Computers and Structures 175 (2016) 74–90

Contents lists available at ScienceDirect

Computers and Structures

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

Fluid-structure interaction in straight pipelines: Friction coupling

mechanisms
David Ferràs a,b,⇑, Pedro A. Manso a, Anton J. Schleiss a, Dídia I.C. Covas b
a
Laboratory of Hydraulic Constructions (LCH), École Polytechnique Fédérale de Lausanne, Switzerland
b
CERIS, Instituto Superior Técnico, Universidade de Lisboa, Portugal

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

Article history: The present paper approaches fluid-structure interaction by means of a 4-equation model. Experimental
Received 21 December 2015 data collected from a straight copper pipe-rig lying directly on the lab floor is used for the model valida-
Accepted 16 June 2016 tion in terms of wave shape, timing and damping. The main focus lies on the friction coupling modelling
Available online 31 July 2016
considering skin and dry friction. For skin friction three approaches are analysed: quasi-steady, Brunone’s
and Trikha’s unsteady friction. For dry friction Coulomb’s model is added in the beam momentum con-
Keywords: servation equation. Results present a good fitting between experimental and numerical data, showing
Fluid-structure interaction
the dissipative effect of dry friction phenomenon which complement that of skin friction, specially in
Junction coupling
Poisson coupling
the short term simulation.
Friction coupling Ó 2016 Elsevier Ltd. All rights reserved.
Skin friction
Dry friction

1. Introduction viscoelasticity are hard to distinguish [9] and, to the knowledge

Fluid-structure interaction (FSI) in pressurized hydraulic tran-
sients analyses is frequently approached by considering the first
two pipe vibration modes (i.e., pressure wave propagation in the
fluid and axial stress wave propagation in the pipe-wall). For the
description of pressure waves in pipe systems, one-mode or two-
mode solutions are sufficient [26]. Two-mode models can be
implemented either by using MOC-FEM procedure (i.e., the method
of characteristics for the fluid and finite element method for the
structure) [40] or MOC procedure (i.e., the method of characteris-
tics for both the fluid and the structure) [41]. Lavooij and Tijsseling
[20] applied the two approaches to solve the four basic conserva-
tion equations in the time domain, concluding that for straight
pipe problems the MOC procedure is more accurate. Thus, a 4-
equation model represents a suited tool to describe the ideal
reservoir-pipe-valve system in its basic FSI configurations, namely
either considering an anchored or non-anchored downstream
valve.
Several authors combined FSI with other wave dissipating
phenomena, such as: FSI and pipe-wall viscoelasticity [39,37,25];
FSI and cavitation [30,29,26]; and the most complete including
FSI, column separation and unsteady friction (UF) in a viscoelastic
pipe [19]. However, the effects of unsteady friction and pipe-wall
⇑ Corresponding author at: Laboratory of Hydraulic Constructions (LCH), École
Polytechnique Fédérale de Lausanne, Switzerland.
of the authors, unsteady friction effect has never been separately
assessed in a two-mode FSI model. Due to FSI, the pipe-wall
vibrates axially at a different rate than the fluid, hence, the relative
velocity between both (V r ) must be considered for skin shear stress
assessment. The higher the Mach number (V r =af ) is, the greater the
wall shear stress effects are [16]. Therefore, unsteady friction
effects may be increased when fluid-structure interaction is
important.
Besides, in the implementation of a 4-equation model a major
question may arise: Is there movement in the pipe supports?
Anchorages of pipelines aim to avoid the pipe-wall movement
essentially by means of dry friction [11]. However, from Newton
principles, when a system is loaded, null deformation/displace-
ment by means of only resistance is not possible. Pipe supports
are never entirely stiff or entirely inert when loaded by impacts
[27]. Thus, movement occurs. Dry friction is proportional to the
normal force, hence, for a high normal force, important energy
might be dissipated from the structure to its supports/surround-
ings. Furthermore, in this context, it is crucial to define with good
criteria the stick-slip transitions.
Tijsseling and Vardy [28] included Coulomb’s dry friction in a 4-
equation model with the goal to describe the behaviour of pipe
racks, proposing a quantitative guideline equation aiming at
assessing when dry friction forces may be relevant during hydrau-
lic transients. In the present paper, dry friction is approached dif-
ferently not at a single point but distributed all throughout the

http://dx.doi.org/10.1016/j.compstruc.2016.06.006
0045-7949/Ó 2016 Elsevier Ltd. All rights reserved.
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90 75

Notation

Af fluid cross-sectional area (m2) N normal force (N)

af pressure wave speed (m s1) nk exponential sum coefficients (–)
As pipe-wall cross-sectional area (m2) p fluid pressure (Pa)
as axial stress wave speed (m s1) Re Reynolds number (–)
CH Vardy decay coefficient (–) r inner radius of the pipe-wall (m)
D pipe inner diameter (m) t time (s)
E pipe-wall Young’s modulus (Pa) tv valve closure time (s)
e pipe-wall thickness (m) U pipe-wall velocity (m s1)
f Darcy skin friction coefficient (–) V fluid mean velocity (m s1)
fs steady Darcy coefficient (–) Vr relative velocity (m s1)
fu unsteady Darcy coefficient (–) W Zielke weighting function (–)
F force acting in the system (N) x distance along the pipe axis (m)
F df dry friction force (N) Yk Trikha function (–)
g gravity acceleration (m s2) m Poisson’s ratio (–)
H hydraulic head (m) l Coulomb dry friction coefficient (–)
HJk Joukowsky hydraulic head (m) ls static Coulomb coefficient (–)
hf skin friction losses (m) lk kinetic Coulomb coefficient (–)
hf s steady skin friction losses (m) qs pipe density (kg m3)
hf u unsteady skin friction losses (m) qf fluid density (kg m3)
K bulk modulus of compressibility (Pa) r pipe axial stress (Pa)
k Brunone coefficient (–) rRK Rankine pipe axial stress (Pa)
L pipe length (m) s valve closure degree (–)
Mv valve mass (kg) st dimensionless time-step (–)
mk exponential sum coefficients (–) k eigenvector (m s1)

pipeline. For this purpose, a new right-hand-side term in the
momentum equation of the pipe-wall axial movement was
incorporated.
This research aims at assessing firstly the effect of different skin
friction models during hydraulic transients in a FSI 4-equation
(two-mode) solver. For this purpose, three skin friction models
are assessed: (i) quasi-steady friction; (ii) Brunone’s unsteady fric-
tion formulation, which is based on instantaneous local and con-
vective accelerations; and (iii) Trikha’s unsteady friction model,
which is based on weights of past velocity changes. Secondly, dry
friction is implemented, nesting its computation into the friction
coupling mechanism, and its dissipation effect over the transient
wave is assessed.
Experimental tests were carried out in a straight copper pipe-rig
for different initial conditions and structural scenarios: (a)
anchored downstream pipe-end, and (b) non-anchored down-
stream pipe-end. These tests allowed to corroborate and validate
the modelling assumptions.
The aim of the paper is the assessment of different friction dis-
sipation assumptions in a FSI two-mode model. A 4-equation sol-
ver is implemented including the three basic coupling
mechanisms: Poisson, junction and friction coupling; and the last
one nests the skin friction models (i.e. quasi-steady, Brunone’s
and Trikha’s) and the dry friction model (i.e. Coulomb’s friction).
The innovation of this research is the incorporation of dry friction
computation in the fundamental equations of the two-mode (four-
equation) waterhammer model. This implies a modification of the
pipe-wall momentum equation in the axial direction. The effect of
dry friction is compared with skin friction and results are assessed
by means of experimental data in a straight copper pipe rig.
2. Experimental data collection
A straight copper pipe rig was assembled at the Laboratory of
Hydraulics and Environment of Instituto Superior Técnico (LHE/
IST). The system is composed of a 15:49 m long pipe, with an inner
diameter D ¼ 0:020 m and pipe-wall thickness e ¼ 0:0010 m.
Young’s modulus of elasticity and Poisson’s ratio of the copper
material were experimentally determined by measuring stress-
strain states over a pipe sample for the experimental range of pres-
sures. The pipe segment, with closed ends, was pressurized and
strains measured using strain gauges disposed in the circumferen-
tial and axial directions. By means of stress-strain relations the
Young’s modulus and the Poison’s ratio were determined. The
obtained experimental values were the Young’s modulus of elastic-
ity E ¼ 105 GPa and the Poisson’s ratio m ¼ 0:33; these values are in
agreement with theoretical values from literature. At the upstream
end, there is a storage tank followed by a pump and an air vessel,
and at the downstream end, there is a ball valve pneumatically
operated that allows the generation of fast transient events, with
an effective closing time of tv ¼ 0:003 s. The ball valve together
with the actuator mechanisms and the supporting system have a
mass of M v ¼ 6 kg and it is anchored by glue-in-bolts on the floor
of the laboratory. Downstream the valve there is a hose conveying
the water to the water-tank, thus closing the pipe system circuit.
Three pressure transducers (WIKA S-10) were installed at the
upstream, midstream and downstream positions of the pipe (PT1,
PT2 and PT3). Strain gauges (TML FLA-2-11) disposed in the axial
(SG1 and SG3) and circumferential (SG2 and SG4) directions were
glued at the midstream and the downstream end of the pipe. The
initial discharge was measured for steady state conditions by a
rotameter located downstream of the valve. The sampling fre-
quency was set to 1200 Hz after preliminary tests in order to mea-
sure with accuracy the FSI response of the pipe system during the
waterhammer events. The wave celerity in the fluid was estimated
from pressure measurements obtaining a value of af ¼ 1239 m=s
[14]. The axial stress wave celerity in the pipe-wall was theoreti-
qffiffiffiffi
cally determined as ¼ qE ¼ 3848:4 m=s. Fig. 1 shows an illustra-
s
tion of the experimental set-up, with the location of the pressure
transducers (PT) and strain gauges (SG).
Two supporting configurations have been analysed: (a) the con-
duit anchored against longitudinal movement at both downstream
and upstream ends; and (b) the conduit only anchored against lon-
76 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90

Fig. 1. Simplified schematic view of the straight copper pipe set-up. Pressure transducers and strain-gauges location.

