6374 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 66, NO.

8, AUGUST 2019

Low-Cost, Tiny-Sized MEMS Hydrophone

Sensor for Water Pipeline Leak Detection
Jinghui Xu, Kevin Tshun-Chuan Chai, Member, IEEE, Guoqiang Wu , Beibei Han, Eva Leong-Ching Wai,
Wei Li, Jason Yeo , Edwin Nijhof, and Yuandong Gu

Abstract—In this paper, we present an experimental in- I. INTRODUCTION

vestigation of a water pipeline leak detection system based
UFFICIENT water supply is crucial for sustainable eco-
on a low-cost, tiny-sized hydrophone sensor fabricated us-
ing the microelectromechanical system (MEMS) technolo-
gies. A 10 × 10 element arrayed MEMS hydrophone device
S nomic activities and our daily lives. A large portion of water
is lost during transportation and distribution, commonly known
with chip size of 3.5 × 3.5 mm2 was used in the experiment. as nonrevenue water (NRW). It is reported that the NRWs of
The hydrophone device is packaged with a customized on-
board preamplification circuit using an acoustic transparent most of the large cities in the developing countries and a lot of
material. The overall package size of the MEMS hydrophone old cities in the developed countries are above 30% in 2011; it
is Φ1.2 × 2.5 cm. The packaged MEMS hydrophone achieves is even higher than 50% in some cities, such as Jakarta, Delhi,
an acoustic sensitivity of −180 dB (re: 1 V/µPa), a bandwidth Adana, etc. [1]–[4]. Although the NRW indicator has decreased
from 10 Hz √to 8 kHz, and a noise resolution of around 60 dB in the past five years since the leak detection systems are gradu-
(re: 1 µPa/ Hz) at 1 kHz. A section of ductile iron water ally adopted, the NRW indicators are still above 30%. In 2016,
pipeline with an internal diameter of 10 cm, wall thickness
of 0.73 cm, and length of 30 m is constructed as the test the World Bank reported that roughly 45 million m3 of water are
bed for the water leak detection. Two different leak sizes lost daily in developing countries, an amount of which is suffi-
with leak flow rates of about 30 and 180 L/min are designed cient to supply 200 million people or worth an economic value
along the pipe, which is pressurized at 3.2 bar. Analysis of over USD 3 billion/year [5]. What exacerbates the problem
of the transient signals and spectrograms shows that the is that the countries with serious water loss are usually those
MEMS hydrophone can capture the key acoustic informa-
tion of the water leak, i.e., identifying the leak and locating which lack water resource. The World Bank also puts the global
the leak position. The measurement results demonstrate estimate of physical water losses at 32 billion m3 each year,
the feasibility to construct an affordable, highly efficient, twice of that occurring in developing countries.
real-time, and permanent in-pipe pipeline health monitoring The NRW is attributed to a variety of sources, including me-
network based on the MEMS hydrophones due to their high tering errors, accounting errors, water theft, and pipeline leak-
performance, low cost, and tiny size.
age. The water loss due to pipeline leak is as high as 30% of
Index Terms—Aluminum Nitride (AlN), hydrophone, the total flow, which is the main component of the water loss
leak detection, mirco-electro-mechanical systems (MEMS), [6]–[9]. Water pipelines are mainly classified into two cate-
pipeline, ultrasonic sensor.
gories, namely distribution pipelines and transmission pipelines.
The leak in the transmission pipeline is usually caused by the
Manuscript received October 25, 2017; revised August 30, 2018; ac- cracks. This kind of leak can be found immediately since the
cepted September 18, 2018. Date of publication October 12, 2018; date
of current version March 29, 2019. This work is part of MEMS Technology cracks will suddenly induce the large variations of the water
commercialization project supported by Exploit Technologies Pte. Ltd. supply pressures, which can be quickly alarmed by the pres-
(ETPL), and JCS Venture Lab, the corporate incubator of JCS Group, sure meters. It is reported that more than 80% of the water leak
Singapore. (Jinghui Xu, Kevin Tshun-Chuan Chai, and Guoqiang Wu
contributed equally to this work.) (Corresponding author: Yuandong Gu.) through pipeline is from the distribution pipeline. Therefore,
J. Xu is with the Institute of Microelectronics, Agency for Science, Tech- efficient pipeline leak detection is the only viable long-term so-
nology and Research (A*STAR) Singapore 138634, and also with also lution, and over the last two decades, it has been widely studied.
with MEMSound Pte. Ltd., Singapore 823270 (e-mail:, xuj@ime.a-star.
edu.sg). There are two systematic leakage-control programs [9]–[14].
K. T.-C. Chai, G. Wu, B. Han, E. L.-C. Wai, and Y. Gu are with the One is water audits and the other is leak-detection surveys. Al-
Institute of Microelectronics, Agency for Science, Technology and Re- though water audit gives a good indication of the amount of
search (A*STAR), Singapore 138634 (e-mail:,chaitc@ime.a-star.edu.sg;
wugq@ime.a-star.edu.sg; hanb@ime.a-star.edu.sg; wailc@ime.a-star. water losses, it provides no information about the locations of
edu.sg; guyd@ime.a-star.edu.sg). these losses. Leak-detection surveys are able to determine the
W. Li is with the Exploit Technologies Pte. Ltd. (ETPL), Agency for exact location of the leaks using acoustic or nonacoustic tech-
Science, Technology and Research (A*STAR), Singapore 138632, and
also with JCS Group, Singapore 738068 (e-mail:, Li_Wei@etpl.sg). niques. Listening rods, aquaphones, and ground microphones
J. Yeo and E. Nijhof are with the JCS Group, Singapore 738068 are used to pinpoint the leak by listening on all accessible con-
(e-mail:, jason@jcsauto.com.sg; edwin@jcstech.com.sg). tact points with the distribution system. As another option, sus-
Color versions of one or more of the figures in this paper are available
online at http://ieeexplore.ieee.org. pected pipeline leaks can be pinpointed automatically by us-
Digital Object Identifier 10.1109/TIE.2018.2874583 ing modern acoustic correlators. In this method, the suspected

