condition assessment of water pipelines using a modified...
TRANSCRIPT
Condition Assessment of Water Pipelines using a Modified Layer 1
Peeling Method 2
3
Wei Zeng1, Jinzhe Gong2, Aaron C. Zecchin3, Martin F. Lambert4, Angus R. 4
Simpson5, Benjamin S. Cazzolato6 5
1PhD Candidate; School of Civil, Environmental and Mining Engineering, University of 6
Adelaide, SA 5005, Australia; Email: [email protected] 7
2Postdoctoral Research Fellow; School of Civil, Environmental and Mining Engineering, 8
University of Adelaide, SA 5005, Australia; Email: [email protected] 9
3Senior Lecturer; School of Civil, Environmental and Mining Engineering, University of 10
Adelaide, SA 5005, Australia; Email: [email protected] 11
4Professor; M.ASCE; School of Civil, Environmental and Mining Engineering, University of 12
Adelaide, SA 5005, Australia; Email: [email protected] 13
5Professor; M.ASCE; School of Civil, Environmental and Mining Engineering, University of 14
Adelaide, SA 5005, Australia; Email: [email protected] 15
6Professor; School of Mechanical Engineering, University of Adelaide, SA 5005, Australia; 16
Email: [email protected] 17
18
ABSTRACT 19
Pipe wall condition assessment is critical for the targeted maintenance and for failure 20
prevention in water distribution systems. This paper proposes a novel approach for 21
condition assessment of water pipelines by adapting the layer peeling method. This 22
method was previously developed for, and applied to, tubular musical instruments. In 23
the proposed approach, the impulse response function (IRF) of a pipeline is obtained 24
using measured pressure traces resulting from transient events. The original layer 25
peeling method is further developed for application to water transmission pipelines by 26
1) modifying the end boundary from being an acoustic source tube to a closed valve 27
boundary condition; 2) incorporating the effects of unsteady friction and pipe wall 28
viscoelasticity into the layer peeling algorithm; and 3) incorporating frequency-29
dependent wave reflections and transmissions. Using the IRF and the modified layer 30
peeling method, the impedance of a pipeline can be estimated section by section from 31
downstream (the dead-end) to upstream of the pipeline. The distribution of wave speeds 32
and wall thickness can then be determined. In this study, numerical verifications were 33
conducted using the pipeline pressure responses simulated by the method of 34
characteristics (MOC). The deteriorated pipe sections (sections with changes in 35
impedance) were accurately detected using the new approach. Experimental 36
verification of the result was conducted on a laboratory copper pipeline. A short section 37
of pipe with a thinner wall thickness was successfully detected. 38
Keywords: hydraulic transient; water hammer; pipeline condition assessment; impulse 39
response function; layer peeling method; unsteady friction; viscoelastic effects. 40
1. Introduction 41
Water distribution systems (WDS) are important infrastructure assets. They normally 42
consist of buried pipeline networks that are massive in scale, and often old and 43
deteriorated. However, the condition of these pipes is extremely difficult and expensive 44
to determine. Over the past two decades, a number of non-invasive, hydraulic, transient-45
based methods have been developed for fault detection in water pipelines (Chaudhry 46
2014). Hydraulic transient-based fault detection in a pipeline system is conducted with 47
a transient disturbance, typically a pulse or a step pressure wave introduced by abruptly 48
operating a valve. The resulting transient waves propagate along the pipeline, and any 49
physical changes or anomalies such as leaks or blockages may result in specific wave 50
reflections (Lee et al. 2007). Anomalies in the pipes are located by analyzing time series 51
and wave speed (Brunone 1999). The size of a leak or a blockage can be determined by 52
examining the magnitude of the wave reflection. 53
Most of the transient-based methods focus on the detection of discrete elements, such 54
as leaks (Brunone 1999; VΓtkovskΓ½ et al. 2007; Ferrante et al. 2009; Shamloo and 55
Haghighi 2009; Duan et al. 2011; Gong et al. 2014a) and blockages (Wang et al. 2005; 56
Lee et al. 2008; Sattar et al. 2008; Meniconi et al. 2011). In addition to discrete faults, 57
extended blockages caused by tuberculation (Duan et al. 2012) and extended sections 58
of pipe degradation caused by spalling of the cement mortar lining and widespread 59
corrosion are common in ageing WDS; however, research on transient-based pipe wall 60
condition assessment is limited. Distributed pipe deterioration may reduce the water 61
transmission efficiency (Tran et al. 2010), create water quality problems (Vreeburg 62
and Boxall 2007) and may develop into bursts or severe blockages over time 63
(Zamanzadeh et al. 2007). Thus, developing cost-effective techniques for pipeline 64
condition assessment is essential in enabling strategically targeted pipe maintenance, 65
replacement and rehabilitation. 66
The inverse transient analysis (ITA) (Stephens et al. 2013) and the reconstructive 67
method of characteristics (MOC) analysis (Gong et al. 2014b) are two available, 68
transient-based techniques for continuous pipe-wall condition assessment. The inverse 69
transient method (ITA) was first proposed by Liggett and Chen (1994) for detecting 70
leaks in a pipe network. It was further developed by VΓtkovskΓ½ et al (2000; 2007) and 71
Covas et al.(2010) and was first applied to pipeline condition assessment by Stephens 72
et al. (2008; 2013). 73
In ITA, the transient responses of a numerical pipeline model with assumed parameters 74
(e.g. wave speeds) are simulated by the MOC. Then, an optimization process is 75
conducted to minimize error between numerical and experimental values. By 76
iteratively modifying the parameters in the model, the best possible match of the 77
numerical pressure traces with the traces measured from a real pipe can be obtained. 78
The numerical pipe model that provides the best match is used to interpret the condition 79
of the real pipeline. However, ITA is not computationally efficient, especially for 80
pipelines of substantial length which involves a large number of parameters to calibrate. 81
There are also many practical issues (e.g. wave dissipation and dispersion) that affect 82
the identifiability of the technique (VΓtkovskΓ½ et al. 2007). 83
The reconstructive MOC analysis assesses the condition of a pipeline section by section 84
by inverting the conventional MOC calculation (Gong et al. 2014b). The method is 85
computationally efficient, because the pipeline properties are calculated analytically 86
instead of through an iterative, optimizer-driven model calibration process. However, 87
