Leak Detection in Pipelines using the Damping of Fluid

Transients
Xiao-Jian Wang1; Martin F. Lambert2; Angus R. Simpson, M.ASCE3; James A. Liggett4;
and John P. Vı́tkovský5

Abstract: Leaks in pipelines contribute to damping of transient events. That fact leads to a method of finding location and magnitude
of leaks. Because the problem of transient flow in pipes is nearly linear, the solution of the governing equations can be expressed in terms
of a Fourier series. All Fourier components are damped uniformly by steady pipe friction, but each component is damped differently in
the presence of a leak. Thus, overall leak-induced damping can be divided into two parts. The magnitude of the damping indicates the size
of a leak, whereas different damping ratios of the various Fourier components are used to find the location of a leak. This method does
not require rigorous determination and modeling of boundary conditions and transient behavior in the pipeline. The technique is successful
in detecting, locating, and quantifying a 0.1% size leak with respect to the cross-sectional area of a pipeline.
DOI: 10.1061/共ASCE兲0733-9429共2002兲128:7共697兲
CE Database keywords: Leakage; Pipelines; Damping; Transient flow.

Introduction
Leakage from pipelines has the potential to cause significant en-
vironmental damage and economic loss. While pipelines are de-
signed and constructed to maintain their integrity, it is difficult to
avoid the occurrence of leakage in a pipeline system during its
lifetime 共Hovey and Farmer 1999兲. Often, accurate leak detection,
enabling a quick response, is necessary to minimize damage.
Leak detection methods previously proposed are: reflected wave
or timing methods 共Brunone 1999; Covas and Romas 1999兲; vol-
ume balance methods 共Griebenow and Mears 1989, Liou 1994兲;
pressure or flow deviation methods 共Griebenow and Mears 1989;
Liou and Tian 1995兲; acoustic methods 共Fuchs and Riehle 1991兲;
pig-based monitoring and on-line surveillance methods 共Black
1992; Weil et al. 1994; Furness and Reet 1998兲; frequency analy-
sis methods 共Jönsson and Larson 1992; Mpesha et al. 2001兲; in-
1
Postgraduate Student, Dept. of Civil & Environmental Engineering,
Univ. of Adelaide, Adelaide SA 5005, Australia. E-mail:
xwang@civeng.adelaide.edu.au
2
Senior Lecturer, Dept. of Civil & Environmental Engineering,
Univ. of Adelaide, Adelaide SA 5005, Australia. E-mail:
mlambert@civeng.adelaide.edu.au
3 leak occurrence, not by trying to model a transient event, as is
Associate Professor, Dept. of Civil & Environmental Engineering,
Univ. of Adelaide, Adelaide SA 5005, Australia. E-mail:
asimpson@civeng.adelaide.edu.au
4
Professor Emeritus, School of Civil & Environmental Engineering,
Cornell Univ., Ithaca, NY 14853-3501 共corresponding author兲. E-mail:
jal8@cornell.edu
5
Research Associate, Dept. of Civil & Environmental Engineering,
Univ. of Adelaide, Adelaide SA 5005, Australia. E-mail:
jvitkovs@civeng.adelaide.edu.au
Note. Discussion open until December 1, 2002. Separate discussions
must be submitted for individual papers. To extend the closing date by
one month, a written request must be filed with the ASCE Managing
Editor. The manuscript for this paper was submitted for review and pos-
sible publication on March 12, 2001; approved on February 13, 2002.
This paper is part of the Journal of Hydraulic Engineering, Vol. 128,
No. 7, July 1, 2002. ©ASCE, ISSN 0733-9429/2002/7-
697–711/$8.00⫹$.50 per page.
verse techniques 共Pudar and Liggett 1992; Liggett and Chen
1994; Vı́tkovský 2001兲; and a genetic algorithm method 共Vı́tk-
ovský et al. 2000兲. However, no single method can always meet
operational needs from an accuracy and cost point of view 共Fur-
ness and Reet 1998兲. Each of these leak detection techniques has
its advantages and disadvantages in different circumstances. Liou
共1998兲 used a pseudorandom binary signal 共PRBS兲 sequence as a
transient tool and showed that change of the spatial damping of
the PRBS sequence along the pipeline can be used to detect a
leak. A leak detection, location and quantification method that
uses damping of a transient event by a leak is presented in this
paper.
Transient response of a pipeline with a distributed leak was
investigated by Wiggett 共1968兲. He found that the transients in a
pipeline were greatly affected by the magnitude of the distributed
lateral flow. To investigate the effects of the demands on the tran-
sients in a field pipe network test, McInnis and Karney 共1995兲
used a similar distributed leak model based on the method of
characteristics. Recent experimental and numerical work at the
Univ. of Adelaide has demonstrated that attenuation of transients
in a pipeline due to a small leak is significant, as shown in Fig. 1.
That observation has lead to this work, which attempts to detect
required in inverse transient analysis 共Liggett and Chen 1994兲,
but by analyzing transient damping or decay of a pressure signal,
a much simpler process. The technique does not require rigorous
determination and modeling of boundary conditions and other
transient behavior. It follows the approach used in pressure mea-
surement where the 共often large兲 common-mode pressure is re-
moved through the use of differential rather than absolute pres-
sure measurement devices. Baseline studies of pipelines that
appear to be leak free can be used to increase the accuracy of the
process, but are not necessary to apply the basic method. Tran-
sient pressure waves are used for the attenuation study because
measurement of pressure in pipelines is more accurate 共and con-
siderably less expensive兲 than measurement of flow.
In the following section, the governing equations for unsteady
pipeline flow with a leak are derived and nondimensional param-

JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 697

 ␦ 共 x⫺x L 兲 ⫽ 再 ⬁
0
if x⫽x L
otherwise

and

lim
␧→0
冕 x L ⫹␧

x L ⫺␧
␦ 共 x⫺x L 兲 dx⫽1 (3)

where ␧⫽a small distance on the either side of the leak. Note that
␦(x⫺x L ) has dimension of length⫺1. Considering the compress-
ibility of the water and the elasticity of the pipe wall with some
simplifications 共Wylie and Streeter 1993兲, Eq. 共2兲 is expressed in
the more usual water-hammer-equation form

⳵H Q ⳵H a 2 ⳵Q a 2
⫹ ⫹ ⫹ Q ␦ 共 x⫺x L 兲 ⫽0 (4)
⳵t A ⳵x gA ⳵x gA L
Fig. 1. Damping caused by presence of leak on pipeline transient in
in which H⫽piezometric head; Q⫽flow rate in the pipeline; a
laboratory test
⫽wave speed in the fluid; and g⫽gravitational acceleration.
Similarly, conservation of momentum for a leak that emits fluid
normal to the pipe axis is
eters describing the behavior of pipeline transients are estab-
lished. Analytical solutions for friction damping and leak damp- ⳵H 1 ⳵Q Q ⳵Q f Q2 QQ L ␦ 共 x⫺x L 兲
⫹ ⫹ ⫹ 2⫺ ⫽0 (5)
ing are then obtained from the linearized equations. The leak ⳵x gA ⳵t gA ⳵x 2DgA
2
gA 2
detection method is developed in the next section followed by
two numerical examples. Finally, results of experimental tests are where f ⫽friction factor and D⫽pipe diameter. The last term in
presented. Eq. 共5兲 is caused by the mass flow discontinuity at the leak in the
pipeline. Eq. 共5兲 assumes that pipe friction during a transient
event is described by a constant steady-state Darcy–Weisbach
Governing Equations friction factor, which is a common assumption. However, pipe
friction during unsteady events can be significantly larger than
A control volume located between points 1 and 2 in Fig. 2 is used that predicted by the Darcy–Weisbach equation. The effects of
for the derivation of the unsteady flow equations 共continuity and unsteady friction are considered later in the paper.
momentum兲 with leakage. The pipe is considered to be horizontal Leak discharge is a function of pressure in a pipe and size of a
with a leak located as shown in Fig. 2, where Q L is the total leak and is expressed by the orifice equation
discharge out of the leak located at x⫽x L .
Adapting the nonleak equation of Wylie and Streeter 共1993兲, Q L ⫽C d A L 冑2g⌬H L (6)
conservation of mass in the control volume gives
where ⌬H L ⫽H L ⫺z L ⫽pressure head at a leak 共assuming pres-
⳵ ⳵ sure outside of the pipe is atmospheric兲; H L ⫽piezometric head in
共 ␳A 兲 ⌬x⫹ 共 ␳AV 兲 ⌬x⫽⫺␳Q L (1)
⳵t ⳵x the pipeline at the leak; z L ⫽pipe elevation at the leak; C d ⫽leak
where x⫽distance along the pipeline; t⫽time; ␳⫽fluid density; discharge coefficient; and A L ⫽leak area. The following dimen-
A⫽cross-sectional area of the pipe; V⫽velocity of flow; and sionless quantities are used to nondimensionalize Eqs. 共4兲, 共5兲,
Q L ⫽the flow rate of the leak. Dividing Eq. 共1兲 throughout by ⌬x and 共6兲:
and letting ⌬x approach zero gives
H t x Q
⳵ ⳵ H *⫽ , t *⫽ , x *⫽ , Q *⫽
共 ␳A 兲 ⫹ 共 ␳AV 兲 ⫽⫺␳Q L ␦ 共 x⫺x L 兲 (2) H1 L/a L Q0
⳵t ⳵x
and
The Dirac delta function is defined as
␦ 共 x * ⫺x L* 兲 ⫽␦ 共 x⫺x L 兲 L (7)

