Download as pdf or txt
Download as pdf or txt
You are on page 1of 7

A straightforward method for estimating the size of leaks in water pipelines using

acoustic transients
Wenjie Wang, Zhao Li, Liwen Jing, Pedro Lee, and Ross Murch

Citation: The Journal of the Acoustical Society of America 144, EL404 (2018); doi: 10.1121/1.5078765
View online: https://doi.org/10.1121/1.5078765
View Table of Contents: https://asa.scitation.org/toc/jas/144/5
Published by the Acoustical Society of America

ARTICLES YOU MAY BE INTERESTED IN

Guided acoustic wave interaction with flanged junctions in water-filled steel pipelines
The Journal of the Acoustical Society of America 144, 2824 (2018); https://doi.org/10.1121/1.5078592

Alternating direction method of multipliers for weighted atomic norm minimization in two-dimensional grid-free
compressive beamforming
The Journal of the Acoustical Society of America 144, EL361 (2018); https://doi.org/10.1121/1.5066345

Three-dimensional printable ultrasound transducer stabilization system


The Journal of the Acoustical Society of America 144, EL392 (2018); https://doi.org/10.1121/1.5066350

An approximate inverse scattering technique for reconstructing blockage profiles in water pipelines using
acoustic transients
The Journal of the Acoustical Society of America 143, EL322 (2018); https://doi.org/10.1121/1.5036957

Acoustic radiation force exerted on a small spheroidal rigid particle by a beam of arbitrary wavefront: Examples
of traveling and standing plane waves
The Journal of the Acoustical Society of America 144, EL453 (2018); https://doi.org/10.1121/1.5080529

Nonlinear ultrasound parameter to monitor cell death in cancer cell samples


The Journal of the Acoustical Society of America 144, EL374 (2018); https://doi.org/10.1121/1.5066348
Wang et al.: JASA Express Letters https://doi.org/10.1121/1.5078765 Published Online 19 November 2018

A straightforward method for estimating the size


of leaks in water pipelines using acoustic transients
Wenjie Wang,1 Zhao Li,1,a) Liwen Jing,1 Pedro Lee,2 and Ross Murch1
1
Department of Electronic and Computer Engineering, The Hong Kong University of
Science and Technology, Hong Kong, China
2
Department of Civil and Natural Resources Engineering, University of Canterbury,
Christchurch, New Zealand
wwangbk@connect.ust.hk, eelizhao@ust.hk, ljing@connect.ust.hk,
pedro.lee@canterbury.ac.nz, eermurch@ust.hk

Abstract: A straightforward explicit expression is derived for estimat-


ing the size and position of leaks in a water pipeline using the acoustic
transient response. Experimental results are provided for a water pipe-
line with multiple leaks to demonstrate the usefulness and performance
of the proposed approach. The form of the leak estimation expression
can be further combined with an existing explicit expression derived for
blockage detection. Using this combined expression it is also demon-
strated that the proposed approach can detect and estimate leaks and
blockages when they are simultaneously present in pipelines.
C 2018 Acoustical Society of America
V
[PG]
Date Received: July 23, 2018 Date Accepted: October 25, 2018