0278-0046 © 2018 IEEE. Personal use is permitted, but republication/redistribution requires IEEE permission.
See http://www.ieee.org/publications standards/publications/rights/index.html for more information.

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
XU et al.: LOW-COST, TINY-SIZED MEMS HYDROPHONE SENSOR FOR WATER PIPELINE LEAK DETECTION 6375

leak is detected by two hydrophones, and the time lag between

the received acoustic signals acquired by hydrophones is used
to identify location of the leak. Besides the above two acous-
tic detection approaches, several nonacoustic techniques, such
as pressure transient, tracer gas, infrared imaging, and satellite
scanning, could be also used to detect the pipeline leaks [9], [10].
However, these nonacoustic techniques are still very limited and
not as effective as the well-established acoustic approaches.
Among the leak detection techniques, the acoustic correlator
method has been increasingly used because it is more efficient Fig. 1. Fabricated Mo-AlN-Mo piezoelectric stack. (a) SEM image
and accurate than the listening devices. The acoustic correlator showing the excellent crystal orientation. (b) Measured XRD texture of
the AlN film with FWHM value of 1.39°.
approach requires two hydrophones to measure the acoustic
signals. However, the existing piezoceramic-based hydrophones
suffer from large size, high cost, and high power consumption.
Recently, permanent acoustic sensor networks are proposed
for real-time pipeline health monitoring [15]–[17]. The net-
works required a large number of hydrophone sensors to be
installed inside the pipeline for reducing the influence the ex-
ternal environmental noise. Therefore, there is a strong need
for a low cost and compact size hydrophone sensor, in order
to construct an affordable sensor network for the real-time leak
detections of the pipelines. In this work, a feasibility study of
leak detection is performed by using the microelectromechani- Fig. 2. Fabricated MEMS hydrophone device. (a) 10 × 10 device with
cal system (MEMS) hydrophones, which feature compact size, chip size 3.5 × 3.5 mm2 . (b) Zoomed in view showing the electrical
low cost, and small power consumption. The reported MEMS connection.
hydrophones exhibit great potential applications in the pipeline
leak detection networks.
This paper presents an experimental investigation of a wa-
ter pipeline leak detection, for the first time, using a low-cost,
tiny-sized hydrophone sensor based on MEMS technology. The
fabrication, packaging, and characterization of the MEMS hy-
drophone sensor are described in Section II. The experiment
setup for the pipeline leak detection is illustrated in Section III.
The experiment results and detailed analysis are reported in
Section IV. Section V reported the evaluation of reliability and
long-term stability of the reported MEMS hydrophones, and
Fig. 3. Schematic diagram of a voltage-mode amplification circuit.
finally, Section VI concludes the paper.

is about 3.5 × 3.5 mm2 , which is much smaller than the

II. MEMS HYDROPHONE
traditional piezoceramic hydrophone counterparts by one order
The reported MEMS hydrophone sensor is fabricated based of magnitude [21].
on an 8-in CMOS-compatible piezoelectric integration plat- A customized signal conditioning and amplification read-out
form. The hydrophone consists of a cavity-silicon-on-insulator circuit is designed to pick up the electrical charges generated
(CSOI) wafer with a sandwiched piezoelectric layer of by the MEMS hydrophone during operation. Fig. 3 shows the
Molybdenum-Aluminum Nitride-Molybdenum (Mo-AlN-Mo) schematic diagram of a voltage-mode amplification circuit used
on top of it [18]. A sandwiched piezoelectric is used to sense the in our work. The circuit is designed to have a gain of 40 dB and
acoustic signals caused by the water leak in the pipeline [19]. features a flat response from 10 Hz to 10 kHz. The voltage output
Fig. 1(a) shows the cross-sectional view of the fabricated (Vo ), lower cutoff frequency fL and upper cutoff frequency fH
piezoelectric Mo-AlN-Mo stack. The sputtered AlN thin film are given by the following [22]:
is fully c-axis oriented with a full width at half maximum  
Vdd Q0 Rf
(FWHM) value of 1.39°, which is obtained from the X-ray V0 = + × 1+
diffraction (XRD) rocking curve, as shown in Fig. 1(b) [20]. 2 C0 + Cp Rg
Fig. 2(a) shows a scanning electron microscopy (SEM) im- R0 + Rb
age of the fabricated 10 × 10 element device, and Fig. 2(b) fL =
2π(R0 × Rb )(C0 + Cp )
shows the zoomed in view of each element, which clearly
1
indicates the membrane structure and electrical connections. fH = (1)
The dimension of the fabricated 10 × 10 hydrophone chip 2πRf Cf

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
6376 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 66, NO. 8, AUGUST 2019

Fig. 4. Hydrophone assembly and packaging. (a) MEMS hydrophone

chip is attached on a PCB. The MEMS chip and bonded wires are pro-
tected using an acoustically transparent polyurethane material. (b) PCB
with MEMS hydrophone is potted into a stainless steel shell using the
same polyurethane material.