in which H 1 ⫽reference piezometric head 共e.g., the head at a

tank兲; L⫽pipe length; and Q 0 ⫽reference flow rate. Substituting
Eq. 共6兲 and applying the dimensionless quantities in Eq. 共7兲 to
Eqs. 共4兲 and 共5兲 gives

⳵H * V 0 Q * ⳵H * aQ 0 ⳵Q *
⫹ ⫹
⳵t * a ⳵x * gAH 1 ⳵x *

aQ 0 C d A L
⫹ 冑2gH 1 冑⌬H L* ␦ 共 x * ⫺x L* 兲 ⫽0 (8)
gAH 1 Q 0
Fig. 2. Pipe section with leak

698 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002

 gAH 1 ⳵H * ⳵Q * V 0 ⳵Q * f LQ 0 and using
⫹ ⫹ Q* ⫹ 共 Q*兲2
Q 0 a ⳵x * ⳵t * a ⳵x * 2DAa 1 ⳵q * ⳵h * h*
⫽⫺ ⫺M ␦ 共 x * ⫺x L* 兲 (16)
V 0 C d A L 冑2gH 1 F ⳵x * ⳵t * 2 冑⌬H L0
*
⫺ Q * 冑⌬H L* ␦ 共 x * ⫺x L* 兲 ⫽0 (9)
a Q0 from the continuity equation, Eq. 共14兲, results in
Because V 0 /a is normally small, the second term in Eq. 共8兲 and
the third and the last terms in Eq. 共9兲 can be neglected. The
dimensionless equations become
⳵ 2h * ⳵ 2h *
⳵x * 2⫽
⳵t * 2 ⫹ 2R⫹M 冋
␦ 共 x * ⫺x L* 兲 ⳵h *
2 冑⌬H L0 * ⳵t *
册
⳵H * 1 ⳵Q *
⫹ ⫹M 冑⌬H L* ␦ 共 x * ⫺x L* 兲 ⫽0 (10) h * ␦ 共 x * ⫺x L* 兲
⳵t * F ⳵x * ⫺2RM (17)
2 冑⌬H L0
*
⳵H * ⳵Q *
F ⫹ ⫹RQ * 2 ⫽0 (11) Eq. 共17兲 simplifies to
⳵x * ⳵t *
⳵ 2h * ⳵ 2h * ⳵h *
2⫽ 2 ⫹ 关 2R⫹F L ␦ 共 x * ⫺x L
where * 兲兴
⳵x * ⳵t * ⳵t *
f LQ 0 C dA L 2a H1
R⫽ , M⫽ , F⫽ ⫺2RF L ␦ 共 x * ⫺x L* 兲 h * (18)
2aDA A 冑2gH 1 HJ
and H J ⫽aV 0 /g⫽the Joukowsky pressure head rise, resulting in which F L ⫽M /2冑⌬H L0 * is the leak parameter. Since ⌬H L0
*
from an instantaneous reduction of velocity V 0 to zero. The di- ⫽(H L0 ⫺z L )/H 1 , if z L ⫽0, the leak parameter is
mensionless quantities R, M, and F are used to characterize the C dA L 2a
leak problem.
A 冑2gH 1 C dA L a
F L⫽ ⫽
冑
(19)
Linearized Solutions H L0 A 冑2gH L0
2
H1
Expressing H * and Q * as steady-state values plus small transient
quantities gives where H L0 ⫽the steady-state piezometric head at the leak.
Consider a pipeline connecting two reservoirs with constant
H * ⫽H *
0 ⫹h * , Q * ⫽Q *
0 ⫹q * (12) water elevations. The boundary conditions are
where h * ⫽nondimensional head deviation; H 0* ⫽nondi- h * 共 0,t * 兲 ⫽0
mensional steady head; q * ⫽nondimensional flow deviation; and
and
Q*0 ⫽nondimensional steady flow. When only linear terms are
retained, the square root in Eq. 共10兲 is expressed as h * 共 1,t * 兲 ⫽0 (20)
冑⌬H L* ⫽ 冑H L* ⫺z L* ⫽ 冑H L0
* ⫹h * ⫺z L* Alternatively, the problem of a pipeline connecting an upstream
reservoir and a downstream valve can be considered by using a
h* mirrored imaginary pipeline. The application of this technique is
⫽ 冑⌬H L0
* ⫹h * ⬇ 冑⌬H L0
*⫹ (13)
2 冑⌬H L0
* presented in the second numerical example later in the paper.
If a known transient is initiated in the pipeline, the initial con-
where ⌬H L* ⫽H L* ⫺z L* ⫽dimensionless pressure head at the leak; ditions are given as
H L* ⫽dimensionless piezometric head in the pipeline at the
h * 共 x * ,0兲 ⫽ f 共 x * 兲
leak; z L* ⫽dimensionless elevation at the leak; and H L0 *
⫽dimensionless steady-state piezometric head at the leak. The and
dimensionless linearization error in Eq. 共13兲 is
⳵h * 共 x * ,0兲

e T⫽
冉
冑⌬H L* ⫺ 冑⌬H L0
*⫹
h*
2 冑⌬H L0
* 冊 ⳵t *
⫽g 共 x * 兲

in which f (x * ) and g(x * ) are known piecewise continuous func-

(21)

冑⌬H L* tions in the range of 0⭐x * ⭐1. The head variation in the pipeline
is obtained by solving Eq. 共18兲 subject to the boundary and initial
Substituting Eqs. 共12兲 and 共13兲 into Eqs. 共10兲 and 共11兲 and ne- conditions of Eqs. 共20兲 and 共21兲. By applying a Fourier expan-
glecting e T and (q * ) 2 terms yields sion, the solution to Eq. 共18兲 is
⬁
⳵h * 1 ⳵q * h*
⫹
⳵t * F ⳵x *
⫹M
2 冑⌬H L0
*
␦ 共 x * ⫺x L* 兲 ⫽0 (14) h * 共 x * ,t * 兲 ⫽ 兺
n⫽1
兵 e ⫺ 共 R⫹R nL 兲 t *

⳵h * ⳵q * ⫻ 关 A n cos冑共 n␲ 兲 2 ⫺4RR nL ⫺ 共 R⫹R nL 兲 2 t *

F ⫹ ⫹2Rq * ⫽0 (15)
⳵x * ⳵t *
⫹B n sin冑共 n␲ 兲 2 ⫺4RR nL ⫺ 共 R⫹R nL 兲 2 t * 兴
Applying the operation
⳵
⳵t *
关 Eq. 共 14兲兴 ⫺
⳵x * F 冋
⳵ Eq. 共 15兲
册 ⫻sin共 n␲x * 兲 其
in which R nL ⫽F L sin (n␲xL*) or2
(22)

JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 699

 C dA L a In Eq. 共23兲 the leak-induced damping coefficient R nL depends
R nL ⫽ sin2 共 n␲x L* 兲 共 n⫽1,2,3,...兲
A 冑2gH L0
where R nL ⫽leak-induced damping factor for component n, in
which x L* ⫽dimensionless location of the leak along the pipeline.
Since values of R and R nL are normally much smaller than unity,
Eq. 共22兲 is approximated as
⬁ sient event.
h * 共 x * ,t * 兲 ⫽ 兺
n⫽1
兵 e ⫺ 共 R⫹R nL 兲 t *
⫻ 关 A n cos共 n␲t * 兲 ⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其
The Fourier coefficients A n and B n are
冕 1
A n ⫽2 f 共 x * 兲 sin共 n␲x * 兲 dx * 共 n⫽1,2,3,...兲
0 E (i)
B n⫽
2
n␲ 冕 0
1
g 共 x * 兲 sin共 n␲x * 兲 dx * ⫹
共 R⫹R nL 兲 A n
n␲
共 n⫽1,2,3,...兲
Note that the friction damping coefficient R(⫽ f LQ 0 /2aDA) of
Eq. 共24兲 does not depend on n provided f is constant. For this
case, the Fourier components are damped exponentially by fric-
tion, and that damping is equal for all components. In fact, e ⫺Rt *
can be taken outside of the summation in Eq. 共24兲. In contrast, the
leak-induced damping factor R nL of Eq. 共24兲 depends on n and is
different for each component; it cannot be removed from the sum-
mation sign in Eq. 共24兲. A separate analysis, not reported in this
paper, found that leak damping is approximately exponential
when applied to the entire transient. Eq. 共24兲 indicates that leak
damping is exactly exponential when applied to a distinct Fourier
component.
Application to Leak Detection
The solution in Eq. 共24兲 shows that any measured pipeline tran-
sient is the summation of a series of harmonic components that
are each exponentially damped with damping rate R⫹R nL (n
⫽1,2,3,...). If unsteady friction effects are negligible, the friction
damping factor R is a function of steady flow conditions only and
its value is known. Alternatively, the value can also be experi-
mentally determined by measuring the damping rate of a transient
in a leak-free pipeline. Therefore, for a measured pipeline tran-
sient, if the damping rate R⫹R nL for an individual component
can be obtained, the leak-induced damping factor R nL can be
found by subtracting R from R⫹R nL .
Two types of measurement data, space domain and time do-
main, can be used for calculation of the damping coefficient R
⫹R nL . Space-domain data are obtained by measuring the tran-
sient pressure head at a number of locations along the pipeline at
a single time 关in Eq. 共24兲 t * ⫽constant and 0⬍x * ⬍1兴. Time-
domain data are obtained by measuring the transient pressure his-
tory at a single pipeline location 关in Eq. 共24兲 x * ⫽constant and
t * ⬎0兴. Since measurement of pressure history at a specific loca-
tion is easier, time-domain data are used for calculation of damp-
ing rates. A separate study has shown that the damping rates of
the harmonic components obtained using both space-domain and
time-domain data are identical. The feasibility of analyzing pipe-
line transient period by period is presented in the Appendix.
700 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002
(23) on leak size C d A L , location of a leak x L* , and harmonic compo-
nent n. An algorithm is developed in this section to locate a leak
utilizing the different damping of separate harmonics and to quan-
tify a leak using the magnitude of damping. The process of leak
detection, location, and quantification is as follows:
1. Set up steady flow in the pipeline and then introduce a tran-
2. Measure the variation of pressure head with time at one or
more points along the pipe.
3. Divide the pressure trace into separate periods 共Period 1,
Period 2, etc. as shown in Fig. 14 of the Appendix兲 so that
each can be analyzed individually.
(24) 4. Using a Fourier transformation, such as the discrete or fast
Fourier transforms 共Press et al. 1992兲, decompose one period
of the transient into its separate harmonic components and
(25) calculate the amplitude of each component. The amplitude
n of the nth harmonic for the ith period is expressed as
E共ni兲⫽E共n1兲e⫺共R⫹RnL兲共i⫺1兲T* (27)
where
(26)
e⫺共R⫹RnL兲共t*0 ⫹T*兲⫺e⫺共R⫹RnL兲t*0
E共n1兲⫽⫺ sin共n␲x*兲冑A 2n ⫹B 2n
共R⫹RnL兲T *
is the amplitude of nth harmonic component at the first pe-
0 ⫽dimensionless starting time of the analysis; and
riod; t *
T * ⫽dimensionless period, defined as T * ⫽T/(L/a), in
which T⫽the natural period of the pipeline. For the
reservoir–pipeline–reservoir problem T * ⫽2.0, and for the
reservoir–pipeline–valve problem, T * ⫽4.0. Details of deri-
vation of Eq. 共27兲 are given in the Appendix.
5. Repeat step 4 period by period along the pressure trace.
6. For each component n, plot the amplitude E (i) n expressed in
Eq. 共27兲 versus period in terms of L/a. Compute the damp-
ing coefficient R⫹R nL from the plotted data using an expo-
nential fitting function in the form of Eq. 共27兲 where both
initial amplitude E (1)
n and the damping coefficient R⫹R nL
can be calculated.
7. Analyze the damping rates of the separate components to
determine occurrence, location, and magnitude of the leak.
Determining existence, location, and size of a leak using damping
rates of separate components is now considered in detail.
Presence of Leak
For a pipeline without a leak, C d A L ⫽0 in Eq. 共23兲 and hence
R nL ⫽0; damping of each component is independent of compo-
nent number n and is only dependent on friction damping factor
R. Therefore, given the same steady flow conditions followed by
a transient event, the presence of a leak is indicated by: 共1兲 the
damping rates R⫹R nL of the decomposed harmonic components,
determined in step 6, are significantly different from each other,
and 共2兲 the damping rates for some components are larger than
the friction-damping factor R.
Leak-induced damping depends strongly on the position of a
leak and on n through a sine squared function in Eq. 共23兲. Differ-
ent components have different responses to a leak. For example,
for a leak located in the middle of a pipeline (x L* ⫽0.5), compo-
nents n⫽1 and 3 have the maximum response, whereas the re-
sponse of component n⫽2 is zero. Therefore, in practice more
than one harmonic component should be used to detect a leak.
Fig. 3 shows the relative response of the first three harmonics
 Fig. 4. Damping ratios 关Eq. 共29兲兴 of harmonic components

Fig. 3. Sensitivity of leak position on different harmonic components
(n⫽1, 2, and 3兲 to the different leak locations along the pipeline.
leak locations. For example, a damping ratio of R 3L /R 1L ⬍1.0
corresponds to four possible leak locations. Therefore, only har-
monic components of n⫽1,2,3 are used for leak detection analy-
sis in this study.

