Download as pdf or txt
Download as pdf or txt
You are on page 1of 11

Computers and Chemical Engineering 35 (2011) 651–661

Contents lists available at ScienceDirect

Computers and Chemical Engineering


journal homepage: www.elsevier.com/locate/compchemeng

Leak detection in gas pipeline networks using an efficient state estimator. Part-I:
Theory and simulations
H. Prashanth Reddy a , Shankar Narasimhan b,∗ , S. Murty Bhallamudi a , S. Bairagi c
a
Department of Civil Engineering, I.I.T. Madras, Chennai 600036, India
b
Deparment of Chemical Engineering, I.I.T. Madras, Chennai 600036, India
c
GAIL (INDIA) Ltd., New Delhi, India

a r t i c l e i n f o a b s t r a c t

Article history: Dynamic simulation models can be used along with flow and pressure measurements, for on-line leak
Received 23 January 2010 detection and identification in gas pipeline networks. In this two part paper, a methodology is pro-
Received in revised form 25 May 2010 posed for detecting and localizing leaks occurring in gas pipelines. The main features of the proposed
Accepted 9 October 2010
methodology are: (i) it is applicable to both single pipelines and pipeline networks and (ii) it considers
Available online 15 October 2010
non-ideal gas mixtures. In order to achieve the desired computational efficiency for on-line deployment,
an efficient state estimation technique based on a transfer function model, previously developed by the
Keywords:
authors, is embedded in a hypothesis testing framework. In Part-I of this paper, a detailed description
Gas pipeline networks
Leak detection
of the methodology is presented, and its performance is evaluated using simulations on two illustrative
Dynamic simulation pipeline systems. The proposed method is shown to perform satisfactorily even with noisy measurements
State estimation and during transient conditions, provided there is sufficient redundancy in the measurements.
Transfer function © 2010 Elsevier Ltd. All rights reserved.
Generalized likelihood ratio method

1. Introduction vide very accurate leak location at the expense of high system costs
and complexity of installation (Geiger et al., 2003). Also, some of the
Pipelines are used extensively all over the world for transporta- hardware based leak detection systems (pigging, acoustic method,
tion and distribution of water, natural gas and other light petroleum etc.) are used only periodically to test the integrity of the pipeline. In
products. Natural gas and petroleum products are carried over long contrast, software based solutions allow continuous on-line moni-
distances from oil fields and refineries to customers and commu- toring and rapid detection of leaks, and are therefore being actively
nities. Majority of these pipelines are buried and pass through sought these days.
crowded cities and remote areas such as forests and farms. One of Several model based solutions for leak detection in pipelines
the most difficult problems affecting the safe operation of pipeline have been proposed in the literature, particularly for water dis-
systems is development of rupture leaks, caused by corrosion and tribution networks. Although these methods cannot be directly
pressure surges. A challenging task for the operators of these sys- applied to gas pipeline networks due to the compressible nature
tems is to detect leaks as and when they occur, and subsequently of gases, some of these methods are reviewed here due to close
locate them. similarities in the dynamic simulation models for water and gas
Leak detection methodologies can be broadly classified into pipelines. Brunone and Ferrante (2001), Beck, Curren, Sims, and
hardware based and software based systems. Hardware based leak Stanway (2005), and Misiunas, Vitkovsky, Olsson, Simpson, and
detection systems include pigging (Furness & Reet, 1998), acous- Lambert (2005) have developed leak detection methods based on
tic methods (Sharp & Campbell, 1997; Watanabe & Himmelblau, the analysis of pressure transients induced or reflected at a location
1980), tracer gas methods (Tracer Research Corporation, 2003), where there is a leak. In these methods, every singularity of the
sensor cable method (Sandberg, Holmes, McCoy, & Koppitsch, system, such as junctions, nodes and bends reflects incident waves
1989), fiber optic methods (McLean et al., 2003), infrared photog- giving misleading information on the location of real leaks (Covas
raphy methods (Eidenshink, 1985), and radar methods (Gopalsami & Romas, 1999). Liggett and Chen (1994), and Vitkovsky, Simpson,
& Raptis, 2001). Hardware based leak detection systems often pro- and Lambert (2000) developed leak detection methods for water
distribution networks, based on inverse transient analysis. These
inverse transient methods detect leakages only at nodal points.
∗ Corresponding author. Tel.: +91 44 22574165; fax: +91 44 22574152. They also require significant computational effort. In recent years,
E-mail addresses: naras@iitm.ac.in, shan1908@yahoo.com (S. Narasimhan). leak detection approaches based on the analysis of how the tran-

0098-1354/$ – see front matter © 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compchemeng.2010.10.006
652 H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661

