. Njihof, T. T. Huang, K. T. C. Chai,
J. H. H. Yeo, and S. T. Lim. 2020. “AIN-based MEMS (micro-electro-
mechanical system) hydrophone sensors for IoT water leakage detection
system.”Water 12 (11): 2966. https://doi.org/10.3390/w12112966.
Pinnington, R., and A. Briscoe. 1994. “Externally applied sensor for ax-
isymmetric waves in a fluid filled pipe.”J. Sound Vib. 173 (4):
503–516. https://doi.org/10.1006/jsvi.1994.1243.
Puust, R., Z. Kapelan, D. Savic, and T. Koppel. 2010. “A review of methods
for leakage management in pipe networks.”Urban Water J. 7 (1):
25–45. https://doi.org/10.1080/15730621003610878.
Sun, P., Y. Gao, B. Jin, and M. J. Brennan. 2020. “Use of PVDF wire sen-
sors for leakage localization in a fluid-filled pipe.”Sensors 20 (3): 692.
https://doi.org/10.3390/s20030692.
Xu, J. H., K. T. C. Chai, G. Q. Wu, B. B. Han, E. L. C. Wai, W. Li, J. Yeo,
E. Nijhof, and Y. D. Gu. 2019. “Low-cost, tiny-sized MEMS hydro-
phone sensor for water pipeline leak detection.”IEEE Trans. Ind. Elec-
tron. 66 (8): 6374–6382. https://doi.org/10.1109/TIE.2018.2874583.
Zaman, D., M. K. Tiwari, A. K. Gupta, and D. Sen. 2020. “Areviewof
leakage detection strategies for pressurised pipeline in steady-state.”
Eng. Fail. Anal. 109 (Jan): 104264. https://doi.org/10.1016/j.engfailanal
.2019.104264.
Zhao, M., C. Zhang, H. Liu, G. Fu, and Y. Wang. 2020. “Optimal sensor
placement for pipe burst detection in water distribution systems using
cost-benefit analysis.”J. Hydroinf. 22 (3): 606–618. https://doi.org/10
.2166/hydro.2020.158.
Zheng, A., and A. Casari. 2018. Feature engineering for machine learning:
Principles and techniques for data scientists. San Francisco: O’Reilly
Media.
© ASCE 03121002-12 J. Pipeline Syst. Eng. Pract.
J. Pipeline Syst. Eng. Pract., 2021, 12(4): 03121002
Downloaded from ascelibrary.org by Beenish Bakhtawar on 07/24/21. Copyright ASCE. For personal use only; all rights reserved.