Location of Leak
Size of Leak
Applying a Fourier transform to a measured pipeline transient,
Once the position of a leak has been determined, the magnitude of
leak location is calculated from the ratios of the different damping
a leak can be easily calculated using Eq. 共28兲. It is
rates of a pair of harmonic components. Consider n⫽n 1 and n
⫽n 2 , for which leak damping factors are R nL A 冑2gH L0
C dA L⫽ 共 n⫽1,2,3,...兲 (30)
C dA La a sin2 共 n␲x L* 兲
R n1L⫽ sin2 共 n 1 ␲x L* 兲
A 冑2gH L0 where n⫽any one of the components. Theoretically, leak magni-
tude calculated using different components should be the same.
and Different measurement positions and different forms of transients
C dA La can be used for added confirmation and to increase accuracy, if
R n2L⫽ sin2 共 n 2 ␲x L* 兲 (28) necessary.
A 冑2gH L0
The ratio of these two terms is Sensitivity Analysis
R n2L sin2 共 n 2 ␲x L* 兲 In the initial derivation, q * 2 was neglected and the orifice equa-
⫽ (29) tion was linearized in Eq. 共13兲 in arriving at Eqs. 共14兲 and 共15兲.
R n1L sin 共 n 1 ␲x L* 兲
2
Reconsidering the equations without linearization gives
which is a function only of leak location x L* and not a function of
⳵ 2h * ⳵ 2h * ⳵h *
the leak size C d A L .
2⫽ ⬘ ␦ 共 x * ⫺x L* 兲兴
2 ⫹ 关 2R ⬘ ⫹F L
For a measured pipeline transient, the damping rate R⫹R nL ⳵x * ⳵t * ⳵t *
for each harmonic component n can be calculated from its ampli-
⫹2R ⬘ F L⬘ h * ␦ 共 x * ⫺x L* 兲 ⫹E␦ 共 x * ⫺x L* 兲 ] (31)
tude E (i)
n by analyzing each period. Since friction damping factor
can be calculated from steady flow conditions, which are nor- where F L⬘ ⫽F L (1⫹e T ); R ⬘ ⫽R(1⫹q * ); and E⫽2e T R ⬘ M 冑H L0
*.
mally known, leak-induced damping R nL for any component n is Assuming a constant q * , an approximate solution for Eq. 共31兲 is
easily obtained by subtraction, giving the ratio of any two leak- ⬁
induced damping rates as in Eq. 共29兲. The solution of Eq. 共29兲 for
x L* yields leak location. Fig. 4 is a plot of the theoretical relation-
h * 共 x * ,t * 兲 ⫽ 兺
n⫽1
⬘ 兲t*
兵 e ⫺ 共 R ⬘ ⫹R nL
ship between the damping ratios of harmonic components n 2
⫽2, n 1 ⫽1 and harmonic components n 2 ⫽3, n 1 ⫽1 with the cor- ⫻ 关 A n cos共 n␲t * 兲 ⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其
responding leak locations in a pipeline.
Due to the symmetric nature of the sine squared function, the (32)
relationship between the damping ratio of two harmonic compo- in which R nL ⬘ ⫽F L⬘ sin (n␲xL*)⫽(1⫹eT)RnL . From Eq. 共32兲, leak
2

nents and leak location is not unique. Two leak locations corre-
spond to one value of the damping ratio R 2L /R 1L except for x L*
⫽0.5. For damping ratios of higher harmonic components, one
damping ratio may correspond to a greater number of possible
damping is only influenced by the value of e T , which is the
linearization error of the orifice equation. Neglecting q * 2 has no
direct influence on the location and quantification of a leak. No-
⬘ is
ticing that the ratio of any two leak damping coefficients of R nL

JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 701

 Fig. 6. Sensitivity of leak location on sensitivity parameter S
Fig. 5. Influence of linearization of orifice equation on leak size

independent of parameter e T , then the location of a leak is not sin2 共 n 2 ␲x L* 兲

affected by the error in the linearization of the orifice equation. ⫹S
sin2 共 n 1 ␲x L* 兲 sin2 关 n 2 ␲ 共 x L* ⫹␧ L 兲兴
The error of leak size induced by the orifice linearization is de- ⫽ (36)
fined as 1⫹S sin2 关 n 1 ␲ 共 x L* ⫹␧ L 兲兴

共 C d A L 兲 real⫺ 共 C d A L 兲 apparent where S⫽Rq * /R n 1 L . The leak location error ␧ L is a function of

␧ s⫽ (33) parameter S and real leak location x L* . Variation of ␧ L with pa-
共 C d A L 兲 real
rameter S and leak location x L* is presented in Fig. 6 if the first
Substituting Eq. 共30兲 into Eq. 共33兲 gives two Fourier components, n 1 ⫽1 and n 2 ⫽2, are used. Due to sym-
R nL ⫺ 共 1⫹e T 兲 R nL metry, only half of the pipeline is plotted. Fig. 6 indicates that
␧ s⫽ ⫽⫺e T large values of parameter S cause large errors in the location of
R nL
the leak, and leaks at different locations have different sensitivi-
冑1⫹h * /H L0
* ⫺ 共 1⫹0.5h * /H L0
*兲 ties to the parameter S. Leaks close to x L* ⫽0.33 are least influ-
⫽⫺ (34) enced by error in the friction damping when the first two harmon-
冑1⫹h * /H L0
* ics are used.
Variation of relative leak size error with parameter of h * /H L0* is Leak size error, assuming that pressure at the leak is little
*
plotted in Fig. 5. To avoid negative pressure in a pipeline, h * /H L0 influenced by leak location error, gives
must be less than 1.0. Within this range, the error in the size of a 共 C d A L 兲 real⫺ 共 C d A L 兲 apparent
leak caused by the orifice linearization is less than 6% and thus ␧ s⫽
共 C d A L 兲 real
not significant.
Neglecting the q * 2 term has no direct influence on location 共 1⫹S 兲 sin2 共 n␲x L* 兲
and quantification of a leak, but it does affect friction damping. ⫽1⫺ (37)
sin2 关 n␲ 共 x L* ⫹␧ L 兲兴
Because leak damping may be obtained by measuring total damp-
ing and subtracting friction damping, linearization may indirectly
lead to an experimental error in leak damping. That error shows
up in the ratio of the Fourier components and thus influences the
calculation of leak location.
By including the term of q * 2 , the total damping is R nL
⫹R(1⫹q * ). If leak damping is obtained by subtracting friction
damping R that is calculated from the steady state, the calculated
leak damping coefficient is R nL ⫹Rq * , in which R nL is the real
leak damping. Then the ratio of leak damping of any two Fourier
components is
R n 2 L ⫹q * R sin2 关 n 2 ␲ 共 x L* ⫹␧ L 兲兴
⫽ (35)
R n 1 L ⫹q * R sin2 关 n 1 ␲ 共 x L* ⫹␧ L 兲兴
in which ␧ L ⫽dimensionless distance away from a real leak loca-
tion. Substituting
R n2L sin2 共 n 2 ␲x L* 兲
⫽
R n1L sin2 共 n 1 ␲x L* 兲
Fig. 7. Sensitivity of leak size on sensitivity parameter S
into Eq. 共35兲 and rearranging gives

702 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002

 Fig. 8. Pipeline connecting two reservoirs

Fig. 9. Fourier series analysis of transients measured from pipeline without leak 共Case 1兲 and with leak 共Case 2兲 of C d A L /A⫽0.1% at x L*
⫽0.25 (T * ⫽2.0): 共a兲 time history of measured pipeline transients 共method of characteristics兲; 共b兲 Fourier series analysis of first period transient
(0⬍t * ⬍2.0); and 共c兲 damping of harmonic components with period 共period number i⫽1,2,3,...兲

JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 703

Table 1. Results of Fourier Transform Analysis on Transients Presented in Numerical and Experimental Examples
Harmonic components
Examples Cases n⫽1 n⫽2 n⫽3
Reservoir-pipeline-reservoir problem 共Fig. 9兲 Case 1 共no leak兲 R 0.0742 0.0742 0.0742
Case 2 共with a leak兲 R⫹R nL 0.0991 0.1232 0.0992
Leak damping R nL 0.0249 0.0490 0.0250

Reservoir-pipeline-valve problem 共Fig. 11兲 Case 3 共no leak兲 R 0.0022 N/Aa 0.0022
Case 4 共with a leak兲 R⫹R nL 0.0088 N/Aa 0.0410
Leak damping R nL 0.0066 N/Aa 0.0388

Experimental verification 共Fig. 13兲 Test I 共no leak兲 R 0.0244 0.0382 0.0563
Test II 共with a leak兲 R⫹R nL 0.0624 0.1180 0.0891
Leak damping R nL 0.0380 0.0798 0.0328
a
Even harmonic components are equal to zero for the reservoir–pipeline–valve problem.

