Arab J Sci Eng (2014) 39:2669–2675

DOI 10.1007/s13369-013-0942-1

RESEARCH ARTICLE - CIVIL ENGINEERING

Experimental and Numerical Simulation of Water Hammer

M. R. Nikpour · A. H. Nazemi · A. Hosseinzadeh Dalir ·
F. Shoja · P. Varjavand

Received: 22 February 2012 / Accepted: 31 December 2012 / Published online: 8 February 2014
© King Fahd University of Petroleum and Minerals 2014

Abstract Water hammer as an important phenomenon in

water distribution networks was studied using experimental
and numerical models in two different setups, with and with-
out the surge pipe. The data collected for different discharges
and arbitrary time intervals were compared with the results
of Computational Fluid Dynamics (CFD). After determin-
ing the appropriate grid and turbulence model, the effect of
sudden stop of water flow on fluid motion and pipeline was
studied by means of FLUENT. The maximum and minimum
values of pressure head on the laboratory model were com-
puted within an accuracy range of 0.12–2.64 %. According to
the qualitative and quantitative comparisons of results, it was
found that CFD could be employed successfully in modeling
of the water hammer phenomenon. 1 Introduction

Keywords Appropriate grid · CFD · FLUENT · Water hammer is one of the destructive hydraulic phenom-
Turbulence model · Water hammer ena that occurs in the form of a rapid pressure wave, which
is propagated in the pipeline and leads to severe damages.
These pressure fluctuations can be severe enough to rupture
the pipeline and the pump. Many researches have numer-
ically studied the hydraulic transient flows such as water
hammer with different methods. Ghidaoui and Kolyshkin [1]
performed linear stability analysis of the base flow velocity
profiles for the laminar and turbulent water hammer flows.
M. R. Nikpour (B) · A. H. Nazemi · A. H. Dalir · F. Shoja · P. Varjavand
They found that the main parameters that govern the stability
Department of Water Engineering, University of Tabriz, Tabriz, Iran of transient flows are Reynolds number and a dimensionless
e-mail: rezanikpoor@yahoo.com time scale. Ghidaoui et al. [2] accomplished two turbulence
A. H. Nazemi water hammer models, namely the two-layer and the five-
e-mail: ahnazemi@tabrizu.ac.ir layer eddy viscosity and evaluated the accuracy of their quasi-
A. H. Dalir steady and axisymmetric assumptions. It is found that the
e-mail: ahdalir@tabrizu.ac.ir results of both models are in good agreement, confirming that
F. Shoja the turbulence modeling of water hammer flows is insensitive
e-mail: shoja_fari@yahoo.com to the magnitude and distribution of the eddy viscosity within
P. Varjavand the pipe core. Izquerio and Iglesias [3] developed a computer
e-mail: pvarjavand@yahoo.com program to simulate hydraulic transients in a simple pipeline

