You are on page 1of 11

Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2, pp.

319329 (2014)
319
PARTIALLY-AVERAGED NAVIER-STOKES MODEL FOR PREDICTING
CAVITATING FLOW IN CENTRIFUGAL PUMP

Houlin Liu, Jian Wang*, Yong Wang, Haoqin Huang and Linglin Jiang

Research Center of Fluid Machinery Engineering and Technology, Jiangsu University,
Zhenjiang 202013, China
*E-Mail: kin.jian.wang@gmail.com (Corresponding Author)


ABSTRACT: Cavitation is a common phenomenon in pump industries, which leads to severe problems, like
vibration and noise. It may degrade the pump performance and even damage the solid surface. So it is significant to
give a precise prediction of the pump cavitation performance. The original k- model is widely used in the past years.
However, it is reported that high viscosity of the original k- model dampens cavitation instabilities and hence makes
it difficult to capture the detachment of the bubbles. Aiming at improving the predictive capability, the partially-
averaged Navier-Stokes (PANS) is employed in this paper to predict the pump cavitation performance. Experiments
on a centrifulgal pump with twisted blades are carried out to validate the simulations. The results show that,
compared with the original k- model, the PANS model with lower f
k
value gives a more accurate prediction and can
reduce the eddy viscosity in the cavity region, leading to capturing the unsteady bubble shedding phenomenon. The
experimental visualizations are performed and the evolution of the cavitation inception and development are
obtained exactly at the impeller inlet. Comparisons with the transient numerical simulations are made, which
demonstrates the PANS model can successfully capture the cavitation detachment. Finally, the blade load pressure,
the pressure distribution in impeller and the pressure fluctuations are analyzed. Good agreement is noticed between
simulations and experiment. So it can be concluded that the PANS model can effectively reduce eddy viscosity in
cavitating flow in centrifugal pumps and improve the numerical simulation prediction of pump cavitation
performance.

Keywords: partially-averaged Navier-Stokes model, centrifugal pump, cavitating flow, visualizations
1. INTRODUCTION

Cavitation is well recognized as a phenomenon
that may cause serious pump malfunctioning, due
to improper pump inlet conditions or increasing
rpm (Revolution Per Minute) (Liu et al., 2012),
such as vibration (Benaouicha and Astolfi, 2010),
noise (Cudina, 2006) and even damage the solid
surface (Bruno and Frank, 2009). For pump
industries, it is much desired to predict pump
cavitation performance accurately in a
preliminary study, and also make it clear how the
bubbles develop and collapse in a pump.
Benefited from the contributions of rapid
development of the computational fluid dynamics
(CFD) technology, researchers could obtain a
deep understanding of cavitating flow field (Bilus
and Andrej, 2009; Chang and Wang, 2012).
Therefore, it is very appropriate to use CFD
approach to analyze the pump cavitation
performance. However, the accuracy of numerical
simulations is strongly dependent on users
experience and numerical models (Morgut and
Nobile, 2011; Morgut et al., 2011).
Because cavitation is an unsteady, multiphase
flow, it makes the simulation much more difficult
to capture the transient process, like the cavitation
inception and shedding off, and the effective
liquid viscosities are important at high Reynolds
numbers, especially in pumps. To get precise
computational results, an appropriate turbulence
model is required. In the last decades,
considerable effort has been devoted to the
Reynolds-averaged Navier-Stokes equations-
based turbulent models. Due to their robustness
and reasonable accuracy, such as the two-equation
models. The original k- model was proposed by
Harlow and Nakayama (1967) and then refined by
Launder and Spalding (1974). However, the k-
model noticeably over-predicts turbulent
production and hence the effective viscosity in
stagnation flow regions. It fails to get the
unsteady properties between quasi-periodic
larger-scale and turbulent chaotic small-scale
feature of the flow field (Wang et al., 2011).
Furthermore, it has come short when coping with
flows with large streamline curvatures and time
dependent characteristics, such as cavitating flows
in pumps (Bilus, et al., 2005). Raiesi et al. (2011)
evaluated some turbulence models by using direct
numerical simulations (DNS) and large-eddy
simulation (LES). It indicated that k- model was