Variation of relative leak size error ␧ s with parameter S and leak simulated transients calculated from the MOC. Two types of
location x L* is presented in Fig. 7. Since leaks close to the ends of problems, a reservoir–pipeline–reservoir system and a reservoir–
the pipeline cause less damping, a small error in the value of leak pipeline–valve system, are considered.
damping can cause a large error in the calculation of the leak size
despite small S. In the application of the proposed leak detection
Reservoir – Pipeline – Reservoir Problem
method, small values of parameter S can be achieved by using
small magnitudes of pipeline transient or small values of friction In this case a transient in the pipeline is initiated by closing a
damping. The latter is achieved by using a low steady-state flow side-discharge valve located 750 m (x * ⫽0.75) away from the
rate in the pipeline. Alternatively, friction damping can be ob- upstream reservoir as shown in Fig. 8, while the valve at the
tained from a test in a leak-free pipeline or calculated using a downstream reservoir is fully open with negligible head loss. The
numerical model. An example of such a situation is given in the magnitude of the side-discharge valve coefficient is C d A S /A
second numerical application later in the paper. ⫽0.001, where A S ⫽area of the side-discharge valve. The closing
time of the valve is 0.05 s. For the first test 共Case 1兲 the leak at
x L* ⫽0.25 is removed (C d A L /A⫽0) while for Case 2 a leak of
Unsteady Friction Damping
relative size of C d A L /A⫽0.001 is assumed. The transient head in
In the derivation of Eq. 共5兲, friction loss was assumed to be rep- the pipeline is calculated numerically by a standard MOC pro-
resented by a steady-state friction relationship. For rapidly vary- gram using 16 pipe reaches. The calculations are started from
ing flow, experiments have shown that damping of transients is steady state. The Darcy–Weisbach friction factor is calculated as
greater than that predicted by the Darcy–Weisbach head loss f ⫽0.015 using the Swamee–Jain formula. The steady-state
equation. This difference can be addressed using unsteady friction friction-damping factor is R⫽0.0742 共also R s 兲 with a steady-flow
models such as those by Zielke 共1968兲, Brunone et al. 共1991兲, Reynolds number of 3.96⫻105 . The sensitivity parameter S is
Vardy and Hwang 共1991兲, and Bergant et al. 共2001兲. Applying an calculated as 0.0063. Based on previous sensitivity analysis, the
unsteady friction model, the total friction factor f is expressed as leak location error ␧ L and leak size error ␧ s are both less than 1%.
Fig. 9 presents the leak detection analysis using the ‘‘mea-
f ⫽ f s⫹ f u (38)
sured’’ transient. Measured pressures at x * ⫽0.75 are shown in
where f s ⫽the quasisteady contribution and f u ⫽an additional Fig. 9共a兲 for cases of no leak 共Case 1兲 and with a leak 共Case 2兲.
contribution due to unsteadiness. To account for the unsteady fric- The amplitudes for separate components n are obtained by apply-
tion damping, the friction-damping factor R in Eq. 共24兲 is re- ing a discrete Fourier transform algorithm to analyze the mea-
placed by sured data in Fig. 9共a兲, the results of which are presented in Fig.
R⫽R s ⫹R u (39) 9共b兲 for the first period signal (t * ⫽0.0– 2.0). The transient sig-
nals were analyzed period by period with an interval of t * ⫽2.0.
in which R s ⫽damping by steady friction and R u ⫽damping by Fig. 9共c兲 shows computed amplitudes of the Fourier series of the
unsteady friction. In contrast to steady friction damping R s the first three harmonic components plotted against period in terms of
value of unsteady friction damping R u is not constant and is dif- L/a. In Fig. 9共c1兲—the no leak case—the damping rate R⫹R nL
ferent for different Fourier components. For this case, the e ⫺Rt * for all three harmonic components is determined to be R
term in Eq. 共24兲 cannot be taken outside the summation sign. The
value of R u 共and hence R兲 can be determined by experimental
tests or by a numerical model that incorporates an appropriate Table 2. Results using Transients from Different Measurement Lo-
unsteady friction model, e.g., the modified Brunone model 共Vitk- cations (R⫽0.0742)-Case 1
ovský 2001兲. Measurement
position R 1L R 2L R 3L R 2L /R 1L R 3L /R 1L
x * ⫽0.375 0.0248 0.0489 0.0249 1.972 1.004
Numerical Verification
x * ⫽0.5 0.0247 0.0314 0.0247 1.271 1.000
x * ⫽0.625 0.0248 0.0489 0.0251 1.972 1.012
The leak detection method discussed above is tested numerically
x * ⫽0.75 0.0249 0.0490 0.0250 1.968 1.004
in an artificial pipeline, shown in Fig. 8, using the results of

704 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002

Fig. 10. Pipeline connecting an upstream reservoir and downstream valve, and added imaginary symmetric pipeline 共D⫽0.2 m, a
⫽1000 m/s, ␧⫽0.023 mm兲
⫽0.0742 by fitting Eq. 共27兲 to the decaying amplitudes. In Fig.
9共c2兲—in which a leak was present—the damping rates R⫹R nL
are 0.0991, 0.1232, and 0.0992 for the first three components. The
second value is significantly different from the first and third and
all are larger than the friction-damping factor. The leak-induced
damping rates R nL for components n⫽1, 2, and 3 are 0.0249,
0.0490, and 0.0250, which are obtained by subtracting R from
R⫹R nL 共see Table 1兲. The ratios of two leak-induced damping
rates defined in Eq. 共29兲 are R 2L /R 1L ⫽1.97 and R 3L /R 1L
⫽1.00. Using these two ratios in Fig. 4, corresponding leak loca-
tions are either x L* ⫽0.25 or x L* ⫽0.75, the former being the real
leak location. Applying Eq. 共30兲 based on either of the calculated
leak locations and R 1L ⫽0.0249, the calculated leak size is
C d A L /A⫽0.001, which is identical to the real magnitude of the
leak used to generate the MOC transient data.
Following the same procedure transient damping, measured at
different pipeline positions x * ⫽0.375, 0.5, and 0.625, was ana-
lyzed. Damping rates for separate harmonic components are pre-
sented in Table 2. The Fourier transform analysis of transients
measured at different locations give almost identical results for
the damping rate for each component except for R 2L at x * ⫽0.5,
where a large error is introduced since the amplitude of the com-
ponent n⫽2 is close to zero.
The results in Table 2 for R 1L and R 3L show that the analysis
can be performed for the measurement site located anywhere (0
⬍x * ⬍1) along the pipeline. The results of the Fourier series
analysis of numerical and experimental verifications presented in
Figs. 9, 12, and 13 are included in Table 1.
Reservoir – Pipeline – Valve Problem
A reservoir–pipeline–valve case provides an additional numerical
example. For the pipeline of 1000 m 共between x * ⫽0.0 and 1.0兲
in Fig. 10, a transient is initiated by closing the downstream
valve. The initial steady flow in the pipeline is Q 0 ⫽2.0 L/s,
which can be achieved by a partially opened downstream valve.
The Reynolds number of the flow in the pipeline is 11,160. The
Darcy–Weisbach friction factor is calculated as f ⫽0.0302 using
the Swamee–Jain formula and the friction damping factor R s
⫽0.0048. The leak detection method presented earlier in the
paper was developed from the general solution presented in Eq.
共24兲 based on boundary conditions of two constant-head reser-
voirs. Thus, it cannot be directly applied to this particular ex-
ample; however, it can be applied by adding an imaginary sym-
metric section 共dashed portion兲 to the original pipeline as shown
in Fig. 10.
Two cases, referred to as Cases 3 and 4, are considered. In
Case 3, no leak is present in the pipeline, and in Case 4, a leak of
C d A L /A⫽0.001 is present at x * ⫽0.25. The transient pressures,
generated by MOC for the original pipeline, are measured at x *
⫽0.75 for the two cases that are presented in Fig. 11共a兲. The
presence of the leak has very obvious effects on the transient
damping and shape of the pressure trace. Note that the period of
the transients is 4L/a for Cases 3 and 4, while for Cases 1 and 2
the transient period is 2L/a.
The amplitudes of different harmonic components of the first
period transient (t * ⫽0.0– 4.0) are given in Fig. 11共b兲 for Cases 3
and 4, respectively. Amplitudes of all even harmonic components
(n⫽2,4,6,...) are close to zero. Damping of the first two odd
harmonics, fitted using Eq. 共27兲, are presented in Fig. 11共c兲. Due
to the large magnitude of q * (q * ⫽1.0) in this case, sensitivity
parameter S is calculated as 0.35. Therefore, as stated in the sen-
sitivity analysis, the accuracy of leak detection will be signifi-
cantly affected if the value used for friction damping is incorrect.
As a result, the accurate value of the friction damping is obtained
from a test in a leak-free pipeline 共see Case 3兲. For Case 3, the
friction damping rates determined from a leak-free pipeline are
R⫽0.0022 for harmonic components n⫽1 and 3. For Case 4,
friction plus leak-induced damping rates for n⫽1 and 3 are R
⫹R 1L ⫽0.0088 and R⫹R 3L ⫽0.0410. Therefore, leak-induced
damping for harmonic components n⫽1 and 3 are R 1L ⫽0.0066
and R 3L ⫽0.0386 共see Table 1兲, and the damping ratio for the first
and the third harmonics is R 3L /R 1L ⫽5.879. The corresponding
leak positions are determined using Fig. 4 as x̂ * 1 ⫽0.124 or x̂ *
2
⫽0.876. Applied to the real pipeline, the leak position of x̂ * 1
⫽0.124 becomes x * ⫽0.248, which is close to the real leak loca-
tion of x * ⫽0.25. The other possible leak is located in the imagi-
nary symmetric section. As a result, the location detected using
this type of transient system is unique. Applying R 1L ⫽0.0066 and
x L* ⫽0.124 in Eq. 共30兲, the magnitude of the leak is calculated as
C d A L /A⫽0.0010, which is identical to the actual leak size
C d A L /A⫽0.001.
Experimental Verification
Experimental tests were conducted in a single pipeline in the
Robin Hydraulics Laboratory at the Univ. of Adelaide. The pipe-
line is a straight 37.2 m copper pipe with inner diameter of 22
mm between two pressurized tanks as shown in Fig. 12. Five
pressure transducers are located at equidistant points along the
pipeline and two 1/4-turn ball valves are installed at both ends for
JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 705
Fig. 11. Fourier series analysis of the transients measured from pipeline without leak 共Case 3兲 and with leak 共Case 4兲 of C d A L /A⫽0.1% at
x L* ⫽0.25 by closing downstream valve (T * ⫽4.0): 共a兲 time history of measured pipeline transients 共method of characteristics兲; 共b兲 Fourier series
analysis of first period transient (0⬍t * ⬍4.0); and 共c兲 damping of harmonic components with period 共period number i⫽1,2,3,...)