sient regime in the frequency domain is affected by the presence of normal transient operating conditions has not been demonstrated
a leak have been gaining attention (Ferrante & Brunone, 2003; Kim, explicitly. In this work, we propose a reliable and computation-
2005; Lee, Vitkovsky, Lambert, Simpson, & Liggett, 2005; Mpesha, ally efficient method for detection and identification of leaks in gas
Chaudhry, & Gassman, 2002). These methods are applicable only pipeline networks. The proposed method uses the available pres-
for well defined boundary conditions, and are very sensitive to the sure and flow rate measurements, sampled at regular intervals and
topology of the system. Accuracy of leak detection is affected by is based on an efficient state estimation technique, developed by
the presence of other singularities and free damping of the system. the authors (Reddy, Narasimhan, & Bhallamudi, 2006). The key fea-
The fact that leaks in pipelines dampen the transient events was tures of the proposed methodology are: (i) it is applicable to both
used by Wang, Lambert, Simpson, Liggett, and Vitkovsky (2002) to single pipelines as well as pipeline networks, (ii) it is applicable to
develop a method for finding leak location and magnitude in water non-ideal gas mixtures, (iii) it can detect and identify leaks even
pipelines. Application of the above methods to networks is com- when they occur during transient conditions created by normal
plex and in many cases, also interferes with the normal operations operations, and (iv) it can be applied on-line because it is based
of the pipeline. on a state estimation technique which uses a computationally effi-
One of the earliest and most popular methods for detecting cient transfer function model of a pipe segment. Performance of
leaks in gas pipelines is the volume balance method (Griebenow the proposed methodology is evaluated using simulations on two
& Mears, 1989; Liou & Tian, 1994). In this method, flows entering illustrative pipeline systems. These include a real life 204.7 km long
and leaving the pipeline, and other process variables like pres- series pipeline system, and a hypothetical eight-node-nine-pipe
sure and temperature are measured. A leak is detected when the network. The effect of measurement noise and the redundancy in
mass of the fluid exiting from the pipeline differs from the esti- the availability of measured data on the performance of the method
mated mass entering the pipeline, after accounting for the line are also studied. In a companion paper, the proposed methodology
pack. These methods can be used only for detecting a leak and is validated using experimental data obtained on a laboratory scale
not for identifying the leak location. Leak detection methods based model and operating data obtained from tests conducted on a real
on the use of state estimation models have been developed in the life pipeline carrying natural gas.
past for water distribution networks (Andersen & Powell, 2000;
Mukherjee & Narasimhan, 1996) and gas pipelines (Benkherouf &
Allidina, 1988; Billmann & Isermann, 1987; Emara-Shabaik, Khulief, 2. Methodology
& Hussaini, 2002). Billmann and Isermann (1987) used a non-linear
state estimation procedure to detect and localize small leaks in 2.1. State estimation
pipelines carrying compressible fluids. In this method, measured
flow and pressure data are compared with those predicted using a An efficient state estimation method for a gas pipeline network
transient process model (state estimator), and a leak is detected if forms the basic building block to develop an on line methodology
the discrepancy between the two is greater than prescribed limits. for leak detection. The state estimation model provides flow and
This method requires pressure and flow measurements at the inlet pressure profiles for the entire pipeline network, which best fit the
and outlet of the pipeline. Also, this method has been applied to measured pressures and flows available at some locations. The state
only single pipeline systems. Benkherouf and Allidina (1988) used estimation models takes into account the measurement noise and
an Extended Kalman filter for simultaneous state and parameter exploits the redundancy in measurements to obtain the estimates.
estimation in gas pipelines. The leak was modeled using multiple The proposed leak detection methodology is based on a state esti-
parameters representing the unknown constant leak magnitude at mation model developed by the authors earlier (Reddy et al., 2006).
every node in the pipeline. This method has been developed assum- Only a brief outline of this method is presented here, and readers
ing pressure wave velocity to be constant, which is valid for only are referred to the original paper for details.
ideal gases and isothermal conditions. Liu, Zang, and Zhou (2005) The conservation of mass and momentum equations for the
improved the accuracy of the above method by using an adap- unsteady subsonic flow of a gas through a constant diameter, rigid
tive particle filter (APF) algorithm. However, applicability of this pipe are (Osiadacz, 1989; Reddy et al., 2006):
method was demonstrated for the case of leaks occurring under Continuity equation:
steady state conditions. Also, only single pipelines were considered.
∂M
 A  ∂p
Emara-Shabaik et al. (2002) also used a Modified Extended Kalman
+ =0 (1)
Filter (MEKF) state estimation technique, in conjunction with a ∂x c2 ∂t
transient model. Ideal gas assumption is used and a backward
time-centered space discretization was also used for solving the Momentum equation:
governing partial differential equations. However, this approach  
was demonstrated only on a single pipeline system and simula- ∂p (gA sin )p c 2 M M  ∂M
tion results for only a 10% leak case were presented. Furthermore, A + + + =0 (2)
∂x c2 2DAp ∂t
the method detects leaks only at the discretized nodes. Based on
a detailed survey of currently available leak detection methods Equation of state:
for oil and gas pipelines, Scott and Barrufet (2003) concluded that
conventional material balance methods remain the most widely p
= zRT = c 2 (3)
used methods for leak detection in commercial software packages. 
Occasionally, these are supplemented with momentum balance
methods. They concluded that there exists a need for independent where M is mass flow rate, p is pressure, A is cross sectional area
verification and demonstration of capabilities of these methods. of the pipe, c is the pressure wave velocity, x is distance along the
Above literature review indicates that although many methods pipeline, t is time, g is acceleration due to gravity,  is inclination of
have been proposed for model based leak detection in gas pipelines, the pipe,  is coefficient of friction, D is inner diameter of the pipe,
none of these are applicable to networks. Furthermore, they have and  is density. Compressibility factor, z is a function of pressure
not considered either non-ideal behavior of the gas arising from and temperature for a given gas. In this work, AGA model (American
high pressures or composition of the natural gas. Also, applica- Gas Association, 1994) is used to compute the compressibility fac-
bility of the methods for detecting leaks that may occur during tor of the natural gas. The friction factor  is determined using an
H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661 653

explicit equation (White, 1986). L2 u


 1 1 2

 T = e /2 1− +
1.11 
(12)
 2Dc
2 6 24
1 6.9 ε/D
√ = −1.8 log10 + (4)
c̄ 2 = z̄RT̄ (13)
 Re 3.7
M0 c̄02
where ε is the roughness height of the pipe; and Re is the Reynolds ū = (14)
p̄0 A
number.
Typically, the above equations are solved using numerical Also, in Eqs. (5) and (6), current time t = NTs , and Ts is the sam-
(e.g. finite-difference) schemes to simulate dynamic conditions in pling interval. The summation terms on the right hand side of the
pipeline networks. Finite-difference approaches require excessive above equations are weighted sums of the discrete values of pres-
computational time and may not be appropriate from the point sures and flows at all times from the initial to the current time. Eqs.
of view of on-line deployment. This difficulty can be overcome by (5) and (6) for all the pipes are combined with (i) junction compat-
using a transfer function model (Kralik, Stiegler, Vostry, & Zavorka, ibility conditions, (ii) mass balance and pressure drop equations
1984; Reddy et al., 2006) to represent the governing equations of a for valves, (iii) equations describing compressor operation, and (iv)
single pipe. the boundary conditions to formulate the complete discrete model
In a transfer function model, the pressure and the mass flow rate in the time domain for the entire network. This formulation for a
at the upstream end are related to their values at the downstream network consisting of only pipe elements results in a linear system
end. Eqs. (1) and (2) are first linearized in terms of deviation vari- of equations (Reddy et al., 2006).
ables (M = M − M0 , P = P − P0 ) from steady state flow rate, M0
Am + Bu = 0 (15)
and average steady state pressure in the pipeline, P0 . Laplace trans-
forms are applied to linearized PDEs to convert them into linear In Eq. (15), the vector m consists of all measured variables cor-
differential equations, which are then solved analytically to obtain responding to time instants 1 to NTs , and the vector u consists of
the relations between the flows and pressures at the ends of the all unmeasured variables corresponding to time instants 1 to NTs .
pipe in the Laplace domain. The following time domain solution is The matrices A and B change with time and depend on the pipe
obtained by an approximate inversion of Laplace domain solution parameters, sampling period, compressibility factor and friction
(Kralik et al., 1984; Reddy et al., 2006). factor.
In the state estimation problem, the state which minimizes

N
p2 (NTs ) = [k1 e−(N−i)(Ts /T ) (1 − e−(Ts /T ) )p1 (iTs )] the quadratic difference between the measured values and the
estimated values is considered as the best estimate of the state,
i=1
subject to satisfying Eq. (15). In our state estimation model, the

N
state estimate for the current time instant NTs is obtained in recur-
+ [−k2 e−(N−i)(Ts /T ) (1 − e−(Ts /T ) )M2 (iTs )] sive manner by utilizing the estimates obtained for all the previous
i=1 times, by solving the following weighted least squares estimation
 problem.
−k2 T2 N
− (1 − e−(Ts /T ) ) [ M2 (iTs )e−(N−i)(Ts /T ) ] Min (ȳ − m̄)T Q̄ −1 (ȳ − m̄)
T m̄,ū
i=1 (16)
st
k2 T2