Fig. 5. Frequency response of the resulting voltage noise density of a

where typical MEMS hydrophone.
Q0 = Charge generated by the MEMS hydrophone.
C0 = Initial capacitance of the MEMS hydrophone.
R0 = Device dc resistance.
Cp = Parasitic capacitances.
Rf = Feedback resistance.
Cf = Feedback capacitance.
Rg = Gain resistance.
Rb = DC filter resistance.
Vdd = Supply dc voltage.
It is reported that the sound signal emitted by pipeline leak
contains a very low frequency (around 10 Hz) component [23],
[24]. Therefore, the bandwidth of the circuit must be extended to
a low enough frequency range in order to capture the leak signal
efficiently. To achieve this objective, R0 , Rb , C0 , Cp should
be as large as possible according to (1). The dc resistance R0
usually is huge (>10 GΩ), which can be ignored and assumed
to be an open circuit. The parasitic capacitance Cp cannot be too
Fig. 6. Comparison of measured and theoretical equivalent pressure
high since a large Cp will decrease the output signal. A large dc noise density of the MEMS hydrophone.
resistor Rb can be purchased off the shelf. Therefore, to achieve
a low cutoff frequency, the initial capacitance of hydrophone C0
should be as large as possible. A 10 × 10 arrayed hydrophone the detectable acoustic pressure. The main noise sources of the
device with C0 as large as 340 pF is chosen in this work. MEMS hydrophone in our work are: tanδ noise or the dielectric
Fig. 4 shows an assembled MEMS hydrophone chip on loss, input noise voltage of the preamplifier, input noise cur-
the printed circuit board (PCB), where the hydrophone chip rent of the preamplifier, and the Johnson (thermal) noise. The
and bonded wires are protected by an acoustically transparent frequency response of the resulting voltage noise density of a
polyurethane film. The assembled PCB is then potted into a typical MEMS hydrophone is shown in Fig. 5. Johnson noise of
stainless steel shell using the same polyurethane material. The the high-megohm resistor dominates at a large frequency range
final packaged hydrophone is shown in Fig. 4(b), which has a (up to 400 Hz). At medium frequencies (400 Hz to 1 kHz),
size of Φ1.2 × 2.5 cm. It should be noted that a commercial tanδ noise, voltage noise, and Johnson noise dominate the re-
steel shell container is used for testing purpose. The packaged sultant noise density. At high frequencies (>1 kHz), the voltage
hydrophone size can be further reduced significantly by using noise of the preamplifier specifies the resultant voltage noise
a customized container and an application-specific integrated density.
circuit (ASIC) instead of the PCB. Fig. 6 shows the comparison of measured and theoreti-
The packaged MEMS hydrophone is characterized with a cal equivalent acoustic pressure noise density of the MEMS
standard anechoic water pool in a national lab using the Brüel hydrophone. The measured equivalent pressure noise densi-
& Kjær Type 8103 hydrophone as the reference. The fre- ties are in good agreement with the theoretical values. The
quency response measurement results show that the packaged MEMS hydrophone √ shows an output noise resolution about
MEMS hydrophone has a very flat response with a sensitivity of 60 dB (re: 1 μPa/ Hz) at 1 kHz. A comparison of the per-
−180 dB ± 1 dB (re: 1 V/μPa) over the bandwidth of 10 Hz formance of the developed MEMS hydrophone with the state-
to 8 kHz. The noise sources of the MEMS hydrophone limit of-art commercially available hydrophones is shown in Table I.

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
XU et al.: LOW-COST, TINY-SIZED MEMS HYDROPHONE SENSOR FOR WATER PIPELINE LEAK DETECTION 6377

TABLE I
PERFORMANCE COMPARISON OF THE DEVELOPED MEMS HYDROPHONE WITH THE COMMERCIALLY AVAILABLE HYDROPHONES

Fig. 7. Schematic diagram of the pipeline and measurement setup.

There are two valves for water leak mimic with flow rate of 30 and
180 L/min, respectively. Two packaged MEMS hydrophones are installed
in the pipeline using the customized adapters. The sound signals de-
tected by the hydrophones are collected by a laptop for signal analysis
and processing.

The reported MEMS hydrophone is shown to be a competi-

tive high-performance device compared with the commercial
hydrophones.