Received: 14 Jun. 2013; Revised: 15 Jan. 2014; Accepted: 24 Feb. 2014
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
320
incapable of predicting correctly the
nonequilibrium separated flow and failed to
model the perturbation, which would be important
for capturing the unsteady process of cavitating
flows. Ding et al. (2011) applied the transient
approach to simulate the cavitating flow in an
axial-flow pump and the k- model was used to
close the equations. The cavitation inception and
development were obtained. However, the
cavitation detachment process was hardly
captured.
Attempts have been made to obtain the highest
accuracy results by employing the DNS
(Sandham et al., 2001). Unfortunately, although
DNS gives the most precise results, the
computational time is prohibitive for practical
applications. The prevailing route to simulate
time dependent flows is the LES method
(Nobuhiro et al., 2003; Kato et al., 2003; Ji et al.,
2013) proposed by Smagorinski (1963).
Typically, LES resolves all the dynamically
important scales of motion and a significant
portion of the inertial scales. However, it also
requires much greater computational effort and
longer simulation times, which is too expensive
for engineering purposes.
Recently, several approaches have been made to
blend RANS and LES models (Batten et al., 2002;
Senocak and Shyy, 2004a and 2004b). Grimaji
presented a partially-averaged Navier-Stokes
method, which is a suite of turbulence closure
models of various filter widths ranging from
RANS to DNS (Girimaji et al., 2006; Girimaji
and Suman (2011); Basara, et al., 2010). The
control filters of the PANS model are resolved-to-
unresolved kinetic energy f
k
and resolved-to-
unresolved dissipation f

. By specifying these two


parameters, the model provides a smooth
transition from RANS to DNS.
For getting a good understanding of the evolution
of cavitating flows in pumps, considerable
research has been performed via experimental
visualization method (Rafael et al., 2011; Erfan et
al., 2010). Duplaa et al., (2010) carried out an
experimental study of cavitating flow in a
centrifugal pump. Visualizations were performed
with a high speed camera to provide a rough
estimation of the cavitation development, but the
visual angel was on the side of the inlet tube, not
perpendicular to the pump inlet, and as a result,
the detailed cavitating flow evolution might not
be observed clearly.
This paper aims at evaluating the time dependent
PANS model used to predict the cavitating flow
in a centrifugal pump and obtain a better
visualization of the development of cavitating
flow at the impeller inlet. The results are
compared with experiment data. The performance
of the PANS model is discussed and the vapor
distribution with different filter widths is
analyzed. Also, the pressure distributions on the
blade surface, the mid-plane cross the impeller
and the pressure fluctuations were discussed
between non-cavitation condition and cavitation
condition.



Fig. 1 Sketch of closed test rig.



Fig. 2 Pump model.



Fig. 3 Water tank.
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
321

Fig. 4 Layout of experimental devices.
2. EXPERIMENTAL SETUP AND PUMP
MODEL

The experiment was performed on a closed circuit
test rig in the Research Center of Fluid Machinery
Engineering and Technology of Jiangsu
University. Fig. 1 presents the sketch of the
closed circuit test rig. The centrifugal pump
model tested in experiment is manufactured by
plexiglas for visualization, which has five twisted
blades, as shown in Fig. 2. The impeller of the
pump has a diameter of D
2
=169mm and an outlet
width of b
2
=10mm. A vane guide is mounted
between the impeller and volute. The pump
operates at 1450 rpm with a design flow rate and
head of Q=32.8m
3
/h and H=5.8m. The images of
the cavitation growth process are captured by a
high-speed camera Y-series 4L, with a spatial
resolution of 10241024 pixels. In this work, the
shooting rate is set to 3000 frames/s, which means
the impeller rotates about 3 between two frames.
Additionally, to get a better view of the
development of bubbles on the leading edge of
the twisted blades, a water tank, also made by
plexiglas, is attached to the pump entrance (Fig.
3) and the high-speed camera is placed just
against it. The detailed layout of the experimental
devices is presented in Fig. 4. A LED lamp and
two halogen lamps are used to illuminate the test
pump. The pressure fluctuation data of the pump
inlet and outlet are measured by pressure
transmitters and collected by a pump tester.

3. NUMERICAL SIMULATION

3.1 Governing equations

The set of governing equation comprises the mass
continuity Eq (1) and momentum equations Eq
(2), and the flow with possible coexistence of
liquid and vapor is treated as a homogeneous
mixture.
( ) 0
m
m i
i
u
t x


c c
+ =
c c
(1)


The mixture density and mixture viscosity are
defined by the vapor volume fraction, expressed
as:
(1 )
m v v v
l
o o = + (3)
(1 )
m v v v
l
o o = + (4)
where p is the pressure, is mixture density, u is
the velocity, and
t
stand for the laminar
viscosity and turbulent viscosity respectively, and

v
is the volume fraction. The subscripts m,l,v
indicate the mixture, liquid and vapor,
respectively.