+ M2 (NTs ) (5) Ām̄ + B̄ū = c


T
where Ā, B̄ are the corresponding sub-matrices of A and B at
current time instant NTs . m̄ and ū are the measured and unmea-
N
sured variables corresponding to the current time instant. Vector
M1 (NTs ) = [e−(N−i)(Ts /T ) (1 − e−(Ts /T ) )M2 (iTs )]
c contains the weighted sum of the estimated flows and pres-
i=1
sures at previous times, ȳ is the vector of measurements at current
 
T1 N time instant NTs and Q̄ is the covariance matrix of errors in mea-
+ − (1 − e−(Ts /T ) )[ p1 (iTs )e−(N−i)(Ts /T ) ] surements. An efficient approach to solve the above optimization
T
i=1 problem is to make use of Crowe’s Projection Matrix technique as
discussed by Narasimhan and Jordache (2000).
T1
+ p1 (NTs ) (6)
T
2.2. Leak detection and isolation
In Eqs. (5) and (6), subscripts 1 and 2 refer to the values at the
The procedure for leak detection and isolation that is developed
upstream and downstream ends of the pipe, respectively, and
  in this work is based on the Generalized Likelihood Ratio (GLR)
L u u gL sin  method as applied to non-linear systems (Narasimhan & Jordache,
= − (7) 2000). Generally this method has been used for fault diagnosis in
2D c 2 c
2
steady state processes or systems described by ordinary differen-
k1 = e ; (8) tial equations. In this work, this approach has been extended to a
  gas pipeline network where the process is modeled using nonlin-
/2 L 1 2
k2 = e u 1+ ; (9) ear partial differential equations (PDEs). This opens the possibility
DA 24
AL
 1 2
 for applying the technique for fault diagnosis in other distributed
T1 = e /2 2
1+ ; (10) parameter systems.
c 24
At each time t, the optimization problem for state estimation
⎛   ⎞
(refer Eq. (16)) is solved, based on the window of measurements
D 1 L2 u 1 [t, t − nTs ]. The time window, n, is chosen such that the weight
T2 =   + ⎝  ⎠ (11)

 u 6 Dc
2
2 factors for past measurements beyond this window are negligi-
1+ 24
ble. The value of the objective function indicates how well the
654 H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661

P1, M1 P2, M2 momentum balance equations for the pipe segment i, where a leak
is hypothesized, are deleted since this pipe is divided into two seg-
xl L-xl ments. Instead of these two equations, we now add six equations.
DL (unknown leak magnitude) Among these, four equations correspond to the two transfer func-
tion equations each for the two pipe segments into which pipe i
Fig. 1. Branch i with leak of magnitude DL at position xl . is divided, and the remaining two equations arise due to the com-
patibility conditions (mass balance equation and pressure balance
equation) at the newly created node. The unknown demand cor-
measured values fit the constraints (pipeline network model). It
responding to the leak magnitude is an additional unmeasured
should be noted that the pipeline network model has been derived
variable that needs to be estimated along with other state variables.
under the assumption that no leak is present in the network. Thus,
The constraint equations for the optimization problem corre-
if the objective function value is small, then it can be concluded
sponding to the state estimation are then assembled in the form of
that no leak is present. On the other hand, if the objective func-
Eq. (16) and the optimization problem is solved. The unknown vari-
tion value is greater than a pre-specified threshold value, then it
able DL (t) is estimated as part of the solution and is contained in the
can be concluded that a leak has occurred. It should be noted that
vector ū. At every sampling instant in the time window [t, t + WTs ],
the state estimation method can be applied online using a mov-
the leak magnitude is estimated, and the corresponding objective
ing window approach, as and when measurements are received at
function value is noted. The cumulative objective function value in
each time instant. The time instant at which the objective func-
the time window [t, t + WTs ] gives the objective function value cor-
tion value exceeds a specified threshold can be used to detect the
responding to that assumed leak location and is an indicator of how
time of occurrence of the leak. It can be shown that, under the
well the measured values in the time window fit the model corre-
assumption that the measurement errors follow a Gaussian dis-
sponding to the hypothesized leak location in the chosen branch
tribution with zero mean, the objective function value follows a
i. The hypothesized location of the leak in the assumed branch i is
chi-square distribution with degrees of freedom equal to ne − nu ,
varied and the location for which the least value of the above objec-
where ne is the number of constraints and nu is the number of
tive function is obtained is taken as the most probable leak location.
unmeasured variables (assuming that all unmeasured variables
This optimal value of xl is determined using the univariate golden
can be uniquely estimated). Therefore, the threshold value can be
section search method (Press, Flannery, Teukolsky, & Vetterling,
chosen from the chi-square distribution for a given level of signifi-
1998). It should be noted that best fit estimate of the time varying
cance. In the present study, the threshold objective function value
leak magnitude profile is obtained as part of the state estimation
was 4.0 per measurement (which is approximately the value of a
procedure. The above procedure is repeated for each pipe segment
chi-squared variable with one degree of freedom, at 95% level of
or branch of the network, in which a leak is hypothesized in turn,
significance). In practice, due to plant model mismatch the thresh-
and the least objective function value among all the hypotheses
old value needs to be readjusted in order to reduce false alarm rate
is determined. The solution (the branch pipeline, the optimal leak
below an acceptable limit.
location and the leak magnitude for that branch) corresponding to
In order to estimate the location of the leak and estimate its mag-
the least objective function value is accepted as the solution for leak
nitude we propose a method which makes use of the pressure and
identification problem. It may be noted that the above procedure
flow profiles subsequent to the occurrence of the leak. Thus, in the
for leak identification requires the state estimation problem to be
proposed method, if a leak is detected at time t, measurements for
solved for every time instant in the chosen window [t, t + WTs ] for
a time period [t, t + WTs ] (W = a user defined window size for leak
every pipe segment, and every hypothesized location (provided by
identification) are collected and used to (i) identify the pipeline
the golden section in each iteration of the search procedure). It is
branch in which leak has occurred, (ii) estimate its location and
here that maximum computational advantage is derived from the
(iii) estimate its magnitude. The value of W (number of measure-
authors’ (Reddy et al., 2006) efficient state estimation procedure
ments collected to identify the leak) should be chosen such that the
based on the transfer function model of the network.
computational time taken by leak detection method is reasonable,
without compromising on the accuracy in the estimation of leak
location and leak magnitude. The following methodology based on
3. Results and discussion
testing different hypotheses for the leak location is used for leak
identification. Suppose that a leak of unknown magnitude, DL (t) is
The online leak detection and identification method proposed
hypothesized in a particular branch “i” of the pipeline network, at
in this work consists of the following objectives: (a) detection of
a distance xl from the upstream end of the pipeline as shown in
the time at which a leak has occurred by continually monitoring
Fig. 1. The location where the leak is hypothesized is modeled as an
the measured pressures and flows; (b) identification of the pipe
additional node and the unknown leak magnitude as an unknown
segment or branch where the leak has occurred; and (c) estimation
demand at this node (Fig. 2). The time domain form of the transfer
of the leak location and magnitude. Among the above objectives,
function model for the pipeline network under this assumption is
it is important to detect the time of occurrence of a leak without
now re-constructed. For all pipes j = / i, the transfer function equa-
significant delay. It is also important to identify the leak location
tions will be assembled as in Eq. (16), that is the coefficients of
as accurately as possible so that further investigation and correc-
matrices Ā and B̄ in the rows corresponding to the flow and momen-
tive action can be initiated. While estimating the magnitude of
tum balance equations will be computed as before. The flow and
the leak is useful, it is relatively a less important issue. Thus we
consider the performance of the leak detection method to be sat-
P1, M1 PL, ML isfactory if leaks can be detected quickly (within 15 min of the
P2, M2
occurrence of the leak) and the estimated location of the leak is
xl L-xl within ±5% (of the pipeline length) from the actual leak location.
A large error in the estimated leak magnitude, though unsatisfac-
tory, is not considered to be a failure of the proposed method. In
DL (unknown leak magnitude) this paper, accuracy and applicability of the proposed approach for
leak detection and isolation is evaluated through simulations on
Fig. 2. Modeling of pipeline with leak. a real life series pipeline system and a hypothetical pipeline net-
H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661 655