III. EXPERIMENT SETUP

Fig. 8. Assembled pipeline leak measurement setup. (a) Assembled
It is reported that the water leakage is mainly from the dis- pipeline water leak measurement system. (b) Inlet of the pipeline with
tribution pipeline instead of the transmission pipeline. There- a pressure meter and a water pump. (c) Customized adapter with the
fore, in this paper, the experimental study is conducted with a packaged MEMS hydrophone. (d) Outlet of the pipeline for water cycling.
purpose-built pipeline, which is with the same dimension as a
typical neighborhood distribution pipeline with internal diam-
eter of 10 cm and wall thickness of 0.73 cm. A 30-m long, the experiment, these mimic leak events are captured by the
ductile iron pipe is constructed based on the design schematic MEMS hydrophone and the signals are transmitted to a laptop
as shown in Fig. 7. The pipeline is divided into five sections. for subsequent signal analysis and processing. Fig. 8 shows
The locations of the mimic leak holes and hydrophone instal- the assembled pipeline leak detection testing bed installed in a
lation holes (adapters) are shown in Fig. 7. The leak holes and warehouse.
hydrophone installation holes can be exchanged for different
testing purposes. IV. RESULTS AND DISCUSSIONS
These sections are connected using the customized bolts and The MEMS hydrophones are installed inside the pipeline,
nuts. A mechanical pump is installed in the inlet which could as shown in Fig. 7. During the experiment, the water supply
pump the water and make the supply pressure up to 4 bar. This pressure is kept at 3.2 bar. The small leak (valve 1) and big
is consistent with the real municipal water distribution pressure leak (valve 2) are opened and closed based on the following
of most of the cities globally. A pressure meter is added in the sequence: a) open valve 1; b) open valve 2; c) close valve 2; d)
loop to monitor the real-time water pressure. The flow rate of close valve 1; e) open valve 2; f) open valve 1; g) close valve 1;
the small and big leaks are maintained at 30 and 180 L/min, and h) close valve 2.
respectively, while the water supply pressure is maintained at
3.2 bar during the testing. The packaged MEMS hydrophones
A. Water Leak Detection
are installed into the pipeline using the customized adapters.
In an event of leak, a friction noise will be generated and In order to obtain a stable result of each sequence, the two
propagates along the water column and the pipeline. During adjacent steps are separated in the time domain. After getting a

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
6378 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 66, NO. 8, AUGUST 2019

Fig. 9. Time-domain measurement results of the water pipeline leak system using the leak opening and closing sequence (a)–(h).

Fig. 10. Signal historical spectrogram of the MEMS hydrophones located at (a) 15 m, and (b) 25 m, respectively. Color in red indicates the highest
PSD and the color in blue indicates the lowest PSD.

Fig. 11. Comparison of the output voltages from the MEMS hydrophone located at 15 m using the inside and outside detection approaches.
It shows that these two approaches have very similar signal profile. While there is a about 30 dB signal attenuation using the outside detection
approach.

stable result causing by the current sequence, the next sequence Color in red indicates the highest signal power spectral density
is triggered. The overall water leak measurement takes about 5 (PSD) and the color in blue indicates the lowest PSD in the
min and the captured time-domain signals are shown in Fig. 9, spectrogram.
where the environmental vibration noise is removed by signal Based on the measurement results, as shown in Figs. 9
processing. Fig. 10 shows the spectrogram of the result of Fig. 9. and 10, the following may be concluded.

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
XU et al.: LOW-COST, TINY-SIZED MEMS HYDROPHONE SENSOR FOR WATER PIPELINE LEAK DETECTION 6379