flow control. A side-discharge orifice, used to simulate a leak, is
installed at the 1/4 position 共point B兲 with exchangeable leak
orifice sizes of 1, 1.5, and 2 mm diameter. To initiate a transient,
another side-discharge valve is installed at point D. More details
of this experimental apparatus are found in Vı́tkovský 共2001兲.
Two tests were conducted. Test I is a no-leak case and in Test
II a 1 mm leak is located at point B 共Fig. 12兲. The flow conditions
are as follows: wave speed a⫽1,320 m/s, head at tank 1 H 1
⫽23.6 m, head at Tank 2 H 2 ⫽22.8 m, and lumped leak param-
eter C d A L ⫽5.00⫻10⫺7 m2 . The steady flow velocity in the pipe-
line is V 0 ⫽0.567 m/s, and the steady friction damping factor is
calculated as R s ⫽0.0109.
In Test I, the valves at locations A and E and the side-discharge
valve at D are opened and steady state achieved. The side-
discharge valve at D is then closed quickly. In Test II, valves at
locations A and E, the leak at B, and the side-discharge valve at D

706 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002


Fig. 12. Experimental pipeline apparatus
are open until steady state is obtained. The side-discharge valve at
D is then sharply closed. During the tests, pressure was measured
by five pressure transducers at points A, B, C, D, and E. Only the
measured pressures at point D (x * ⫽0.75) for Tests I and II are
used for leak detection analysis and they are plotted in Fig. 13共a兲.
The leak-induced damping of Test II is obvious compared to the
transient in Test I that has no leak.
Fig. 13共b兲 shows the computed amplitudes of different har-
monic components for the first period of the transient. The am-
plitude of each component is less in Test II than in Test I because
the magnitude of the transient in Test II is less than that in Test I
due to the energy loss at the leak.
Friction damping factors obtained from the no-leak case 共Test
I兲 are presented in Fig. 13共c兲. Damping factors of the first three
harmonic components (n⫽1,2,3) are R 1 ⫽0.0244, R 2 ⫽0.0382,
and R 3 ⫽0.0563 共see Table 1兲, all being larger than the steady
friction damping factor R s ⫽0.0109, calculated using steady state
friction. The differences between the measured and the calculated
damping values are due to unsteady friction. It accounts for 50,
71, and 80% of the total for the first three components. As a
result, despite a small value of sensitivity parameter S (S
⫽0.00036), the friction damping in this case cannot be calculated
from the steady-state conditions, and must be obtained either
from a leak-free measurement or from numerical analysis. As
previously indicated, values of unsteady friction damping are dif-
ferent for different Fourier components. In Fig. 13共c兲, regression
coefficients of the fitted curves of the damping factors are larger
than 0.99, experimentally confirming the analysis that damping
for each component including unsteady friction is exponential.
In Tests I and II, since the steady flow conditions were similar
and both the transients were initiated by closing the side-
discharge valve at approximately the same speed, steady and un-
steady friction effects are similar. Leak-induced damping rates for
the first three harmonic components (n⫽1,2,3) are R 1L
⫽0.0380, R 2L ⫽0.0798, and R 3L ⫽0.0328 共see Table 1兲. The ra-
tios of damping rates R 2L and R 1L , and R 3L and R 1L are
R 2L /R 1L ⫽2.10, and R 3L /R 1L ⫽0.863. Corresponding leak loca-
tions for these damping ratios are x L* ⫽0.242 共or x L* ⫽0.758兲 and
x L* ⫽0.255 共or x L* ⫽0.745兲 by applying these two ratios in Fig. 4.
Averaging of these results gives the location as x L* ⫽0.249, almost
the exact real location of the leak. Using R 1L ⫽0.0380, R 2L
⫽0.0801 and leak location x L* ⫽0.242 and x L* ⫽0.255, the mag-
nitude of the leak is calculated, from Eq. 共30兲, as C d A L ⫽4.912
⫻10⫺7 and C d A L ⫽4.917⫻10⫺7 m2 , which are about 1.7%
smaller than the real leak size of C d A L ⫽5.00⫻10⫺7 m2 .
Conclusions
Transients in pipelines are damped by both pipe friction and
leaks. Steady state friction damping is independent of the har-
monic component of the transient, but leak damping is different
for different Fourier components. This fact can be used success-
fully to detect presence of leaks, to compute their location, and to
find their size. The damping rate is useful for finding the quantity
of leak discharge and the ratio of damping rates between different
harmonic components is used to find leak location. Leaks of 0.1%
of a pipeline’s cross-sectional area or smaller can be detected and
located.
A linearized analysis of the governing equations indicates that
steady Darcy–Weisbach friction damping is exactly exponential,
and leak damping is exponential for each of the individual har-
monic components but only approximately exponential for an en-
tire transient. Sensitivity analysis shows that linearization gener-
ates an insignificant error in both leak location and quantification.
Inaccurate steady-state friction determination 共if it is used to find
the leak damping by subtraction from total damping兲, on the other
hand, may or may not be significant, depending on the parameters
of the pipeline and flow and the location of a leak. Also, if sub-
traction is used to find leak damping, the added damping caused
by unsteady flow may become important. Both nonlinear numeri-
cal analyses using the method of characteristics and laboratory
experiments have verified the accuracy of the linearized solution
in specific cases. The analysis of a linearized set of equations has
provided significant insight into a leak detection technique.
Leak detection techniques based on analyzing pressure mea-
surement during transient events in a pipe network have the ob-
vious advantage of being orders of magnitude less expensive than
field investigations. Although the leak detection technique pre-
sented in this paper is simple to use and apply, and has some
significant advantages for leak detection in pipelines, it does not
have the generality of some other methods 共e.g., the inverse tran-
sient method兲, and is not generally applicable to complex systems
such as pipe networks. Difficulties in complicated geometries in-
clude the complex waveforms created by branches and loops and
demands, which may be difficult to distinguish from leaks.
Acknowledgments
The writers thank Professor E. O. Tuck of the Department of
Applied Mathematics, the University of Adelaide for the valuable
discussion on the mathematical approach for the Dirac delta func-
tion. Financial support from the Australian Research Council to
the second, third, and fourth writers and scholarships provided by
the Australian Government and the University of Adelaide to the
first writer are gratefully acknowledged.
Appendix: Fourier Analysis of the Time-Domain
Pressure Variation Data
When a time-domain transient 共measured at a particular location
x * 兲 is divided into sections period by period as shown in Fig. 14,
each transient period can be expressed as a Fourier series based
on Eq. 共24兲.
For the first period P 1 :
⬁
h * 共 x * ,t * 兲 ⫽ 兺
n⫽1
兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲
⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其
for t 0* ⬍t * ⬍t *
0 ⫹T * (40)
JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 707
Fig. 13. Experimental transients and Fourier series analysis results: 共a兲 time history of experimental transients; 共b兲 Fourier series analysis results
of first period transient (0⬍t * ⬍2); and 共c兲 damping of harmonic components with periods 共T * ⫽2.0, and period number i⫽1,2,3,...兲