when the pipeline configuration is either complex, or wave dissipation is significant or 88
dispersion needs to be considered, it would be impossible to handle these situations 89
using the reconstructive MOC algorithm. 90
A technique that shares a similar principle to the reconstructive MOC analysis is the 91
layer peeling method, which was developed in the acoustics research field. An early 92
technique named acoustic pulse reflectometry was proposed by Ware and Aki (1969) 93
as a seismological technique for observing stratifications in the earthβs crust. The layer 94
peeling method was built on this, and applied reconstructions of the geometry of short 95
air ducts with varying cross sections (Amir and Shimony 1995b, a). It was then applied 96
to bore reconstruction of musical wind instruments in order to inspect their qualities 97
(Sharp and Campbell 1997). In these applications, an acoustic source tube with 98
properties designed to prevent or control wave reflections is attached to one end of the 99
duct/instrument to extract the impulse response function (IRF) of the system. The IRF 100
is then analyzed to determine the acoustic impedance and reflection coefficients moving 101
section by section away from the acoustic source. Further research increased the length 102
of reconstruction (Sharp 1998), enhanced the robustness (Forbes et al. 2003), increased 103
the axial resolution (Li et al. 2005), and coupled the higher mode acoustic waves in the 104
method (Hendrie 2007). However, there is no application of the layer peeling concept 105
to pipeline condition assessment using hydraulic transients to date. 106
The research reported in this paper develops a novel pipeline condition assessment 107
approach that uses the IRF of a pipeline and a modified layer peeling algorithm. The 108
key innovations include: the use of a dead-end boundary condition instead of an 109
acoustic tube and the determination of the directional IRF at the dead-end; the 110
incorporation of wave dissipation and dispersion induced by unsteady friction and pipe 111
wall viscoelasticity; and the incorporation of the frequency-dependent wave reflection 112
and transmission at cross-sections with an impedance change. 113
The new technique enables the distribution of pipeline impedance to be determined, 114
from which wall thickness and wave speed along the pipe can be reconstructed. 115
Compared with the ITA method (Stephens et al. 2013), the proposed approach is more 116
efficient, because it does not need a time-consuming, iterative, forward-modelling and 117
optimization process. Compared with the reconstructive MOC analysis, the proposed 118
approach is more accurate, because it can conveniently incorporate frequency-119
dependent effects, such as wave dissipation and dispersion. 120
To validate the new layer-peeling-based approach, extensive numerical simulations 121
have been conducted for pipes with and without both friction or viscoelasticity; with 122
uniformly and non-uniformly distributed deteriorations; and with and without 123
measurement noise. For all the numerical cases, the distribution of the wave speed along 124
the pipe was accurately reconstructed. Experimental verification was also conducted on 125
a copper pipeline in the laboratory. A pipe section with a thinner wall thickness was 126
clearly identifiable in the reconstructed pipe impedance. Towards the end of this paper, 127
a discussion of the limitations and practical challenges of the new method is presented. 128
Finally conclusions are made. 129
2. The Layer Peeling Method for Pipeline Condition Assessment 130
The principle of the new approach is illustrated in Fig. 1, which is a block diagram 131
describing the wave propagation and reflection process in a pipeline. In the figure, the 132
superscripts (+ and β) illustrate forward and backward directions respectively, the 133
subscripts 1, 2, and i represent the number of the pipe section, and the subscripts l and 134
r represent the left side and right side of the section. 135
The method includes the four following main steps. First, the IRF of a pipeline system 136
(P1,l ), which is the input to the modified layer peeling algorithm, is determined through 137
transient analysis first, and pre-processed to yield the directional IRF (π1,π+ and π1,π
β ). 138
Second, the dissipation, and the dispersion of pressure waves during the propagation in 139
each pipe section, is formulized as a transfer function (hi). Thirdly, the wave 140
transmission and reflection at the interface between pipe sections is represented using 141
reflection ratios (ri,i+1 and ri+1,i) and transmission ratios (si,i+1 and si+1,i). Finally, a 142
recursive procedure is applied to reconstruct the reflection and transmission ratios 143
section by section, using the IRF and the propagation functions. These four components 144
are discussed in detail in subsequent sections. 145
146
147
Fig. 1 Block diagram describing the wave transmission and reflection process in a 148
pipeline. 149
2.1 Impulse response function (IRF) 150
This section illustrates the two steps to get the directional IRF that is the input of the 151
layer peeling method. First, a singular value decomposition is used to obtain the IRF of 152
the pipeline system. The formulae to transfer the system IRF to the directional IRF are 153
then given. 154
2.1.1 Methods for obtaining the system IRF 155
The IRF π§(π‘) is defined as the response measured at the output when an ideal impulse 156
input is injected into a system. In a linear transmission system, a real signal input π₯(π‘) 157
with N data samples, such as a pulse, can be treated as N scaled impulses with different 158
starting times. Each scaled impulse generates a scaled response that starts at the same 159
time as the input impulse. Thus, the overall response of the system π¦(π‘) is the sum of 160
these scaled responses, which may be written as (Smith 1999) 161
π¦(π) = β π₯(π β π)π§(π)βπ=ββ (1) 162
The method used to obtain the IRF in this research is based on deconvolution in the 163
time domain using the singular value decomposition (AgullΓ³ et al. 1995). Equation (1) 164
can be rewritten in the matrix form π² = ππ³, in which π is a triangular matrix of π₯ππ =165
π₯πβπ+1 for π β€ π and π₯ππ = 0 for π > π , and π² is the column vector of the wave 166
reflections. Applying the singular valve decomposition to π gives π = ππ²ππ, where 167
π, π are orthonormal matrices composed of column vectors π©π and πͺπ, respectively, 168
and π² is a diagonal matrix composed of the singular values Ξ»π sorted in descending 169
order of size. Then π can be written as π = β Ξ»ππͺππ©πππ
π=1 and the IRF π³ = πβ1π² can be 170
written as (AgullΓ³ et al. 1995) 171
π³ = (β ππβ1π©ππͺπ
πππ=1 )π² (2) 172
However, a small value of Ξ»π when i approaches N makes a significant contribution 173