Fig. 3. Schematic diagram of a series pipeline system carrying natural gas.

work. The simulations are chosen to investigate (i) the minimum (Reddy et al., 2006). It may be noted here that the state estima-
size of leak that can be detected and isolated, (ii) whether additional tion model works as a simulation model if just sufficient number
instrumentation can improve the performance, and (iii) the appli- of specifications (pressure variation at the upstream end, demand
cability of the methodology to detect and isolate leaks occurring in variation at the downstream end, and demand variation at the leak
networks. Performance evaluation of the proposed methodology location as a step function) are provided. This is referred to as the
using laboratory experiments and field tests is discussed in detail forward problem. The solution of the forward problem was used to
in the companion paper. generate the simulated measurements at all the intermediate loca-
tions, wherever such measurements were assumed to be made. The
3.1. Series pipeline system proposed methodology for the leak detection and identification was
then used to detect the leak. In the leak detection and identifica-
A real life pipeline system (Fig. 3) used for transporting natural tion runs (the inverse problem), all the synthetically generated data
gas from a field to a power plant in India is considered for illus- at the measurement locations, obtained using the forward prob-
tration of proposed methodology for leak detection. The pipeline lem, were given as input. Data for leak location, time of occurrence
is 204.7 km long and has a uniform circular cross section of diam- of the leak, and the leak magnitude estimated by the leak detec-
eter 0.443 m. It consists of 8 full bore sectionalized ball valves. The tion method were then compared with their true values used in
roughness height of the pipeline, calibrated using steady flow data, the simulation in order to evaluate performance of the proposed
is 250 ␮m. In the present study, this pipeline system was discretized approach. All the leak detection runs were carried out in conjunc-
into 24 pipeline segments, with 25 nodes as shown in Fig. 4. Node tion with a transient state prevailing due to a demand variation
1 was the source node at which the gas enters the system and the at the consumption node. In the generation of simulated data, a
node 25 was the consumption node where the gas exits from the constant source pressure equal to 45.0 kg/cm2 was considered. Ini-
system. Discretization was carried out such that a node is located tial demand at the consumption node was taken as 60500.0 SCMH
at every point where the measurements (either flow or pressure (standard cubic meters for hour). The transient conditions were cre-
or both) are available. Further, additional intermediate nodes are
created in order to limit the length of any segment to 10 km. This
Table 1
choice was based on preliminary tests for trade off between compu- Nodal elevations, series pipeline.
tational time and accuracy. Elevations of the nodes and the pipeline
characteristics for all the segments are presented in Tables 1 and 2, Node number Elevation (above msl)

respectively. Composition of the natural gas is given in Table 3. The 1 2.844


viscosity of the gas was assumed to be 0.0000125 N s/m2 . A con- 2 10.4
3 15.872
stant ambient temperature of 302 K was specified at all points in
4 15.162
the network. 5 11.381
Three levels of instrumentation were considered for testing 6 14.173
the performance of the leak detection method. The first level of 7 17.665
instrumentation (LI-1) corresponds to the currently available mea- 8 22.16
9 27.685
surements, consisting of six pressure measurements at nodes 1, 3, 10 27.423
9, 12, 18 and 25, one mass flow measurement at node 1, and one 11 30.798
demand measurement at node 8. In the second level of instrumen- 12 37.81
tation (LI-2), pressure measurements were assumed to be available 13 57.8776
14 67.14
also at the nodes 5, 7, 10, 11, 14, 15, 16, 19, 20 and 22, in addition to
15 65.59
the L1-1 measurements. Thus, a total of 16 pressure measurements, 16 44.68
one demand measurement at the node 25, and one flow rate mea- 17 35.695
surement at the node 1 were available in this case. In the third level 18 25.6
of instrumentation (LI-3), it was assumed that pressure measure- 19 36.3
20 42.8
ments were available at all the nodes, in addition to one demand 21 56
measurement at the node 25, and one flow measurement at the 22 63.8
node 1. In all cases, measurements were sampled at 10 s intervals. 23 70.3
Measurements corresponding to various leak scenarios were 24 27
25 36.9
simulated using the transfer function based state estimation model
656 H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661

CH. 45566.0
1 1 2 2 3 3 4 4 5 5 6 6
7

8166
8750 8750 2065 8917.5 8917.5 8166 7
CH. 0 CH. 8750.0 CH. 17500.0 CH. 19565.0 CH. 28482.5 CH. 37400.0
CH. 53732.0 8

8166
CH. 120837.6
18" pipeline 8

CH. 61898.0 9

8100.66
9
16 15 15 14 14 13 13 12 12 11 11 10
10
9347.66

9347.66 8430 8430 8430 8100.66 8100.66


16

CH. 69998.66
CH. 111490.0 CH. 103060.0 CH. 94629.98 CH. 86199.98 CH. 78099.32
17 CH. 130185.3
9347.66

17

18 CH. 139533.0
9460.33

18

19 19 20 20 21 21 22 22 23 23 24 24 25

9460.33 9460.33 9195.75 9195.75 9195.75 9195.75


CH. 148993.3 CH. 158453.6 CH. 167914.0 CH. 177109.7 CH. 186305.5 CH. 195501.2 CH. 204700.0

Intermediate Consumption Pipe element


Source node node
node
All Dimensions are in meters

Fig. 4. Discretization of the series pipeline system shown in Fig. 3.

ated by varying the demand at the consumption node. The demand Table 3
Natural gas composition.
was increased linearly at the 25th sampling instant from its orig-
inal value by 10% over 25 sampling intervals, held constant for 25 S. no. Component Mole percentage
sampling intervals, and subsequently reduced linearly back to its 1 Methane 92.562
original value in the next 25 sampling instants (see Fig. 5). At the 2 Nitrogen 0.104
30th sampling instant, a leak of specified magnitude at a specified 3 Carbondioxide 2.173
location was created using a step change in demand at that location 4 Ethane 2.035
5 Propane 1.629
(see Fig. 6).
6 i-Butane 0.348
7 n-Butane 0.440
8 i-Pentane 0.152
Table 2 9 n-Pentane 0.136
Pipe element properties, series pipeline. 10 n-Hexane 0.420
Pipe element number From node number To node number Length (m)