1. Introduction
Globally, water leakage from pipelines in urban water supply systems is estimated to
result in 20%–30% water loss.1 Significant effort has been devoted to developing water
leak detection techniques. One approach is to transmit transient acoustic waves inside
the pipelines and measure the resulting reflections to find the location and size of the
leaks.1–3 Detecting leaks can be considered an acoustic inverse problem4,5 where scat-
tered waves resulting from an acoustic transient are used to determine pipeline defects.
Inverse problems in one-dimension have a long history but have focused on the imped-
ance or blockage reconstruction problem.4
In this work we formulate leak detection as an inverse scattering problem.
Our novel contributions include (1) deriving a straightforward approximate explicit
expression for the size and position of leaks in terms of the acoustic transient response,
(2) proposing a method for reconstructing leaks and blockages simultaneously by com-
bining our formulation with an existing explicit expression for blockage reconstruc-
tion,6 and (3) utilizing experimental and simulation results to demonstrate the effective-
ness of the approach. The work is based on previous results for electrical transmission
lines7 but incorporates an acoustic formulation for leak reconstruction and demon-
strates the possibility of removing the effect of blockages.
2. Background
To describe the method we use an example pipeline of length LT and inner diameter
Dp with two leaks as shown in Fig. 1. The pipeline is connected to reservoirs at both
pipe boundaries and two circular orifices with diameter DL are drilled in the pipe to
simulate leaks as Leaka and Leakb. An acoustic transient is generated by rapidly clos-
ing and opening the solenoid valve at the left side of the water pipeline to impose
changes in the mean flow.8 The resulting reflections of the transient by the leaks are
acquired by a pressure sensor (Sensor 3) and used to deduce the size and location of
the leaks. We define the location of Sensor 3 as the origin (z ¼ 0) of the axial coordi-
nate z. The solenoid valve closure provides a transient of around 4 ms and therefore a
bandwidth of around 250 Hz. Given that the wavelength k is much larger than the
diameter of the pipeline, the acoustic transient propagates along the axial coordinate z
as a plane wave in a waveguide. The continuity and momentum equations in the water
column can therefore be used to model the acoustic waves as6,8
@Uðz; kÞ cðz; kÞ
þ Pðz; kÞ þ G ðzÞPðz; kÞ ¼ 0; (1a)
@z Zðz; kÞ

a)
Author to whom correspondence should be addressed.

EL404 J. Acoust. Soc. Am. 144 (5), November 2018 C 2018 Acoustical Society of America
V
Wang et al.: JASA Express Letters https://doi.org/10.1121/1.5078765 Published Online 19 November 2018

Fig. 1. Experimental configuration and parameters for the experiments.

@Pðz; kÞ
þ cðz; kÞZðz; kÞUðz; kÞ ¼ 0; (1b)
@z
where acoustic impedance Z(z, k) and propagation constant c(z, k) are given by
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
jkAp ðzÞR
cðz; kÞ ¼ a þ jb ¼ k 2 þ ; (2a)
qa
qacðz; kÞ
Zðz; kÞ ¼ ; (2b)
jkAp ðzÞ
and where a is the acoustic waves speed, q is water density, Ap(z) is the cross-sectional
area of the pipeline, k is the wavenumber, and R is the steady friction term obtained
from the Darcy–Weisbach friction equation8 if the effects of unsteady friction are
ignored. Volume velocity U(z, k), pressure P(z, k), and steady state leak loss G(z) are
z-dependent where8
pffiffiffiffiffiffiffiffiffiffiffi
Cd AL ðzÞ 2gHo
G ðzÞ ¼ ; (3)
2qgHo
in which Cd is the coefficient of discharge, AL(z) is cross-sectional area of the leak, g is
the gravitational acceleration, and Ho is the average value of the head. Equation (3)
captures the effect that variations in volume velocity are directly related to the size of
the leak and the pressure head at the leak.8 The acoustic wave speed a for k  Dp can
be found from Halliwell9
sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
K
a¼  ; (4)
q 1 þ Dp K=Ee
where e and E are the wall thickness and Young’s modulus of the pipeline, respec-
tively, and K is the bulk modulus of water. For a copper pipeline with dimensions
shown in Fig. 1, e ¼ 1.6 mm, Dp ¼ 22.1 mm, E ¼ 120 GPa, q ¼ 998 kg/m3, and K ¼ 2.2 GPa
so that a ¼ 1325 m/s. Using Eq. (4) our method can be used with a variety of pipeline
materials and diameters.
If the pipeline is assumed to have low friction and no cross-sectional area vari-
ation [Ap(z) ¼ Ap(0) ¼ Ao], Z(z, k) and c(z, k) can be well approximated7 by the first
two terms of their Taylor series expansion in k giving
Ao R
cðz; kÞ ¼ a þ jb ’ þ jk; (5a)
2qa
qa R qa
Zðz; kÞ ¼ Zo ’ þ ’ : (5b)
Ao 2jk Ao
Writing the acoustic transient as an incident field Pinc(z, k), and taking it as an impulse,
our inverse problem is to determine AL(z) from the reflected waves given R and Ao.
3. Leak reconstruction
Substituting Eq. (5) into Eq. (1) and rearranging we obtain a wave equation with a
source
P00 ðz; kÞ  ða þ jkÞ2 Pðz; kÞ ¼ ða þ jkÞZo GðzÞPðz; kÞ; (6)
0
where the prime ( ) denotes derivative with respect to z. To allow us to separate the
incident excitation from the reflections caused by the leaks, pressure P(z, k) is