3.2 PANS turbulence model

The PANS turbulence model is first derived by
Girimaji et al. (2006) based on the original k-
model, aiming at resolving different cases
depending on the flow geometry and physical
effects. In Girimaji et al. (2006), this model can
be changed smoothly and seamlessly from RANS
to DNS with various filter widths, which is
accomplished by correcting the model
coefficients of the original k- model. The two
filter widths are resolved-to-unresolved kinetic
energy f
k
and resolved-to-unresolved dissipation
f

, defined as:
u
k
k
f
k
=
(5)
u
f
c
c
c
=
(6)
where k and represent the total turbulent kinetic
energy and dissipation rate respectively, and the
subscript u stands for unresolved scales. The
parameter f
k
controls the cut-off ratio between
resolved and unresolved scales. That is to say, the
smaller is the f
k
, the greater is the physical
resolution: f
k
=1 represents RANS and f
k
=0
indicates DNS. The parameter f

determines the
unresolved flow Reynolds number. In the case of
high Reynolds number flow, the f

can be set to 1.
In the opposite case, for low Reynolds number
flow, f

= f
k
. In this work, the aim is to figure out
the influence of f
k
on unsteady cavitating flows in
centrifugal pumps, so the f

is set to be 1, which
implies the unresolved dissipation scales of PANS
and RANS are identical. Then the PANS model
could be summarized as:
m
u j
u u u
u u
j j j ku
k u
k k
P
t x x x

c
o
(
| |
(
|
( |
\ . (

c
c c c
+ = + +
c c c c
(7)
2
*
1 2
m
u j
u u u u u
u
u u u j j j
u
C P C
t x x x k k
c c
c
c
c c c c

o
(
| |
(
|
|
(
\ .

c
c c c
+ = + +
c c c c

(2)
(8)
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
322
where the modified coefficients are as follows:
2
k
ku k
f
f
c
o o = (9)
2
k
u
f
f
c c
c
o o = (10)
*
2 1 2 1
( )
k
f
C C C C
f
c c c c
c
= + (11)
The details of the derivation of this model, based
on the original k- model, can be found in
Girimaji et al. (2006) and Girimaji and Suman
(2011). The differences are the modified
coefficients
ku
,
u
and
*
2
C
c
, while the other
coefficients are identical to those of the original k-
model:
1
1.44 C
c
= ,
2
1.92 C
c
= , 1.0
k
o = and
1.3
c
o = .

3.3 Cavitation model

Over the years, the transport equation model
(TEM) derived from the homogeneous
equilibrium model (HEM) has become a very
popular approach to deal with cavitating flows
(Zwart et al., 2004; Kunz et al., 2000; Singhal et
al., 2002). Different modeling concepts
embodying different source terms m
+
and m

,
which indicate the condensation and evaporation
rates. In this work, the Zwart et al. (2004) model,
implemented into the CFX software, is employed,
which has been validated by many researchers,
for the reason that it has a precise cavitating
prediction performance and a good convergence
behavior. It can be described as:
( ) v
j
v
j
u
m m
t x
o
o
+
c
c
+ = +
c c
(12)
3 (1 ) 2
3
nuc v v v
vap
B l
r P P
F
R
m
o

+

= , if P<P
v
(13)
3 2
3
v v v
cond
B l
P P
F
R
m
o


= , if P>P
v
(14)
P
v
=P
sat
+P
turb
/2 (15)
P
turb
=0.39k (16)
where F
vap
and F
cond
are empirical calibration
coefficients of evaporation and condensation,
respectively. And r
nuc
is the nucleation site
volume fraction, R
B
stands for the bubble radius,
P
v
represents the water vaporization pressure, P
sat

is the vapor saturation pressure and P
turb
is defined
as the turbulent pressure fluctuations. In this
study, these coefficients are set as defaults, as
recommended by Zwart et al. (2004) and Bilus
and Andrej (2009): F
vap
=50, F
cond
=0.01,
r
nuc
=510
-4
, R
B
=110
-6
m.


Fig. 5 Pump computational grids.

Fig. 6 Yplus on blade surface along streamwise
coordinate at span=0.5.

Fig. 7 Streamwise coordinate at span=0.5.



Fig. 8 Pump cavitation performance curve.
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
323
3.4 Grids and simulation method