to πβ1 , which results in distorted deconvolution. A combination of a truncation 174
regularization and a Tikhonovβs regularization gives (Forbes et al. 2003) 175
π³ = (βΞ»π
ππ2+πΌπ
π©ππͺπππ½
π=1 ) π² (3) 176
where J is the truncation point and J<N, πΌπ is an regularization parameter. A smaller J 177
or a larger πΌπ results in a smoother but less sharp IRF. The two parameters can be 178
determined by trial and error until satisfactory results are achieved (Forbes et al. 2003). 179
The algorithm in Eq. (3) is used in this paper. 180
181
2.1.2 From the system IRF to the directional IRF 182
In the conventional layer peeling method, which has previously been applied to musical 183
instruments (Sharp 1996), a long, uniform, source tube, which is two times longer than 184
the musical instrument, was attached to one end of the instrument, as shown in Fig. 2 185
(a). The source tube enables the identification and separation of the forward signal input 186
and the backward wave reflections. Thus, the IRF calculated from the system response 187
is directional. It represents the impulse reflections from the instrument only (i.e. no 188
impact from the source tube). 189
In a water transmission line systems, it is not feasible to connect a long source tube to 190
a water pipeline. Instead, a dead-end boundary condition is considered by closing an 191
inline valve, as shown in Fig. 2 (b). A pressure wave generator and a transducer are 192
installed close to the dead-end. 193
194
Fig. 2 Schematic diagram of the testing systems; (a) musical instrument, and (b) 195
water pipeline. 196
Due to the dead end, the directional impulse reflection π1,πβ would be further reflected 197
by the end boundary, and will again enter into the pipeline system, as shown in the first 198
dashed box in Fig. 1. Neglecting any loss in the reflection at the dead-end, i.e. the 199
reflection coefficient is unity, the magnitude of the system IRF at the dead end [π1,π, 200
may be directly determined from Eq. (3)] is effectively two times that of the directional 201
impulse reflections from the pipe discontinuities (backward propagating wave, π1,πβ ), i.e. 202
π1,π = 2π1,πβ (4) 203
As shown in Fig. 1(a), the forward propagating wave into the pipeline is described as 204
π1,π+ = πππππ’ππ π
+ + π1,πβ (5) 205
where πππππ’ππ π+ represents an impulse signal. 206
The two directional IRF signals, which represent the impulse reflections from the pipe 207
discontinuities and the forward-propagating wave into the pipeline respectively, can be 208
obtained through Eqs. (4) and (5), and written as 209
π1,πβ = 0.5π1,π (6) 210
π1,π+ = πππππ’ππ π
+ + 0.5π1,π (7) 211
2.2 Wave dissipation and dispersion 212
For a fluid-filled pipe, the wave speed (π) in the fluid depends on the properties of the 213
fluid and the pipe wall. For deteriorated sections of a metallic pipeline, the change in 214
wall thickness affects the wave speed governed by the following formula (Wylie and 215
Streeter 1993), 216
π = βπΎ πβ
1+(πΎ πΈβ )(π· πβ )π1 (8) 217
where πΎ represents the bulk modulus of the water; π is the density of water; πΈ is the 218
Youngβs modulus of elasticity of the pipe wall; π· is the pipeβs inner diameter; π is the 219
wall thickness of the pipe; and π1 is the pipeline restraint factor. The value of π1 220
depends on whether the pipe is thin walled (D/e > 25) or thick walled (D/e β€25) (Wylie 221
and Streeter 1993). For a structurally degraded, viscoelastic pipeline, the wave speed 222
may also change due to the degradation of the pipe elasticity. 223
Pressure waves in pipelines experience frequency-dependent dissipation and dispersion 224
due to friction and viscoelasticity from the pipe wall (Gong et al. 2015a). The wave 225
dissipation and dispersion in the i th pipe section (within which the properties are 226
assumed uniform) can be described by a transfer function βπ such that 227
ππ,π+ = ππ,π
+ β βπ (9) 228
ππ,πβ = ππ,π
β β βπβ1 (10) 229
where β means the convolution in the time domain, π is the Fourier transformation of 230
π, and the inverse in (10) refers the inverse convolution, that is, for function a(t), π β231
πβ1 = πΏ(π‘). 232
The following part of this section illustrates the processes used to obtain the analytical 233
expression of the transfer function. 234
The oneβdimensional (1D) momentum equation for transient fluid flow (Wylie and 235
Streeter 1993) is given as 236
1
π΄
ππ
ππ‘+
ππ
πππ₯+ πβππ = 0 (11) 237
in which Q represents the volumetric flow rate, g is acceleration due to gravity, A is the 238
internal cross-section area of the section, and the head loss βππ, which is the sum of a 239
steady-state component βππ and an unsteady-state component βππ’. 240
The continuity equation for the 1D transient flow incorporating the viscoelastic 241
behavior of the pipe wall is given as (Covas et al. 2005) 242
π΄
π2
ππ
ππ‘+ π
ππ
ππ₯+ 2ππ΄
πππ
ππ‘= 0 (12) 243
where νπ is the retarded circumferential strain in the pipe wall. 244
Using the concept of steady-oscillatory flow with h, q and νπβ representing the time 245
varying oscillatory components of flow, head and retarded strain, Eq. (11) and Eq. (12) 246
can be transformed into the frequency domain (Gong et al. 2016) as 247
1
π΄πππ +
ππ
πππ₯+ π ππ = 0 (13) 248
π΄
π2 πππ + πππ
ππ₯+ 2ππ΄ππνπ
β(ππ) = 0 (14) 249
Where j is the imaginary unit, π is the linearized resistance per unit length, and can be 250
described by a summation of the steady friction part π π and unsteady friction part π π’, 251
i.e. 252
π = π π + π π’ (15) 253
where π π = ππ0/(ππ·π΄2), with f representing the Darcy-Weisbach friction factor. The 254
expression of π π’ depends on the selection of the unsteady friction model in the time 255
domain. If the Vardy and Brown unsteady friction model for smooth-pipe turbulent 256
flow (Vardy and Brown 1995) is chosen, the expression of π π’ can be written based on 257
the derivation by VΓtkovskΓ½ et al. (2003) as 258
π π’ =2ππ
ππ΄(
πππ·2
4π+
1
πΆβ)β1 2β
(16) 259
where π is kinematic viscosity of fluid and πΆβ is the shear decay coefficient. πΆβ =260
0.00476 for laminar flows and πΆβ = 7.41/π ππ for smooth pipe turbulent flow with 261
π = log10(14.3/π π0.05) and Re representing the Reynolds number. 262
With the multi-element Kevin-Voigt model (Covas et al. 2005) for viscoelastic effects, 263
and the corresponding frequency domain transformation (Gong et al. 2016), νπβ(ππ) in 264
Eq.(14) can be written as 265
νπβ(ππ) = πΆβ β
π½π
ππππ+1
ππ=1 (17) 266
where πΆ = π1π·ππ/2π, π½π is the creep-compliance of the k th Kevin-Voigt element 267
and ππ is the retardation time of the dashpot of the k th Kevin-Voigt element. 268
Combining Eqs. (13), (14) and (17) yields 269
π2π
ππ₯2 + (π2 β ππ΄π ππ) (2πΆ
πβ
π½π
ππππ+1
ππ=1 +
1
ππ2) π = 0 (18) 270
The equation can be also written as 271
π2π
ππ₯2+
π2
ππ2 π = 0 (19) 272
with the general solution for the oscillatory pressure wave 273