J. Acoust. Soc. Am. 144 (5), November 2018 Wang et al. EL405
Wang et al.: JASA Express Letters https://doi.org/10.1121/1.5078765 Published Online 19 November 2018
jkzaz
partitioned as an incident wave with attenuation Pinc ðz; kÞ ¼ e and a reflected
wave Ps(z, k) giving
Pðz; kÞ ¼ Pinc ðz; kÞ þ Ps ðz; kÞ ¼ ejkzaz þ Ps ðz; kÞ: (7)
Since Pinc(z, k) is the field inside the pipeline without defects (and satisfying a homoge-
neous Helmholtz equation), substituting Eq. (7) into Eq. (6) gives
P00s ðz; kÞ  ða þ jkÞ2 Ps ðz; kÞ ¼ ða þ jkÞZo GðzÞ½Pinc ðz; kÞ þ Ps ðz; kÞ: (8)
ðaþjkÞjzyj
The solution of Eq. (8) can be expressed using Green’s function e =½2ða þ jkÞ
as
ð
ða þ jk Þ LT  
Ps ðz; kÞ ¼  Zo G ð yÞ P0inc ð y; kÞ þ P0s ð y; kÞ eðaþjkÞjzyj dy: (9)
2ða þ jk Þ 0
This equation expresses the reflected field as an integral of the leak defect and the
derivative of the incident and scattered fields. If the leak is mild Ps(z, k) will be negligi-
ble compared to the incident wave, so that P(z, k) can be approximated by Pinc(z, k)
and the second term in the brackets of Eq. (9) can be dropped. This is known as the
Born approximation5 and allows Eq. (9) to be written as
ð
Zo L T
Ps ðz ¼ 0; k Þ ¼  G ð yÞe2ay e2jky dy: (10)
2 0
This can be expressed in the form of Fourier transform, by combining Eq. (3) with Eq.
(10), as
pffiffiffiffiffiffiffiffiffiffiffi
AL ðzÞ 2 2gHo
¼  F 1 ½Ps ð0; kÞeaz ð2zÞ: (11)
Ao Cd a
It can be deduced that the reflections caused by leaks have opposite sign to the incident
transient and their amplitude can be directly related to the orifice area. This equation pro-
vides a straightforward method for estimating the normalized area of leaks as a function
of z using the reflected signal in the time domain [Fourier Transform of Ps(0, k)].
The accuracy of Eq. (11) is determined by how well the incident wave approxi-
mates the total wave [see Eq. (9)] inside the pipeline. For a single leak, Eq. (11)
becomes the exact solution since there is no approximation. For multiple leaks, a
validity condition can be based on the maximum size of the scattering. Taking the ker-
nel of Eq. (10) to be ZoG(z)/2 < 1/4 as a condition for the Born approximation to be
accurate and given Cd ¼ 0.6, a ¼ 1325 m/s, g ¼ 9.8 m/s2, and Ho ¼ 38.4 m, for example,
then the diameter of each leak should not be larger than around 1/8 of the diameter of
the pipe for accurate reconstruction.
Measurement noise will also affect the reconstruction and could be modeled as
a Gaussian process (zero-mean with variance r2) added to the right- hand side of Eq.
(11). If the noise variance is known then a hypothesis test can be performed at every
point along z to determine if the detected leak is above the expected noise level.
It is useful to compare Eq. (11) with an explicit expression for the reconstruc-
tion of blockages.6 Characterizing blockages by pipeline cross-sectional area Ap(z) this
is written as6
dAp ðzÞ=dz
¼ 2F 1 ½Ps ð0; kÞeaz ð2zÞ: (12)
Ap ðzÞ
If there are both leaks and blockages in the pipeline we will receive scattering Ps(0, k)
from each of them. The total combined reflected wave can then be written as
AL ðzÞ Cd a dAp ðzÞ=dz
 pffiffiffiffiffiffiffiffiffiffiffi þ ¼ F 1 ½Ps ð0; kÞeaz ð2zÞ; (13)
Ao 2 2gHo 2Ap ðzÞ
which is the sum of the individual leak and blockage reflections since we have invoked
the Born approximation.7 A difference between leaks and blockages is that the
response due to blockages contains a spatial derivative dAp(z)/dz. This shows that on a
common coordinate system the response to an excitation from the left side of the
blockage will be inverted compared to that of an excitation from the right side. To sep-
arate leaks from blockages we can use an additional source (valve) and sensor at the
right side of the leaks (at location zR, for example) in addition to the sensor and source
at z ¼ 0. The reflections obtained at the right sensor due to the source at the right are
then