Table 1 motion of the pipe-wall and the pressure in the fluid occurring
Characteristics of the experimental tests. by means of the Poisson’s effect; (ii) friction coupling arises from
Test ID V 0 ðm=sÞ Re0 H0 ðmÞ Hmax ðmÞ Hmin ðmÞ Valve the shear stress between the pipe-wall and the fluid; and (iii) junc-
SCP01 0.26 5276 43.65 77.25 10.66 Fixed tion coupling results from unbalanced local forces and by changes
SCP02 0.36 7253 42.39 88.67 2.95 Fixed in the fluid momentum that occur in pipe bends, T-junctions or
SCP03 0.41 8206 42.87 94.88 8.67 Fixed cross section changes.
SCP04 0.26 5276 44.16 83.17 5.97 Released During the development of the model the three coupling mech-
SCP05 0.36 7253 42.27 94.69 8.34 Released
anisms were analysed and implemented. Friction coupling was
SCP06 0.41 8206 42.55 100.04 9.95 Released
addressed with the goal to embed and analyse the combination
of fluid-structure interaction with unsteady friction models and
dry friction. Junction coupling was considered to characterize the
gitudinal movement at the upstream end. Throughout the pipe anchoring conditions of the downstream valve.
there were no anchorages nor supports; the pipe layed directly
over the floor of the laboratory. 3.2. Unsteady skin friction
Table 1 summarizes the tests carried out in these experimental
configurations, displaying the initial flow velocity and Reynolds Friction has frequently a non-linear behaviour in physical sys-
number, the initial piezometric head and the maximum and mini- tems. Understanding the relation between the parameters defining
mum piezometric heads measured immediately upstream the the unsteadiness of friction has motivated researches of water-
valve. hammer during the last decades. Fast transients have a strong 2D
Fig. 2a and b depict the transient pressure traces at the down- nature of the flow field [5,6]. Consequently, a dimension-
stream section of the pipe for both anchoring conditions. Fig. 3a reduction problem must be added to the non-linear problem of
and b depict the transient circumferential and axial strain traces, the friction.
respectively, at the downstream pipe section for both anchoring Unsteady friction is one of the factors that generates dissipa-
conditions. As it can be observed, strain and pressure traces pre- tion, dispersion and shape-change of the pressure wave. Its impor-
sent a similar response during the waterhammer wave being in tance depends on the system considered and the operating
pressure and circumferential strains in phase, and axial strains fol- conditions. In most of the laboratory water-hammer test rigs made
lowing the expected behaviour according to Poisson effect. Either of metal pipes, unsteady friction dominates over steady friction [2],
in terms of pressure (Fig. 2) or strain, (Fig. 3), the system response in particular for fast transients.
is very different when the downstream pipe-end is released, pre- Unsteady friction models can be classified according to their
senting greater maximum pressures and a noticeable wave shape basic modelling assumptions in two main groups: (i) acceleration
change. based models [4,36,21], and (ii) convolution based models [44,31,
18,22,32,35,33,34,42,43]. With the aim to analyse the effect of
unsteady friction on the two-mode FSI model, friction coupling
3. Background theory
was implemented considering Brunone [5] and Trikha [31]
unsteady friction formulations, which, as mentioned, belong to
3.1. Fluid-structure interaction
two different families of approaches.
3.1.1. The four-equation model
Skalak [24] introduced the theoretical basis for FSI in straight 3.2.1. Brunone’s unsteady friction model
pipes and extended Joukowsky’s classical theory [17] by means Typically, the skin friction losses result from the combination of
of the classical beam theory equations [38], for the vibration of thin the steady, hf s , and the unsteady, hf u , components:
cylindrical tubes, with waterhammer equations [8] for the pressure hf ¼ hf s þ hf u ð1Þ
wave propagation in pressurized conduits. Skalak [24] found an
infinite number of wave propagation modes. The first two modes where
are dominant in straight pipes during hydraulic transients. The first
f s VjVj
mode corresponds to the pressure wave in the fluid and the second hf s ¼ ð2Þ
to the axial stress wave in the pipe-wall. Two-mode models (i.e. 4- 2gD
equation models) allow the description of the interaction between With the idea that during fast transients both, local and convec-
the two modes [13]. The 1D 4-equation non-dispersive model tive accelerations are correlated to friction forces, Brunone et al. [4]
developed by Lavooij and Tijsseling [20] in all the basic coupling proposed a single expression to calculate the unsteady component
mechanisms (Poisson, junction and friction coupling) has been that requires an empirical coefficient k:
implemented and used in this research.  
k @V @V
hf u ¼  af ð3Þ
2g @t @x
3.1.2. Coupling mechanisms
There are three basic kinds of coupling mechanisms [26]: (i) In the present study the k coefficient is computed as suggested
Poisson coupling describes the interaction between the axial by Bergant et al. [1]:
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90 77

Fig. 2. Pressure data acquired at the downstream end (PT3) of the straight copper pipe for an anchored (a) and for a non-anchored (b) downstream end.

Fig. 3. Circumferential (solid lines) and axial (dashed lines) strain data acquired at the downstream section of the straight copper pipe for an anchored (a) and for a non-
anchored (b) downstream end.

pffiffiffiffiffiffi
CH rule and the acceleration term is approximated using a central dif-
k¼ ð4Þ ference. However, this scheme is very expensive from the computa-
2
tional point of view. Trikha [31] simplified this computation
where the Vardy’s shear decay coefficient C H is given by [35]: reducing the weighting function to the summation of three expo-
nential terms and eliminating the need for convolution with an
C H ¼ 0:00476 ð5Þ approximate recursive relationship:
for laminar flow, and 16m Xn
hfu ðtÞapp: ¼ Y k ðtÞ ð8Þ
7:41 gD2 k¼1
CH ¼ log 14:3
ð6Þ
Re Re0:05
where Y k is a function that represents the exponential terms:
for turbulent flow. Y k ðt þ DtÞ ¼ mk ½Vðt þ DtÞ  VðtÞ þ enk Dst Y k ðtÞ; ð9Þ
By means of this method the k coefficient is computed as func-
tion of the Reynolds number and there is no need for calibration. n ¼ the number of exponential terms (n ¼ 3 in the case of
Trikha formulation); st ¼ is the dimensionless time step,
3.2.2. Trikha’s unsteady friction model st ¼ 4DmD2 t ; nk and mk ¼ coefficients of the exponential summation,
The weighting-function models take into account the 2D nature nk ¼ ð26:4; 200; 8000Þ and mk ¼ ð1; 8:1; 40Þ.
of the velocity profile that causes the frequency-dependent atten-
uation and dispersion of the hydraulic transient. The first model of 3.3. Dry friction
this kind was proposed by Zielke [44], who developed an analytical
solution for unsteady friction for laminar flows, where the In a system where the pipe is allowed to move, not only viscous
unsteady head loss term is the convolution of the past fluid accel- dissipation but also dry dissipation between the pipe-wall and its
erations with a weighting function (full convolution method) supports, occur during fast hydraulic transients [11].
described by: For dry friction computation the Coulomb’s law is usually
  applied (Eq. (10)), and it is assumed that the friction force F df is
16m @V
hfu ðtÞexact ¼  W ðtÞ ð7Þ proportional to the normal force N acting between the surfaces
gD2 @t and opposite to the pipe-wall movement:
where  indicates convolution and W the weighting function. The F df ¼ lN signðUÞ ð10Þ
convolution in Zielke’s model is approximated using the rectangular
78 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90

in which l is the friction coefficient, signðÞ is a function that returns derivation of the compatibility equations is explained in Appen-
1 or 1 according to the pipe-wall movement direction and N is the dices A.2–A.4.
normal force corresponding to the weight of the pipe segment filled 0 1
with water: a2f
dV B 1 r af C dp a2 dU
N ¼ ½qs As þ qf Af   Dx  g @ þ 2m 2 A þ 2m s a2
ð11Þ dt af qf eE a2s
1 dt dt
a2f 1  a2
f

The friction coefficient increases from 0 until a maximum value 0 1 s

is reached when the movement is imminent, and then it drops to a B 2m af 2mC dr f