π(π₯, π‘) = πππ(π‘βπ₯ ππ)β (20) 274
Thus, the pressure wave after propagating a distance of βπ₯ can be written as 275
π(π₯ + βπ₯) = π(π₯)πππβπ₯ ππβ (21) 276
in which ππ is the complex wave speed described by 277
ππ = βπ2
(π2βππ΄π ππ)(2πΆ
πβ
π½πππππ+1
ππ=1 +
1
π2) (22) 278
If friction is neglected (Gong et al. 2016) and then π = 0, then 279
ππ = β1
(2πΆ
πβ
π½πππππ+1
ππ=1 +
1
π2) (23) 280
If viscoelastic behavior of the pipe wall is neglected, then the expression for the 281
complex wave speed simplifies to 282
ππ = πβ1
(1βππ΄π π πβ ) (24) 283
If both the friction and the viscoelastic effects of the pipe wall are neglected, the 284
expression of the complex wave speed ππ collapses to the elastic wave speed π. 285
The complex wave speed ππ can be also treated in the following form ππ = ππ + πππ 286
(Suo and Wylie 1990) such that ππ and ππ in Eq.(25) can represent the frequency 287
dependent wave speed and attenuation respectively with the following expressions: 288
ππ =|ππ|
ππ πππ ππ =
πππ
|ππ| (25) 289
For a pressure wave propagating from the left boundary to the right boundary of the i 290
th pipe section, the following equation can be obtained through Eqs. (21) and (25). 291
ππ,π+ = ππ,π
+ β πβ ππβπ₯ππβππβπ₯π ππβ (26) 292
in which T/2 is the time needed for the wave to pass through the i th section, βxi is the 293
length of the section, πβ ππβπ₯π represents the wave dissipation and πβππβπ₯π ππβ 294
represents the wave dispersion. The transfer function in Eqs. (9) and (10) is given by 295
βπ = πβ ππβπ₯ππβππβπ₯π ππβ (27) 296
The transfer function βπ represents the wave dissipation and dispersion, and will be 297
used in the reconstruction process outlined in the following section 2.4. 298
2.3 Wave transmission and reflection 299
If an incident pressure wave p (in the frequency domain) meets a discontinuity in the 300
pipe (an interface with an impedance change), a reflected wave pr will be generated. 301
The amplitude of the incident wave will change to ps after passing the discontinuity. 302
The reflection coefficient π (the ratio of the complex amplitude of the reflected wave 303
to that of the incident wave) and the transmission coefficient π (the ratio of the complex 304
amplitude of the transmitted wave to that of the incident wave) are determined by the 305
pipeline characteristic impedance π΅ . They can be calculated using the formulae 306
(Chaudhry 2014) 307
ππ,π+1 =ππ
π=
π΅π+1βπ΅π
π΅π+1+π΅π (28) 308
π π,π+1 =ππ
π=
2π΅π+1
π΅π+1+π΅π= 1 + ππ,π+1 (29) 309
in which 310
π΅ = ππ ππ΄β (30) 311
is the complex characteristic impedance. According to Eq. (23), it can be seen that ππ,π+1 312
and π π,π+1 are frequency-dependent for the viscoelastic model. For the unsteady friction 313
model, the complex wave speed ππ is proportional to the elastic wave speed π if A is 314
assumed to be constant in Eq. (24). Thus, ππ,π+1 and π π,π+1 are frequency-independent 315
for the unsteady friction model when Eq. (24) is substituted into Eqs. (28) and (29). 316
If the incident wave propagates in the opposite direction, then the reflection coefficient 317
and the transmission coefficient can be calculated as: 318
ππ+1,π = βππ,π+1 (31) 319
π π+1,π = 1 β ππ,1+1 (32) 320
At the interface of two sections as shown in Fig. 3, the transmitted waves and reflected 321
waves are represented by the dot-dashed arrows and dashed arrows, respectively. 322
According to the directions of the arrows, it follows that the forward-propagating wave 323
ππ+1,π+ travelling into section i+1 is the sum of the transmitted wave of ππ,π
+ and the 324
reflected wave of ππ+1,πβ . The backward-propagating wave ππ,π
β travelling into section i 325
is the sum of the reflected wave of ππ,π+ and the transmitted wave of ππ+1,π
β . Combining 326
Eqs. (28) to (32), the following formulae can be written based on the analysis above: 327
ππ+1,π+ = ππ+1,π
β (βππ,π+1) + ππ,π+ (1 + ππ,π+1) (33) 328
ππ,πβ = ππ,π
+ (ππ,π+1) + ππ+1,πβ (1 β ππ,π+1) (34) 329
The relationship of the wave transmission and reflection can also be represented by the 330
third dashed box in Fig. 1. By rearranging the two formulae, the relationship between 331
the travelling waves at either side of a section interface can be summarized as (Amir 332
and Shimony 1995b) 333
[ππ+1,π
+
ππ+1,πβ ] =
1
1βππ,π+1[
1 βππ,π+1
βππ,π+1 1] [
ππ,π+
ππ,πβ ] (35) 334
335
Fig. 3 Wave transmission and reflection at a junction 336
In the time domain, the space-time diagram of the wave reflection and transmission are 337
shown in Fig. 4 for a pipeline discretized into sections. The forward-propagating waves 338
along the diagonal in the diagram are defined as the main transmitted waves ππ,π+ (ππ 2β ) 339
(thick solid lines). The backward propagating waves, at the same time and position as 340
the main transmitted waves, are defined as the initial reflected waves ππ,πβ (ππ 2β ) 341
(dashed lines). If the frequency-dependent characteristics of the wave reflection and 342
transmission are neglected in the time domain, the initial reflected waves are only 343
caused by the reflection of the main transmitted waves, as shown in Fig. 4, which means 344
ππ,π+1π‘ =
ππ,πβ (ππ 2β )
ππ,π+ (ππ 2β )
(36) 345
in which ππ,π+1π‘ is the elastic wave reflection ratio in the time domain. If the elastic 346
characteristic impedance π΅ππ‘ = ππ/ππ΄π is known, the real characteristic impedance 347
π΅π+1π‘ can be estimated according to Eq. (28) as 348
π΅π+1π‘ =
1+ππ,π+1π‘
1βππ,π+1π‘ π΅π
π‘ (37) 349
350
351
352
Fig. 4 Space-time diagram of the wave propagation 353
2.4 Procedures of the Modified Layer Peeling Method 354
The steps for reconstructing a pipeline with N sections using the modified layer peeling 355
method are shown in Fig. 5 (Note π΅1 is obtained as in Gong et al. (2014b)) and 356
described as follows. 357
358
359
Fig. 5 Main steps of the modified layer peeling method for pipeline condition 360
assessment. 361
Step 1: Use experimental data to calculate the system IRF through Eq. (3) 362
Step 2: Use the system IRF to calculate the directional IRF π1,πβ and forward-363
propagating wave π1,π+ through Eqs. (6) and (7). 364
Step 3: Use the waves at the left side of the ith (i=1 for the first step) section ππ,π+ and ππ,π
β 365
to calculate the waves at the right side of the ith section ππ,π+ and ππ,π
β through Eqs. (9) 366
and (10). 367
Step 4: Use the waves at the right side of the ith section ππ,π+ and ππ,π
β (transferred to the 368
time domain) to calculate the elastic reflection ratio ππ,π+1π‘ through Eq. (36), and then 369
obtain the elastic characteristic impedance of the next section π΅π+1π‘ through Eq. (37), as 370
well as the complex characteristic impedance π΅π+1 through Eqs.(22) and (30). 371
Step 5: Use the complex π΅π and π΅π+1 to calculate the frequency-dependent wave 372
reflection ratio ππ,π+1. Then, use the waves at the right side of the i th section ππ,π+ , ππ,π