123
2670 Arab J Sci Eng (2014) 39:2669–2675

system by mathematical modeling. Izquerio and Iglesias [4] under adverse and favorable pressure gradients was applied.
presented another model to simulate fluid behavior in com- They reported that the results of the numerical method are
plex systems. Their model produced good numerical results in good agreement with test data obtained from both steady
with desirable accuracy. This model was later generalized to and unsteady turbulent pipe flows. The main objectives of this
include a pumping station fitted with check valve, delivery work were to: (1) solve 2D water hammer equations using
valve and two air vessels. Filion and Karney [5] presented the fourth-order Runge–Kutta scheme; (2) examine the capa-
a numerical method to analyze transient flows occurrence in bility of the k-ω turbulence model to predict unsteady flow
pipe networks. Their method leads to a significant increase due to water hammer; and (3) analyze turbulent flow struc-
of accuracy in simulation, but requires more system infor- ture arising from water hammer problems. They observed a
mation and computational effort. Ghidaoui [6] formulated slight phase shift which is produced after several wave cycles
first- and second-order explicit finite volume (FV) methods of between the pressure values of experimental and numeri-
Godunov-type for water hammer problems. They compared cal models and they stated that this behavior is probably
the performances of FV scheme and method of characteris- arisen from an error in the wave speed estimation. Keramat et
tic (MOC). It was found that the first-order FV Godunov- al. [13] studied fluid–structure interaction (FSI) due to water
scheme produces the same results with MOC. It was also hammer in a pipeline which had visco-elastic wall behav-
shown that the second-order Godunov scheme is simple to ior. The governing equations were solved by two different
implement and stable for a Courant number less than or equal approaches, namely the method of characteristics-finite ele-
to one. Wood [7] and Wood et al. [8] investigated capability ment method (MOC-FEM) and full MOC. Both numerical
of MOC and wave characteristics method (WCM) to analyze methods gave the same results, thus confirming the correct-
the transient flow in water distribution networks including of ness of the solutions.
pipe friction. They showed that for the same modeling accu- The recent developments in computer sciences have pro-
racy, the WCM requires less execution time. In addition, they moted the use of computational fluid dynamics (CFD) for
showed that the number of calculations per time step required modeling the fluid flow [14]. So in this study, the water ham-
by WCM does not increase when more accuracy is required mer phenomenon was simulated as a case of sudden closure
while for the MOC, the number of calculations per time step using FLUENT, then the obtained results were compared
is roughly proportional to the accuracy. It was found that use with those of experimental models.
of the WCM would be more suitable for analyzing large pipe
networks. Saikia and Sarma [9] presented a numerical model
using MOC and Barr’s explicit friction factor for simulation 2 Experiments
of the water hammer problems. The proposed model was
examined for rapid valve closure in downstream of a long 2.1 Experimental Model
conduit with a reservoir upstream. Stability and accuracy of
the method was compared with those of the explicit finite Figure 1 shows the experimental apparatus used in this study.
difference method. Kim [10] used impulse response method This model had two similar tanks with dimensions of 90 ×
(IRM) to analyze pipeline systems equipped with protection 90 × 30 cm. The upper tank was connected to a recharge
devices, such as surge tank and air chambers. Transient analy- pipe with diameter of 20 mm. A pump with discharge of
sis revealed that the IRM results were in good agreement with
those of simulations of MOC in the presence of the hydraulic
device. Kwon and Lee [11] studied transient flows in a pipe
using both experimental and computer models. Three differ-
ent numerical models: the method of characteristics model,
the axisymmetrical model, and the implicit scheme model
are utilized and compared. Experiments for transient flow in
a simple pipeline have been conducted to verify the results The second valve
The first valve
from the computer models. It was found that head loss coef-
ficient for the 1D models, such as the method of charac-
teristics model and the implicit scheme model, should be
much bigger than the Darcy–Weisbach frictional coefficient.
They reported that the results of different computer models
have good agreement with the experimental data. Raisi et
al. [12] have studied the behavior of unsteady turbulent pipe
flow resulting from water hammer numerically. An accurate
k-ω turbulence model for two-dimensional boundary layers Fig. 1 Experimental apparatus

123
Arab J Sci Eng (2014) 39:2669–2675 2671

the second valve and fixing the water level in the lower tank.
Similar to the first model after reaching the steady flow, water
hammer was produced by sudden closing of the first valve
and values of pressure head were recorded in arbitrary time
intervals.

3 Numerical Model

FLUENT is one of the most popular and suitable software of

CFD that provides a wide range of advanced physical models
for fluid flow and heat transfer including multiphase flow. It
can exchange 2D and 3D dominant differential equations to
Fig. 2 Variations of computation error versus the number of meshes
in the first and second model Algebraic equations by using the finite volume method. Cur-
rently, thousands of companies throughout the world benefit
from the use of FLUENT as an integral part of the design
15–70 l/min supplied tanks with water and a rotameter with and optimization of their products.
5–50 l/min range was used for discharge measurement. The The hydraulic effects of sudden closure of the valve were
lower tank was recharged by the upper tank. A PVC pipeline studied after determining the optimum grid and turbulence
was constructed to measure pressure waves associated with model. In order to simulate turbulent flow, 2D RSM and k-
sudden valve closure. The diameter and total length of ε models were used and volume of fluid (VOF) model was
pipeline were 0.04 and 4 m, respectively. In order to fit the used to simulate two-phase flow of water and air. It should
total pipe length in the available laboratory space, six 90◦ be mentioned that the k-ε model has three types called Stan-
elbows were used as shown in Fig. 2a. There were two valves dard, RNG, and Realizable. The Standard model has two
at the end of the pipeline. The valve close to the surge pipe independent equations based on k and ε, where k shows tur-
was used for sudden closure (the first valve in Fig. 1) and the bulent kinetic energy and ε is its dissipation rate. The RNG
second with larger closure range (the second valve in Fig. 1), model has an additional term in its ε equation that signifi-
was used as an outlet valve. In order to achieve a steady- cantly improves the accuracy for rapidly strained flows; fur-
state condition, the water head in the lower tank was kept thermore, the effect of swirl on turbulence is included in
equal to the water head in the surge pipe. At a distance of the RNG model, enhancing accuracy for swirling flows. The
0.23 m upstream the first valve, a plastic surge pipe of 2.38 m Realizable model is a relatively recent development and dif-
long and 0.063 m inner diameter was used to dissipate the fers from the Standard model in two important ways: (1)
pressure waves caused by the water hammer. At this point, a the Realizable model contains a new formulation for the tur-
transducer was installed to record changes of pressure head. bulent viscosity; (2) a new transport equation for the ε has
Two sets of experiments were carried out with flow rate of been derived from an exact equation for the transport of the
Q = 20–50 l/min. mean-square vorticity fluctuation [15].

