Professional Documents
Culture Documents
To cite this article: Amgad S. Elansary , M. Hanif Chaudhry & Walter Silva (1994) Numerical and
experimental investigation of transient pipe flow, Journal of Hydraulic Research, 32:5, 689-706,
DOI: 10.1080/00221689409498709
Taylor & Francis makes every effort to ensure the accuracy of all the information (the
“Content”) contained in the publications on our platform. However, Taylor & Francis,
our agents, and our licensors make no representations or warranties whatsoever as to
the accuracy, completeness, or suitability for any purpose of the Content. Any opinions
and views expressed in this publication are the opinions and views of the authors,
and are not the views of or endorsed by Taylor & Francis. The accuracy of the Content
should not be relied upon and should be independently verified with primary sources
of information. Taylor and Francis shall not be liable for any losses, actions, claims,
proceedings, demands, costs, expenses, damages, and other liabilities whatsoever
or howsoever caused arising directly or indirectly in connection with, in relation to or
arising out of the use of the Content.
This article may be used for research, teaching, and private study purposes. Any
substantial or systematic reproduction, redistribution, reselling, loan, sub-licensing,
systematic supply, or distribution in any form to anyone is expressly forbidden. Terms
& Conditions of access and use can be found at http://www.tandfonline.com/page/
terms-and-conditions
Downloaded by [North Carolina State University] at 16:22 30 November 2014
Numerical and experimental investigation of transient
pipe flow
Etude numérique et expérimentale d'un écoulement
transitoire dans une conduite
AMGAD S. ELANSARY
Assistant Professor,
Irrigation and Hydraulics Dept., WALTER SILVA
Faculty of Engineering, Assistant Professor,
Cairo University, Dept. of General Engineering,
Downloaded by [North Carolina State University] at 16:22 30 November 2014
M. HANIF CHAUDHRY
Professor,
Dept. of Civil and Environmental Engineering,
Washington State University,
Pullman WA
ABSTRACT
Two mathematical formulations for the computation of transient flow in piping systems are compared with
experimental data. The formulations are: a four-equations fluid structure interaction model (FSI) that includes
Poisson coupling, and a two-equations model for the fluid. Both models are solved numericaly using the
method of characteristics. A partial-closure of a valve located at an intermediate point in a pipeline is used to
create transient flow. The two-equations model computed the maximum pressure peak satisfactorily but the
FSI model gave an overall better simulation. An unsteady-friction model, added to the FSI model, did not
influence the final results significantly. The experimental procedures followed to obtain the valve characteris
tics and the pressure history along the pipeline are explained in detail. Excellent numerical results at the valve
are obtained when experimental data is used to simulate the time-dependent boundary condition.
RÉSUMÉ
Deux formulations mathématiques pour Ie calcul d'un écoulement transioire dans des systèmes de conduites
sont comparées a des résultats expérimentaux. Les formulations sont un modèle d'intéraction fluide - structure
(FSI) a quatre éguations qui comprend Ie couplage de Poisson, et un modèle a deux equations pour Ie fluide.
Les deux modèles sont résolus numériquement par la methode des cractéristiques. Une fermeture partielle
d'une vanne située a un point intermediaire de la conduite est utilisée pour créer un transitoire. Le modèle a
deux equations a calculé de facon satisfaisante le pic de pression, mais le modèle FSI a donné une simulation
encore meilleure. Un modèle de frottement non permanent, ajouté au modèle FSI, n'a pas influence significa-
tivement les résultats finaux. Les procédures expérimentales suivies pour obtenir les cractéristiques de la
vanne de fermeture et revolution de la pression le long de la conduite sont présentées en détail. D'excellents
résultats numériques au niveau de la vanne sont obtenus lorsque les données expérimentales sont utilisées
comme condition a la limite en fonction du temps.