β 373
and ππ,π+1 to calculate the waves at the left side of the (i+1)th section ππ+1,π+ and ππ+1,π
β 374
through Eq. (35). 375
Step 6: Repeat steps 3 to 5 for i = 2, β¦, N-1 to calculate the characteristic impedances, 376
wave speeds and wall thicknesses for the remaining sections. 377
3. Numerical Verification 378
Numerical simulations were conducted on several reservoir-pipeline-valve systems to 379
verify the proposed approach for pipeline condition assessment. The MOC (Wylie and 380
Streeter 1993) has been used to obtain the transient pressure responses induced by a 381
flow fluctuation. For large pipes with a constant outer dimeter, a small change in wall 382
thickness slightly alters the inner diameter, but can cause large change in the wave 383
speed according to Eq.(8).Thus, only the wave speed varies along the pipeline to 384
simulate extended deterioration, and the inner diameter stays constant in the numerical 385
cases. 386
3.1 Case 1: Frictionless pipe 387
The first case study was conducted for a frictionless metallic pipeline with a uniformly 388
deteriorated section and a non-uniformly deteriorated section. The inner diameter of the 389
frictionless pipeline is assumed to be 600 mm throughout, and the wave speed in the 390
normal sections of the pipeline is 1000 m/s. The pipeline configuration and the 391
properties of the deteriorated sections are given in Fig. 6. The wave speed in the 392
uniformly deteriorated section is 800 m/s. The wave speed in the non-uniformly 393
deteriorated section has a gradual change from 1000 m/s to 800 m/s and back to 1000 394
m/s following one period of a cosine pattern (as shown in Fig. 8) of degradation over a 395
length of 108.3 m. A pressure pulse wave, shown in Fig. 7(a), induced by a flow pulse 396
was injected into the pipeline at the upstream face of the closed valve, and the wave 397
reflections were simulated using a frictionless MOC model (time step = 0.001 s for all 398
the numerical cases) that was given in Fig. 7 (b). The primary reflections marked in the 399
figure were directly induced by the deteriorated sections, and the other much smaller 400
reflections were caused by high-order reflections. Note that reflections from the closed 401
valve (dead-end) contributed to the signals seen in Fig. 7(b). 402
The proposed layer-peeling-based technique applied to the signals in Fig. 7 accurately 403
yields reconstructed wave speed distributions that are almost identical to the theoretical 404
values (as shown in Fig. 8). To determine how robust the proposed method is against 405
measurement noise, the reflection signal was contaminated with white Gaussian noise, 406
with a signal-to-noise ratio of 10. The mixed signal shown in Fig. 9(a) was then used 407
to calculate the system IRF [Fig. 9(b)] of the pipeline using Eq. (3). The modified layer 408
peeling method, without considering any wave dissipation or dispersion, was then 409
applied to reconstruct the pipeline. The outcome shown in Fig. 10 demonstrates that the 410
proposed technique can reconstruct the pipeline condition (wave speed in this case) for 411
both uniformly and non-uniformly deteriorated sections, even when measurement noise 412
is present. 413
414
Fig. 6 Pipeline configuration 1 415
416
Fig. 7 Input and reflected signals 417
418
Fig. 8 Wave speed reconstructed from the modified layer peeling method for the 419
frictionless pipe without measurement noise. (The two plots are virtually coincident) 420
421
422
Fig. 9 Reflection with noise and IRF 423
424
425
Fig. 10 Wave speed reconstructed from the modified layer peeling method for the 426
frictionless pipe with measurement noise. 427
3.2 Case 2: Effects of unsteady friction 428
A reservoir-pipeline, closed-valve system with one uniformly deteriorated section 429
(shown in Fig. 11) was considered to analyze the effect of the unsteady friction. The 430
inner diameter of the pipeline is 50.6 mm and the wave speed of a normal pipeline is 431
1000 m/s. The steady-state flow rate π0 is 0.1 L/s (V = 0.05m/s) and the Darcy-432
Weisbach factor π is 0.02. A smaller-sized pipe (compared to the pipe in Fig. 6) was 433
chosen to highlight the effect of unsteady friction [unsteady friction effects are 434
insignificant and typically negligible for the primary reflections in large pipes (Stephens 435
et al. 2013)]. A pulse wave with the same duration and waveform as case 1 was injected 436
into the pipeline just upstream of the closed valve, and the pressure wave reflections at 437
the same point were simulated using an unsteady friction MOC model (Vardy and 438
Brown 1995). 439
The pipeline was initially reconstructed using the modified layer peeling method which 440
is used in the frictionless case, without considering any signal dissipation or dispersion 441
(thus ignoring unsteady friction), and the result is plotted as the dashed line in Fig. 12. 442
Obvious errors are shown between 200 m to 350 m. 443
Another reconstruction was conducted using the modified layer peeling method 444
incorporating the transfer function that describes the unsteady friction using Eq. (24), 445
(25) and (27). It needs to be emphasized that the transfer function was renewed in every 446
step using the wave speed (and wall thickness in the experimental case), calculated in 447
the previous time step. The result shown as the dot-dashed line in Fig. 12 illustrates that 448
the error was eliminated along the pipeline except at points where the impedance 449
changed sharply (200 m and 300 m). The error at the impedance change interfaces was 450
induced by the ripples of the IRF when transformed from the wave reflections. 451
452
Fig. 11 Pipeline configuration 2 453
454
455
Fig. 12 Wave speed reconstructed from the modified layer peeling method for the unsteady 456
friction case. 457
3.3 Case 3: Effect of viscoelasticity 458
The pipeline system shown in Fig. 13 was used to analyze the viscoelastic effects of the 459
pipe wall. Friction was not considered, so that the viscoelastic effects would be 460
highlighted. The inner diameter of the pipeline was 50.6 mm and the wave speed of a 461
normal pipeline is 393 m/s. The physical details of the pipeline are adapted from the 462
experimental pipeline at Imperial College as reported in Covas et al. (2005), and the 463
viscoelastic parameters used are listed in Table 1. A pulse wave with the same duration 464
and waveform as case 1 was injected into the pipeline, and the pressure wave reflections 465
were simulated using a four-element Kevin-Voigt viscoelastic MOC model (Covas et 466
al. 2005). 467
468
Table 1 Viscoelastic parameters used in the numerical case 469
Retardation time
ππ (s)
Creep coefficients π½π
(Pa-1)
0.005 1.048E-10
0.5 1.029E-10
1.5 1.134E-10
5 8.083E-12
The pipeline was initially reconstructed using the modified layer peeling method, 470