2.1.1 First Model: With Surge Pipe

3.1 Numerical Model Calibration
By using rotameter and the second valve, we reached steady-
state condition in the experimental model so water surface in To check for grid convergence, the grid numbers were suc-
the surge pipe and lower tank were fixed in the same level. cessively increased in four different ranges for constant dis-
The hydraulic transient flow was generated by sudden closure charge and turbulence model. The lowest required meshes
of the first valve. Depending on the generated pressure, the which had the least computational error for the first and
surge pipe acted as an energy dissipation system. The process second models were determined 12,000 and 9,810 meshes,
of energy dissipation continued until the water surface in the respectively, as shown in Fig. 3
surge pipe reached the tranquil condition. In this study, unsteady flow simulation firstly started with
time steps equal to 0.001 s and after the stability of solution
2.1.2 Second Model: Without Surge Pipe procedure, time steps gradually increased up to 0.01 s. The
k-ε and RSM models were utilized to analyze turbulent flow
The goal of the second series of experiments without surge and the suitable model was determined on the basis of higher
pipe was to find the real values of head pressure generated simulation accuracy and lesser time-consuming calculation
in water hammer. The steady flow was created by adjusting under the same simulation conditions.

123
2672 Arab J Sci Eng (2014) 39:2669–2675

Fig. 3 a Scheme of the

numerical model. b The definite
boundary conditions in the
numerical model

(a)

(b)

3.2 Boundary Conditions the surge pipe were taken solid, so wall boundary condition
was defined to them. A code was utilized as a user-defined
Figure 2a, b shows the scheme of the numerical model and function (UDF) in FLUENT. According to this UDF, the val-
the definite boundary conditions used in the numerical model, ues of velocity at the boundaries 2 and 3 (inlet and outlet)
respectively. As shown in Fig. 2a, b 2D grid was used for mesh become zero after the system has stabilized. In fact, when
generation which consisted of Tri (pave) elements for elbows the velocity equals zero, it is similar to the condition when
and Quad (map) elements for other parts of the structure. the pump is turned off or the valve is suddenly closed in the
The upper tank played no role in calculation, so it was elim- experimental model. In order to determine the time needed
inated in the numerical model. Since the boundaries 1 and to reach the steady flow before the occurrence of water ham-
4 are exposed to the air, pressure outlet boundary condition mer, the models were primary run without water hammer
with atmospheric value was assigned to them. In addition, for occurrence and the time needed was extracted. FLUENT has
boundaries 2 and 3, which act as an input and output of the time-record ability for recording hydraulic parameters such
model, the velocity inlet and velocity outlet were considered as pressure head in unsteady flows. This parameter has been
as a boundary condition, respectively. Walls of the pipe and recorded in experimental model by transducer. The consid-

123
Arab J Sci Eng (2014) 39:2669–2675 2673

Numerical Experimental
ered time step for recording the pressure head values in the
280
numerical model and in the experimental setup was the same
and equal to 0.1 s.