where t 0* ⫽starting time of analysis, and T * ⫽dimensionless pe- Setting t 2* ⫽t * ⫹T * and noticing the periodic property of the si-
riod defined as T * ⫽T/(L/a), in which T⫽natural period of pipe- nusoid functions give
line. For the case in Fig. 14, T * ⫽2.0.
⬁
For the second period P 2 :
⬁
h * 共 x * ,t * ⫹T * 兲 ⫽ 兺
n⫽1
e ⫺ 共 R⫹R nL 兲 T * 兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲

h * 共 x * ,t 2* 兲 ⫽ 兺
n⫽1
兵 e ⫺ 共 R⫹R nL 兲 t 2* 关 A n cos共 n␲t 2* 兲
⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其

⫹B n sin共 n␲t 2* 兲兴 sin共 n␲x * 兲 其 0 ⬍t * ⬍t *

for t * 0 ⫹T * (42)

0 ⫹T * ⬍t *
for t * 2 ⬍t *
0 ⫹2T * (41) For the ith period P i

708 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002

 冕
⬁
1 0 ⫹T *
兺
t*
C 共0i 兲 ⫽ e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T *
T* t*
0 n⫽1

⫻ 兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲

⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其 dt * (49)

Thus
⬁
1
C 共0i 兲 ⫽
T * n⫽1 兺
sin共 n␲x * 兲 e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T *

⫻ 冕 t 0* ⫹T *

t 0*
兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲

⫹B n sin共 n␲t * 兲兴 其 dt * (50)

Fig. 14. Time-domain pipeline transient 共generated by method of The integral term in Eq. 共50兲 is zero 共Abramowitz and Stegun
characteristics兲 1972兲. Therefore,
C 共0i 兲 ⫽0 (51)
Substituting the theoretical solution of h * (x * ,t * ) from Eq. 共44兲
⬁ into Eq. 共47兲 gives
h * 共 x * ,t i* 兲 ⫽ 兺 兵 e ⫺ 共 R⫹R nL 兲 t i* 关 A n cos共 n␲t i* 兲
冕
⬁
2
兺 e ⫺共 R⫹R
n⫽1 t 0* ⫹T *
C 共mi 兲 ⫽ nL 兲共 i⫺1 兲 T *
T* t 0* n⫽1
⫹B n sin共 n␲t i* 兲兴 sin共 n␲x * 兲 其

0 ⫹ 共 i⫺1 兲 T * ⬍t i* ⬍t *
for t * 0 ⫹iT * (43) ⫻ 兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲 ⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其

i ⫽t * ⫹(i⫺1)T * gives
Setting t * ⫻cos共 m␲t * 兲 dt * (52)
⬁
Thus
h * 共 x * ,t * ⫹ 共 i⫺1 兲 T * 兲 ⫽ 兺 e ⫺共 R⫹R
n⫽1
nL 兲共 i⫺1 兲 T *
⬁
2
⫻ 兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲
C 共mi 兲 ⫽
T* 兺
n⫽1
sin共 n␲x * 兲 e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T *

⫹B n sin共 n␲t * 兲兴 sin共 n␲x * 兲 其 ⫻ 冕 t*

0 ⫹T *
t*
兵 e ⫺ 共 R⫹R nL 兲 t * 关 A n cos共 n␲t * 兲
0 ⬍t * ⬍t *
for t * 0 ⫹T *
0
(44)
A similar Fourier series is now fitted period by period to the ⫹B n sin共 n␲t * 兲兴 其 cos共 m␲t * 兲 dt * (53)
‘‘measured’’ transient data given in Fig. 14 共created by MOC兲. The integral term in Eq. 共53兲 is zero except when n⫽m
For the ith period transient the fitted Fourier series is 共Abramowitz and Stegun 1972兲. Then Eq. 共53兲 is expressed as
⬁

h * 共 x * ,t * ⫹ 共 i⫺1 兲 T * 兲 ⫽C 共0i 兲 ⫹ 兺 关 C 共mi 兲 cos共 m␲t * 兲

C 共ni 兲 ⫽⫺
e ⫺ 共 R⫹R nL 兲共 t 0* ⫹T * 兲 ⫺e ⫺ 共 R⫹R nL 兲 t 0*
m⫽1
共 R⫹R nL 兲 T *
⫹D 共mi 兲 sin共 m␲t * 兲兴 ⫻sin共 n␲x * 兲 A n e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T * (54)
0 ⫹⬍t * ⬍t *
for t * 0 ⫹T * (45) Similarly by substituting Eqs. 共44兲 into 共48兲 the Fourier coeffi-
The Fourier coefficients C (i) , C m(i) and D m(i) are defined as cient D m(i) is expressed as
0

C 共0i 兲 ⫽
1
T* 冕 t 0*
0 ⫹T *
t*
h * 共 x * ,t * ⫹ 共 i⫺1 兲 T * 兲 dt * (46) D 共ni 兲 ⫽⫺
e ⫺ 共 R⫹R nL 兲共 t *0 ⫹T * 兲 ⫺e ⫺ 共 R⫹R nL 兲 t *0
共 R⫹R nL 兲 T *

C 共mi 兲 ⫽
2
T* 冕
t*
0
t 0* ⫹T *
h * 共 x * ,t * ⫹ 共 i⫺1 兲 T * 兲 cos共 m␲t * 兲 dt *
The amplitude E (i)
⫻sin共 n␲x * 兲 B n e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T * (55)

(47) n for a component n at the ith period P i is

D 共mi 兲 ⫽
2
T* 冕
t*
0
0 ⫹T *
t*
h * 共 x * ,t * ⫹ 共 i⫺1 兲 T * 兲 sin共 m␲t * 兲 dt * E 共ni 兲 ⫽⫺
e ⫺ 共 R⫹R nL 兲共 t 0* ⫹T * 兲 ⫺e ⫺ 共 R⫹R nL 兲 t 0*
共 R⫹R nL 兲 T *
(48)
⫻sin共 n␲x * 兲 冑A 2n ⫹B 2n e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T * (56)
Substituting the theoretical solution of h * (x * ,t * ⫹(i⫺1)T * )
from Eq. 共44兲 into 共46兲 gives or

JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 709

 E 共ni 兲 ⫽E 共n1 兲 e ⫺ 共 R⫹R nL 兲共 i⫺1 兲 T *
where
e ⫺ 共 R⫹R nL 兲共 t *0 ⫹T * 兲 ⫺e ⫺ 共 R⫹R nL 兲 t *0
E 共n1 兲 ⫽⫺
共 R⫹R nL 兲 T *
is the amplitude of the component n at the first period.
Each Fourier component is exponentially damped in time. The
damping rate of component n is (R⫹R nL ), which is the param-
eter to be determined.
Notation
The following symbols are used in this paper:
A ⫽ inner pipe cross-sectional area;
A L ⫽ leak area;
A S ⫽ area of side-discharge valve;
A n , B n ⫽ Fourier coefficients;
a ⫽ wave speed;
C ⫽ constant;
C d ⫽ leak orifice discharge coefficient;
C m(i) , D m(i) ⫽ Fourier coefficients of ith period transient;
0 ⫽ Fourier coefficient of ith period transient;
C (i)
D ⫽ diameter of pipe;
E ⫽ parameter defined as E⫽2e T R ⬘ M 冑H L0 *;
E (i)
n ⫽ amplitude of nth harmonic component of ith
period transient;
E (1)
n ⫽ amplitude of nth harmonic component of
first period transient;
e T ⫽ linearization error of orifice equation;
F ⫽ dimensionless head;
F L , F L⬘ ⫽ leak parameter;
f ⫽ friction factor;
f s ⫽ Darcy–Weisbach steady friction factor;
f u ⫽ unsteady friction factor;
g ⫽ gravitational acceleration;
H ⫽ piezometric head;
H * ⫽ dimensionless head⫽H/H 1 ;
H J ⫽ Joukowsky head rise;
H L ⫽ piezometric head at leak;
H L0 ⫽ steady piezometric head at leak;
* ⫽ dimensionless steady head at
H L0
leak⫽H L0 /H 1 ;
H 0 ⫽ steady state piezometric head;
H 1 ⫽ reference head in pipeline;
h * ⫽ dimensionless head disturbance;
L ⫽ length of pipeline;
M ⫽ leak parameter;
m, n ⫽ component number in Fourier series;
P n ⫽ nth period of transient;
p ⫽ pressure;
Q ⫽ flow rate;
Q * ⫽ dimensionless flow rate⫽Q/Q 0 ;
Q L ⫽ flow rate through leak;
Q 0 ⫽ steady state flow rate;
q * ⫽ dimensionless flow rate disturbance; friction
damping factor;
R, R ⬘ ⫽ pipeline friction damping factor;
R nL , R nL⬘ ⫽ leak damping factor for nth harmonic (n
⫽1,2,3,...);
R s ⫽ steady friction damping factor;
R u ⫽ unsteady friction damping factor;
710 / JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002
(57) S ⫽ parameter used in sensitivity analysis;
T ⫽ natural period of pipeline;
T* ⫽ dimensionless period of transient⫽T/(L/a);
t ⫽ time;
sin共 n␲x * 兲 冑A 2n ⫹B 2n t* ⫽ dimensionless time⫽t/(L/a);
t* ⫽ dimensionless reference time;
0
V ⫽ flow velocity in pipe;
V0 ⫽ steady flow velocity in pipe;
x ⫽ distance along pipeline;
x* ⫽ dimensionless distance⫽x/L;
xL ⫽ position of leak;
x L* ⫽ dimensionless leak position⫽x L /L;
x̂ *
1 ,x̂ *
2 ⫽ dimensionless leak position in combined
real and imaginary pipeline;
⌬H L ⫽ pressure head at leak⫽H L ⫺z L ;
⌬H L* ⫽ dimensionless pressure head at leak⫽H L*
⫺z L* ;
⌬x ⫽ distance interval;
␦ ⫽ Dirac delta function;
␧ ⫽ roughness height for pipe wall; small dis-
tance from leak;
␧L ⫽ dimensionless leak location error;
␧S ⫽ dimensionless leak size error; and
␳ ⫽ density of fluid.
References
Abramowitz, M., and Stegun, I. A. 共1972兲. Handbook of mathematical
functions with formulas, graphs, and mathematical tables, National
Bureau of Standards, Washington, D.C.
Bergant, T., Simpson, A. R., and Vı́tkovský, J. 共2001兲. ‘‘Developments in
unsteady pipe friction modelling.’’ J. Hydraul. Res., 39共3兲, 249–257.
Black, P. 共1992兲. ‘‘A review of pipeline leak detection technology.’’ Pipe-
line systems, B. Coulbeck and E. Evans eds., Kluwer Academic, Dor-
drecht, The Netherlands, 287–297.
Brunone, B. 共1999兲. ‘‘Transient test-based technique for leak detection in
outfall pipes,’’ J. Water Resour. Plan. Manage., 125共5兲, 302–306.
Brunone, B., Golia, U. M., and Greco, M. 共1991兲. ‘‘Modelling of fast
transients by numerical methods.’’ Proc., Int. Meeting on Hydr. Tran-
sients with Column Separation, IAHR, Valencia, Spain, 201–209.
Covas, D., and Romas, H. 共1999兲. ‘‘Leakage detection in single pipeline
using pressure wave behaviour.’’ Water industry system: modelling
and optimization application, Baldock, Hertfordshire, England, 287–
299.
Fuchs, H. V., and Riehle, R. 共1991兲. ‘‘Ten years of experience with leak
detection by acoustic signal analysis.’’ Appl. Acoust., 33共1兲, 1–19.
Furness, R. A., and Reet, J. D. 共1998兲. ‘‘Pipe line leak detection tech-
niques.’’ Pipe line rules of thumb handbook, E. W. McAllister, ed.,
Gulf, Houston, 476 – 484.
Griebenow, G., and Mears, M. 共1989兲. ‘‘Leak detection implementation:
modeling and tuning methods.’’ J. Energy Resour. Technol., 111, 66 –
71.
Hovey, D. J., and Farmer, E. J. 共1999兲. ‘‘DOT states indicate need to
refocus pipeline accident prevention.’’ Oil & Gas J., 97共11兲, 52–53.
Jönsson, L., and Larson, M. 共1992兲. ‘‘Leak detection through hydraulic
transient analysis.’’ Pipeline systems, B. Coulbeck and E. Evans, eds.,
Kluwer Academic, Dordrecht, The Netherlands, 273–286.
Liggett, J. A., and Chen, L.-C. 共1994兲. ‘‘Inverse transient analysis in pipe
networks.’’ J. Hydraul. Eng., 120共8兲, 934 –955.
Liou, C. P. 共1994兲. ‘‘Mass imbalance error of waterhammer equations and
leak detection.’’ J. Fluids Eng., 116, 103–109.
Liou, C. P. 共1998兲. ‘‘Pipeline leak detection by impulse response extrac-
tion.’’ J. Fluids Eng., 120, 833– 838.
Liou, C. P., and Tian, J. 共1995兲. ‘‘Leak detection-transient flow simulation
approaches.’’ J. Energy Resour. Technol. , 117, 243–248.
McInnis, D., and Karney, B. 共1995兲. ‘‘Transients in distribution networks:
field tests and demand models.’’ J. Hydraul. Eng., 121共3兲, 218 –231.
Mpesha, W., Gassman, S. L., and Chaudhry, M. H. 共2001兲. ‘‘Leak detec-
tion in pipes by frequency response method.’’ J. Hydraul. Eng.,
127共2兲, 134 –147.
Press, W. H., Teukolsky, S. A., Vetterling, W. T., and Flannery, B. P.
共1992兲. Numerical recipes: the art of scientific computing, Cambridge
University Press, Cambridge, U.K.
Pudar, R. S., and Liggett, J. A. 共1992兲. ‘‘Leaks in pipe networks.’’ J.Hy-
draul. Eng., 118共7兲, 1031–1046.
Vardy, A. E., and Hwang, K.-L. 共1991兲. ‘‘A characteristics model of tran-
sient friction in pipes.’’ J. Hydraul. Res., 29共5兲, 669– 684.
Vı́tkovský, J. 共2001兲. ‘‘Inverse analysis and modelling of unsteady pipe
flow: theory, applications and experimental verification.’’ Ph.D. thesis,
Univ. of Adelaide, Adelaide, Australia.
Vı́tkovský, J., Simpson, A. R., and Lambert, M. F. 共2000兲. ‘‘Leak detec-
tion and calibration using transients and genetic algorithms.’’ J. Water
Resour. Plan. Manage., 126共4兲, 262–265.
Weil, G. J., Graf, R. J., and Forister, L. M. 共1994兲. ‘‘Remote sensing
pipeline rehabilitation methodologies based upon the utilization of
infrared thermography.’’ Proc., Urban Drainage Rehabilitation Pro-
grams and Techniques, ASCE, New York, 173–181.
Wiggert, D. C. 共1968兲. ‘‘Unsteady flows in lines with distributed leak-
age.’’ J. Hydraul. Div., Am. Soc. Civ. Eng., 94共HY1兲, 143–162.
Wylie, E. B., and Streeter, S. L. 共1993兲. Fluid transients in systems,
Prentice–Hall, Englewood Cliffs, N.J.
Zielke, W. 共1968兲. ‘‘Frequency-dependent friction in transient pipe flow.’’
J. Basic Eng., 90, 109–115.
JOURNAL OF HYDRAULIC ENGINEERING / JULY 2002 / 711