EL406 J. Acoust. Soc. Am. 144 (5), November 2018 Wang et al.
Wang et al.: JASA Express Letters https://doi.org/10.1121/1.5078765 Published Online 19 November 2018

AL ðzR  zÞ Cd a dAp ðzR  zÞ=dz   


 pffiffiffiffiffiffiffiffiffiffiffi þ ¼ F 1 Ps ðzR ; kÞeaðzR zÞ 2ðzR  zÞ : (14)
Ao 2 2gHo Ap ðzR  zÞ
It can be observed that the spatial derivative of the blockage profile dAp(zR  z)/dz in
Eq. (14) will have an opposite sign to that in Eq. (13) due to the reversal of the coordi-
nates. However leak estimation in Eq. (11) does not have a derivative and therefore if
the left and right responses are added together at each location z, the blockage
responses will cancel while the leak responses will add.7 On the other hand subtracting
the two results would leave us with blockages only.7 We investigate this in Sec. 4.
4. Experimental results for leak estimation
Experimental results from Refs. 10 and 11 are used to verify Eq. (11). The configura-
tion of the pipeline is as in Fig. 1, where dimensions are shown, and the pipeline is
composed of copper. Measurement data consist of configurations with no leaks, indi-
vidual leaks at Leaka or Leakb and simultaneous leaks, Leaksa,b. The sampling fre-
quency is 2000 Hz and this provides a temporal resolution of 0.5 ms or approximately
0.66 m since a ¼ 1325 m/s. Given the impulse duration is 4 ms each leak will appear in
8 consecutive samples. To allow Eq. (11) to be applied we therefore normalize the
measurements so that the sum of the incident transient samples is unity across its eight
sample duration. We estimate noise variance directly from the normalized measure-
ment without leaks as r2 ¼ 2.2  105. For negative signals 10 dB below this noise
threshold we accept them as a leak (<3.2r) [corresponding to a required received sig-
nal-to-noise ratio (SNR) of greater than 10 dB].
The normalized impulse response acquired at Sensor 3 is shown in Fig. 2(a).
The leaks can be identified as where the responses cross the negative threshold and
their locations from the peaks of those negative responses. The estimated locations are
9.5 and 31.0 m for Leaka and Leakb as compared to the exact locations of 9.265 and
30.665 m, respectively. The spatial resolution of the system is 0.66 m because we can
distinguish the peak of the 4 ms duration impulse using the 0.5 ms sampling rate. The
estimation for the leak orifice area is found from Eq. (11). Since the normalized tran-
sient impulse duration is eight samples we add together eight consecutive samples con-
taining the peak in the calculation of the orifice size. The normalized orifice area (AL/
Ao) estimation for Leaka only is 0.0038, Leakb only is 0.0053, and simultaneous leaks
Leaksa,b is 0.0038 and 0.0046, respectively (the actual normalized orifice area of each
leak is 0.0046). The orifice estimates are within 618% of the actual orifice area and
this is competitive with established experimentally verified techniques.2 The error in
size is related to the noise and modeling errors of the system.
The pipeline length or range this technique can operate depends on meeting
the SNR > 10 dB requirement. This depends on the size of the leak, incident pulse
power, and pipeline attenuation. Assuming pipeline attenuation at these frequencies is
low (0.05 dB/m) and noting that leakb in Fig. 2(b) is approximately 15 dB higher than
the noise, we can expect an additional 100 m round trip distance could be successfully
achieved.
Modeling errors may arise from the presence of sensors and joints in the pipe-
line, which could appear as small blockages. These can be removed by acquiring
impulse responses from both the left and right sides of the pipeline as discussed in Sec.
3. In this example we are able to capture Ps(zR ¼ 19.2 m, k) from the right of Leaka by
using Sensor 2. The transient generated at the left side travels the length of the pipeline
and is reflected from the right end, where it becomes a left traveling transient. This left
traveling transient can then be captured at Sensor 2 as well as the resulting reflections