without considering any signal dissipation or dispersion (Model 1) (thus ignoring 471
viscoelasticity). The reconstructed results plotted as the solid line in Fig. 14 include 472
significant errors at the deteriorated section. Another reconstruction was then 473
conducted using the model incorporating the viscoelastic effects using Eqs. (23), (25) 474
and (27), but excluding the frequency-dependent wave reflection and transmission 475
(Model 2). This means that the complex characteristic impedances in Eq. (28) were 476
replaced by the real characteristic impedances in this model. The results shown as the 477
dashed line in Fig. 14 illustrate that the error was reduced but still distinct. The third 478
reconstruction was conducted by further considering the frequency-dependent wave 479
reflection and transmission by using complex characteristic impedances in Eq. (28) 480
(Model 3). A high correlation was achieved between the reconstructed result (the dot-481
dashed line) and the theoretical values in Fig. 14. A small error remains in the part of 482
the pipeline close to the reservoir. This error was caused by neglecting the frequency-483
dependent characteristics of the wave reflection and transmission in Eqs. (36) and (37) 484
in the time domain. 485
486
487
Fig. 13 Pipeline configuration 2 488
489
490
Fig. 14 Wave speed reconstructed from the modified layer peeling method for the 491
viscoelastic case (Model 2 - with viscoelasticity and frequency-independent wave reflections 492
and transmissions; Model 3- with viscoelasticity and frequency-dependent wave reflections 493
and transmissions). 494
4. Experimental Verification 495
Laboratory experiments have been conducted on a single copper pipeline system in the 496
Robin Hydraulics Laboratory at the University of Adelaide to verify the proposed 497
technique for pipeline condition assessment. The experimental data was used to verify 498
the use of a direct wave, reflection-based technique (Gong et al. 2013), and the 499
reconstructive MOC analysis method (Gong et al. 2014b) for detecting a distributed 500
deterioration in a pressurized pipeline. 501
4.1 Experimental pipeline layout 502
The layout of the experimental pipeline system is given in Fig. 15. The pipeline was 503
connected to a pressurized tank at the upstream side, and a dead-end was created by 504
closure of the in-line valve at the downstream side. The basic geometry parameters are: 505
length L = 37.46 m, internal diameter D0 = 22.14 mm and wall thickness e0 = 1.63 mm 506
(D/e =13.6). A pipe section with a thinner pipe wall with L1 = 1.649 m, D1 = 22.96 mm, 507
e1 = 1.22 mm (D/e=18.8) and same material with the original pipeline was placed 508
17.805 m upstream from the in-line valve. It represents a pipe section with a uniform 509
wall thickness reduction due to internal corrosion. A side-discharge solenoid valve was 510
located 144 mm upstream from the closed in-line valve, for the generation of transient 511
waves. 512
The wave speed of the pipeline can be calculated using Eq. (8) with the following 513
parameters: E = 124.1GPa, K = 2.149 GPa, π = 999.1 kg/m3 and π1 = 1.006 which is 514
assumed to be uniform for a thick-walled pipe. The theoretical wave speed calculated 515
using Eq. (8) for the intact pipeline was a0 = 1319 m/s and a1 = 1273 m/s for the thinner-516
walled section. 517
518
Fig. 15 System layout of the experimental pipeline system 519
4.2 Experimental data 520
A transient wave was generated by sharply closing the side-discharge solenoid valve, 521
which was adjacent to the closed in-line valve. The pressure traces were collected by a 522
Druck PDCR 810 pressure transducer with a 2 kHz sampling rate. Three experiments 523
were conducted using the same configuration, with the pressure traces shown in Fig. 524
16. 525
The pressure trace in the first 6 ms, as shown in Fig. 15, covered the full wave front and 526
was defined as the input signal to the system. The reflection signal could then be 527
obtained by subtracting the input signal from the original pressure trace. With the input 528
and reflection signals of each experiment, the system IRF of each experiment was 529
obtained using Eq. (3), represented as the dash-dotted lines in Fig. 17. To reduce the 530
background noise, an averaged IRF (represented as the solid line in Fig. 17) was 531
obtained by averaging these three sets of IRF traces. 532
533
Fig. 16 Experimental pressure traces 534
535
Fig. 17 Individual IRFs and averaged IRF 536
4.3 Reconstruction of the pipeline 537
The pipeline was reconstructed using the proposed approach, with the individual and 538
averaged IRFs. The reconstructed pipelines using the individual IRFs, with the model 539
incorporating the unsteady friction, are plotted in Fig. 18. A reconstruction using the 540
averaged IRF, using both frictionless and unsteady friction models, is shown in Fig. 19. 541
A clear dip in the reconstructed wall thickness, and a clear dip in the reconstructed wave 542
speed, which matches well with the theoretical values, can be observed. 543
Perturbations in the estimated wall thickness and wave speed of the reconstructed 544
pipelines are also illustrated in Fig. 18 and Fig. 19. They are caused by the joints in the 545
pipeline, natural variations in the pipeline parameter, fluid-structure interactions, and 546
other uncertainties associated with the experiments. 547
Slight differences can be seen in Fig. 20, in which the reconstructed results using the 548
frictionless model and the unsteady friction model are compared. The figure illustrates 549
that the frictionless model gives a conservative result. As the base flow in the pipeline 550
is zero (after the closure of the side-discharge valve), and the pipeline is short in length, 551
the effect of the unsteady friction is marginal in this experimental case. Overall, the 552
experimental results have validated the effectiveness of the proposed layer-peeling-553
based pipeline condition assessment technique. 554
555
Fig. 18 Reconstruction using individual IRFs: (a) reconstructed wall thickness 556
distribution compared with the theoretical values; and (b) reconstructed wave speed 557
distribution compared with the theoretical values. 558
559
Fig. 19 Reconstruction using the averaged IRF: (a) reconstructed wall thickness 560
distribution compared with the theoretical values; and (b) reconstructed wave speed 561
distribution compared with the theoretical values. 562
563
564
Fig. 20 An enlarged view of Fig. 19 565
5. Discussion 566
In this section, some practical issues when applying this new approach to field pipelines 567
are discussed. Strategies to refine this approach to enhance its practicality are also 568
included. 569
5.1 Resolution 570
The proposed technique enables continuous reconstruction of pipeline conditions, 571