Pressure Head (cm)

180

4 Results and Discussion

4.1 Qualitative Comparison of Results 80

This study is divided into two parts: experimental and com-

putational. In the experimental part, flow discharge was in -20 0 10 20 30 40
the range of 20–50 l/min and totally 31 tests were carried Time (s)
out in the first and second models. All of the experiments
Fig. 5 Variations of pressure head for the second model (Q = 35 l/min)
were simulated by numerical model. Figures 4 and 5 show
the variations of pressure head at the upstream of the first the real values of pressure head produced by water hammer
valve in the experimental and the numerical models for the are reported, so the needed time for energy dissipation is
first and second models at a discharge equal to 35 l/min. lower. This behavior is due to lack of the surge pipe in the
These figures show that as soon as the system approached second model, so its pressure wave is stronger than the first
the steady-flow condition, outlet flow is stopped by sudden model and there is high momentum exchange between fluid
closing of the first valve. A strong positive pressure wave elements or between fluid and solid boundary which is caused
is generated and propagated along the pipe toward the tank more energy dissipation and therefore the needed time for
and then reflected. In other words, another wave is produced wave dissipation for second model is lesser than the first
which returns along the pipe to the starting point of the first model while wave intensity for the second model is higher.
wave. Approaching this wave to the valve, a negative wave Due to two phase flow in the surge pipe, simulation accu-
is propagated from the valve to the tank. This process takes racy of the first model (Fig. 4) is less than the second model
place successively along the pipe and energy is gradually dis- (Fig. 5). In this model, pressure wave speed is different from
sipated so waves intensity attenuated and faded. The role of its theoretical value and that is because of presence the surge
the surge pipe is to discharge water from the pipe at the time pipe which causes that the numerical simulation did not accu-
of positive pressure wave movement and injecting water to rately reproduce the timing of the peaks in the pressure wave.
it at the time of negative pressure wave presence. Therefore, In the beginning of water hammer occurrence, difference
the surge pipe decreases the variations of pressure and the between the numerical and experimental values of Hmax and
intensity of produced waves. Hmin is remarkable but reduced by dissipating of the pres-
Figures 4 and 5 also show the time needed to dissipate sure wave. It is probably due to high speed of water surface
excess energy of the waves caused by water hammer in the fluctuations in the surge pipe in the beginning of water ham-
first and second models. As it can be seen, in the second mer occurrence (short wave period and high amplitude in
model the maximum and minimum values of pressure head the Fig. 4) in comparison with their fluctuations speed after
(Hmax and Hmin ) are so higher than the first model but the dissipation of pressure wave. When the fluctuations speed of
time needed to dissipate the energy is lower. In the second water surface is high, Courant number is high too, so the accu-
model there is no facility for dissipating the energy, as a result racy of simulation is decreased. In the second model (Fig. 5)
during dissipation progress, a phase shift is produced after
some wave cycles probably arising from an error in the wave
Numerical Experimental
speed estimation, as Raisi et al. [12] have mentioned to this
130 behavior.
Pressure Head (cm)

120 4.2 Quantitative Comparison of Results

110
Figure 6 shows the values of the mean relative error (MRE)
for different turbulence models using the surge pipe by the
following formula:
100
0 25 50 75 100
Time (s)  N  Hn −Ho 
i=1  Ho 
Fig. 4 Variations of pressure head for the first model (Q = 35 l/min) MRE(%) = × 100 (1)
N

123
2674 Arab J Sci Eng (2014) 39:2669–2675

1 0.75
0.68
0.63
0.59
MRE (%)

0.5

0
1
Turbulence Model
Standard RNG Realizable RSM

Fig. 6 Values of MRE for different turbulence models using surge pipe
Fig. 8 Values of Hmax and Hmin for the first model
0.98
0.93

0.78
1 0.75
MRE (%)