@ þ af A ¼  V r jV r j
relatively constant value during motion for low velocities [38]. qs a2s a2s
1 E dt 4r
In Davis [10] approximate values are given for Coulomb’s fric- a2f

tion coefficient for copper sliding over cast iron as ls ¼ 1:05 for a2f    
the static friction and lk ¼ 0:29 for the kinetic/sliding friction. a2s qf Af f qf Af
þ 2m V jV j þ 1 þ g l signðUÞ ð16Þ
The static friction coefficient is used to conduct the stick-slip con- 1
a2
f qs As 4r r r qs As
a2s
dition [7] that balances the resultant force from axial stress in the
pipe-wall section with the resultant force from dry friction. Hence, 0 1
if the force due to axial stress is higher than the static friction force, a2f

then the stick condition is true and friction losses due to dry fric-
mrqf a2s dV Bmr as r m C dp dU
 @ þ as A þ
eqs a2f dt eE a2f eE dt dt
tion are considered 0; in the opposite case, the slip condition is true 1  a2 1  a2
0 1
s s
and friction losses are computed according to Coulomb’s law (Eq.
r m2 af as A dr
2 2
(10)) using the kinematic friction coefficient. For a more robust 1
 @  2qf 2
computation, discontinuities in the dry friction force are avoided qs as 2
eE 1  af dt
by a imposing a transition between slip and stick states propor- as

tional to pipe-wall axial stress. a2f

r qf a2s f qf Af f
¼m V r jV r j þ V jV j
e qs a2f 4r qs As 4r r r
4. Numerical model development 1  a2
s
 
qf Af
4.1. Fundamental equations þ 1þ g l signðUÞ ð17Þ
qs As
The following set of equations (Eqs. (12)–(15)) is based on The compatibility Eqs. (16) and (17), are only valid along the
Lavooij and Tijsseling [20] four fundamental conservation equa- characteristic lines with slopes: 1=af for Eq. (16), and 1=as for
tions with the momentum equation (Eq. (14)) of the pipe-wall in Eq. (17).
the axial direction adapted in order to include dry friction (v.i. Fig. 4 depicts the adopted numerical scheme at the interior
H). Their derivation is presented in Appendix A.1. nodes and the domain boundaries, where: ‘P’ represents the space
@V 1 @p f and time coordinates in the grid where the computation is carried
þ ¼  ðV r ÞjV r j ð12Þ out, ‘A’ represents the information source brought by the positive
@t qf @x 4r
characteristic line in the pipe-wall, ‘B’ the information source
brought by the positive characteristic line in the fluid, ‘C’ the infor-
@V 1 @p 2m @ r mation source brought by the negative characteristic line in the
þ ¼ ð13Þ
@x qf a2f @t E @t fluid and ‘D’ the information source brought by the negative char-
acteristic line in the pipe-wall. Notice that time interpolations are
 
@U 1 @ r qf Af f qf Af necessary in the nodes close to the extreme-end boundaries.
 ¼ V r jV r j þ 1 þ g l signðUÞ ð14Þ
@t qs @x qs As 4r qs As The previous ordinary differential equations (Eqs. (16) and (17))
|fflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflfflffl} can be integrated according to the schemes presented in Fig. 4. The
H
n coefficients are presented in Table 2.
@U 1 @r rm @p
 ¼ ð15Þ nf p pP þ nf V V P  nf r rP þ nf U U P þ Cpf ¼ 0 ð18Þ
@x qs a2s @t eE @t
 nf p pP þ nf V V P þ nf r rP þ nf U U P þ Cnf ¼ 0 ð19Þ
It is the right-hand-side terms of Eqs. (12)–(15) that make the
system of equations nonlinear. The right-hand-side term of the  nsp pP  nsV V P  nsr rP þ nsU U P þ Cps ¼ 0 ð20Þ
fluid continuity equation (Eq. (13)) describes the interaction with nsp pP  nsV V P þ nsr rP þ nsU U P þ Cns ¼ 0 ð21Þ
the pipe-wall by means of Poisson coupling mechanism. Similarly,
the right-hand-side of the pipe-wall continuity equation (Eq. (15)) where Eqs. (18) and (19) correspond to the positive and negative
describes the interaction with the fluid. On the other side, the characteristic equations of the fluid pressure wave and Eqs. (20)
right-hand-side terms of the momentum equations (Eqs. (12) and and (21) correspond to the positive and negative characteristic
(14)) represent the friction losses. Skin friction loss affects oppo- equations of the axial stress wave. The values Cpf ; Cnf ; Cps and
sitely the fluid and the pipe-wall, while dry friction (v.s. H) only Cns enclose the information from the previous time-step:
affects the pipe-wall momentum equation, as this occurs between
the pipe and the outer surrounding. Cpf ¼ nf p pB  nf V V B þ nf r rB  nf U U B þ SF f B þ DF f B ð22Þ
Cnf ¼ nf p pC  nf V V C  nf r rC  nf U U C þ SF f C þ DF f C ð23Þ
4.2. Compatibility equations
Cps ¼ nsp pA þ nsV V A þ nsr rA  nsU U A þ SF sA þ DF sA ð24Þ
Eqs. (12)–(15) represent a linear hyperbolic system of four first- Cns ¼ nsp pD þ nsV V D  nsr rD  nsU U D þ SF sD þ DF sD ð25Þ
order partial differential equations. Due to its hyperbolic nature,
the system can be converted into a set of four ordinary differential where SF and DF correspond, respectively, to the skin and the dry
equations (Eqs. (16) and (17)) by the MOC method [15,20]. The friction losses terms (v.i. Table 2).
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90 79

Fig. 4. Numerical scheme of the 4-equation model. Characteristic lines at different sections of the pipe.

Table 2
Cpf þCnf
Compatibility equations coefficients for FSI, skin friction and dry friction terms.  CpsnþCn
nf
s

U n

sU
VP ¼  n
ð27Þ
Fluid-structure interaction 2 nf V þ nssV
No FSI fU U

nf p ¼ q 1af n sp ¼ 0
f
Cpf Cnf
nf V ¼ 1 n sV ¼ 0
nf p
þ CpsnCn s

rP
sp
nf r ¼ 0 n sr ¼ 1
qs as ¼   ð28Þ
n n
nf U ¼ 0 n sU ¼ 1 2 nf r þ nssr
fp p
FSI
nf p ¼ af1q þ 2m2 eE
r f a a2
f
a2
f s 1
nsp ¼ qm
a2 Cpf þCnf
þ CpsnþCn
a2
r s s
f e s as a2
1 f2 nf
V n 
sV
a
s UP ¼  ns
ð29Þ
nf V ¼ 1 a2
f 2 n U þ nsU
f
mq r f a2s fV
n sV ¼ q e s a2
V

1 f2
a
s

nf r ¼ q2am2 a2 f þ af 2Em
a 2 2
rm2 af as
s s s 1 nsr ¼ q 1as  2qf eE 2 a2 4.4. Boundary conditions
s
a2 1 afs
f

a2
f n sU ¼ 1
nf U ¼ 2m
a2
s
4.4.1. Upstream reservoir
a2
1 f2
a
As shown in Fig. 4a only the negative characteristic lines reach
s
the upstream boundary. The boundary condition for a constant
Skin friction
Frictionless level reservoir in the upstream pipe-end is given by:
SFf ¼ 0 SFs ¼ 0
pP ¼ pres ð30Þ
Steady friction
   
qA
SFf ¼ nf V  nf U qf Afs 4rf Vr ij jVr ij jdt SFs ¼ nsV 4rf
qA
 nsU qf Afs 4rf Vr ij jVr ij jdt
UP ¼ 0 ð31Þ
s s

Brunone unsteady friction Substituting Eqs. (30) and (31) into Eqs. (19) and (21) expres-
f ¼ fs þ fu
sions for V P and rP are obtained.
Vrij Vrj1 Vrij Vri1
j
f u ¼ K b jD j
Vr jVr j dt
 af dx
i
n ns

 i i    Cn
þ ns p pres  n f þ Cn
fp s
qf Af f qA nf r ns
SFf ¼ nf V  nf U q As 4r Vr i jVr ij jdt
j
SFs ¼ nsV 4rf  nsU qf Afs 4rf Vr ij jVr ij jdt r fr r
s s VP ¼ nf ns
ð32Þ
Trikha unsteady friction
nf r
V
þ nsV
mk ¼ ð1; 8:1; 40Þ r

nk ¼ ð26:4; 200; 8000Þ n ns


Cn
Þ þ enkw dst Yki w  ns p pres  n f  Cn
j j1 fp
Yki w ¼ mkw ðVr ij  Vrj1 s
 i
  nf ns
SFf ¼ dt16 lP
Yk
j
þ f j j
4r Vr i jVr i jdt
qA
nf V  nf U qf Afs rP ¼ V V
nf r n
fV V
ð33Þ

D2 iw
  s
 þ nssr
lP j qA nf
SFs ¼ dt16
D2
Yki w þ 4rf Vr ij jVr ij jdt nsV  nsU qf Afs V V
s

Dry friction
Frictionless 4.4.2. Downstream valve
DFf ¼ 0 DFs ¼ 0 A solution is derived for a non-instantaneous valve closure, with
Coulomb’s friction a non-anchored pipe-end and taking into account the valve inertia.
 qA
  qA

DFf ¼ 1 þ qf Afs g lnf U signðU ij Þdt DFs ¼ 1 þ qf Afs g lnsU signðU ij Þdt As shown in Fig. 4b only the positive characteristic lines reach the
s s

downstream boundary. The boundary conditions for such assump-

tions are expressed in Eqs. (34) and (35), where the first determi-
nes the discharge rate through the valve and the second the
4.3. Interior nodes balance of forces at the valve section by means of Newton’s second
law of motion.
From the linear system of equations (Eqs. (18)–(21)) the follow- sffiffiffiffiffiffiffiffi
ing expressions are obtained determining the dependent variables Dp
V P ¼ sðtÞ þ UP ð34Þ
along the interior nodes: qf g
 