1) A transient acoustic pulse signal is detected each time a TABLE II

ACOUSTIC PROPERTIES OF MATERIALS IN USE
mimic leak occurs.
2) Both the small and big leaks can be detected by the
MEMS hydrophones at different locations. The output
voltage of the MEMS hydrophone at 15 m is about two
times larger than that of the MEMS hydrophone at 25 m.
This is to say, there is roughly a 0.6 dB/m transmission
loss when assuming the loss is linear with the propagated
distance.
3) The output voltage of the MEMS hydrophone at 15 m is
±300 mV when only valve 2 is opened. It indicates that
about 300 Pa acoustic pressure is emitted at the location
which is 10 m away from the water leak.
4) Both the existing leak and new leak can be detected. As
shown in Fig. 9, the initiation of a leak is received by both
MEMS hydrophones as a pulse signal. This pulse signal
can be used as a indicator of a new leak. As the leak
continues, the MEMS hydrophones continuously receive
the emitted acoustic signals. It illustrated that the existing
leak can also be detected by the MEMS hydrophones.
Both the existing and new leak detections are important,
because both of these leak identifications are key missions Fig. 12. Correlation results of the measured output signals from the
in the pipeline leak detection. two MEMS hydrophones. It shows that the two signals have very similar
profile while there is time delay between the two signals.
5) The spectrograms of the signals indicate that the fre-
quency of the acoustic signals emitted by the leak can be
divided into three subbands. First, the frequency range surements with the MEMS hydrophones installed outside the
up to 200 Hz corresponds to the longitudinal resonance pipeline are conducted.
frequency of the pipe, since longitudinal resonance of For the outside detection approach, the MEMS hydrophones
the pipe always exists regardless of whether a leakage is are adhered to the pipeline surface tightly. The opening and
present. Second, the pipeline leak signals mainly concen- closing of valves followed the same sequence as that in the
trate in the frequency range of 200 Hz–2 kHz. Finally, the previous inside detection approach. The measurement results
signals with frequency above 2 kHz might be caused by are recorded and are compared with the results of the inside
the ambient noises, such as vehicular motions and run- detection approach. Fig. 11 shows the comparison of the mea-
ning machinery. The power of the signals beyond 2 kHz surement results from the MEMS hydrophone located at 15 m
will decrease rapidly. between the inside and outside detection approaches.
As a stress-sensitive device, the hydrophone responds to all It can be observed that the signals captured using the two de-
the sound signals, including the leak and external environmental tection approaches show excellent correlation. The signal profile
noises. Therefore, it is impossible for hydrophone to decouple using the outside detection approach is almost the same as that
the leak signals and the external noises with similar frequen- using the inside detection approach. Only the magnitudes are
cies. However, the leak signals have featured characteristics in significantly different. There is an around 30-dB signal atten-
frequency domain, which could be analyzed with signal process- uation using the outside detection approach. One of the main
ing schemes. Therefore, signal processing or machine learning sources of this large attenuation is the acoustic impedance mis-
based artificial intelligence (AI) could be an effective way to match at the interface between the pipeline and the MEMS
analyze the receiving signals from the hydrophone and identify hydrophone. Hence, only a small portion of acoustic power is
the leak signals. transmitted from the pipeline to the MEMS hydrophone attached
on the surface of the pipeline. Table II lists the material proper-
B. Comparison of the Inside and Outside Detection ties of the ductile iron used for the pipeline and the polyurethane
Approaches covered on the hydrophone chip. Assuming that the incident
sound signal is normal to the interface, the transmitted sound
To implement the leak detection, the installation of the MEMS pressure coefficient Tp can be calculated as follows [29]:
hydrophones outside the pipeline is preferred, since it is much
easier to install, repair, or maintain the detection systems from 2Z2 2ρ2 V2
the outside than inside. However, the iron pipe may degrade Tp = = (2)
Z1 + Z2 ρ1 V1 + ρ2 V2
the acoustic signal and the environmental noise may also de-
grade the accuracy of the leak detections. In order to investigate where Z1 and Z2 , ρ1 and ρ2 , and V1 and V2 are the acoustic
the feasibility of the water leak detection based on the reported impedances, mass densities, and sound speeds of medium 1 and
MEMS hydrophone using the outside detection approach, mea- 2, respectively.

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
6380 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 66, NO. 8, AUGUST 2019