1 1 2 8750.00 Leak detection tests carried out in this study can be divided
2 2 3 8750.00
into four categories, depending on the noise level in the measure-
3 3 4 2065.00
4 4 5 8917.50 ments and the level of instrumentation (Table 4). In category-1,
5 5 6 8917.50 level of instrumentation is LI-1, and the noise in the measurements
6 6 7 8166.00
7 7 8 8166.00
8 8 9 8166.00 67000
9 9 10 8100.66 66000
demand (SCMH)

10 10 11 8100.66
65000
11 11 12 8100.66
12 12 13 8430.00 64000
13 13 14 8430.00 63000
14 14 15 8430.00
15 15 16 9347.66 62000
16 16 17 9347.66 61000
17 17 18 9347.66
60000
18 18 19 9460.33
19 19 20 9460.33 59000
20 20 21 9460.33 1 28 55 82 109 136 163 190 217 244 271 298 325 352 379
21 21 22 9195.75
22 22 23 9195.75
sampling instants (sampling interval = 10 sec)
23 23 24 9195.75
Fig. 5. Demand variation at the consumption node, series pipeline.
24 24 25 9195.75
H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661 657

1400 3.1.2. Performance evaluation: effect of measurement noise


1200 Performance of the proposed method, in the presence of mea-
demand (SCMH)

surement noise, was evaluated through category-2 runs. The level


1000
of instrumentation in these runs was LI-1. Results from these tests
800 are presented in Table 6. It is clear from this table that the error in
600 leak detection and identification increased due to the presence of
measurement noise. There was a delay in detecting the leak (almost
400
half an hour in case of 2% leak), which, in turn, affected the accuracy
200
in the estimation of the leak magnitude and the location. Maximum
0 error in the estimation of the leak location was 4.08 km, while the
1 27 53 79 105 131 157 183 209 235 261 287 313 339 365 391 maximum error in the estimation of the leak magnitude was more
sampling instants (sampling interval = 10 sec) than 100%. In general, the performance of the method was better if
the leak magnitude was higher.
Fig. 6. 2% leak as a step function in demand variation to generate measurements.

3.1.3. Performance evaluation: effect of redundancy in


measurements
Numerical leak detection tests under category-3 and category-
was negligible. In category-2, leak detection runs were similar to
4 were conducted to evaluate the performance of the proposed
those in category 1, but measurements are noisy. In category-3
method as the level of instrumentation increases. It may be noted
and category-4 runs, the number of pressure measurements was
here that the proposed method is based on the premise that the
increased (LI-2 and LI-3) to investigate the dependence of the per-
state of the system is better estimated if there is sufficient redun-
formance of the proposed leak detection methodology on the level
dancy in the measurements, even when they contain noise. Results
of instrumentation. Measured data in these runs also contained
from the numerical leak detection tests under category-4 are pre-
noise. Standard deviation of noise in the flow rate measurement
sented in Table 7. It may be recalled here that ten more pressure
was 175 SCMH, which was equal to 0.25% accuracy in 0–70,000
measurements were considered in these tests as compared to
SCMH flow rate range. Standard deviation of noise in the pres-
the tests in category-1. It can be observed from Tables 6 and 7
sure measurement was 0.0155 kg/cm2 , which was equal to 0.0275%
that increased redundancy in measurements improved the accu-
accuracy in 0–45.0 kg/cm2 pressure range. To test the applicabil-
racy of the leak detection method. In most cases, the error in the
ity of the proposed methodology, leak magnitudes of 2%, 5% and
estimation of the leak location was less than 1.6 km (maximum
10% (percent of total flow rate through the pipeline) were consid-
value = 3.03 km). Error in the estimation of the leak magnitude
ered. Also, three different possible locations for the leak occurrence
remained less than 20% in most of the cases. Maximum error in
were considered. Distances from the source node to Location-1,
the leak magnitude estimation (48.6%) occurred for 2% leak at
Location-2 and Location-3 for the leak were 53.73 km, 103.06 km
53.73 km. There was not much improvement in the leak detection
and 195.50 km, respectively. Although, the estimated leak mag-
time, although delay in leak detection did not exceed 10 min when
nitude varies with time, the average value of the estimated leak
leak magnitude is more than or equal to 5%.
magnitudes in the time window [t, t + WTs ] is reported, since a
Numerical leak detection tests under category-4 considered
constant magnitude leak is simulated in all the simulation runs.
twenty-five pressure measurements and two mass flow measure-
ments at both ends of the pipeline. These tests were carried out
to confirm that the performance of the proposed model improves
3.1.1. Performance evaluation: no measurement noise with an increase in the redundancy in the measurements. Results
Leak detection tests under first category were carried out to from these tests are presented in Table 8. A comparison of results
study the ability of the proposed method to detect and identify presented in Tables 7 and 8 confirms that the performance of the
even small size leaks, if very accurate measurements are available. proposed leak detection model improves if more measurements
Results from leak detection runs under this category are presented are available.
in Table 5. It can be seen from this table that the leak was detected
almost as soon as it occurred. The error in the estimation of the 3.2. Leak detection tests on a network
leak magnitude was less than 2% and the error in the estimation of
the leak location was less than 0.5 km, in a total length of 204.7 km. Accuracy and applicability of the proposed approach for leak
Maximum error in leak identification occurred when the leak mag- detection and identification were also evaluated through numeri-
nitude was as low as 2%. It may be noted that the computational cal leak detection tests on a hypothetical pipeline network shown
time required for estimating the leak location and leak magnitude is in Fig. 7. This network (Fig. 7) consisted of 8 nodes (including one
less than one sampling period (10 s). The delay in reporting the esti- source node and two delivery nodes) and 9 pipes. Data for the
mated leak location and magnitude arises only due to the window nodes and pipe elements are given in Tables 9 and 10, respec-
size W. These leak detection tests demonstrate that the proposed tively. Composition of the natural gas flowing through the network
methodology for the leak detection and identification performs was same as that given in Table 3. The viscosity of natural gas was
very well when the errors in the measurements are negligible. 0.0000125 N s/m2 . AGA gas model was used to compute the com-

Table 4
Categories of leak detection runs, series pipeline.

Category Level of noise Level of instrumentation


2 2 2
Variance in pressure (kg/cm ) Variance in flow rate (SCMH)

1 Negligible (10−8 ) Negligible (10−8 ) LI-1


2 2.4 × 10−4 3.06 × 104 LI-1
3 2.4 × 10−4 3.06 × 104 LI-2
4 2.4 × 10−4 3.06 × 104 LI-3
658 H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661

Table 5
Results of leak detection tests under category-1, series pipeline system.