Cpf Cnf
 CpsnCn 1 DU
rP Af ð1  sðtÞÞðpP  p0 Þ  Mv
s
nf r ¼ ð35Þ
pP ¼  n
sr
 ð26Þ As Dt
ns
2 n p þ ns p
f

fr r
80 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90

A non-linear dependency arises between the valve displace- combined two at a time. Table 3 summarizes how this verification
ment and the valve closure, yet, it is a second degree relation. Sub- tests have been carried out according to the combination of each
stituting Eqs. (34) and (35) into Eqs. (18) and (20) expressions for one of the modelled phenomena (v.s. Table 2 for activation and
pP and U P are obtained. deactivation of each phenomenon in the numerical code). The
pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi!2 geometry of the simulated pipe corresponds to the experimental
 2
b þ b  4a c facility described in Section 2.
pP ¼ þ p0 ð36Þ Thereafter, in the following subsections fluid-structure interac-
2a
tion, skin friction models and Coulomb’s dry friction are separately
where verified and assessed. The aim is not only to show the right perfor-
mance of the implemented code but also the sensitivity of the
Af ð1sðtÞÞ Af ð1sðtÞÞ !

nf p  nf r As
n sp þ n sr As
numerical output to each phenomenon in terms of wave shape,
a ¼ 
nf V þ nf r AMs Dvt þ nf U nsV  nsr AMs Dvt  nsU timing and damping.
0 1
nf V psðtÞ
ffiffiffiffiffiffi
qf g
nsV psðtÞ
ffiffiffiffiffiffi
qg 5.2. Fluid-structure interaction verification
b ¼@ A
 f

nf V þ nf r AMs Dvt þ nf U nsV  nsr AMs Dvt  nsU
! With skin and dry friction deactivated a verification of the fluid-

nf p p0  nf r AMs Dvt U P1 þ Cpf nsp p0 þ nsr AMs Dvt U P1  Cps structure interaction was carried out for the two basic configura-
c ¼ 
nf V þ nf r AMs Dvt þ nf U nsV  nsr AMs Dvt  nsU tions: anchored (FSI-1) and non-anchored (FSI-2) pipe-ends. The
first allows the assessment of Poisson coupling, as the Poisson’s
and effect throughout the pipe-wall generates axial displacements.

  pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
Af ð1sðtÞÞ
nf p  nf r As
ðpP  p0 Þ þ nf V psðtÞ
ffiffiffiffiffiffi pP  p0 þ nf p p0  nf r AMDvt U P1 þ Cpf
qg s
f
UP ¼  ð37Þ
nf V þ nf r AMs Dvt þ nf U

For the anchored valve condition a very high fictitious mass
value is considered so the downstream section becomes
motionless.
Notice that, as it can be inferred from Fig. 4a and c, temporal
interpolation have to be carried out at the nodes located in the
vicinity of the computational domain boundaries as a consequence
of the leaps adopted in the numerical scheme.
5. Model testing
5.1. Introduction
The implemented model describes three physical phenomena
occurring during hydraulic transients in pipe-flow: fluid-
structure interaction, skin friction and dry friction. To ensure and
verify the implemented code outputs according with the modelling
assumptions, the three phenomena are assessed one at a time by
deactivating their functionality in the code; afterwards they are
The second allows the assessment of Junction coupling due to the
valve movement. Fig. 5a and b depict the numerical outputs from
these simulations. Both show the effect of the axial stress waves
propagating at a celerity three times faster than the waterhammer
wave.
The apparent amplification of the transient pressure of Fig. 5
can be further analysed by longer simulation periods. For this pur-
pose simulation periods of 3 s were launched. Results are shown in
Fig. 6.
Fig. 6a depicts a phenomenon called the Poisson-coupling beat,
which was already documented and described by Tijsseling [27].
The phenomenon is not present for a non-anchored pipe end as
shown in Fig. 6b. Comparing with Joukowsky overpressure
(DHJK ¼ af DV=g) both simulations show that FSI does not introduce
directly damping into the pipe system and the maximum pressure
may be much higher than Joukowsky overpressure.
5.3. Skin friction verification

Table 3
The verification of the skin friction models was carried out by
Summary of simulations carried out for model verification. (U) activated mechanism,
(–) deactivated mechanism. deactivating FSI and activating, according to Table 2, the different
skin friction computations once at a time: (i) quasi-steady friction
Test Fluid-structure interaction Skin friction Dry
(SF-1), (ii) Brunone’s unsteady friction (SF-2), and (iii) Trikha’s
ID friction
Poisson Junction Steady Brunone Trikha Coulomb unsteady friction (SF-3).
coupling coupling Fig. 7 compares both unsteady friction models with quasi-
Cl-1 – – – – – – steady friction at the downstream end of the pipe; Joukowsky
FSI-1 U – – – – – overpressure (DHJK ) is also presented in this figure. Brunone’s
FSI-2 – U – – – – unsteady friction model introduces a higher damping and delay
SF-1 – – U – – –
on the transient wave, while Trikha’s model affects rather the wave
SF-2 – – – U – –
SF-3 – – – – U –
shape.
DF-1 – – – – – U To analyse the overall dissipation effect of skin friction in the
DF-2 U – – – – U pipe system, a longer simulation period of 3 s was run. Fig. 8 shows
DF-3 – U – – – U the output for the three different skin friction models. Unsteady
Note: Cl = classical model; FSI = fluid-structure interaction; SF = skin friction; friction models significantly increase the pressure wave damping,
DF = dry friction. specially Brunone’s model with the k coefficient calculated accord-
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90
Fig. 5. Frictionless 4-equation FSI model output for (a) anchored and (b) non-anchored downstream pipe-end vs. classical theory. Simulation period of 0.3 s.
Fig. 6. Frictionless 4-equation FSI model output for (a) anchored and (b) non-anchored downstream pipe-ends vs. Joukowsky overpressure. Simulation period of 3 s.
ing to Eq. (4). A small phase shift is also observed in the pressure
wave.
5.4. Dry friction models verification
Deactivating FSI as depicted in Table 2 allows cutting out the
interaction between the pressure wave in the fluid and the axial
stress wave in the pipe-wall, yet the system is still composed of
four equations: the two classic mass and momentum conservation
equations of waterhammer theory and the beam equations for the
pipe axial vibration.
In the real system dry friction directly affects the momentum
dissipation of the pipe-wall and indirectly the fluid, and vice versa.
Hence, in order to isolate, and simplify, the assessment of the dry
friction phenomenon a different transient simulation is performed.
FSI and skin friction are deactivated, fluid density is set to qf ¼ 0,
and only the structure is excited (DF-1), so the pipe vibrates unaf-
fected by the inner fluid (as if it was empty) but affected by the dry
friction between the pipe-wall and the outer media (i.e. pipe sup-
ports). As explained in Section 3.3, the Coulomb’s dry friction coef-
ficient value used corresponds to the one for copper sliding over
cast iron, which is lk ¼ 0:29. In this analysis no stick-slip condition
is considered. Notice that in this set-up the normal force N is
reduced as only the mass of the pipe-wall affects dry friction com-
putation. The transient is generated by hammering the pipe at the
upstream boundary, which is set fixed after the impact while the
downstream boundary is free to move. Fig. 9 shows the output of
this simulation compared with the Rankine solution
(DrRK ¼ qs as DU), which is equivalent to Joukowsky’s expression
81
but for axial stress transients. A linear decrease in the wave ampli-
tude can be observed in Fig. 9. Unlike the traditional logarithmic
decrement associated to viscous dissipation, dry dissipation
involves a rather linear damping [12].
The effect of dry friction on pressure head for a fluid-filled con-
duit is depicted in Fig. 10 for the two anchoring conditions. In both
simulations the dry friction dissipation affects the pressure tran-
sient wave in a similar manner, by smoothing the edges of the
stepped transitions caused by the pipe-wall axial movement. This
effect is much more evident when the valve is released, as pipe-
wall displacements are higher.
6. Model application
6.1. Simulation of combined effects
In the present analysis the experimental tests presented in Sec-
tion 2 are simulated using the numerical model developed. Fluid-
structure interaction is considered in the numerical simulations
and the main input parameters are presented in Table 4.
A preliminary analysis has been carried out to identify which of
the two UF models described better the observed dynamic beha-
viour combined with FSI effects. The simulations considering
anchored (a) and non-anchored (b) pipe-ends are depicted in
Fig. 11. Both Brunone’s (blue1 lines) and Trikha’s (red lines)
unsteady skin friction losses are analysed and dry friction omitted.
1
For interpretation of color in Fig. 11, the reader is referred to the web version of
this article.
82 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90
Fig. 7. Quasi-steady friction (SF-1) vs. (a) Brunone’s (SF-2) and (b) Trikha’s (SF-3)
unsteady friction. The horizontal dashed line represents Joukowsky overpressure.
Fig. 8. Quasi-steady friction (SF-1) vs. Brunone’s (SF-2) and Trikha’s (SF-3) unsteady
friction models. Simulation period of 3 s.
Results are compared with pressure measurements (black lines) at
the downstream end section for a simulation period of 3 s. First,
the overall wave dissipation is underestimated in both simulations,
with Brunone’s results being slightly closer to measurements in
terms of damping and wave timing. Second, in the first 10 pressure
wave cycles UF has hardly any effect being indifferent the use on
any of the models.
A similar analysis has been carried out combining dry friction
with FSI effects and using Brunone’s unsteady friction model;
results are presented in Fig. 12. A much higher damping effect
due to dry friction can be observed in the released set-up, specially
for a long simulation period. The reason is the higher energy dissi-
pation caused by the higher axial pipe-wall displacements in the
released set-up. Coulomb’s dry friction is a force affecting the
Fig. 9. Axial stress of a vibrating fluid-emptied conduit excited by hammering and
damped by dry friction. The horizontal dashed line represent Rankine axial stress
rise.
momentum of the pipe-wall. Dry friction force times pipe-wall dis-
placement is the energy dissipated by the pipe system to its
surroundings.
In Fig. 13 the numerical output, considering the combination of
the three phenomena, is compared with the experimental data for
all the assessed discharges and for (a) anchored pipe-end and (b)
non-anchored pipe-end. Skin friction losses are computed using
Brunone’s model and dry friction by Coulomb’s law. The simulation
period of 0.15 s allows the assessment of the wave shape. As it can
be observed, during the first wave cycles FSI is the dominant effect,
specially in the case of the released set-up.
6.2. Discussion of results
The results shown in Section 6.1 enable the analysis of the reli-
ability of the modelling assumptions concerning fluid-structure
interaction, skin friction and dry friction. Notice the coherent pat-
tern followed by all the assessed time series (v.s. Fig. 13): for the
different initial flow velocities (0.26, 0.36, and 0.41 m/s), and for
both anchoring set-ups (fixed and moving downstream valve).
The accurate fitting of the first pressure cycle, either in the tests
with anchored pipe-end (Fig. 13a) or, specially, in the ones with
non-anchored pipe-end (Fig. 13b), indicate the good performance
of the 4-equations model in describing the pipe system structural
behaviour and its rebound on the pressure transient wave. Hence
the rightness of the hypothesis of considering the pipe-wall axial
vibration is confirmed, being FSI dominant over UF and DF at the
beginning of the transient. However, as the wave propagates (sec-
ond and third wave cycles) the numerical output, in the case of
non-anchored pipe-end, tends to detach from the measured data
revealing that modelling assumptions can be further improved.
Moreover, the long term simulations (v.s. Fig. 11) show that
Brunone’s unsteady friction computation outputs slightly higher
wave damping as compared with the experimental tests. However,
if dry friction is not considered in the numerical model, such
damping is even further from the one observed in the real pipe sys-
tem. When dry friction is included (v.s. Fig. 12) the wave damping
rate is improved, yet remains insufficient in the set-up with
anchored pipe-end and overestimated in the released pipe-end.
Anchoring conditions are deeper analysed in the following
Fig. 14 for the tests SCP03 (anchored pipe-end) and SCP06 (non-
anchored pipe-end), revealing the different FSI behaviour depen-
dent on the anchoring conditions for longer simulation periods.
Junction coupling at the downstream pipe-end section plays a
very important role in the system response (e.g. Fig. 14b). The evi-
dent wave shape change is a result from the interaction between
the two first pipe vibration modes, related with the fluid pressure
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90 83