The structured hexahedral grids are used in the
present study. The fluid domain of the pump is
shown in Fig. 5, generated by GridPro
commercial software. The grids near the blade
surface region layer are refined, which is locally
zoomed in Fig. 5. And the Y
plus
, on the blade
surface along the streamwise coordinate at
span=0.5, is plotted in Fig. 6. The streamwise
coordinate follows the blade surface and it ranges
from 0 at the leading edge to 1 at the trailing edge
of the blade. And the span represents the
dimensionless distance (between 0 and 1) from
the hub to the shroud (Fig. 7). As shown, the Y
plus

value ranges from 11 to 11.5. To get a relatively
stable inlet and outlet flow, two prolongations,
four times of the pipe diameter, are assembled on
the impeller and volute. A grid independence test
is conducted based on the pump head H under
non-cavitation condition. It is found that when the
cell number is over five millions, the discrepancy
of the pump head is within 3%. Ultimately,
considering the simulation time and accuracy, the
total cell number of all the parts is 6.510
6
.
A multiple reference frame (MRF) approach is
adopted, where the impeller is put into a rotating
reference frame and the other domains use the
translating reference frame. The boundary
conditions of pressure, P
inlet
=1atm, and the mass
flow rate, Q=32.8 m
3
/h, are imposed at the inlet
and outlet, respectively. No slip boundary
condition is imposed on the solid surface of the
pump. The simulation is first conducted under
non-cavitation situation to obtain the pump
performance and steady result, which will be used
as an initial flow filed to predict the cavitating
flow. Then, the pressure loaded on the inlet is
gradually reduced when the calculation is
converged at a given pressure value. In the
meantime, the transient simulation is also
executed to compare with the experimental
visualizations. The total time is set to 10T, where
T denotes the cycle time of the centrifugal pump.
And the step time T is set to T/120, which
implies that for one period time, the calculation
will be conducted at every 3.

4. RESULTS AND DISCUSSIONS

For the convenience of comparing the results, a
couple of dimensionless parameters are defined
as:
Pump head coefficient
( )
2
2
2 H u g =

(17)
Cavitation number
( )
2
2
0.5
v in l
P P u o =

(18)
Pressure coefficient
2
2
( ) / 0.5
pt in l
C P P u = (19)

where u
2
is the circumferential velocity of
impeller outlet and P
in
represents the static
pressure of the inlet.

4.1 Pump cavitation performance

Fig. 8 plots the comparison of cavitation
performance between experiment and numerical
simulation results, which were calculated by the
original k- model and the PANS model with four
different filters f
k
= 0.9, 0.7, 0.4 and 0.2. It is
clearly indicated that, with the decreasing f
k
value,
the pump head, obtained by the PANS model, is
getting closer to the experimental data. For higher
f
k
value, such as 0.9, the performance of the
PANS model seems similar to that of the original
k- model.
At the cavitation number =1.2, when there is no
vapor generating in the pump, the pump head
coefficient is about 0.84 and 0.83 according to
the original k- model and the PANS model with
f
k
=0.9, respectively. As for the PANS model
where f
k
drops to 0.2, the value is 0.76,
compared with 0.71 tested by experiment.
With the decreasing pressure, we can find that the
resemblance of cavitation inception, for all the
simulation results, approximately occurs at
=0.45. But for the original k- model and the
PANS model with higher f
k
, the declining rate of
the pump head is smaller than those with lower f
k
.
To have a quantified interpretation, a critical
cavitation number
c
, which is defined as the
value when the pump head drops by 3%, are
summarized in Table 1. As seen, the PANS model
with f
k
=0.2 shows better prediction results:

c
=0.36 compared with 0.39 from experiment.
Whereas, the value obtained by the original k-
model is 0.26.

Table 1 Comparison of critical cavitation number
c

between original k-, PANS and experiment.


Original k-

PANS

Exp.
fk=0.9 fk=0.7 fk=0.4 fk=0.2
0.26 0.29 0.30 0.34 0.36 0.39

4.2 Cavity distribution

Fig. 9 shows the comparison of vapor volume
fraction distribution between experimental
visualizations and simulations at the inlet region
of the impeller. For the computed results, the
isosurfaces of 10% vapor volume fraction are
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
324
shown based on previous experience, which
indicates that it relates best to the real cavity
shapes (Okita and Kajishima, 2002). Four
different cavitation numbers are selected ranged
from 0.32 to 0.45 to present the cavity growth.
Apparently, the numerical simulations show a
good agreement with experiment, especially the
PANS model with the filter width of f
k
=0.2. The
image shows that the cavitation inception
approximately emerges at the leading edge of the
blade at = 0.45, demonstrating the conclusion
arrived before in Fig. 8. As for the pressure
dropping, compared to the original k- model, the
PANS model with lower f
k
value captures more
detached bubbles at the rear of the cavity. The
reason is that more over-predicted eddy viscosity
is filtered out under higher filter f
k
, which is the
primary factor affecting cavity detachment. For
f
k
=0.9, the results are similar to those of the
original k- model. It is important to note that the
asymmetrical cavity distribution is mainly caused
by the interaction between the impeller and vane


=0.45


=0.41


=0.37


=0.32

(a) Exp. (b) original k- (c) f
k
=0.9 (d) f
k
=0.7 (e) f
k
=0.4 (f) f
k
=0.2

Fig. 9 Comparison of cavity distribution at impeller inlet between experimental visualizations and simulations.