Fig. 13. Comparison of the output voltages from the MEMS hydrophone and the commercial hydrophone. Both of the hydrophone sensors are
located 15 m away from the leak and using the inside detection approach. The two hydrophones have very similar signal profile.

Substituting the related material properties into (2), we ob- material, dimension, and the amount of the air bubbles inside
tain Tp ≈ 0.14. In other words, there is a direct 17-dB signal the water [29].
attenuation when the sound signal propagate from ductile iron In this work, both of the MEMS hydrophones are located at
to the MEMS hydrophone sensor. The remaining 13-dB signal the same side of the leak position. Therefore, this leak detec-
attenuation may be caused by the transmission loss since the tion configuration cannot be used to ascertain the leak location.
sound propagation from inside to outside the pipe is not an ideal However, the experiment clearly demonstrates the signal de-
plane wave. lay, hence proving that leak positioning using the report MEMS
Different sensor installation positions away from leak posi- hydrophones is feasible.
tion are investigated and they show similar signal attenuation.
It should be noted that the length of each section of the real
water distribution pipeline is about 100 m. Assuming a MEMS D. Comparison With the Commercial Hydrophone
hydrophone is installed at each junction, the maximum trans- In our work, a commercial piezoceramic hydrophone H2a,
mission loss of the inside and outside detection approaches will from Aquarian Audio & Scientific, is used as the reference hy-
be about 60 and 90 dB, respectively. This huge loss will severely drophone. This commercial hydrophone is currently used for
affect the signal fidelity and cause potential false alarms. There- pipeline leak detection by some industry players. The reference
fore, the inside detection approach is better in terms of the signal hydrophone is installed 15 m away from the leak, together with
fidelity. the MEMS hydrophone. Fig. 13 shows the comparison of the
output voltages from the MEMS hydrophone and the reference
C. Leak Positioning Evaluation hydrophone. Both the reference hydrophone and the MEMS
hydrophone capture very similar signal profiles, but different
Besides identifying the leak, another key task of pipeline signal magnitudes due to different acoustic sensitivities. The
leak detection system is to ascertain the leak location. This is experimental result verifies the feasibility of the MEMS hy-
usually realized by using multisensor correlation [30]. In this drophone for the pipeline leak detection, which features smaller
paper, we performed the leak positioning testing using the two size, lower cost, and decent performance than the commercial
deployed MEMS hydrophones as shown in Fig. 7. The output piezoceramic hydrophone.
signals from the two MEMS hydrophones are synchronized to
the same data acquisition software. From the correlation results
of the measured data, as shown in Fig. 12, it is observed that V. EVALUATION OF RELIABILITY AND LONG-TERM STABILITY
both MEMS hydrophones capture similar signal profile while
The packaged MEMS hydrophones are evaluated by the high-
there is a time delay between the two signals. This is because
temperature storage (HTS), low-temperature storage (LTS), and
the acoustic signal travels different distances before it arrives at
temperature cycling (TC) tests in order to assess the reliability
the two MEMS hydrophones. The sound speed in the pipeline
of the packaged devices. Fig. 14 shows the measured sensitiv-
can be deduced as
ity variations of five packaged MEMS hydrophones after the
d HTS, LTS, and TC tests. The sensitivity drifts of all the five
vs = (3) hydrophones are less than ±0.5 dB (±0.56%). No significant
Δt
performance drift or failure mechanism is observed after the reli-
where vs is the sound speed in water, and d = 10 m is the distance ability tests. These measured reliability evaluation results verify
between the two MEMS hydrophones. Assuming Δt = 9 ms the reliability and robustness of the MEMS hydrophones and
is the time delay between the received signals from the two the packaging. The estimated life time of the packaged MEMS
MEMS hydrophones, according to (3), the sound speed in water hydrophone is 10.4 years based on the Coffin–Manson model.
is calculated to be 1111 m/s. The calculated sound speed is lower The long-term stability measurements have been performed
than the theoretical value of 1450 m/s by 23.4%, which may on the MEMS hydrophone, in order to investigate the stability of
be caused by the setup imperfection, such as synchronization the packaged MEMS hydrophone over time. Fig. 15 shows the
error of the two signals. In addition, the sound speed in the measured sensitivity drifts of a packaged MEMS hydrophone for
water filled in the pipeline is highly dependent on the pipeline over 1200 d. The sensitivity drift is less than ±0.3 dB (±0.15%).