Fig. 10. 4-equation FSI model output for (a) anchored and (b) non-anchored downstream pipe-ends; for a frictionless systems (FSI-1,2) and taking into account only dry
friction (DF-1,2).

Table 4
Input parameters for the simulation of combined effects.

Parameters for pipe system

L ðmÞ D ðmÞ e ðmÞ Mv ðkgÞ tv ðsÞ

15.49 0.02 0.001 6 0.003

Parameters for fluid and pipe-wall materials
qf ðkg m3 Þ qs ðkg m3 Þ E ðPaÞ K ðPaÞ g ðms2 Þ m ls lk
1000 7900 1:17  1011 2:2  109 9:81 0:33 1:05 0:29

Wave celerities and domain discretization

af ðms2 Þ as ðms2 Þ Dx ðmÞ Dt ðsÞ

1239 3717 0:304 2:45  104

Fig. 11. Numerical output considering Brunone’s and Trikha’s unsteady friction Fig. 12. Experimental measurements vs. numerical output considering Brunone’s
losses vs. experimental measurements for: (a) test SCP03 with anchored pipe-end; unsteady friction losses and including and excluding dry friction for: (a) test SCP03
and (b) test SCP06 with non-anchored pipe-end. Output form the downstream end with anchored pipe-end; and (b) test SCP06 with non-anchored pipe-end. Output
section. Dry friction is not considered. form the downstream end section.
84 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90
Fig. 13. Numerical output (solid lines) considering FSI, Brunone’s unsteady friction losses and dry friction vs. experimental measurements (dashed lines) for the tests SCP01
and SCP04 (black), SCP02 and SCP05 (dark grey), SCP03 and SCP06 (light grey). (a) For anchored pipe-end, (b) for non-anchored pipe-end. Pressure history at the downstream
pipe-end.
Fig. 14. Numerical output (solid lines) considering FSI, Brunone’s unsteady and
Coulomb’s dry friction losses vs. experimental measurements (dashed lines) for: (a)
test SCP03, with anchored downstream end; and (b) test SCP06, with non-anchored
downstream end.
and the axial pipe movement. This characteristic wave shape
change due to the pipe vibration mode for a released downstream
valve was described by Bergant et al. [3] from a numerical point of
view and by Ferrás et al. [14] from an experimental standpoint.
In addition, in the case of a non-anchored pipe-end a clear wave
delay can be observed with respect to the anchored set-up (cp.
Fig. 14a and b). This wave delay occurs due to the vibration of
the pipe when this is released, and it can be observed straight from
the comparison of experimental time series. However, after some
wave cycles (t ’ 0:4), numerical output tends to overestimate this
delay (cp. time series SCP06, solid and dashed lines from Fig. 14b).
FSI does not directly affect the wave dissipation; momentum is
transferred from the fluid to the structure but also the other way
around. Hence it is by means of dry friction that the momentum
transferred to the structure is finally dissipated. Consequently,
the reason why the numerical output mismatches wave dissipation
and phase after some wave cycles lies in the dry friction modelling,
specially in the stick-slip condition, which is distributed through-
out the pipe and is not treated differently at the valve section.
The valve is heavier and, consequently, dry friction losses should
be higher and the stick-slip effect more intensive. In the case of
the released pipe-end set-up, dry friction is underestimated in
the first wave cycles and subsequently overestimated during fur-
ther propagation. FSI should be diminished according to stick-slip
condition in a way that junction coupling would be dominant only
during the firsts oscillations. Additionally, the hose connected
downstream the valve could affect as well the experimental pipe
rig behaviour, increasing the inertia of the pipe downstream end.
With regard to skin friction head losses, Fig. 11 depicts the
numerical output considering Brunone’s and Trikha’s unsteady
friction model in comparison with measurements from the test
SCP03 (anchored pipe-end) and SCP06 (non-anchored pipe-end).
Both unsteady friction models are in agreement as they offer a
pretty similar wave damping. In terms of wave timing, Brunone’s
adds more wave delay than Trikha’s. In the case of anchored
pipe-end this delay matches with good accuracy the wave propa-
gation. In the test SCP03 fluid-structure interaction is highly con-
strained and unsteady skin friction rather isolated as being the
only damping mechanisms. Observing a zoom between t ¼ 1 s
and t ¼ 1:5 s, depicted in Fig. 15, Brunone’s model seems to be
more faithful to the real system behaviour than Trikha’s.
With regard to dry friction, Fig. 12 confronts the numerical out-
put either including or excluding dry friction losses. For anchored
pipe-end, the dry friction affects the wave shape (v.i. Fig. 16a)
and the additional damping adjusts and develops in a closer man-
ner to the real wave, though, this is still underestimated for the
fixed valve set-up, as lower momentum is transferred from the
fluid to the structure. On the contrary, in the case of a non-
anchored pipe-end the effect of Coulomb friction is much more evi-
dent, where the dry dissipation fully dampens the wave in a long
term simulation (v.i. Fig. 16b). At this stage the model overesti-
mates wave damping because it fails on describing the stick-slip
phenomenon. The reason is that the implemented distributed dry
friction does not allow to locally adjust dry friction dissipation at
the valve section. It is also important to highlight that in both
anchoring conditions dry friction does not directly influence the
wave timing but only shape and amplitude.
An adjustment of the static dry friction coefficient would either
enhance the output for the anchored valve set-up (Fig. 16a) and
worsen the non-anchored one (Fig. 16b), or the other way around.
This is a sign that stick-slip condition merits further improvement
by means of imposing null pipe movement when stick condition is
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90
Fig. 15. Numerical output for Brunone’s and Trikha’s friction losses vs. experimental measurements for the test with anchored downstream end (SCP03). Time window
between t ¼ 1 s and t ¼ 1:5 s.
true. Though, this would involve the implementation of internal
conditions and describing dry friction as junction coupling rather
than friction coupling, which is out of the scope of the present
research.
7. Conclusions
The present research is based on the implementation of a 1D 4-
equation MOC solver. Experimental data collected from a straight
copper pipe-rig are used for the model validation of the main mod-
elling assumptions in terms of wave shape, timing and damping.
Fluid-structure interaction, skin friction and dry friction are the
main phenomena to be assessed. Essentially, the implemented
FSI code includes three coupling mechanisms: Poisson, junction
and friction coupling and the last one nests the skin friction models
(i.e. quasi-steady, Brunone’s and Trikha’s) and the dry friction
model (i.e. Coulomb’s friction).
With regard to fluid-structure interaction, two different set-ups
are assessed: (i) anchored downstream pipe-end, and (ii) non-
anchored downstream pipe-end. In both cases the pipe is lying
over the floor of the laboratory. In the first set-up Poisson coupling
dominates the FSI physical phenomenon (cf. Fig. 13a), while in the
second set-up junction coupling at the downstream valve section
strongly affects the transient wave (cf. Fig. 13b). Numerical results
satisfactorily fit measured pressure head at the downstream sec-
tion for the first wave cycle, when FSI phenomenon is dominant.
However, numerical output tends to detach for a long term simu-
lation, specially in the set-up for a non-anchored downstream
pipe-end (cf. Fig. 14). Skin and dry friction dissipation are the dom-
inant damping phenomena in the long term simulation.
Unsteady friction effects are rather isolated in the anchored set-
up as less momentum is transferred to the structure and conse-
quently FSI and dry friction effects are much lower. For this set-
up Brunone’s unsteady friction model gives a better performance
in the account of both wave timing and damping (cf. Fig. 11). How-
ever, without the incorporation of dry friction, a clear additional
wave damping is missing.
The implementation of Coulomb’s dry friction model aims at
describing this additional wave damping. Dry friction arises from
the shear between the pipe-wall and its surroundings. Conse-
quently, its effect is strongly related with the FSI occurring during
the transient event. The higher momentum is transferred to the
pipe-wall, the greater the effect of dry friction is (c.p. Fig. 12a
and b). The inclusion of dry friction allowed a clear improvement
of the numerical model output, proving the importance of consid-
ering such phenomenon in hydraulic transient analyses. Nonethe-
less, the present approach of nesting the Coulomb model in the FSI
friction coupling does not allow for a fully satisfactory description
of the observed pressure signal (Figs. 14 and 16), specially in the
case of released downstream pipe-end. The dry friction dissipation
occurring at the valve section is crucial for the accurate description
85
of the pipe system behaviour, and this should be treated differently
for two main reasons. First, the valve section is heavier and pipe-
wall displacements are higher than any other section, hence kine-
matic dry friction dissipation is greater at the downstream section.
Second, fluid-structure interaction effect is more intensive at the
downstream end, and consequently more sensitive to stick-slip
condition.
At this stage of research the authors conclude that a more sat-
isfactory means to represent junction coupling merits being inves-
tigated. In fact, assumptions on dry friction computation need a
further upgrade as, in the present paper, Coulomb’s model was
nested in the FSI friction coupling but not considered in junction
coupling. Junction coupling arises from a balance of forces at pipe
junctions, tees, elbows, boundaries, etc. Dry friction force can be as
well considered in such a balance. Using this approach stick-slip
condition could be as well improved by imposing null pipe-wall
movement when stick criterion is met.
Acknowledgements
This research is supported by the Portuguese Foundation for
Science and Technology (Fundação para a Ciência e a Tecnologia)
through the project ref. PTDC/ECM/112868/2009 ‘‘Friction and
mechanical energy dissipation in pressurized transient flows: con-
ceptual and experimental analysis” and the PhD grant ref. SFRH/
BD/51932/2012 also issued by FCT under IST-EPFL joint PhD initia-
tive. The second author is founded by the Swiss Competence Center
for Energy and Research Supply of Electricity (SCCER-SoE).
Appendix A
A.1. Modification of the momentum conservation equation of the pipe-
wall
In order to include dry friction dissipation in the physical sys-
tem the fundamental conservation equations must be adapted.
Dry friction affects the momentum conservation equation of the
pipe axial movement as depicted in Fig. 17. In the present appendix
the momentum conservation equation of the pipe axial movement
is derived.
Similarly to waterhammer fundamental conservation equations
derivation [8], beams equation for axial stress waves can be devel-
oped from the Reynolds transport theorem:
Z
dBsys d
¼ bqs d8 þ ½bqs As ðU  WÞ2  ½bqs As ðU  WÞ1 ð38Þ
dt dt cv
where the subscripts after the square brackets indicate the control
volume boundary sections, W their respective velocities, B is the
extensive property of the system, which in this case is the pipe-
wall momentum (M ¼ ms U), and b is the intensive property:
86 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90