Revision received April 26, 1994. Open for discussion till April 30, 1995.
the steady-state equation for friction do not simulate the shear stress accurately during transient
flow. Zielke (1968) developed an expression for unsteady laminar fluid friction. Trikha (1975) pro
posed a simplification of Zielke's equation that have been used as an approximation for unsteady
friction computations in turbulent transient flow (Vardy, 1992; Budny et al., 1990).
Wiggert et al. (1985) and Budny et al. (1990) validated their fluid-structure interaction models by
using a sudden closure of a valve as the transient generator. Also, most of the new methodologies
for the computations of unsteady flow friction in laminar and turbulent flow have been verified
with experimental data for sudden valve closures (Zielke, 1968; Suo and Wylie, 1989; Vardy and
Hwang, 1991). In numerical simulations the valve was assumed to close completely in less than
one calculation time interval. The major attributes of a sudden valve closure as a boundary condi
tion to generate waterhammer are the creation of a well-defined pattern of pressure heads with time
and the easy handling of the boundary conditions in the numerical computations. Possible limita
tions of a boundary condition like this in the verification of certain numerical models is the fact that
the flow descelerates to zero velocity and reverses direction very quickly, not allowing a sustained
transient flow in the forward direction. Also, the boundary condition at the valve becomes inde
pendent of time and the application of some numerical models could be restricted by changes at the
boundary. Finally, a sudden closure is not a common situation in real life applications because all
the flow regulating devices take some time to complete the movement.
The pressure oscillations generated by a partial valve-closure depend on the valve geometry, the
valve location in the system, the valve movement sequence and the percentage of closure. A partial
valve closure maintains flow in the forward direction and turbulent conditions for a longer period of
time, when compared to a total closure. If the valve is located at the downstream end of a pipe, with
its end open to the atmosphere or to a constant-head tank, any partial-closure movement will pro
duce only one single pressure peak. If the valve is located in an intermediate position (hereafter
called intermediate valve) between two constant head-tanks, pressure oscillations can be generated
depending upon the percentage of closure and the distance from the valve to both tanks. These
oscillations will damp faster than those occuring from a total closure.
This paper presents experimental results obtained during transient conditions created by a control
led partial-closure of an intermediate valve. This particular configuration provides a boundary that
maintains turbulent flow and can produce pressure oscillations for several cycles. At the same time,
no severe flow reverse is allowed because the valve remains partially open at all times. A motor
ized butterfly valve with these characteristics was assembled and installed in an experimental
setup built in the R. L. Albrook Hydraulics Laboratory at Washington State University.
Numerical Simulations
Two formulations are used to solve the governing differential equations. Experimental information
at the valve is used to model this boundary condition. The treatment of the boundary condition is
described in the next section.
The first formulation is described in Chaudhry (1987). The equation of motion and continuity for
one-dimensional unsteady pipe flow are given by:
Downloaded by [North Carolina State University] at 16:22 30 November 2014
with
xw, = p,/uM/8 0)
where u = average flow velocity; p = pressure intensity; R = inside pipe radius; xws - steady state
shear stress at the wall; pf= fluid density; af= wave speed in fluid, and ƒ = the Darcy-Weisbach
friction factor.
By using the method of characteristics, the two partial differential equations, (Eqs. 1 and 2), are
transformed into the following ordinary differential equations:
dp du xws (4)
dx
which is valid along the positive characteristic line given by — = +af and,
dp du t,„ /5)
dt * ' dt ' R
dx
which is valid along the negative characteristic line given by — = -a f .
dt '
The second formulation is an axially-coupled model presented by Wiggert et al. (1985). Four par
tial differential equations are solved for the state variables: fluid pressure, p , fluid axial velocity, v,
pipe axial velocity, ü, and the axial stress, ax. In the following analysis, an axisymmetric, linear
elastic pipe wall material, small deformation and a thin-wall pipe are assumed for the structural
equations. The fluid is described as being linear, homogeneous and isotropic and the pressure
always remains above vapor pressure. With these assumptions the governing equations becomes
at ox ax
dv dp KD _ (-i\
lus of elastisity for the pipe wall; K is the bulk modulus of fluid; v is the Poisson ratio and e is the
pipe wall thickness.