Fig. 2. Experimental results of leak detection. (a) Signals at Sensor 3 when Leaka, Leakb, and Leaksa,b are pre-
sent and (b) signals at Sensors 2 and 3 and their addition for Leaka detection. Arrows indicate the actual loca-
tion of the leaks and dashed horizontal lines are the decision threshold.

J. Acoust. Soc. Am. 144 (5), November 2018 Wang et al. EL407
Wang et al.: JASA Express Letters https://doi.org/10.1121/1.5078765 Published Online 19 November 2018

after it passes through Leaka. Using Eq. (11) we can therefore obtain another estimate
of Leaka using the data from Sensor 2 and this is plotted in Fig. 2(b). Adding the two
results together will result in a better estimate of the leak size and this result is also
shown in Fig. 2(b). The decision threshold is also reduced by 3 dB due to averaging
(<2.2r). While it can be observed that blockages are not playing a major role the
leak estimate is improved and the estimated normalized cross-sectional area is
AL(za ¼ 9.5 m)/Ao ¼ 0.0039. If left traveling transients were generated directly on the
right by another valve we believe further improvement would occur.
5. Simulation and experimental results for leak and blockage estimation
Simulation results for leaks and blockages are presented in Fig. 3 with the same leak,
excitation, and sampling rate as for the experimental configuration described in Sec. 4.
We obtain reflections Ps(z, k) from both sides of the leaks and blockages similarly to
the previous example but this time using Sensors 3 and 1. A simulated blockage with
rectangular profile is added to the pipeline and centered at z ¼ 20 with length 4 m and
cross-sectional area 88% of the pipe. In Figs. 3(a) and 3(b) results are provided without
noise to demonstrate the principle of the leak and blockage separation technique.
From Figs. 3(a) and 3(b) we can observe that we can reconstruct the leak [solid line in
Fig. 3(a)] and the blockage [dashed-dotted line in Fig. 3(b)]. In Fig. 3(a) we can also
observe from the two broken lines that the leak affects the measurements at Sensors 3
and 1 in the same way. However for the blockage the results from Sensor 1 are
inverted to those from Sensor 3 as predicted by Eqs. (13) and (14). Therefore when we
add the results from Sensors 1 and 3 together we arrive at the leak reconstruction as
shown by the solid line in Fig. 3(a) with AL(za ¼ 9.5 m)/Ao ¼ 0.0048 and the effects of
the blockage are removed. However if we subtract the results and apply Eq. (12) we
arrive at the blockage reconstruction as shown by the dashed-dotted line in Fig. 3(b).
If we did not apply this method to separate out blockages, reflections from the leading
edge of the blockage [see Fig. 3(a)] would give the appearance of a leak causing a false
positive.
When noise is added to the simulation at the same SNR level as the experi-
ments we obtain the simulation results shown in Figs. 3(c) and 3(d). Even with this
level of noise, leak and blockage detection is possible, but accuracy is reduced with
AL(za ¼ 9.5 m)/Ao ¼ 0.0051.
In terms of experimental verification we are able to capture Ps(zR ¼ 30.665 m,
k) from the right of Leaka by using Sensor 1 and the results are plotted in Fig. 4. It
can be observed that the bump around z ¼ 20 m in Fig. 4(a) is reduced in Fig. 4(b). At
the location of the bump there is Sensor 2 which could cause the appearance of a mild
blockage. Our method has removed that false positive and we find a leak at 9.5 m with
orifice area estimate for Leaka being 0.0024. While this orifice estimate is not accurate
this is thought to be a result of the left side transient being generated indirectly as a