Dm 1
b ¼ lim U ¼U ð39Þ sws ¼ f qf V r jV r j ð50Þ
Dm!0 Dm 8
Also from the second law of Newton we know that: substituting Eq. (50) into Eq. (49)
X P
dM F @ r f q f Af
¼ F ð40Þ lim ¼ As þ V r jV r j þ ðqs As þ qf Af Þg l signðUÞ ð51Þ
dt Dx!0 Dx @x 4r
Substituting Eqs. (39) and (40) into Eq. (38) Substituting Eq. (51) into Eq. (46) and dividing by qs As :
Z X  
d
U qs d8 þ ½qs As ðU  WÞU 2  ½qs As ðU  WÞU 1 ¼ ð41Þ dU 1 @ r qf Af f qf Af
F  ¼ V r jV r j þ 1 þ g l signðUÞ ð52Þ
dt cv dt qs @x qs As 4r qs As
Applying Leibnitz’s rule of integration and defining dxdt1 ¼ W 1 and Finally expanding the total derivative in Eq. (52) and neglecting
dx2
¼ W2 its convective acceleration term, which is much smaller in most
dt
Z engineering applications, the momentum conservation equation
x2
@
ðqs As UÞdx þ ðqs As UÞ2 W 2  ðqs As UÞ1 W 1 þ ½qs As ðU  WÞU 2 Eq. (53) for the pipe-wall axial movement is obtained:
@t  
@U 1 @ r qf Af f qf Af
x1
X
 ½qs As ðU  WÞU 1 ¼ F ð42Þ  ¼ V r jV r j þ 1 þ g l signðUÞ ð53Þ
@t qs @x qs As 4r qs As
Applying the mean-value theorem to Eq. (42),
@ X A.2. MOC transformation
ðqs As UÞDx þ ðqs As U 2 Þ2  ðqs As U 2 Þ1 ¼ F ð43Þ
@t
The derivation of compatibility equations for the 4-equations
dividing Eq. (43) by Dx and letting Dx approach 0
P model by means of MOC transformation is carried out in the pre-
@ @ F sent appendix and it is based on Shuy and Apelt [23] and Tijsseling
ðq As UÞ þ ðqs As U 2 Þ ¼ lim ð44Þ
@t s @x Dx!0 Dx [30], where dry friction terms have been added.
and expanding the differential terms in the following manner:
 P A.2.1. Matrix notation
@ @ @U @U F The system of equations composed of Eqs. (12)–(15) can be
U ðqs As Þ þ ðqs As UÞ þ qs As þ qs As U ¼ lim ð45Þ
@t @x @t @x Dx!0 Dx expressed in matrix notation as follows:
It can be proved that the term in square brackets of Eq. (45) is @y @y
A þB ¼r ð54Þ
actually the mass conservation equation [8], hence its value is 0 @t @x
and taking the total derivative of the remaining terms Eq. (45) where y is the vector of unknowns
becomes 1 0
P V
dU F BpC
qs As ¼ lim ð46Þ B C
dt Dx!0 Dx y¼B C ð55Þ
@UA
Notice that Eq. (46) is nothing but the second law of Newton
applied in an infinitesimal system.
r
At this point the balance of forces acting on the system may be A and B are the matrices of coefficients
incorporated in Eq. (46). Fig. 17 depicts this balance of forces acting 0 1
1 0 0 0
in the axial direction of the pipe. B C
The forces acting on the control volume in the axial direction of B0 1
qf a2f
0  2Em C
B C
the pipe are: A¼B C ð56Þ
B0 0 1 0 C
@ A
F r1 ¼ r1 As 0 rm
eE
0  q 1a2
s s
F r2 ¼ r2 As
0 1
F w ¼ ðx2  x1 Þqs As g sin H 0 1
qf 0 0
B C
F sws ¼ sws pDðx2  x1 Þ B1 0 0 0 C
B¼B
B0
C ð57Þ
F df ¼ ðx2  x1 Þðqs As þ qf Af Þg cos Hl signðUÞ ð47Þ @ 0 0  q1 CA
s

Notice that F df is the Coulomb’s dry friction force (Eqs. (10) and 0 0 1 0
(11)), F sws is the skin friction force, F w the weight of the pipe wall and r is the right-hand-side vector
and H is the angle of the pipe respect the horizontal plane. Let’s 0 1
consider a horizontal pipe. The balance of forces is then  4rf V r jV r j
X B C
B  0 C
F ¼ r1 As þ r2 As þ sws pDðx2  x1 Þ r¼B  C ð58Þ
B qf Af f
V jV j þ 1 þ
qf Af
g l signðUÞ C
þ ðx2  x1 Þðqs As þ qf Af Þg l signðUÞ ð48Þ @ qs As 4r r r q As s
A
0
and dividing by Dx and letting Dx approach 0:
P Matrices A and B are regular. For r ¼ 0 the system (54) is linear,
F @r whereas for r – 0 the friction terms make the system (54) non-
lim ¼ As þ sws pD þ ðqs As þ qf Af Þg l signðUÞ ð49Þ
Dx!0 Dx @x linear. In Appendix A.3 the solution of the characteristic equation
and assuming quasi-steady skin friction the shear stress between jB  kAj ¼ 0 is derived reaching the following four distinct real
the fluid and the pipe-wall can be computed by means of Darcy- roots and identifying the system as hyperbolic:
Weisbach equation k1 ¼ þaf ; k2 ¼ af ; k3 ¼ þas ; k4 ¼ as ð59Þ
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90 87