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
XU et al.: LOW-COST, TINY-SIZED MEMS HYDROPHONE SENSOR FOR WATER PIPELINE LEAK DETECTION 6381

ACKNOWLEDGMENT
The authors would like to thank N. Jaafar for hydrophone
sensor assembly and also, Q. Xie at Massachusetts Institute of
Technology (MIT), for his valuable discussion in preparing this
manuscript.
REFERENCES
[1] A. C. McIntosh, Asian Water Supplies Reaching the Urban Poor. London,
U.K.: ADB, 2003.
[2] C. Lallana and N. Thyssen, Water Use Efficiency (in Cities): Leakage.
Copenhagen, Denmark: EEA, 2003.
[3] R. Frauendorfer and R. Liemberger, The Issues and Challenges of Reduc-
ing Non-Revenue Water. London, U.K.: ADB, 2010.
[4] The Smart Water Networks Forum, Stated NRW (Non-Revenue Water)
Rates in Urban Networks. Surrey, U.K.: SWAN, 2011.
[5] B. Kingdom, G. Soppe, and J. Sy, “What is non-revenue water? How can
we reduce it for better water service?,” Water Blog. (2016). [Online]. Avail-
able: https://blogs.worldbank.org/water/non-revenue-water-management
Fig. 14. Reliability tests of five packaged hydrophones. The tested -has-it-s-time-finally-come
hydrophones are evaluated by HTS, LTS, and TC tests, respectively. [6] L. W. Mays, Water Distribution Systems Handbook. New York, NY, USA:
McGraw-Hill, 2000.
[7] O. Hunaidi, W. T. Chu, A. Wang, and W. Guan, “Leak detection method
for plastic water distribution pipes,” Seminars Water Sewer Infrastructure
Syst.: Challenges Solutions, pp. 249–270, Apr. 2000.
[8] P. Gopalakrishnan, S. Abhishek, R. Ranjith, R. Venkatesh, and V. Jai
Suriya, “Smart pipeline water leakage detection system,” Int. J. Appl.
Eng. Res., vol. 12, no. 16, pp. 5559–5564, 2017.
[9] D. Chatzigeorgiou, A. Khalifa, K. Youcef-Toumi, and R. Ben-Mansour,
“An in-pipe leak detection sensor: Sensing capabilities and evaluation,” in
Proc. ASME/IEEE Int. Conf. Mechatronics Embedded Syst. Appl., Aug.
2011, pp. 481–489.
[10] O. Hunaidi, A. Wang, M. Bracken, T. Gambino, and C. Fricke, “Detecting
leaks in water-distribution pipes,” Arab Water World, vol. 29, no. 4, pp. 52–
55, May 2005.
[11] Z. Liu and Y. Kleiner, “State-of-the-art review of technologies for pipe
structural health monitoring,” IEEE Sensors J., vol. 12, no. 6, pp. 1987–
1992, Jun. 2012.
[12] S. P. Siebenaler and G. R. Walter, “Detection of small leaks in liquid
pipelines utilizing distributed temperature sensing,” in Proc. Int. Pipeline
Conf., Calgary, Canada, 2012.
[13] S. B. Costello, D. N. Chapman, C. D. F. Rogers, and N. Metje, “Under-
Fig. 15. Long-term stability measurements for the packaged MEMS ground asset location and condition assessment technologies,” Tunnelling
hydrophone. Underground Space Technol., vol. 22, no. 5–6, pp. 524–542, Nov. 2007.
[14] J. Zhang, A. Hoffman, K. Murphy, J. Lewis, and M. Twomey, “Review
of pipeline leak detection technologies,” in Proc. PSIG Ann. Meeting,
Prague, Czech Republic, 2013.
[15] A. Sadeghioon, N. Metje, D. Chapman, and C. Anthony, “SmartPipes:
The measurement results illustrate that the MEMS hydrophone Smart wireless sensor networks for leak detection in water pipelines,”
has excellent long-term stability. J. Sens. Actuator Netw., vol. 3, no. 1, pp. 64–78, 2014.
[16] A. Whittle, M. Allen, A. Preis, and M. Iqbal, “Sensor networks for moni-
toring and control of water distribution systems,” in Proc. 6th Conf. Struct.
Health Monitor. Intell. Infrastruct., Hong Kong, 2013.
VI. CONCLUSION [17] M. Allen, A. Preis, M. Iqbal, and A. Whittle, “Water distribution system
monitoring and decision support using a wireless sensor net-work,” in
This paper experimentally presented a water pipeline leak Proc. 14th ACIS Int. Conf. Softw. Eng. Artif. Intell. Netw. Parallel/Distrib.
detection system, for the first time, using a low-cost, tiny-sized Comput., Jul. 2013, pp. 641–646.
MEMS hydrophone sensor. All the experiments were conducted [18] G. Q. Wu et al., “Wafer-level vacuum-packaged high performance AlN-
on-SOI piezoelectric resonator for sub-100 MHz oscillator applications,”
based on ductile iron pipeline with internal diameter of 10 cm IEEE Trans. Ind. Electron., vol. 65, no. 4, pp. 3576–3584, Apr. 2018.
and wall thickness of 0.73 cm. The experiment results indicate [19] J. Xu, X. Zhang, S. N. Fernando, K. T. Chai, and Y. Gu, “AlN-on-SOI
that both new and existing leaks can be detected even with flow platform-based micro-machined hydrophone,” Appl. Phys. Lett., vol. 109,
no. 3, Jul. 2016, Art. no. 032902.
rate down to 30 L/min. The frequency spectrum of sound signals [20] J. Xu, X. Zhang, S. N. Fernando, S. Merugu, K. T. Chai, and Y. Gu,
emitted by water leak was largely within 2 kHz. The comparison “AlN-on-SOI platform-based MEMS hydrophone with ultra-low opera-
of the inside and outside detection approaches was also reported. tion frequency and ultra-high noise resolution,” Proc. IEEE 29th Int.Conf.
Micro Electro Mech. Syst., 2016, pp. 1086–1089.
The leak positioning evaluation proved the feasibility of the [21] J. Xu, J. M. Tsai, W. Sun, and C. Sun, “Sensor with vacuum-sealed cavity,”
leak positioning using the reported MEMS hydrophones. This U.S. Patent US20140230557, Apr. 26, 2016.
paper demonstrates a highly efficient, real-time, permanent in- [22] J. Karki, “Signal conditioning piezoelectric sensors,” TI, Appl. Rep.
SLOA033A, Dallas, TX, USA, 2000.
pipe pipeline health monitoring technology based on the MEMS [23] Y. Khulief, A. Khalifa, R. Mansour, and M. Habib, “Acoustic detection
hydrophone sensors. It paves a promising path for water leak of leaks in water pipelines using measurements inside pipe,” J. Pipeline
detections. Syst. Eng., vol. 3, no. 2, pp. 47–54, 2012.