S. no. Leak location Estimated leak Error in leak Magnitude of Estimated leak % error in leak Delay in leak detection
(km) location (km) location (km) leak tested magnitude magnitude time (s)
(SCMH) (SCMH)

1 53.73 53.7 0.03 1210.0 1218.0 0.66 10


2 53.73 53.7 0.03 3025.0 3032.2 0.23 0
3 53.73 53.7 0.03 6050.0 6065.7 0.26 0
4 103.06 103.38 0.32 1210.0 1188.0 1.85 30
5 103.06 103.01 0.05 3025.0 2998.1 0.90 10
6 103.06 103.07 0.01 6050.0 6024.7 0.42 10
7 195.50 195.97 0.47 1210.0 1197.4 1.07 0
8 195.50 195.72 0.22 3025.0 2975.8 1.65 0
9 195.50 195.63 0.13 6050.0 5976.6 1.23 0

Table 6
Results of leak detection tests under category-2, series pipeline system.

S. no. Leak location Estimated leak Error in leak Magnitude of Leak magnitude % error in leak Delay in leak
(km) location (km) location (km) leak tested estimated (SCMH) magnitude detection time (s)
(SCMH)

1 53.73 49.65 4.08 1210.0 2410.7 99.2 1650


2 53.73 51.86 1.87 3025.0 2558.7 15.4 270
3 53.73 54.24 0.51 6050.0 6810.1 12.6 160
4 103.06 104.36 1.30 1210.0 1490.9 23.2 1720
5 103.06 104.10 1.04 3025.0 6483.3 114.3 790
6 103.06 104.67 1.61 6050.0 6513.0 7.7 350
7 195.5 191.73 3.77 1210.0 2345.8 93.9 380
8 195.5 193.75 1.75 3025.0 2688.4 11.1 160
9 195.5 196.79 1.29 6050.0 6067.2 0.3 130

Table 7
Results of leak detection tests under category-3, series pipeline system.

S. no. Leak location Estimated leak Error in leak Magnitude of Estimated leak % error in Delay in leak
(km) location (km) location (km) leak tested magnitude estimated leak detection time
(SCMH) (SCMH) magnitude (s)

1 53.73 52.21 1.53 1210 1797.5 48.6 1770


2 53.73 52.24 1.49 3025 3481.0 15.1 480
3 53.73 53.51 0.22 6050 5842.9 3.4 220
4 103.06 104.30 1.24 1210 1382.6 14.3 1910
5 103.06 101.92 1.14 3025 3587.1 18.6 550
6 103.06 103.74 0.68 6050 6313.0 4.4 170
7 195.5 192.47 3.03 1210 1423.5 17.6 1690
8 195.5 193.93 1.57 3025 3400.6 12.4 350
9 195.5 194.36 1.14 6050 5831.0 3.6 200

Table 8
Results of leak detection tests under category-4, series pipeline system.

S. no. Leak location Estimated leak Error in leak Magnitude of Estimated leak % error in Delay in leak
(km) location (km) location (km) leak tested magnitude estimated leak detection time
(SCMH) (SCMH) magnitude (s)

1 53.73 52.84 0.89 1210 1596.3 31.9 1750


2 53.73 53.21 0.52 3025 2735.3 9.6 520
3 53.73 53.97 0.24 6050 6193.8 2.4 210
4 103.06 102.15 0.91 1210 1393.6 15.2 1800
5 103.06 103.83 0.77 3025 3472.5 14.8 490
6 103.06 103.56 0.50 6050 6310.1 4.3 180
7 195.5 192.61 2.89 1210 1020.7 15.7 1280
8 195.5 194.56 0.94 3025 2723.9 10.0 330
9 195.5 195.08 0.42 6050 6123.7 1.2 150

Fig. 7. Schematic diagram of the network.


H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661 659

Table 9 In the leak detection tests, measured data was provided for (i)
Node data for the example network.
pressure and mass flow rates at node 1, (ii) pressure at node 3, (iii)
Node number Elevation (m) Demand (SCMH) pressure and demand at node 5, (iv) pressure at node 7, and (v)
1 35.0 Source node demand at node 8. Measurements were assumed to be sampled at
2 36.0 0 one second interval. Results from the above leak detection tests are
3 37.0 0 presented in Table 11. These results demonstrate the ability of the
4 37.0 0 proposed method to detect and identify leaks in gas pipeline net-
5 38.0 20880.0
works. The delay in leak detection was less than 10 seconds, and
6 38.0 0
7 39.0 0 the pipeline branch in which the leak occurred was also identified
8 40.0 42600.0 correctly. Error in the estimation of the leak magnitude was more
than 10% only in three cases (maximum error = 15.6%). Error in the
estimation of the leak location was more than 300 m only in two
pressibility factor. A constant ambient temperature of 300 K was cases (maximum error = 1.19 km). It may be noted that finite differ-
specified at all points in the network. ence method was used in the forward problem to generate the data
and transfer function method was used in the inverse problem for
3.2.1. Steady state conditions and no noise in the measurements the leak detection. Thus modeling error was present between the
In the first set of tests on the example pipe network, leak forward problem and the inverse problems, although the measure-
detection tests were carried out for steady state conditions. Under ments did not contain any noise. Therefore, these tests show that
these conditions, the constant upstream pressure at node 1 was the proposed methodology for the leak detection and identification
60.0 kg/cm2 . Constant demands of 20880 SCMH and 42600 SCMH performs very well for a given network even though there is some
were maintained at delivery nodes 5 and 8, respectively. To test modeling error due to approximations used in deriving the transfer
whether the proposed methodology performs satisfactorily, leak function model.
magnitudes of 2%, 5% and 10% (percent of total flow rate through
the network) were considered. Also, four different possible loca-
tions for the leak occurrence were considered. These leak locations 3.2.2. Transient state conditions and no noise in the
in the network (Fig. 7) are represented as Location-1 (at the node measurements
2), Location-2 (at the node 4), Location-3 (at the mid-point of pipe The proposed leak detection methodology exploits information
8) and Location-4 (at the mid-point of pipe 9). from the transients induced by the leak. Therefore, if additional
Measurements required for testing various leak scenarios were transients are present due to changes in the operating conditions,
synthetically generated using the complete nonlinear dynamic the accuracy in leak detection and identification could be adversely
model, based on the second order accurate McCormack finite differ- affected. This issue is explored in the present section. In the set of
ence scheme (Chung, 2002). The synthetic measurement data were leak detection tests discussed in this section, transient conditions
obtained using a time step of 1 second, and a spatial discretization of prevailed in the network when the leak occurred. These transient
375 m, which satisfied the requirement for numerical stability and conditions were introduced due to variation in the demand at the
grid convergence. Noise was not present in the synthetically gen- node 8. The synthetic measurements for use in the leak detection
erated measurements used in the leak detection tests discussed in tests were generated as explained in the previous section, but with
this section. the demand variation at the node 8 as given in Fig. 8. All the other

Table 10
Pipe data for example network.