Hyperbolic systems can be transformed by means of multiplica- dv 1

¼ ðTrÞi ; i ¼ 1; 2; 3; 4 ð69Þ
tion by a regular matrix dt
@y @y when it is considered along a line in the x–t plane having the char-
TA þ TB ¼ Tr ð60Þ
@t @x acteristic direction
For square matrices with distinct real eigenvalues there is a dz
matrix S such that ¼ ki ; i ¼ 1; 2; 3; 4 ð70Þ
dt
1 1
S A BS ¼ K ð61Þ
where K Eq. (69) are known as the compatibility equations along the
0 1 characteristic lines with slopes k1 in the x–t plane.
k1 0 0 0 i

B0 Finally, substituting Eqs. (89) and (56) into Eqs. (66) and (89),
k2 0 0C
K¼B
C (58) and (66) into Eq. (69).
B C ð62Þ
@0 0 k3 0A
0 0 0 k4

0
af 2 1
þ 2m2 eEr  f 2 2m  q2am2  f 2 2m
a a
B 1 1
q f af
asaf 2  af C
B 1 s s E
C
B C
as as
af 1 as
af 1
B
af 2 C 0 1
B C V
B C
 q 1a  2m2 eEr  f 2 2m 2m 2m C
a af
B 1
asaf 2   þ a BpC
B qs a2s 2 f C
C B
d C
E
B
ff
1
C dt B C
as as
1 as 1
B af af
B C @UA
B a2f af as C
2
B qf eE
rm rm rm
 q 1as þ 2qf eE rm

a 2 C r
2

  eE
aasf 2 þ eE as 1
B a 2
1 afs s
2
1 afs C
B 1 as C
B C
@ a2 a2f as A
qf eE
rm

af f 2 rm
eE
rm

aasf 2  eE as 1 q as  2qf eE2
1 r m2

af 2
ð71Þ
s
1 as 1 a 1 a
0 s

af 2     1
s

q q
B  4r V r jV r j þ 2m
saf 2 qs As 4r V r jV r j þ 1 þ qs As g l signðUÞ C
f a A
f f f A
f f

B 1 as C
B
af 2  C
B    C
B qf Af f qf Af C
B  4r V r jV r j þ 2m
a 2 q A 4r V r jV r j þ 1 þ q A g l signðUÞ C
f as
B s s s s C
B C
f
1 as
¼B C
B r m af 2     C
qf Af f qf Af
B qf eE
a 2 4r V r jV r j þ q A 4r V r jV r j þ 1 þ q A g l signðUÞ C
f
B f s s s s C
B 1 as C
B     C
@ r m af 2
qf Af f qf Af A
qf eE
a 2 4r V r jV r j þ q As 4r V r jV r j þ 1 þ q As g l signðUÞ
f
f s s
1 as

the T matrix is chosen as which for convenience of integration and implementation, using
1 1 the summary Table 2, the system (71) is presented in the following
T¼S A ð63Þ form:
Substituting Eq. (63) into Eq. (61) leads to the condition 0 10 1 0 1
nf V nf p nf U nf r V SF f þ DF f
TB ¼ KTA ð64Þ B nf nf p nf U C B
n f r CB p CC B C
B V B SF f þ DF f C
B CB C ¼  B C ð72Þ
The transformation matrix T is determined in Appendix A.4. @ nsV nsp nsU nsr A@ U A @ SF s þ DF s A
With condition (64), Eq. (60) becomes the so called normal form nsV nsp nsU nsr r SF s þ DF s
of the system (54)
@y @y A.3. Solution of the characteristic equation
TA þ KTA ¼ Tr ð65Þ
@t @x
The characteristic equation of the system (54) is:
Introducing the vector
jB  kAj ¼ 0 ð73Þ
v ¼ TAy ð66Þ
From Eqs. (56), (57) and (62)
Eq. (65) becomes 0 1
k 1
0 0
@v @v B
qf
C
þK ¼ Tr ð67Þ B 1  k2 k2mC
@t @x B q f af
0 E C
B  kA ¼ B
B
C ð74Þ
1 C
which can be written as a set of four uncoupled equations: B 0 0 k  q C
@ s A

@v i @v i 0  kreEm 1 q ka2
þ ki ¼ ðTrÞi ; i ¼ 1; 2; 3; 4 ð68Þ s s
@t @x
and
Each of Eq. (68) transforms into an ordinary differential
equation
88 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90

Fig. 16. Experimental measurements (black lines) vs. numerical output considering Brunone’s friction losses and with (blue lines) or without (red lines) dry friction for the
tests: SCP03, with anchored downstream end (a); and SCP06, with non-anchored downstream end (b). Time window between t ¼ 1 s and t ¼ 1:5 s. (For interpretation of the
references to colour in this figure legend, the reader is referred to the web version of this article.)

! !
k2 k2 1 1 k2 1 Eq. (78) is a system composed by sixteen equations with sixteen
jB  kAj ¼  þ   þ þ k2
qf af
2 qs a2s qs qf qs a2s qs unknowns tij . Let’s consider row i:
!
ti2 ¼ ki ti1
2m k2 rm !

E eE ti1 ti2 rm
¼ ki þ t
qf qf a2f eE i4
¼0 ð75Þ
ti4 ¼ ki ti3
If the second-order Poisson-term (i.e. m2 -term) is neglected  
t i3 t i2 2m ti4
[23,30] the solutions of the characteristic equations are:  ¼ ki þ ð79Þ
qs E qs a2s
k1 ¼ þaf ; k2 ¼ af ; k3 ¼ þas ; k4 ¼ as ð76Þ
or what is the same
0 10 1 0 1
ki 1 0 0 t i1 0
A.4. Transformation matrix
B 1  qkai 2 rm C
B qf 0 ki eE CB t i2 C
B B0C
B f s
CB C B C
C¼B C ð80Þ
From Appendix A.2 the condition expressed in Eq. (64) must be B 0 1 C
@ 0 ki A@ t i3 A @ 0 A
fulfilled. Hence substituting Eqs. (56), (57) and (62) into Eq. (64).
0 ki 2Em  q1 ki
t i4 0
0 10 0 1 0 0 1 s q a2
s s
t11 t 12 t13 t 14 qf
Bt CB 1 0 0 0 C The system (80) can be written in matrix notation
B 21 t 22 t23 t 24 CB C
B CB C
@ t31 t 32 t33 t 34 AB
@ 0 0 0  q1s A
C ðB  ki AÞT t i ¼ 0 ð81Þ
t41 t 42 t43 t 44 0 0 1 0 If the matrix ðB  ki AÞT is regular, Eq. (81) has only the trivial
0 1
0 10 1 1 0 0 0 solution t i ¼ 0, which leads to a singular matrix T. To obtain a reg-
k1 0 0 0 t11 t 12 t 13 t14
B C
B 0 k 0 0 CB t CB 0 1
0  2Em C ular matrix T, ðB  ki AÞT has to be singular. Hence
B 2 CB 21 t 22 t 23 t24 CB q f a2 C
¼B CB CB f
C jB  ki Aj ¼ 0 ð82Þ
@ 0 0 k3 0 A@ t31 t 32 t 33 t34 AB 0 0 1 0 C
@ A
0 0 0 k4 rm
t41 t 42 t 43 t44 0 eE
0  q 1a2 The equations of the system (80) are dependent so one of them
s s
can be omitted. The second equation of the system (80) is omitted
ð77Þ
if i ¼ 1 or i ¼ 2, and the fort equation is omitted if i ¼ 3 or i ¼ 4.
which is For i ¼ 1 and i ¼ 2 the eigenvector ti is calculated as follows:
0 1
t12 tq11 t 14  tq13  ki t i1 þ t i2 ¼ 0
B f s
C  ki t i3 þ t i4 ¼ 0
B t22 t21 t 24  t23 C
B qf qs C  
B C t i3 t i2 2m t i4
Bt t t C  þ ki þ ¼0 ð83Þ
B 32 q31f t 34  q33s C qs E qs a2s
@ A
t42 tq41 t 44  tq43
 
f s
0  1
t 12 rm t 12 2m
k t k
B 1 11 1 qf a2f eE 14þ t k1 t13 k1 E
þ qt14a2
C ð78Þ
B   C
s s
B  C
B rm t 22 2m C
B k2 t 21 k2 qt22a2 þ eE t 24 k2 t23 k2 E þ q a2 Ct 24
B f f s s C
¼B B   
C
C
Bk t t rm
k3 t32E2m þ qt34a2 C
B 3 31 k3 qf32a2 þ eE t 34 k3 t33 C
B f s s C
B    C
@ t 42 rm t 42 2m t 44
A
k4 t 41 k4 q a2 þ eE t 44 k4 t43 k4 E þ q a2
f f s s