The above four fluid-structure equations are coupled through the Poisson ratio, v, as given by
Equations 6 and 9. The method of characteristics is used to transform the linear, first-order, hyper
bolic partial differential equations, (Eqs. 6-9), to ordinary differential equations. Following For-
sythe and Wasow (1960) Equations 6-9 can be solved to obtain the coupled wave speed for the
fluid and the pipe. The relation for the wave speed in fluid is
l
C) = -[M-,jM2-4a2fa2] (11)
C) = l
-[M+,jM2-4afa2] (12)
where,
M = a~f+a~t + 2v K*R/ep,
2 K*
a, = —
1
Pf
2 E
=
°' P,
The four compatibility equations are given as follows:
1 vR - X 2t„ (13)
±C G C G ±C — + ■—(j/ ^ + C]G -
-dt
t f f^f-P' f f-J-, J ƒ " ƒf +
df K* e >_ dt ry?
2v
Gf = c
P,( r -"')
G -
2v
"/ 2
P, ( c , - a2,)
The four compatibility equations, (Eqs 13 and 14), are integrated along the corresponding charac
teristic lines. The time-line interpolation proposed by Goldberg and Wylie (1983) is used to solve
for p, v, ox, ü .
Downloaded by [North Carolina State University] at 16:22 30 November 2014
The shear stress between the fluid and the pipe wall, T„, is computed from the following equations:
Xm and Xwu are the steady and the unsteady shear stresses at the wall respectively. Expressions for
these stresses are
where
vf= kinematic viscosity; n, = 26.4, 200.0, 8000.0; m, = 1.0, 8.1, 40.0; for i = 1,2,3; and U= x>- ü.
A first-order approximation is used in Equation 18 where the velocity is set equal to the previous
time-step value.
Boundary Conditions
For the pipeline presented in Figure 1, the valve is located at the junction of two series pipes and
connected upstream to a tank which is maintained at a constant pressure head level, Hur. At the
downstream end the system is connected to another constant-head reservoir, Hdr. In the numerical
simulations, only the section from the upstream reservoir to the intermediate valve was modelled.
The section from the valve to the downstream tank includes the flowmeter devices, a long-radius
90° bend and a plexiglass pipe section for visualization. Instead of modelling this section the pres
sure head history downstream of the valve, Hdxi, was recorded and used at the boundary.
At the upstream reservoir, the constant pressure is given by
p = ynur (i9)
where, AH0 = head loss at the valve for the initial steady state, and Q0 = initial steady state dis
Downloaded by [North Carolina State University] at 16:22 30 November 2014
charge.
MgmtÓtd Valve
C3)/®
4
Gait voivt
1 f \££
ü = 0 (21)
Equations 19 and 21 are solved with the two negative characteristic equations to determine Gv and x>
at the tank.
At the downstream end, the valve is fixed to the wall and the orifice equation is valid. Therefore
Equations 20 and 21 are solved together with the two positive characteristic equations to obtain p
and a, at the valve.
Pointer
Downloaded by [North Carolina State University] at 16:22 30 November 2014
Fig. 2. Layout of the Motor-Valve Coupling System (a) Sectional view, (b) Plan view.
The discharge was measured for the initial and final steady state conditions. A thin square-edged
orifice (orifice to pipe diameter ratio = 0.69) and a 60° V-notch were used to measure the dis
charge. The orifice satisfy the standards of the ASME as reported in Bean, (1971). Both instru
ments gave practically the same discharge in the range between 0.0125 m3/s and 0.0015 m-Vs.