Fig. 10 Comparison of unsteady cavity behavior in experimental visualizations and simulations (PANS with f
k
=0.2) at
=0.41.
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2, pp. 319329 (2014)
325
guide, which results in asymmetrical pressure
distribution on the blade surface.
According to the experimental visualizations, the
cavitation grows with the cavitation number
decreasing and the detached cavity is well
captured at the rear of cavity region, as can be
seen when drops to 0.41. Within the PANS
model, for a lower ratio of kinetic energy, such as
f
k
=0.2, the detached cavity can be clearly
observed.
To analyse the evolution of the cavitating flow in
a pump under a constant cavitation number, the
unsteady cavity behavior in experiments and
simulations (PANS with f
k
=0.2) are compared in
Fig. 10 at =0.41. As shown in the visualizations,
the attached cavity can be observed on the leading
edge of blades, or more precisely, it is attached in
the region conjoining the suction side of the
blades with hub and shroud. The bubbles travel
from the hub to the shroud along the blade
surface, which forms a triangular cavity shape.
And some small bubbles are shed from the
attached cavity with the impeller rotating.
Moreover, because of the interaction between the
impeller and vane guide mentioned above,
asymmetrical cavity distribution is observed: the
attached cavity in the upper region of the impeller
is much larger than in the lower part.
Undoubtedly, this asymmetry would induce the
pump vibration (Yang et al., 2011). Hence, with
the impeller rotating, the attached cavity size in
one channel becomes smaller firstly and then
recovers again. The unsteady simulation results,
computed by the PANS model with f
k
=0.2, are
also presented in Fig. 10. The asymmetrical
cavity and detached cavity are well captured.
However, it failed to obtain the triangular
attached cavity shape in the calculations. This
may be due to the Coriolis force and the
centrifugal force, which are not considered in the
turbulence model and cavitation model adopted in
this study. Nevertheless, the PANS model with
f
k
=0.2 still well predicts the unsteady cavity
behavior in the pump.

4.3 Eddy viscosity

As mentioned above, the original k- model over-
predicts the eddy viscosity and so fails to capture
the unsteady properties in the cavitating flow. Fig.
11 compares the eddy viscosity calculated by the
original k- model and PANS model when =0.32
at span=0.5. And for a better comparison between
each cavitation condition, the eddy viscosity data
are normalized. It also should be noted that the
rotating direction in the figure is from bottom to
top.
Obviously, for the original k- model, an over-
predicted high eddy viscosity region is developed
on the suction side of the blade surface as
expected. On the other hand, with decreasing f
k
,
the PANS model effectively reduces the eddy
viscosity in the cavity region and captures some
irregular flow structure in the detached cavity
region. Because with the lower f
k
, the PANS
model reduces the dissipation in the flow, the
excess eddy viscosity region is filtered out,
resulting in more resolved flow features. It
demonstrates that the PANS model can obtain a
much better unsteady cavitating flow in
centrifugal pumps than the original k- model.


Fig. 11 Comparison of eddy viscosity in impeller when
=0.32 at span=0.5 as between original k- and
PANS.



Fig. 12 Comparison of blade load pressure under
various (PANS f
k
=0.2) at span=0.5.

4.4 Pressure distribution

In a bid to obtain a more detailed hydrodynamic
flow structure, the pressure distribution in the

Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
326


Fig. 13 Pressure distribution under various at mid-
plane of impeller.