0.5

0
1
Turbulence Model
Standard RNG Realizable RSM

Fig. 7 Values of MRE for different turbulence models without surge

pipe

In which Hn and Ho are the numerical and observed values

of head pressure, respectively, and N is the total number of
data.
As shown in Fig. 6 the RSM and the standard k-ε models
Fig. 9 Values of Hmax and Hmin for the second model
with 0.59 and 0.75 % errors have the minimum and maximum
values of MRE, respectively. It is seen that the difference
between the MRE values in the Realizable k-ε and RSM where Hmn and Hmo denote the numerical and the observed
models is negligible, but the running time of the latter is much values of Hmax or Hmin , respectively. The results of the numer-
longer. Thus, the Realizable k-ε model was chosen as an ical model are in the acceptable accuracy range of 0.12–
optimum model in the present study. Figure 7 demonstrates 2.64 %. Also the MRE (%) of Hmax and Hmin computation for
values of the MRE for different turbulence models in the the first model were 0.66 and 0.83 %, respectively; whereas,
second model. Again, the Realizable k-ε model was chosen they were 0.93 and 1.04 % for the second model. It can be
as a suitable model. concluded that the numerical model in simulation of the first
Since the most important parameters in the water ham- model has yielded slightly better than the latter.
mer phenomenon are the Hmax and Hmin , they were mea-
sured and computed after implementing the experimental and
the numerical models. These values were plotted in Figs. 8 5 Conclusion
and 9. Values of Hmax and Hmin obtained of the numerical
model were compared with those of the experimental model Because of the importance of water hammer in water con-
by calculation of relative error index (RE) using the following veyance systems and its destructive effects, in this study this
formula: phenomenon was investigated by means of experimental and
 
 Hmn − Hmo  numerical models and also considering two different mod-
RE(%) =    × 100, (2)
H  els, with and without surge pipe. FLUENT as a reliable and
mo

123
Arab J Sci Eng (2014) 39:2669–2675 2675

acceptable CFD software as employed for numerical sim- 6. Zhao, M.; Ghidaoui, M.: Godunov-type solutions for water hammer
ulation. It should be mentioned that after applying the tur- flows. J. Hydraul. Eng. ASCE 130(4), 341–348 (2004)
7. Wood, D.J.: Water hammer analysis—essential and easy (and effi-
bulence models of k-ε and RSM, the k-ε Realizable model cient). J. Environ. Eng. ASCE. 131(8), 1123–1131 (2005)
was found as the best one. Finally, the qualitative and quan- 8. Wood, D.J.; Lindireddy, S.; Boulos, P.F.; Karney, B.; Mcpherson,
titative comparisons of the results showed a good agreement D.L.: Numerical methods for modeling transient flow in distribu-
between experimental and numerical models that reveals the tion systems. J. Am. Water Works Assoc. 97(7), 104–115 (2005)
9. Saikia, M.D.; Sarma, A.K.: Numerical modeling of water hammer
high capability of the numerical model for simulating the with variable friction factor. J. Eng. Appl. Sci. 1(4), 35–40 (2006)
water hammer phenomenon. 10. Kim, S.: Impulse response method for pipeline systems equipped
with water hammer protection devices. J. Hydraul. Eng. ASCE.
134(7), 961–969 (2008)
11. Kown, H.J.; Lee, J.J.: Computer and experimental models of tran-
References
sient flow in a pipe involving backflow preventers. J. Hydraul. Eng.
ASCE 134(4), 426–434 (2008)
1. Ghidaoui, M.S.; Kolyshkin, A.A.: Stability analysis of velocity 12. Riasi, A.; Nourbakhsh, A.; Raisee, M.: Unsteady turbulent pipe
profiles in water hammer flows. J. Hydraul. Eng. ASCE. 127(6) flow due to water hammer using k-θ turbulence model. J. Hydraul.
499–512 (2001) Res. IAHR. 46(4), 429–437 (2009)
2. Ghidaoui, M.S.; Mansour, G.S.; Zhao, M.: Applicability of quasi 13. Keramat, A.; Tijsseling, A.S.; Hou, Q.; Ahmadi, A.: Fluid–
steady and axisymmetric turbulence models in water hammer. J. structure interaction with pipe-wall viscoelasticity during water
Hydraul. Eng. ASCE. 128(10), 917–924 (2002) hammer. J. Fluids Struct. Elsevier. 28, 434–455 (2012)
3. Izquierdo, J.; Iglesias, P.L.: Mathematical modeling of hydraulic 14. Dargahi, B.: Experimental study and 3D numerical simulations for
transients in simple systems. Math. Comput. Model. Elsevier 35(7), a free over flow spillway. J. Hydraul. Eng. ASCE. 132(9), 899–907
801–812 (2002) (2006)
4. Izquierdo, J.; Iglesias, P.L.: Mathematical modeling of hydraulic 15. Malalasekera, W.M.G.; Versteeg, R.: An introduction to compu-
transients in complex systems. Math. Comput. Model. ELSEVIER tational fluid dynamics. Addison Wesley Longman limited, UK
39(4), 529–540 (2004) (1955)
5. Filion, Y.R.; Karney, B.W.: Sources of error in network modeling: a
question of perspective. J. Am. WaterWorks Assoc. 95(2), 119–130
(2003)

123