The maximum difference in the flow computations was 0.0001 mVs. Piezometric tubes were
installed along the pipe to obtain the pressure gradient during the steady state conditions. The
time history of the valve opening angle was obtained using a linear-potentiometer. The voltage
from the potentiometer was converted to degrees using a protractor capable of reading differ
ences in angles up to 1.25°.
Prior to the transient experiments, a steady state calibration was carefully done to obtain the valve
loss coefficients at different opening angles. The head loss at the valve was measured from piezo
metric tubes located at 4.8 and 3.14 diameters upstream and downstream of the valve. A total of 32
points were obtained for the valve loss coefficients and 56 points for the potenciometer calibration.
Careful attention must be given to this calibration in order to reproduce the valve characteristics
and its motion accurately. Watts et. al , (1980), reported accuracy problems in their experiments
near the valve fully closed position. The valve loss coefficients are presented in Figure 3 (0 vs \lk)
*= 24(Hm-HdJ (22)
v'
where, v is the average velocity in the pipe and Hm and Hdv are the pressure heads upstream and
downstream of the valve respectively. As mentioned before, the accuracy of the calibration is par-
ticulary sensitive near the valve fully closed position. In this region, the loss coefficient, as defined
by Equation 22 tends to infinity.
Strain gage pressure transducers (Sensotec, Model TJE with an accuracy of 0.1%) were located at
10.88 m, 21.32 m (approx. 1/3 and 2/3 of the pipe length upstream of the the valve) and at both
Downloaded by [North Carolina State University] at 16:22 30 November 2014
sides of the valve. Their positions are shown in Figure 1. The transducers for the valve were
located at the same position as the piezometric tubes and are designed to record absolute pressures.
The signal was amplified using a Sensotec SA-BII Multichannel Amplifier and recorded on a PC
computer using a Computerscope (ISC-16) Data Acquisition System. The pressure transducers and
the potentiometer were triggered at the moment when the electric switch for the motor power was
activated. Computer software was developed for the conversion of voltages into water head and for
the generation of the valve closure curve as described in the next section. The use of a partial clo
sure of an intermedite valve to generate a transient condition is a significant difference between this
and previous experiments reported in the literature.
lE+oi^— —I
IE+00
IE-01
7—
1E-02
1E-03
1E-04-I- — i — — i — — i — — i — — i — — i —
0 10 20 30 40 50 60 70 80
Valve Angle, degrees
Fig. 3. Valve loss coefficient vs. valve angle.
versus loss coefficient curve it is possible to obtain the nondimensional valve opening, x, at any
time, t, from the following relation:
1 (24)
=i
where k„ was taken as the loss coefficient for the initial steady state condition and k, is the loss coef
ficient at time t. It can be shown that Equation 23 is the same as Equation 24. Linear interpolation
was used to obtain the angle from the potentiometer data and parabolic interpolation was used to
obtain the loss coefficient from the interpolated angle. Figure 4 shows a typical X vs t curve com
puted following this procedure. Smooth and continuous X vs t curves were obtained for partial clo
sures from fully open position to maximum closure angle of about 72°. It must be stressed that, the
type of valve used has a thick rubber seal around the interior section that actually reduces the flow
to almost zero before the 90° angle is reached. It was observed that, after 83° the flow is com
pletely stopped. At 72° the steady state valve discharge is 0.0017 m3/s i.e. 13.6% of the maximum
possible discharge.
1.0-1 -S^ 1
0.9 \
0.8- \
0.7 \
0.6 \
0.5- \
0.4- \
0.3 \.
0.2 N.
0.1 ^ ^ ^ ^
0.0-I 1 1 1 1 1
0 100 200 300 400 500 600
Time, ms
Fig. 4. Effective valve opening curve.
Region 2 is characterized by a linear relationship between x and t. In Figure 4 this region goes from
X = 0.9 to x = 0.45 in about 100 ms. The value of X reduced more than 50% in this short period but,
interestingly enough, this is associated with a change in the closing angle of only 11° (from 0 = 8°
at x = 0.9 to x = 19° at x =0.45). Figure 8a shows that for regions 1 and 2 (a total of 220 ms) the
Downloaded by [North Carolina State University] at 16:22 30 November 2014
pressure at the upstream side of the valve remained practically constant. This is in agreement with
the small change in the valve position.