impeller is analyzed in this section, including the
blade load pressure, the pressure distribution on
impeller, and the pressure fluctuations at the
pump inlet and outlet, which are validated by
experiment. Fig. 12 plots the blade load pressure
at span=0.5 along the streamwise coordinate. The
computational data are from the PANS model
with f
k
=0.2 under various cavitation numbers ,
ranged from 0.59 (non-cavitation condition) to
0.32 (fully developed cavitation condition).
It should be noted that, in this figure, under one
cavitation condition, the upper curves are the data
of the pressure side surface, while the lower ones
are of the suction side. We can find the pressure
load distributions on both pressure and suction
sides are similar under different cavitation
numbers, except for the leading edge of the
suction side. And apparently, with decreasing
cavitation number, the pressure load has a sharp
drop. For high cavitation number situation, like
=0.59, the pressure load on the suction side
changes incrementally from the leading edge to
the trailing edge. It is mainly because there are no
bubbles attached on the blade in this situation.
But when the local pressure drops below the
liquid vaporized pressure, the bubbles firstly form
on the leading edge, which makes the nearby
pressure load on the suction side very low. This
phenomenon can be clearly observed while
decreases to 0.32. Under this condition the low
pressure region is much longer, which can be
explained in Fig. 13. It shows the pressure
coefficient distribution on the mid-plane cross the
impeller, normalized by equation (19).
We can notice that the pressure distribution
gradually increases from the impeller inlet to
outlet. And the low pressure region firstly
emerged on the leading edge of the suction side.
Then, with the decreasing cavitation number, the
region progressively expands downstream along
the blade surface and finally covers the pressure
side at =0.32. Meanwhile, due to the developed
cavitating flow attached on the suction side
surface, the re-entrant flow can be observed in
Fig. 13c, which is highlighted by the red box.
The pressure fluctuations of the pump inlet and
outlet were also measured to validate the transient
simulations. Both of the experiment and the
simulation data in one cycle time are normalized
via equation (19) and plotted in Fig. 14. We can
notice that, during one period, the pressure
fluctuation changes periodically, not only under
the non-cavitation condition =0.59, but also in
the cavitation state =0.32. Five peaks can be
found both in the experiments and numerical
simulations. With decreasing cavitation number,
the pressure coefficient C
pt
declined, however,
because the cavitation inception is slower in the
simulations, the experimental data of the pressure
fluctuations drop a little faster, leading to larger
discrepancy for the lower cavitation number
=0.32. Even so, the agreement between transient
numerical simulations and experiment is good,
demonstrating that the simulations are acceptable.

5. CONCLUSIONS

In this paper, the partially-averaged Navier-
Stokes method, derived by Girimaji based on the
original k- model, is utilized to predict the
cavitating flow in centrifugal pump with twisted
blades. With the constant filter f

=1, the influence


of filter f
k
on simulation results are discussed and
the results are compared with the original k-
model. The experiment and visualizations at the

(a) =0.59 (b) =0.41 (c) =0.32

Fig. 14 Pressure coefficients fluctuations under various .
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2, pp. 319329 (2014)
327
pump inlet are performed to validate the
numerical simulations. In order to obtain the
cavitation inception and its evolution on the
leading edge of the twisted blades, a water tank is
mounted just against the pump entrance. It can be
concluded that the PANS model with lower f
k

value, compared with the original k- model, can
effectively reduce the eddy viscosity in cavitating
flow in centrifugal pumps and obtain unsteady
bubble detachment phenomenon for a better
simulation of pump performance.
Firstly, the performance of the PANS model in
coping with cavitating flow in centrifugal pump
are evaluated, four different filters f
k
= 0.9, 0.7,
0.4 and 0.2, are chosen in the present work. The
simulation results indicate that the PANS model
with lower f
k
value predicts more accurate results.
Besides, resemblance of the performance is
observed between the PANS model with higher
filter value and the original k- model. With
decreasing f
k
value, both the pump head and
cavitation performance are much closer to the
experimental result. For f
k
=0.2 in the PANS
model, the critical cavitation number
c
is 0.35,
and the pump head coefficient is 0.76, compared
with 0.39 and 0.71 obtained by experiment,
respectively. But for the original k- model, the
values are 0.26 and 0.84, respectively.
Experimental visualizations are also carried out to
validate the numerical simulations. The unsteady
cavity behavior is well recorded via a high speed
camera. The attached cavity can be seen in the
region conjoining the suction side of the blades
with hub and shroud on the leading edge. A
triangular cavity shape is observed, which is
probably caused by the Coriolis force and the
centrifugal force due to the rotation. Besides,
because of the interaction between the impeller
and vane guide, asymmetrical cavity distribution
can be seen, which would undoubtedly induce
pump vibration.
The comparison of the experimental
visualizations and calculations is conducted. Both
the development of the cavitating flow with
decreasing cavitation number and under a
constant cavitation number are analyzed. The
cavitation inception approximately occurs at
=0.45 for all the simulation cases. As the pump
inlet pressure decreases, for the lower filter value,
f
k
=0.2, the PANS model captures more detached
cavity and unsteady flow structures at the rear of
the cavity region, because it successfully reduces
the over-predicted eddy viscosity. However, for
higher f
k
value, the results are similar to those of
the original k- model and no such unsteady
phenomenon are captured. Still, there are some
discrepancy between simulations and
visualizations. The reason may be related to the
Coriolis force and centrifugal force, which were
not considered in the turbulence model and
cavitation model in the present study. The
influence of these will be studied in future work.
And the particle image velocimetry (PIV) and
laser induced fluorescence (LIF) technology
should be applied to validate the cavitating flow
structure.
Simultaneously, the eddy viscosity distribution
under =0.32 at span=0.5 are investigated to
validate the ability of the PANS model. With
lower f
k
value, the results show that it can
effectively reduce the eddy viscosity in the cavity
region and captures more irregular flow structure
in the detached cavity region. That is because for
lower f
k
value, the computing dissipation is
reduced in the flow, leading to filtering out the
excess eddy viscosity region and bubbles
shedding off. So the PANS model can
successfully overcome the deficiency of the
original k- model of over-predicting the eddy
viscosity.
Finally, the blade load pressure, the pressure
distribution in impeller and the pressure
fluctuations are analyzed to get a clear view of the
flow structure, by adopting the simulation results
calculated by the PANS model with f
k
=0.2. It is
found that the pressure on blade surface rises
gradually from the leading edge to the trailing
edge, and the cavity region produces a much
wider low pressure region on the suction side,
starting from the leading edge of the blade. As for
the pressure fluctuation, comparison between
transient simulations and experiment is conducted
under three different cavitation numbers. Regular
variation is observed both in simulations and
experiment during one period, and the agreement
is good.