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.
6382 IEEE TRANSACTIONS ON INDUSTRIAL ELECTRONICS, VOL. 66, NO. 8, AUGUST 2019

[24] D. Chatzigeorgiou, K. Youcef-Toumi, A. Khalifa, and R. Ben-Mansour, Beibei Han received the B.S. degree in elec-
“Analysis and design of an in-pipe system for water leak detection,” in tronic science and technology from Shandong
Proc. ASME Int. Des. Eng. Tech. Conf. Comput. Inf. Eng. Conf., 2011, University, China, in 2006, the M.S. degree in
pp. 1007–1016. mechanical engineering from Korea University
[25] Brüel & Kjær, “Hydrophones - Types 8103, 8104, 8105 and 8106,” Sep. of Technology and Education, South Korea, in
2017. 2008, and the Ph.D. degree in microelectronics
[26] Benthowave Instrument Inc., Product Datasheet. [Online]. Available: from Nanyang Technological University (NTU),
https://www.benthowave.com/products/BII-7150Hydrophone.html Singapore, in 2015.
[27] DolphinEar Hydrophones, Product Datasheet. [Online]. Available: From 2008 to 2010, she was a Re-
http://www.dolphinear.com/de200.html search Scientist working on cognitive robotics
[28] H2a Hydrophone User’s Guide, Aquarian Audio, Anacortes, WA, USA. and simulation with Korea Institute of Science
[29] J. M. Hovem, Marine Acoustics: The Physics of Sound in Underwater and Technology, South Korea, and Korea Advanced Institute of Science
Environments. Los Altos, CA, USA: Peninsula Publishing, 2010. and Technology, South Korea. Since 2015, she has been a Scientist with
[30] L. Zhang, Y. Wu, L. Guo, and P. Cai, “Design and implementation of leak the Institute of Microelectronics, A*STAR, Singapore. Her research inter-
acoustic signal correlator for water pipelines,” Inf. Technol. J., vol. 12, est focuses on physical MEMS sensors and acoustic sensor systems.
no. 11, pp. 2195–2200, 2013.
Eva Leong-Ching Wai received the bachelor’s
degree in materials engineering from Nanyang
Technology University (NTU), Singapore, in
Jinghui Xu received the B.Eng. degree in au- 2003.
tomatic engineering from Chang’an University, She worked in semiconductor packaging in-
Xi’an, China, in 2003, and the M.S. and Ph.D. dustrial for two years after graduating from uni-
degrees in mechanical engineering from North- versity. Since then, she has been with Institute
western Polytechnical University, Xi’an, China, of Microelectronics, A*STAR, Singapore, where
in 2006 and 2009, respectively. she is currently the Senior Research Engineer
From 2009 to 2010, he was a Postdoctoral of Interconnect and Packaging Platform.
Researcher with the Department of Microsys-
tems Engineering, Albert Ludwigs Freiburg Uni- Wei Li received the B.Eng and Ph.D. degrees
versity, Freiburg, Germany. Since 2010, he has in microelectronics from Nanyang Technologi-
been with Institute of Microelectronics, Agency cal University, Singapore, in 2005 and 2009, re-
for Science, Technology and Research (A*STAR), Singapore, where he spectively, and the M.B.A. degree from INSEAD,
is currently a Research Scientist. He holds three U.S. patents and com- Fontainebleau, France, in 2012.
pleted two times IP licensing. His research interests include acoustic He has many years of experience in technol-
(hydrophone, microphone, pMUT, etc.) and vibrational MEMS (resonator, ogy development, IP management, technology
oscillator, accelerator, etc.) based on piezoelectric AlN platform. commercialization, start-up incubation, and in-
vestment. He is currently the General Manager
with JCS Venture Lab and the founding CEO of
Kevin Tshun-Chuan Chai (M’16) received CSF Ventures. He is also Adjunct Assistant Pro-
the B.Eng. (Hons.) and the Ph.D. degrees in fessor with Singapore University of Technology and Design (SUTD) for
electronic and electrical engineering from the SUTD Technology entrepreneurship program.
University of Glasgow, U.K., in 2002 and 2007,
respectively, developing tissue cell imaging solu- Jason Yeo received the Diploma of Engineering
tion based on electrical impedance tomography in factory automation from Ngee Ann Polytech-
on a CMOS chip. nic, Singapore, in 1989 and the EMBA degree in
He joined Institute of Microelectronics, business administration from National University
A*STAR, Singapore, in 2008 as a Research Sci- of Singapore, Singapore, in 2015.
entist and developed a silicon nanowire based He founded JCS in 1990 and led JCS to be-
biosensor readout system for the detection of come a multinational group company with grow-
biomarkers in cardiac disease. He has received several competitive ing businesses in a range of deep-tech indus-
A*STAR grants as PI/Co-PI for MEMS sensor-related applications in tem- tries. He is currently the Chairman with JCS
perature, motion and sound detection, cell counting, electronic stetho- Group and JCS Venture Lab. He is also the
scope system for the early detection of diastolic dysfunction in hyperten- Chairman with McLean, a listed company in
sive heart disease, etc. He currently heads a department of more than Malaysia.
30 IC designers working on various topics from AI powered hardware Edwin Nijhof received the Diploma in electri-
accelerators looking at both deep learning and neuromorphic methodolo- cal and electronics engineering from Middel-
gies, compute-in-memory using emerging memories, hardware security bare Technische School, Veghel, Netherlands,
for edge IOT, power management solution with IVR, mmWave IC, and in 1987.
design acceleration techniques using machine learning. He has more than 25 years of working expe-
rience in high tech industry and has worked in
several multinational corporations and startups
Guoqiang Wu received the B.Eng. degree in for their technology development and regional
electrical science and technology from Xidian operations. He is currently the Chief Technology
University, Xi’an, China, in 2008, and the Ph.D. Officer with JCS Group, JCS Venture Lab, and
degree in microelectronics and solid-state elec- the General Manager with JCS Technologies.
tronics from the Shanghai Institute of Microsys-
tem and Information Technology (SIMIT), Chi- Yuandong Gu received the M.E.E. degree in
nese Academy of Sciences, Shanghai, China, electrical engineering and the Ph.D. degree in
in 2013. pharmaceutics from the University of Minnesota,
He is currently a Research Scientist with Minneapolis, MN, USA, in 2001 and 2003,
the Institute of Microelectronics, Agency for respectively.
Science, Technology and Research (A*STAR), He was a Principal Research Scientist with
Singapore. His research interests include microelectromechanical sys- the Honeywell Sensors and Wireless Lab for ten
tems (MEMS) resonators, RF MEMS, inertial MEMS, MEMS integration, years. He is currently the Deputy Executive Di-
and packaging technologies. rector (DED) with the Institute of Microelectron-
Dr. Wu was the recipient of the Best Dissertation Award of Shanghai ics, Agency for Science, Technology and Re-
in 2014. search (A*STAR), Singapore.

Authorized licensed use limited to: Monash University. Downloaded on August 29,2020 at 01:35:38 UTC from IEEE Xplore. Restrictions apply.