which is an advantage over conventional time-domain reflectometry-based techniques 572
that only focus on major reflections and major deteriorations (Gong et al. 2013; Gong 573
et al. 2015b). The spatial resolution of the proposed method is limited by the effective 574
bandwidth of the incident waves, which is determined by the sharpness of the wave 575
front. Theoretically, one can accurately diagnose a deteriorated section with a length 576
longer than πππ/2, where ππ is the duration of the pulse. The duration of the pulse, 577
generated by a side-discharge valve, is typically several milliseconds in the laboratory 578
and tens of milliseconds in the field due to limitations in the maneuverability of the 579
valve. Pressure generators that can generate high-frequency pressure waves, such as the 580
prototype spark-transient generator (Gong et al. 2018), will be helpful in increasing the 581
resolution. 582
5.2 Location of the generator 583
For the approach proposed in this paper, and for some other existing transient-based 584
methods for pipeline condition assessment, the pressure generator and transducer are 585
required to be installed close to the dead-end of the pipe. The dead-end can be achieved 586
by closing an in-line valve, but the installation of a generator and a transducer is not 587
always convenient. Further research will be conducted to extend the new approach to 588
other testing configurations that do not require a dead end. 589
590
591
5.3 Wave dissipation and dispersion 592
This research has demonstrated that errors are likely to occur if the wave dissipation 593
and dispersion are not properly considered in the algorithm. In field pipelines, there are 594
uncertainties and variations in the wave dissipation and dispersion, and they are difficult 595
to predict using theoretical models. Further research is needed to enable the in-situ 596
calibration of the wave dissipation and dispersion. 597
6. Conclusions 598
This paper has proposed a novel approach for pipeline condition assessment. The layer 599
peeling method previously applied to tubular musical instruments has been modified to 600
accommodate the differences between musical instruments and water pipelines. The 601
long source tube, which was used in the original method, has been eliminated. Unsteady 602
friction of the transient flow, and the viscoelastic effects of the pipe wall, have been 603
incorporated into the new method. Frequency dependent wave reflections and 604
transmissions were also incorporated into the method. This research has demonstrated 605
that the wall condition of water pipelines can be assessed using the reconstructed results, 606
using the modified layer peeling method. 607
Numerical simulations conducted in this research have demonstrated that 1) the 608
proposed approach can deal with multiple, deteriorated sections including non-609
uniformly distributed deteriorations; 2) reasonably accurate results can be achieved 610
even when the signal is contaminated by moderate background noise; 3) wave 611
dissipation and dispersion can be accounted for using a transfer function that is 612
constructed to consider the unsteady friction of the transient flow and the viscoelastic 613
effects of the pipe wall; and 4) frequency-dependent wave reflection and transmission 614
can be incorporated in the frequency domain, and that this is important for viscoelastic 615
pipes. 616
The experimental results further validated the new pipeline condition assessment 617
approach. Three sets of experimental data were used to obtain the averaged IRF, which 618
is the input for the modified layer peeling algorithm. The pipeline with a thinner-walled 619
section was successfully reconstructed both with and without the unsteady friction of 620
transient flow. 621
The proposed layer-peeling pipeline condition assessment technique is a promising 622
alternative to other existing methods because of both its computational efficiency and 623
its ability to assess a long pipe section continuously. The method is capable of 624
conveniently incorporating wave dissipation and dispersion, and frequency-dependent 625
reflection and transmission. Thus, it has the potential to allow engineers to conduct 626
reliable condition assessments for real pipelines, including viscoelastic pipes. 627
Acknowledgements 628
The research presented in this paper has been supported by the Australia Research 629
Council through the Discovery Project Grant DP170103715. 630
The author would like to acknowledge Leticia Mooney for her editorial assistance as 631
this paper was finalized. 632
633
References 634
AgullΓ³, J., Cardona, S., and Keefe, D. H. (1995). "Timeβdomain deconvolution to 635
measure reflection functions for discontinuities in waveguides." The Journal 636
of the Acoustical Society of America, 97(3), 1950-1957. 637
Amir, N., and Shimony, U. (1995a). "A discrete model for tubular acoustic systems 638
with varying cross section - the direct and inverse problems. Part 2 639
experiments." Acustica, 1995(81), 463-474. 640
Amir, N., and Shimony, U. (1995b). "A discrete model for tubular acoustic systems 641
with varying cross section - the direct and inverse problems. Part 1 theory." 642
Acustica, 1995(81), 450-462. 643
Brunone, B. (1999). "Transient test-based technique for leak detection in outfall 644
pipes." Journal of Water Resources Planning and Management, 125(5), 302-645
306. 646
Chaudhry, M. H. (2014). Applied Hydraulic Transients, Springer, New York, NY. 647
Covas, D., Stoianov, I., Mano, J. F., Ramos, H., Graham, N., and Maksimovic, C. 648
(2005). "The dynamic effect of pipe-wall viscoelasticity in hydraulic 649
transients. Part II - Model development, calibration and verification." Journal 650
of Hydraulic Research, 43(1), 56-70. 651
Covas, D., and Ramos, H. (2010). "Case studies of leak detection and location in 652
water pipe systems by inverse transient analysis." Journal of Water Resources 653
Planning and Management, 136(2), 248-257. 654
Duan, H.-F., Lee Pedro, J., Ghidaoui Mohamed, S., and Tung, Y.-K. (2012). 655
"Extended Blockage Detection in Pipelines by Using the System Frequency 656
Response Analysis." Journal of Water Resources Planning and Management, 657
138(1), 55-62. 658
Duan, H. F., Lee, P. J., Ghidaoui, M. S., and Tung, Y. K. (2011). "Leak detection in 659
complex series pipelines by using the system frequency response method." 660
Journal of Hydraulic Research, 49(2), 213-221. 661
Ferrante, M., Brunone, B., and Meniconi, S. (2009). "Leak detection in branched pipe 662
systems coupling wavelet analysis and a Lagrangian model." Journal of Water 663
Supply: Research and Technology - Aqua, 58(2), 95-106. 664
Forbes, B. J., Sharp, D. B., Kemp, J. A., and Li, A. (2003). "Singular system methods 665
in acoustic pulse reflectometry." Acta Acustica united with Acustica, 89(5), 11. 666