ACKNOWLEDGEMENTS

The authors would like to thank the support by
the National Natural Science Foundation of China
(Nos. 51239005, 51309120, 51109095 and
51179075), National Science & Technology Pillar
Program of China (Nos. 2011BAF14B03,
2013BAF01B02 and 2013BAK06B02), Natural
Science Foundation of Jiangsu Province of China
(No. BY2011140), Senior Talents Project of
Jiangsu University (No. 12JDG044), Priority
Academic Program Development of Jiangsu
Higher Education Institutions and Jiangsu
Planned Projects for Postdoctoral Research Funds
(No. 1202076C) and Scientific Research
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
328
Innovation Program in Colleges and Universities
of Jiangsu Province (No. CXLX12_0640)

REFERENCES

1. Batten P, Goldberg U, Chakravarthy, S
(2002). LNS-an approach towards embedded
LES. AIAA paper 0427.
2. Basara B, Krajnovi S, Girimaji S (2010).
PANS methodology applied to elliptic-
relaxation based eddy viscosity transport
model. Turbulence and Interactions (TI 2009
Conference), 63-69.
3. Benaouicha M, Astolfi J (2010). A numerical
study of cavitation induced vibration.
Proceedings of the ASME 2010 Pressure
Vessels & Piping Division, 18-22 July,
Bellevue, Washington, USA, 35-42.
4. Bilus I, Andrej P (2009). Numerical and
experimental approach to cavitation surge
obstruction in water pump. International
Journal of Numerical Methods for Heat &
Fluid Flow 19(7):818-834.
5. Bilus I, Skerget L, Predin A, Hribersek M
(2005). A homogenous cavitation transport
model in turbo machinery. Transactions on
Engineering Sciences 50: 25-34.
6. Bruno S, Frank CV (2009). Pump cavitation-
Various NPSHr criteria, NPSHa margins, and
impeller life expectancy. Proceedings of the
Twenty-fifth International Pump Users
Symposium, Texas, USA, 113-144.
7. Campos-Amezcua R, Khelladi S, Mazur-
Czerwiec Z, Bakir F, Campos-Amezcua A,
Rey R (2011). Numerical and experimental
study of mass transfer through cavitation in
turbomachinery. Mass Transfer-Advanced
Aspects (Ed. Nakajima, H):177-206.
8. Chang SP, Wang YS (2012). Cavitation
performance research of mixed- flow pump
based on CFD. Journal of Drainage and
Irrigation Machinery Engineering 30(2):171-
176.
9. Cudina M (2006). Noise as an indicator of
cavitation in a centrifugal pump. Acoustical
Physics 49(4):463-74.
10. Ding H, Visser FC, Jiang Y, Furmanczyk M
(2011). Demonstration and validation of a
3D CFD simulation tool predicting pump
performance and cavitation for industrial
applications. Journal of Fluids Engineering
133 (1): 011101.
11. Duplaa S, Coutier - Delgosha O, Dazin
A, Roussette O, Bois G, Caignaert G
(2010). Experimental study on hydrodynamic