Pipe number From node To node Pipe length (km) Pipe diameter (m) Roughness height (m)

1 1 2 8 0.406 4.6E−05
2 2 3 9 0.406 4.6E−05
3 3 4 10 0.406 4.6E−05
4 4 5 8 0.406 4.6E−05
5 5 6 9 0.406 4.6E−05
6 3 6 10 0.406 4.6E−05
7 5 7 8 0.406 4.6E−05
8 6 7 9 0.406 4.6E−05
9 7 8 10 0.406 4.6E−05

Table 11
Results for leak detection tests on the network (steady state conditions and no noise in the measurements).

S. no. Leak location Magnitude of leak % error in Error in estimated Delay in leak
tested (% of total estimated location (km) detection time (s)
flow) magnitude

1 Location-1 2 0.27 0.95 10


2 Location-1 5 1.68 0.12 10
3 Location-1 10 0.80 0.20 10
4 Location-2 2 6.78 0.09 10
5 Location-2 5 9.74 0.96 10
6 Location-2 10 6.57 1.19 10
7 Location-3 2 15.60 0.27 10
8 Location-3 5 12.41 0.03 10
9 Location-3 10 10.37 0.10 10
10 Location-4 2 2.04 0.00 10
11 Location-4 5 4.57 0.00 10
12 Location-4 10 5.40 0.00 10
660 H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661

52000 nitude was more than 2%. However, there was an error of as much
as 72% in the estimation of the leak magnitude in the case of small
50000 leaks (2%). The transient created by the leak itself was somewhat
masked by the already prevailing transient in the system, and hence
Demand (SCMH)

48000 the deterioration in the estimation of leak magnitude.

46000 3.2.3. Transient state conditions and noise in the measurements


In the set of leak detection tests discussed in this section, tran-
44000 sient conditions prevailed in the network when the leak occurred.
Also, noise was introduced into the synthetically generated mea-
42000 surement data. As in the case of tests discussed in Section 3.2.2,
transient conditions were introduced due to variation in the
40000
demand at the node 8. Noise as per the Gaussian distribution,
0 100 200 300 400 500 600 700 800 900 1000 1100 1200 with zero mean and specified standard deviation was added to the
Time (seconds) above synthetically generated measurements. The standard devia-
tion of the noise in the pressure measurements was 0.0165 kg/cm2
Fig. 8. Demand variation at node 8 to create transient in the network. (=0.0275% of steady state pressure at the source node). The stan-
dard deviation of the noise in the flow and demand measurements
was 0.25% of the respective steady state values. Rest of the condi-
conditions in these leak detection tests were the same as in the tions in these leak detection tests were the same as described in
preceding section. Results from these leak detection tests are pre- the previous section. Results from the above leak detection tests
sented in Table 12. These results demonstrate that the proposed are presented in Table 13. These results demonstrate that the pro-
method detects and identifies leaks in gas pipeline networks satis- posed method detects and identifies leaks in gas pipeline networks
factorily, even when transient conditions (due to demand variation) satisfactorily, only for leaks of magnitude 5% or greater when tran-
prevail in the system. However, as expected, the performance of sient conditions (due to demand variation) prevail in the system
the proposed model was not as good as in the case when transients and there is measurement noise. As expected, the performance of
were not present. The delay in leak detection was less than 35 s, the proposed model was not as good as in the case when there
and the pipeline branch in which the leak occurred was also iden- was no noise. The leak was detected with a maximum time delay
tified correctly. The performance of the model with regard to the of 514 s for a 5% leak. However, the pipeline branch in which the
estimation of the leak location was as good as earlier. Error in the leak occurred was identified correctly. Accuracy in the estimation
estimation of the leak location was more than 500 m only in two of the leak magnitude increased with the size of the leak. There
cases (maximum error = 1.71 km). Error in the estimation of the leak was more than 100% error in the leak magnitude estimation, if the
magnitude was less than 10% in all the cases when the leak mag- leak size was less than 5%. However, the performance of the model

Table 12
Results for leak detection tests on the network (transient conditions and no noise in the measurements).

S. no. Leak location Magnitude of leak % error in estimated Error in estimated Delay in leak
tested (% of total flow) magnitude location (km) detection time (s)

1 Location-1 2 2.97 0.47 13


2 Location-1 5 3.51 0.25 11
3 Location-1 10 1.5 0.12 10
4 Location-2 2 72.38 1.29 26
5 Location-2 5 29.49 0.32 14
6 Location-2 10 0.8 0.11 10
7 Location-3 2 62.47 1.71 33
8 Location-3 5 8.11 0.24 15
9 Location-3 10 0.4 0.11 10
10 Location-4 2 21.45 0 26
11 Location-4 5 9.02 0 14
12 Location-4 10 4.86 0 10

Table 13
Results for leak detection tests on the network (with transient state conditions and noise in the measurements).

S. no. Leak location Magnitude of leak % error in Error in estimated Delay in leak
tested (% of total flow) estimated location (km) detection time (s)
magnitude

1 Location-1 2 234.2 2.8 808


2 Location-1 5 104.8 2.68 342
3 Location-1 10 30.3 0.84 90
4 Location-2 2 179.0 3.1 874
5 Location-2 5 86.0 1.96 389
6 Location-2 10 36.0 1.14 187
7 Location-3 2 105.1 3.27 966
8 Location-3 5 74.6 2.37 371
9 Location-3 10 13.2 1.47 186
10 Location-4 2 242.1 2.63 975
11 Location-4 5 167.0 1.19 514
12 Location-4 10 37.0 1.19 193
H.P. Reddy et al. / Computers and Chemical Engineering 35 (2011) 651–661 661