Gong, J., Simpson, A. R., Lambert, M. F., Zecchin, A. C., Kim, Y. I., and Tijsseling, 667
A. S. (2013). "Detection of distributed deterioration in single pipes using 668
transient reflections." Journal of Pipeline Systems Engineering and Practice, 669
4(1), 32-40. 670
Gong, J., Zecchin, A. C., Simpson, A. R., and Lambert, M. F. (2014a). "Frequency 671
response diagram for pipeline leak detection: comparing the odd and the even 672
harmonics." Journal of Water Resources Planning and Management, 140(1), 673
65β74. 674
Gong, J., Lambert, M. F., Simpson, A. R., and Zecchin, A. C. (2014b). "Detection of 675
localized deterioration distributed along single pipelines by reconstructive 676
MOC analysis." Journal of Hydraulic Engineering, 140(2), 190-198. 677
Gong, J., Zecchin, A. C., Lambert, M. F., and Simpson, A. R. (2015a). "Study on the 678
frequency response function of viscoelastic pipelines using a multi-element 679
Kevin-Voigt model." Procedia Engineering, 119, 226β234. 680
Gong, J., Stephens, M. L., Arbon, N. S., Zecchin, A. C., Lambert, M. F., and 681
Simpson, A. R. (2015b). "On-site non-invasive condition assessment for 682
cement mortar-lined metallic pipelines by time-domain fluid transient 683
analysis." Structural Health Monitoring, 14(5), 426-438. 684
Gong, J., Zecchin, A. C., Lambert, M. F., and Simpson, A. R. (2016). "Determination 685
of the creep function of viscoelastic pipelines using system resonant 686
frequencies with hydraulic transient analysis." Journal of Hydraulic 687
Engineering, 10.1061/(ASCE)HY.1943-7900.0001149, 04016023. 688
Gong, J., Lambert Martin, F., Nguyen Si, T. N., Zecchin Aaron, C., and Simpson 689
Angus, R. (2018). "Detecting Thinner-Walled Pipe Sections Using a Spark 690
Transient Pressure Wave Generator." Journal of Hydraulic Engineering, 691
144(2), 06017027. 692
Hendrie, D. A. (2007). "Development of bore reconstruction techniques applied to the 693
study of brass wind instruments," Ph.D. Thesis, University of Edinburgh. 694
Lee, P. J., Lambert, M. F., Simpson, A. R., VΓtkovskΓ½, J. P., and Misiunas, D. (2007). 695
"Leak location in single pipelines using transient reflections." Australian 696
Journal of Water Resources, 11(1), 53-65. 697
Lee, P. J., VΓtkovskΓ½, J. P., Lambert, M. F., Simpson, A. R., and Liggett, J. A. (2008). 698
"Discrete blockage detection in pipelines using the frequency response 699
diagram: Numerical study." Journal of Hydraulic Engineering, 134(5), 658-700
663. 701
Li, A., Sharp, D. B., and Forbes, B. J. (2005). "Increasing the axial resolution of bore 702
profile measurements made using acoustic pulse reflectometry." Measurement 703
Science and Technology, 16(10), 2011-2019. 704
Liggett, J. A., and Chen, L. C. (1994). "Inverse transient analysis in pipe networks." 705
Journal of Hydraulic Engineering, 120(8), 934-955. 706
Meniconi, S., Brunone, B., and Ferrante, M. (2011). "In-line pipe device checking by 707
short-period analysis of transient tests." Journal of Hydraulic Engineering, 708
137(7), 713-722. 709
Sattar, A. M., Chaudhry, M. H., and Kassem, A. A. (2008). "Partial blockage 710
detection in pipelines by frequency response method." Journal of Hydraulic 711
Engineering, 134(1), 76-89. 712
Shamloo, H., and Haghighi, A. (2009). "Leak detection in pipelines by inverse 713
backward transient analysis." Journal of Hydraulic Research, 47(3), 311-318. 714
Sharp, D. B. (1996). "Acoustic pulse reflectometry for the measurement of musical 715
wind instruments," University of Edinburgh, Edinburgh, UK. 716
Sharp, D. B., and Campbell, D. M. (1997). "Leak detection in pipes using acoustic 717
pulse reflectometry." Acta Acustica (Stuttgart), 83(3), 560-566. 718
Sharp, D. B. (1998). "Increasing the length of tubular objects that can be measured 719
using acoustic pulse reflectometry." Measurement Science and Technology, 720
9(9), 1469-1479. 721
Smith, S. W. (1999). The Scientist and Engineer's Guide to Digital Signal Processing, 722
California Technical Publishing. 723
Stephens, M. L., Lambert, M. F., and Simpson, A. R. (2013). "Determining the 724
internal wall condition of a water pipeline in the field using an inverse 725
transient model." Journal of Hydraulic Engineering, 139(3), 310β324. 726
Stephens, M. L., Simpson, A. R., and Lambert, M. F. (2008). "Internal wall condition 727
assessment for water pipelines using inverse transient analysis." Proceedings 728
of the 10th Annual Symposium on Water Distribution Systems Analysis, 729
ASCE, Reston, VA. 730
Suo, L., and Wylie, E. B. (1990). "Complex wavespeed and hydraulic transients in 731
viscoelastic pipes." Journal of Fluids Engineering, 112(4), 496-500. 732
Tran, D. H., Perera, B. J. C., and Ng, A. W. M. (2010). "Hydraulic deterioration 733
models for storm-water drainage pipes: Ordered probit versus probabilistic 734
neural network." Journal of Computing in Civil Engineering, 24(2), 140-150. 735
Vardy, A. E., and Brown, J. M. B. (1995). "Transient, turbulent, smooth pipe 736
friction." Journal of Hydraulic Research, 33(4), 435-456. 737
VΓtkovskΓ½, J. P., Simpson, A. R., and Lambert, M. F. (2000). "Leak detection and 738
calibration using transients and genetic algorithms." Journal of Water 739
Resources Planning and Management, 126(4), 262-265. 740
VΓtkovskΓ½, J. P., Lambert, M. F., Simpson, A. R., and Liggett, J. A. (2007). 741
"Experimental observation and analysis of inverse transients for pipeline leak 742
detection." Journal of Water Resources Planning and Management, 133(6), 743
519-530. 744
VΓtkovskΓ½, J. P., Bergant, A., Simpson, A. R., and Lambert, M. F. (2003). 745
"Frequency-domain transient pipe flow solution including unsteady friction." 746
Pumps, Electromechanical Devices and Systems Applied to Urban Water 747
Management: Proceedings of the International Conference, A. A. Balkema 748
Publishers, Lisse, The Netherlands, 773-780. 749
Vreeburg, I. J. H. G., and Boxall, D. J. B. (2007). "Discolouration in potable water 750
distribution systems: A review." Water Research, 41(3), 519-529. 751
Wang, X. J., Lambert, M. F., and Simpson, A. R. (2005). "Detection and location of a 752
partial blockage in a pipeline using damping of fluid transients." Journal of 753
Water Resources Planning and Management, 131(3), 244-249. 754
Ware, J. A., and Aki, K. (1969). "Continuous and discrete inverse-scattering problems 755
in a stratified elastic medium. 1 plane waves at normal incidence." The 756
Journal of the Acoustical Society of America, 45(4), 11. 757
Wylie, E. B., and Streeter, V. L. (1993). Fluid Transients in Systems, Prentice Hall 758
Inc., Englewood Cliffs, New Jersey, USA. 759
Zamanzadeh, M., Kirkwood, G. C., Scheinman, S., and Bayer, G. T. (2007). 760
"Corrosion sensors for detecting graphitization of cast iron in water mains." 761
Proceedings of Corrosion Conference and Expo 2007, Curran Associates, 762
Inc., Red Hook, NY. 763
764