performance of a cavitating centrifugal pump
during transient operation. Journal of Fluids
Engineering 132(021301):1-12.
12. Erfan N, Mahjoob MJ, Ardeshir B (2010).
Experimental and numerical study of
cavitation in centrifugal pumps. Proceedings
of the ASME 2010 Biennial Conference on
Engineering Systems Design and Analysis,
12-14 July, Istanbul, Turkey, 1-6.
13. Girimaji SS, Jeong E, Srinivasan (2006).
Partially averaged Navier-Stokes method for
turbulence: Fixed point analysis and
comparison with unsteady partially averaged
Navier-Stokes. Journal of Applied Mechanics
73:422-429.
14. Girimaji SS, Suman S (2011). Partially
averaged Navier Stokes (PANS) method for
turbulence simulations: Theory and practice.
Progress in Hybrid RANS-LES Modelling,
September, Beijing, China, 2943.
15. Harlow FH, Nakayama PI (1967). Turbulence
transport equations. Physics of Fluids 10(11):
2323-2332.
16. Ji B, Luo XW, Peng XX, Wu YL (2013).
Three-dimensional large eddy simulation and
vorticity analysis of unsteady cavitation flow
around a twisted hydrofoil. Journal of
Hydrodynamics 25(4): 510-519.
17. Kato C, Mukai H, Manabe A (2003). Large
eddy simulation of unsteady flow in a mixed-
flow pump. International Journal of Rotating
Machinery 9(5): 245-351.
18. Kunz RF, Boger DA, Stinebring DR,
Chyczewski TS, Lindaua JW, Gibelinga HJ,
Venkateswaranb S, Govindanc TR (2000). A
preconditioned Navier-Stokes method for
two-phase flows with application to cavitation
prediction. Computers and Fluids 29(8): 849-
875.
19. Launder BE, Spalding DB (1974). The
numerical computation of turbulent flows.
Computer Methods in Applied Mechanics and
Engineering 3(2): 269-289.
20. Liu HL, Liu DX, Wang Y, Wu XF, Zhuang
SG (2012). Applicative evaluation of three
cavitating models on cavitating flow
calculation in centrifugal pump. Transactions
of the Chinese Society of Agricultural
Engineering 28(16): 54-59.
21. Morgut M, Nobile E (2011). Numerical
predictions of the cavitating and non-
cavitating flow around the model scale
propeller PPTC. Second International
Symposium on Marine Propulsors, Smp11,
June, Hamburg, Germany, 1-6.

Engineering Applications of Computational Fluid Mechanics Vol. 8, No. 2 (2014)
329
22. Morgut M, Nobile E, Bilu I (2011).
Comparison of mass transfer models for the
numerical prediction of sheet cavitation
around a hydrofoil. International Journal of
Multiphase Flow 37(6): 620-626.
23. Nobuhiro Y, Chisachi K, Yoichiro M (2003).
LES analysis of a rocket turbopump inducer
in non-cavitating and cavitating flows.
Proceedings of ASME FEDSM03, 4th
ASME/JSME Joint Fluids Engineering
Conference, 6-10 July, Honolulu, Hawaii,
USA, 1219-1226.
24. Okita K, Kajishima T (2002). Three-
dimensional computation of unsteady
cavitating flow in a cascade. The 9th of
International Symposium on Transport
Phenomena and Dynamics of Rotating
Machinery, 10-14 February, Honolulu,
Hawaii.
25. Raiesi H, Piomelli U, Pollard A (2011).
Evaluation of turbulence models using direct
numerical and large-eddy simulation data.
Journal of Fluids Engineering 133(2):
021203.
26. Sandham ND, Alam M, Morin S (2001).
Embedded direct numerical simulation for
aeronautical CFD. Aeronautical Journal
105(1046): 193-198.
27. Senocak I, Shyy W (2004a). Interfacial
dynamics-based modeling of turbulent
cavitating flows, Part-1: Model development
and steady-state computations. International
Journal for Numerical Method in Fluids
44(9): 975-995.
28. Senocak I, Shyy W (2004b). Interfacial
dynamics-based modeling of turbulent
cavitating flows, Part 2: Time-dependent
computations. International Journal for
Numerical Methods in Fluids 44(9): 997-
1016.
29. Singhal AK, Athavale MM, Li H, Jiang Y
(2002). Mathemat-ical basis and validation of
the full cavitation model. Journal of Fluids
Engineering 124(3), 617624.
30. Smagorinski J (1963). General circulation
experiments with the primitive equations: I.
The basic experiment. Monthly Weather
Review 91: 99-164.
31. Wang GY, Huang B, Zhang B (2011).
Evaluation of a filter-based model for
computations of cavitating flows. Chinese
Physics Letters 28(2): 026401(1-4).
32. Yang ZJ, Wang FJ, Liu ZQ, Zhang ZM
(2011). Prediction of cavitation performance
of axial-flow pump based on CFD. Journal of
Drainage and Irrigation Machinery
Engineering 29(1): 11-15.
33. Zwart P, Gerber AG, Belamri T (2004). A
two-phase model for predicting cavitation
dynamics. Proceedings of ICMF2004
International Conference on Multiphas Flow,
30 May-3 June, Yokohama, Japan, Paper
No.152.

You might also like