Fig. 3. Simulated estimation results for a leak at 9.5 m and a rectangular blockage centered at z ¼ 20 m. (a)
Signals from Sensors 1 and 3 and their addition for Leaka detection and (b) blockage reconstruction results
using signals from Sensors 1 and 3 and Eqs. (12)–(14). (c) Same as (a) but with noise. (d) Same as (b) but with
noise.

EL408 J. Acoust. Soc. Am. 144 (5), November 2018 Wang et al.
Wang et al.: JASA Express Letters https://doi.org/10.1121/1.5078765 Published Online 19 November 2018

Fig. 4. Experimental results for leak detection. (a) Signals at Sensors 1 and 3 and the identification of a leak and
false positive and (b) detection results for Leaka after the addition of the signals from Sensors 1 and 3. The sig-
nal at Sensor 3 in (a) is the same as in Fig. 2(a) for Leaka.

reflection. If the two profiles were subtracted the phenomena at z ¼ 20 m is increased


to enhance the hypothesized blockage but it is still below the threshold. Nevertheless
we have shown that the effect of blockages can potentially be removed using measure-
ments from both sides of the leaks.
6. Conclusion
An explicit expression for estimating leaks in water pipelines using acoustic transients
was derived and experimental results demonstrated good accuracy. Estimating leaks
and blockages separately when they were simultaneously present in pipelines was also
demonstrated.
Acknowledgments
We would like to thank the Hong Kong Research Grants Council for the Theme-based
Research Scheme Grant No. T21-602/15R.
References and links
1
A. F. Colombo, P. Lee, and B. W. Karney, “A selective literature review of transient-based leak detection
methods,” J. Hydro-environ. Res. 2(4), 212–227 (2009).
2
D. Covas and H. Ramos, “Case studies of leak detection and location in water pipe systems by inverse
transient analysis,” J. Water Resource Plan. Manage. 136(2), 248–257 (2010).
3
B. Zhou, A. Liu, X. Wang, Y. She, and V. Lau, “Compressive sensing-based multiple-leak identification
for smart water supply systems,” IEEE Internet Things J. 5(2), 1228–1241 (2018).
4
A. Bruckstein and T. Kailath, “Inverse scattering for discrete transmission-line models,” SIAM Rev.
29(3), 359–389 (1987).
5
R. Bates, V. Smith, and R. D. Murch, “Manageable multidimensional inverse scattering theory,” Phys.
Rep. 201(4), 185–277 (1991).
6
L. Jing, Z. Li, W. Wang, A. Dubey, P. Lee, S. Meniconi, B. Brunone, and R. D. Murch, “An approxi-
mate inverse scattering technique for reconstructing blockage profiles in water pipelines using acoustic
transients,” J. Acoust. Soc. Am. 143(5), EL322–EL327 (2018).
7
L. Jing, W. Wang, Z. Li, and R. D. Murch, “Detecting impedance and shunt conductance faults in lossy
transmission lines,” IEEE Trans. Antennas Propag. 66(7), 3678–3689 (2018).
8
M. H. Chaudhry, Applied Hydraulic Transients, 3rd ed. (Springer, New York, 2014).
9
A. Halliwell, “Velocity of a water-hammer wave in an elastic pipe,” J. Hydraulics Div. 89(4), 1–21
(1963).
10
P. J. Lee, M. F. Lambert, A. R. Simpson, J. P. Vıtkovsky, and J. Liggett, “Experimental verification of
the frequency response method for pipeline leak detection,” J. Hydro-environ. Res. 44(5), 693–707
(2006).
11
P. J. Lee, “Using system response functions of liquid pipelines for leak and blockage detection,” Ph.D.
thesis, The University of Adelaide, Australia, 2005.

J. Acoust. Soc. Am. 144 (5), November 2018 Wang et al. EL409

You might also like