with regard to the estimation of the leak location did not deterio- Chung, T. J. (2002). Computational fluid dynamics. Cambridge: Cambridge University
rate as much. Maximum error in the estimation of the leak location Press.
Covas, D., & Romas, H. (1999). Leakage detection in single pipeline using pressure
was 3.27 km. It may be noted here that the number of measure- wave behavior. In D. Savic, & G. Walters (Eds.), Water industry system: Modeling
ments provided in all the above leak detection tests was only seven, and optimization application (pp. 287–299). Baldock: Hertfordshire.
while the combined length of pipelines was as much as 81 km. As Eidenshink, J. C. (1985). Detection of leaks in buried rural water pipelines using
thermal infrared images. Photogrammetric Engineering and Remote Sensing, 51,
demonstrated for the case of series pipeline system, the perfor- 561–564.
mance of the leak detection method would definitely improve if Emara-Shabaik, H. E., Khulief, Y. A., & Hussaini, I. (2002). A non-linear multiple-model
more measurements are provided. state estimation scheme for pipeline leak detection and isolation. Proceedings
of the Institution of Mechanical Engineers Part I Journal of Systems and Control
Results from the simulation studies indicate that the proposed Engineering, 216, 497–512.
methodology is able to detect the occurrence of leaks quickly Ferrante, M., & Brunone, B. (2003). Pipe system diagnosis and leak detection by
in single as well as pipeline networks. The branch in which the unsteady state tests. 1. Harmonic analysis. Advances in Water Resources, 26,
95–105.
leak occurs is identified correctly. Estimation of leak location is
Furness, R. A., & Reet, J. D. (1998). Pipeline leak detection techniques. In E. W. McAl-
also satisfactory. However, the estimation of leak magnitude is lister (Ed.), Pipeline rules of thumb handbook (pp. 476–484). Houston: Gulf.
good only when the measurement errors are not significant. The Geiger, G., Werner, T., & Matko, D. (2003). Leak detection and locating—a survey. In
computational time required for estimating the leak location and Pipeline Simulation Interest Group (Ed.), Proceedings of 35th annual meeting of
PSIG held in Bern Switzerland, October 15–17, paper no. 301.
leak magnitude is less than one sampling period (10 s). How- Gopalsami, N., & Raptis, A. C. (2001). Millimeter-wave radar sensing of airborne
ever, detailed studies have not been carried out for benchmarking chemicals. IEEE Transactions on Microwave Theory and Techniques, 49, 646–653.
the proposed methodology (in terms of computational time and Griebenow, G., & Mears, M. (1989). Leak detection implementations: Modeling and
tuning methods. Journal of Energy Research Technology, 111, 66–72.
accuracy) with the other existing model based methods. Such a Kim, S. H. (2005). Extensive development of leak detection algorithm by impulse
study will be very useful and can be undertaken as part of future response method. Journal of Hydraulic Engineering-ASCE, 131, 201–208.
research. Kralik, J., Stiegler, P., Vostry, Z., & Zavorka, J. (1984). Modeling the dynamics of flow
in gas pipeline. IEEE Transactions on Systems, Man, and Cybernetics, 14, 586–
596.
4. Conclusions Lee, P. J., Vitkovsky, J. P., Lambert, M. F., Simpson, A. R., & Liggett, J. A. (2005).
Frequency domain analysis for detecting pipeline leaks. Journal of Hydraulic
Engineering-ASCE, 131, 596–604.
A computationally efficient transfer function based state esti- Liggett, J. A., & Chen, L. C. (1994). Inverse transient analysis in pipe networks. Journal
mation model for dynamic flows has been successfully used in a of Hydraulic Engineering-ASCE, 120, 934–955.
hypothesis testing framework for developing an approach for leak Liou, C. P., & Tian, J. (1994). Leak detection: Transient flow simulation approach.
Pipeline Engineering, 60, 51–58.
detection and identification in gas pipeline networks. Ability of the
Liu, M., Zang, S., & Zhou, D. (2005). Fast leak detection and location of gas pipelines
proposed method to accurately detect and isolate leaks is evalu- based on an adaptive particle filter. International Journal of Applied Mathematics
ated using numerical simulations of a real life 204.7 km long series and Computer Science, 15, 541–550.
McLean, A., Moran, C., Johnstone, W., Culshaw, B., Marsh, D., & Parker, P. (2003).
pipeline, and a hypothetical pipeline network. Leak magnitudes
Detection of hydrocarbon fuel spills using distributed fiber optic sensor. Sensors
considered ranged from 2% to 10% of normal flow rate. Results of and Actuators A, 109, 60–67.
numerical runs indicated that the proposed methodology works Misiunas, D., Vitkovsky, J., Olsson, G., Simpson, A., & Lambert, M. (2005). Pipeline
very well if noise level in the measured data is low. Under nearly break detection using pressure transient monitoring. Journal of Water Resources
Planning and Management-ASCE, 131, 316–325.
steady state conditions, leaks of magnitude even as low as 2% of the Mpesha, W., Chaudhry, M. H., & Gassman, S. L. (2002). Leak detection in pipes by
flow rate can be detected quickly and identified accurately. How- frequency response method using a step excitation. Journal of Hydraulic Research,
ever, when leak occurs during transient conditions, then the leak 40, 55–62.
Mukherjee, J., & Narasimhan, S. (1996). Leak detection in networks of pipelines by the
detection performance is satisfactory for magnitudes larger than generalized likelihood ratio method. Industrial & Engineering Chemical Research,
or equal to 5% of flow rate. In case of noisy data, performance of 35, 1886–1893.
the proposed method improves with increase in redundancy in the Narasimhan, S., & Jordache, C. (2000). Data reconciliation & gross error detection—an
intelligent use of process data. Houton: Gulf Publishing.
measurements. Osiadacz, A. J. (1989). Simulation and analysis of gas networks. Houston: Gulf Pub-
lishing Company.
Acknowledgement Press, W. H., Flannery, B. P., Teukolsky, S. A., & Vetterling, W. T. (1998). Numerical
Recipes in C. Cambridge University Press.
Reddy, H. P., Narasimhan, S., & Bhallamudi, S. M. (2006). Simulation and state esti-
This research work was financially supported by GAIL (India) mation of transient flow in gas pipeline networks using transfer function model.
Ltd. under sponsored project “Development of leak detection meth- Industrial & Engineering Chemical Research, 45, 3853–3863.
Sandberg, C., Holmes, J., McCoy, K., & Koppitsch, H. (1989). Application of a con-
ods in gas pipeline networks”.
tinuous leak detection system to pipelines and associated equipment. IEEE
Transactions on Industry Application, 25, 906–909.
References Scott, S. L., & Barrufet, M. A. (2003). Worldwide assessment of industry leak detection
capabilities for single and multiphase pipelines (OTRC Library Number: 8/03A120).
Austin: University of Texas Austin.
American Gas Association. (1994). Compressibility factor of natural gas and related
Sharp, D. B., & Campbell, D. M. (1997). Leak detection in pipes using acoustic pulse
hydrocarbon gases, report 8. Arlington: AGA.
reflectometry. Acta Acoustics, 83, 560–566.
Andersen, J. H., & Powell, R. S. (2000). Implicit state-estimation technique for water
Tracer Research Corporation. (2003). http://www.tracerresearch.com.
network monitoring. Urban Water Journal, 2, 123–130.
Vitkovsky, J. P., Simpson, A. R., & Lambert, M. F. (2000). Leak detection and calibration
Beck, S. B. M., Curren, M. D., Sims, N. D., & Stanway, R. (2005). Pipeline network fea-
using transients and genetic algorithms. Journal of Water Resources Planning and
tures and leak detection by cross correlation analysis of reflected waves. Journal
Management-ASCE, 126, 262–265.
of Hydraulic Engineering-ASCE, 131, 715–723.
Wang, X. J., Lambert, M. F., Simpson, A. R., Liggett, J. A., & Vitkovsky, J. P. (2002). Leak
Benkherouf, A., & Allidina, Y. A. (1988). Leak detection and location in gas pipelines.
detection in pipelines using the damping of fluid transients. Journal of Hydraulic
IEE Proceedings-D, 135, 142–148.
Engineering-ASCE, 128, 697–711.
Billmann, L., & Isermann, R. (1987). Leak detection methods for pipelines. Automatica,
Watanabe, K., & Himmelblau, D. M. (1980). Detection and location of a leak in a
239, 381–385.
gas-transport pipeline by a new acoustic method. AIChE Journal, 32, 1690–1701.
Brunone, B., & Ferrante, M. (2001). Detecting leaks in pressurized pipes by means of
White, F. M. (1986). Fluid mechanics. New York: McGraw-Hill.
transients. Automatica, 39, 1–9.

You might also like