Fig. 17. Schematic of forces acting in the pipe-wall in the axial direction.
 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90 89

or [3] Bergant A, Tijsseling AS, Vítkovskỳ JP, Covas DI, Simpson AR, Lambert MF.
Parameters affecting water-hammer wave attenuation, shape and timing. Part
t i2 2: case studies. J Hydraul Res 2008;46(3):382–91.
t i1 ¼
ki [4] Brunone B, Golia U, Greco M. Some remarks on the momentum equation for
fast transients. In: Proc, int meeting on hydraulic transients and water column
t i4
t i3 ¼ separation, Valencia; 1991. p. 201–9.
ki [5] Brunone B, Golia U, Greco M. Effects of two-dimensionality on pipe transients
! modeling. J Hydraul Eng 1995;121(12):906–12.
1 1 2m [6] Brunone B, Karney BW, Mecarelli M, Ferrante M. Velocity profiles and unsteady
 ti4 ¼ t i2 ð84Þ
qs k2i qs a2s E pipe friction in transient flow. J Water Resour Plann Manage 2000;126
(4):236–44.
[7] Capone G, D’Agostino V, Della Valle S, Guida D. Influence of the variation
and taking t i2 ¼ ki between static and kinetic friction on stick-slip instability. Wear 1993;161
(1):121–6.
t i1 ¼ 1 [8] Chaudhry MH. Applied hydraulic transients. New York, NY: Springer New
t i2 ¼ ki York; 2014. ISBN 978-1-4614-8537-7.
[9] Covas D, Ramos H, Graham N, Maksimovic C. The interaction between
2m viscoelastic behaviour of the pipe-wall, unsteady friction and transient
t i3 ¼  
pressures. In: International conference on pressure surges, proceedings.
E q 1k2  q 1a2 Chester; 2004.
s i s s
[10] Davis JR. Concise metals engineering data book. ASM International; 1997. ISBN
2m 978-0871706065.
t i4 ¼ ki   ð85Þ [11] Feeny B, Guran A, Hinrichs N, Popp K. A historical review on dry friction and
E q 1k2  q 1a2 stick-slip phenomena. Appl Mech Rev 1998;51(5):321–41.
s i s s [12] Feeny B, Liang J. A decrement method for the simultaneous estimation of
coulomb and viscous friction. J Sound Vib 1996;195(1):149–54.
For i ¼ 3 and i ¼ 4 the eigenvector t i is calculated as follows: [13] Ferràs D, Manso PA, Covas DI, Schleiss AJ. Comparison of conceptual models for
fluid-structure interaction in pipe coils during hydraulic transients. J Hydraul
 ki ti1 þ ti2 ¼ 0 Res 2015 [in preparation].
rm
[14] Ferràs D, Manso PA, Covas DI, Schleiss AJ. Experimental evidence of damping
t i1 ki t i2
  ki ti4 ¼ 0 mechanisms during hydraulic transients in pipe flow. J Fluids Struct 2015
qf qf a2f eE [accepted for publication].
[15] Forsythe GE, Wasow WR, et al. Finite-difference methods for partial
 ki ti3 þ ti4 ¼ 0 ð86Þ differential equations. Wiley; 1960. ISB 978-1258669256.
[16] Ghidaoui MS, Zhao M, McInnis DA, Axworthy DH. A review of water hammer
or theory and practice. Appl Mech Rev 2005;58(1):49–76.
[17] Joukowsky N. On the hydraulic hammer in water supply pipes. Proc Am Water
 ki ti1 þ ti2 ¼ 0 Works Assoc 1904;24:341–424.
! [18] Kagawa T, Lee I, Kitagawa A, Takenaka T. High speed and accurate computing
1 1 rm method of frequency-dependent friction in laminar pipe flow for
 t i2 ¼ qf t i4 characteristics method. Trans Jpn Soc Mech Eng Ser A 1983;49(447):2638–44.
k2i a2f eE
[19] Keramat A, Tijsseling A. Waterhammer with column separation, fluid-
t i4 structure interaction and unsteady friction in a viscoelastic pipe. In:
t i3 ¼ ð87Þ International conference on pressure surges, proceedings; 2012.
ki
[20] Lavooij C, Tijsseling AS. Fluid-structure interaction in liquid-filled piping
and taking t i4 ¼ ki systems. J Fluids Struct 1991;5(5):573–95.
[21] Ramos H, Covas D, Borga A, Loureiro D. Surge damping analysis in pipe
rm 1 systems: modelling and experiments. J Hydraul Res 2004;42(4):413–25.
t i1 ¼ qf [22] Schohl G. Improved approximate method for simulating frequency-dependent
eE 1
k2i
 a12 friction in transient laminar flow. Trans-Am Soc Mech Eng J Fluid Eng
f
1993;115:420–4.
rm 1 [23] Shuy E, Apelt C. Experimental studies of unsteady shear stress in turbulent
t i2 ¼ ki qf
2  a2
eE 1 1 flow in smooth round pipes. In: Conf on hydraulics in civil engineering; 1987.
k
i f p. 137–41.
[24] Skalak R. An extension of the theory of water hammer. New York: Columbia
t i3 ¼ 1 Univ., Dept of Civil Engineering and Engineering Mechanics; 1955.
t i4 ¼ ki ð88Þ [25] Stuckenbruck S, Wiggert D. Unsteady flow through flexible tubing with
coupled axial wall motion. In: 5th international conference on pressure surges,
And finally substituting the eigenvalues ki from Eq. (76) in Eq. Hannover, Germany, proceedings; 1986.
[26] Tijsseling A. Fluid-structure interaction in liquid-filled pipe systems: a review.
(85) for i ¼ 1; 2 and in Eq. (88) for i ¼ 3; 4 yields: J Fluids Struct 1996;10(2):109–46.
0
af 2 1 [27] Tijsseling A. Poisson-coupling beat in extended waterhammer theory. ASME-
2m 2m  f 2
a Publications-Ad 1997;53:529–32.
B 1 af
asaf 2 C
B 1 C [28] Tijsseling A, Vardy A. Axial modelling and testing of a pipe rack. In: 7th
B C
as
as 1
af conference on pressure surges. BHR group conference series, vol.
B
af 2 C
B C 19. Mechanical Engineering Publications Limited; 1996. p. 363–84.
B C
2m 2m  2 C
af
B 1 af
asaf 2 [29] Tijsseling A, Vardy A, Fan D. Fluid-structure interaction and cavitation in a
B C single-elbow pipe system. J Fluids Struct 1996;10(4):395–420.
T¼B
B
1 as as
af 1 C
C [30] Tijsseling AS. Fluid-structure interaction in case of waterhammer with
B C cavitation. Ph.D. thesis. TU Delft, Delft University of Technology; 1993.
B a2 a2f as C
B qf eE
rm

af f 2 qf rm

af 2 1 as C [31] Trikha AK. An efficient method for simulating frequency-dependent friction in
B eE C transient liquid flow. ASME Trans J Fluids Eng 1975;97:97–105.
B 1 as 1 as C
B C [32] Vardy A, Brown J. On turbulent, unsteady, smooth-pipe friction. In: 7th
@ a2f a2f as A
qf eE
af 2
rm
qf eE
rm

af 2 1 as conference on pressure surges. BHR group conference series, vol.
1 as 1 as
19. Mechanical Engineering Publications Limited; 1996. p. 289–312.
[33] Vardy A, Brown J. Transient turbulent friction in smooth pipe flows. J Sound
ð89Þ Vib 2003;259(5):1011–36.
[34] Vardy A, Brown J. Transient turbulent friction in fully rough pipe flows. J Sound
Vib 2004;270(1):233–57.
References [35] Vardy AE, Brown JM. Transient, turbulent, smooth pipe friction. J Hydraul Res
1995;33(4):435–56.
[36] Vitkovsky JP, Lambert MF, Simpson AR, Bergant A. Advances in unsteady
[1] Bergant A, Simpson AR, Vı ` tkovsky J. Developments in unsteady pipe flow
friction modelling in transient pipe flow. In: 8th international conference on
friction modelling. J Hydraul Res 2001;39(3):249–57.
pressure surges, The Hague, The Netherlands; 2000. p. 471–82.
[2] Bergant A, Tijsseling AS, Vítkovskỳ JP, Covas DI, Simpson AR, Lambert MF.
[37] Walker J, Phillips J. Pulse propagation in fluid-filled tubes. J Appl Mech
Parameters affecting water-hammer wave attenuation, shape and timing. Part
1977;44(1):31–5.
1: mathematical tools. J Hydraul Res 2008;46(3):373–81.
90 D. Ferràs et al. / Computers and Structures 175 (2016) 74–90
[38] Weaver Jr W, Timoshenko SP, Young DH. Vibration problems in
engineering. John Wiley & Sons; 1990. ISBN 978-0-471-63228-3.
[39] Weijde P. van der, 1985. Prediction of pressure surges and dynamic forces in
pipeline systems, influence of system vibrations on pressures and dynamic
forces (fluid–structure interaction). In: Transactions of the symposium on
pipelines, Utrecht, The Netherlands; 1985.
[40] Wiggert D. Fluid-structure interaction in piping systems. In: Proceedings
Druckstoberechnung von Rohrleitungssystemen, Haus der Technik. Essen,
Germany; 1983.
[41] Wiggert D, Hatfield F, Stuckenbruck S. Analysis of liquid and structural
transients in piping by the method of characteristics. J Fluids Eng 1987;109
(2):161–5.
[42] Zarzycki Z. Hydraulic resistance of unsteady turbulent liquid flow in pipes. In:
BHR group conference series; 1997. p. 163–78.
[43] Zarzycki Z. On weighting function for wall shear stress during unsteady
turbulent pipe flow. BHR group conference series, vol. 39. Bury St.
Edmunds: Professional Engineering Publishing; 2000. p. 529–46. 1998.
[44] Zielke W. Frequency-dependent friction in transient pipe flow. J Basic Eng
1968;90:109.