The last region, 3, is characterized by an exponential decrease in the x value. The slope of the curve
in this region has a major effect on the pressure rise history at the valve. The valve stops at about
520 ms where x reduces to a constant value of 0.029, (Figure 4). The first peak in pressure at the
upstream side of the valve occurs at this moment, (Figure 8a). Detailed information of the valve
position and the head loss coefficients in region 3 is recommended for accurate simulation of valve
closures in transient flow. Unfortunately, this data is not widely available.
where af is the fluid wave speed and L is the pipe length between the valve and the upstream tank.
For L = 32.0 m the wave speed is 1125 m/s. The test was repeated for three different sampling fre
quencies (200 Hz, 333.33 Hz and 1000 Hz) and the calculated wave speeds ranges from 984 m/s to
1125 m/s. A theoretical value of wave speed for pure water is 1300 m/s. The oxygen concentration
in the tanks was measured as 8.5 mg/1 indicating the presence of dissolved air in the water. This
explains the reduction in the wave speed.
O 100 200 300 400 500 600 700 800 900 1000
Time, msec
Frequency, Hz
Fig. 5. Sudden valve closure data (a) Pressure history, (b) Power spectrum.
Results
The transient conditions generated by the motorized valve in the piping system, shown in Figure 1,
were simulated using the numerical models described in a previous section. The pressure history
downstream of the valve was measured by transducer number 4, (i.e Hdv) with a sampling rate of
2.0 ms, (Figure 6). The minimum pressure head recorded in this test is -6.7 m. The x - 1 curve and
the pressure history downstream of the valve (Figures 4 and 6) are used as an input to the models.
To avoid interpolation in any of the data input files, At was chosen as 2.0 ms and the selected wave
speed in the fluid for computation was adjusted to 900 m/sec. The measured pressure heads at sec
tions 1, 2 and 3, (see Figure 1), are used to compare with the numerical results.
The wave speed in the pipe wall was used as 3 times the wave speed in the fluid to avoid interpola
tion in the x -1 plane. The steady state flow was measured as 0.0125 mVs and the friction factor,
10
5
I
\ i h A/\
Downloaded by [North Carolina State University] at 16:22 30 November 2014
-5
VJ \J V v^
10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.1
Time, sec
Fig. 6. Pressure history downstraeam of the valve.
The results for the two-equations model are presented in Figures 7a, 7b and 7c. Figure 7a presents
measured and simulated results at section 3, (at the valve). The comparison is excellent for the
entire simulation. Only a slight delay in the simulated results is observed in the second and the third
peaks. Figures 7b and 7c, shows the computed and measured pressure heads from transducers
located at 2L/3 and L/3 from the upstream tank respectively. The computed and measured first and
the second peaks compare satisfactorily. The general trends is also identified in the numerical
results, however; the presence of oscillations and reflections is noticed after 0.7 ms. The amplitude
of the computed oscillations is larger than the measured values.
The results for the FSI, with the frequency dependent friction model, are shown in Figures 8a, 8b,
and 8c. Figure 8a presents the measured and simulated results at section 3. Similar to the two-
equations model, the results at the valve show a very good agreement. Figures 8b and 8c show a
better numerical results than the first model and the higher frequency component of the pressure
wave is reduced giving a better overall simulation. There is some delay in the pressure history but
the damping rate is improved. A significant reduction in the first negative pressure peak is
observed in Figure 8b, when compared with Figure 7b.
For the simulations with the FSI model, the Young modulus of elasticity, e, was reduced to account
for the presence of the PVC pipe section located near the valve. There is some evidence that this
small part could act as a damping mechanism, (Ghilardi and Paoletti, 1986). Even though the pipe
was fixed to the wall, a small movement was noticed during the experiment. No measurements of
the pipe velocity or stress were taken, therefore it is impossible to quantify the effects of these
uncertainties.
When the frequency dependent friction effect is added to the FSI model, the differences can only
be noticed at the peak values. Figure 9 presents the results of the FSI without frequency dependent
friction at section 3. In Figure 9 a difference of about 0.5m is noticed between the numerical and
the measured results. This model has a stronger effect for fluids with higher viscosity.
15
•a
ra
a»
I
Downloaded by [North Carolina State University] at 16:22 30 November 2014
-10 1
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.1
Time, sec
15
Downloaded by [North Carolina State University] at 16:22 30 November 2014
-10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Time, sec
Fig. 7. Computed and measured results for sections 3, 2, and 1 using two equations model.
25
Measured Values
Computed Values
-10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
a. Time, sec
Computed Values
20
15
■a
ra
01
X
0)
I/>
V)
0)
Downloaded by [North Carolina State University] at 16:22 30 November 2014
-5
-10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.1
Time, sec
25
Measured Values
Computed Values
-5
-10
0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Time, sec
Fig. 8. Computed and measured results for sections 3, 2 and 1 using FSI with frequecy dependent friction.
Computed Values
•a
ra
a>
I
01
Downloaded by [North Carolina State University] at 16:22 30 November 2014
-10
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 1.1
Time, sec
Fig. 9. Computed and measured results for section 3 using FSI and frequency-dependent friction.
Based on the experimental test and the numerical simulations, the following observation can be
made:
1. Either the two-equations or the FSI formulation can reproduce the pressure history at the valve
accurately, even for the smaller amplitude oscillations. This confirms that the measured pressure
head data and the computed x — t curve can be used to model the boundary condition imposed by
a partial-closure of an intermediate valve.
2. The two-equations formulation is adequate if the purpose of the computations is to determine the
first maximum positive pressure heads. After the second peak, the amplitude of the oscillations
was higher than the measured and low frequency oscillations appeared in the simulations.
3. The FSI formulation gave overall better numerical results. When the interaction between the
pipe and the fluid is modelled through Poisson coupling the damping of the oscillation is
improved considerably. Another improvement can be achieved when the frequency dependent
friction is added to the FSI model.
Acknowledgements
This work was supported in part by the National Science Foundation, through grant No. INT-9107009.
References
BEAN, H. S., 1971, Fluid Meters: Their Theory and Application, 6th ed., ASME publication, New York, NY.
BUDNY, D. D., WlGGERT, D. C, and HATFIELD, F. J., 1990, "Energy Dissipation in the Axially Coupled Model
for Transient Flow", BHRA Pressure Surges-Proceedings of the 6th International Conference, pp. 15-26.
Notations:
A Cross-sectional area, m2
Av Opening area of valve, m2
a Speed of sound, m2
cd Orifice discharge coefficient
c Coupled wave speed, m/sec
D Pipe diameter, m
E Young's modulus of elasticity of pipe wall material, Pa
e Pipe wall' thickness, m
F Natural frequency, Hz
f Darcy-Weisbach friction factor.
G Parameter in the FSI formulation,
H Pressure head, m
AH Head loss at the valve, m
K Bulk modulus of fluid, Pa
k loss coefficient,
L Pipe length, m
P Pressure, Pa
R Pipe inside radius, m
t Time, sec
U Relative fluid velocity, (v - u), m/sec
u Pipe axial velocity, m/sec
V Fluid velocity, m/sec
X Axial coordinate
Subscripts
dr Downstream reservoir
dv Downstream of the valve
Downloaded by [North Carolina State University] at 16:22 30 November 2014
ƒ liquid
5 Steady state conditioner
t Pipe
ur Upstream reservoir
uv Uptream of the valve
u Unsteady-state
x Axial direction
o Initial condition