Cav03-OS-4-010

Fifth International Symposium on Cavitation (CAV2003)

Osaka, Japan, November 1-4, 2003

CAVITATION CFD IN A CENTRIFUGAL PUMP

Motohiko Nohmi / Ebara Research Co. /
nohmi20181@erc.ebara.co.jp

Akira Goto / Ebara Research Co. /

goto05296@erc.ebara.co.jp

Yuka Iga / Tohoku University

Toshiaki Ikohagi / Tohoku University

ABSTRACT
The cavitation flow in a low specific centrifugal pump is
analyzed by using two types of cavitation CFD codes. A locally
homogeneous model of compressible air-vapor-liquid two-phase
medium is applied to calculate the objective flow. Constant
Enthalpy Vaporization model is also applied. Numerical results
are compared to measured results. At the cavitation breakdown
of the BEP flow rate, the throat is choked by cavities on both
suction surface and pressure surface and head drops steeply.
The steep head drop is well captured by both CFD codes. The
NPSH value at the breakdown point can be predicted reasonably.
At the high flow rate further modification is necessary for both
codes to achieve stable and accurate results.
INTRODUCTION
In recent years, ability of CFD predictions advances rapidly.
One of the new features is cavitation CFD. Many cavitation
modelings have been developed and are now available in
commercial codes such as CFD-ACE, STAR-CD, FLUENT, FINETURBO and CFX-TASCflow. For pump industries it is much
desired to predict pump suction performance accurately in
preliminary study. However accuracy of predictions and
robustness of computation look very case-dependant from
authors experience. It is still necessary to evaluate and modify
its scheme and modeling based on the comparison with
experimental results. Not so many experimental data are available
for this purpose in open literature. Data from Shen and
Dimotakis s work [1] have been widely used to evaluate
commercial codes. Their data show the pressure distribution over
hydrofoil in the GALCIT cavitation tunnel. There is only one
pressure tap over pressure side in their hydrofoil, so the
information of the flow around the pressure surface is limited.
More detailed data were measured by Drei and Kosyna [2], and
Friedrichs and Kosyna [3]. They installed piezoresistive
transducers inside the impeller blade of a centrifugal pump and
measured pressure distribution under cavitation.
In this study flow field with cavitaion in a centrifugal pump
was measured for the validation of cavitation CFD. The flow field
was predicted by using a numerical method based on a locally
homogeneous model of compressible gas -liquid two-phase

medium by authors (Iga et al. [4]) and by using commercial code

CFX-TASCflow. Computed results are compared to the
experimental results.

NOMENCLATURE
Q = pump volume flow rate [m3/min]
NPSH = NPSH available at pump inlet [m]
p = pressure [Pa]
= density [kg/m3]
T = temperature [K]
R = Gas constant [m2/s 2K]
p c = pressure constant [Pa]
K = Liquid constant [m2/s 2K]
T0 = temperature constant [K]
Y = mass fraction of air in mixture media
Y = mass fraction of vapor in vapor and liquid water
(mass ratio of vaporization)
= void faction
Subscripts
a = air
w = water
v =vapor
p & without subscripts are mean values of mixture media

EXPERIMENTAL APPARATUS
Test loop for a pump under cavitation was constructed. It
consists of a reservoir tank, an objective pump, a booster pump
and a controlled valve. The reservoir tank is connected to a
vacuum pump and a chilling machine which control NPSH and
water temperature
Low specific speed centrifugal pump of 0.44
(nondimentional) or 180 [m, m3/min, rpm] with two-dimensional
blade was adopted as the objective pump. Its blade is
perpendicular to the hub and is installed with pressure
measurement devices. Design specification of the pump impeller
is shown in Table 1. The impeller blades and the pump
schematics are shown in Fig.1 and Fig.2. The pump has an inlet
1

volute casing and inlet guide vanes. An acrylic-resin transparent

and flat window was installed downstream of the guide vanes.

Table 1. Design of Impeller

Impeller Inlet Radius (mm)
91
Impeller Outlet Radius
174
(mm)
Span Width (mm)
18
Inlet Blade Angle (deg)
8
Outlet Blade Angle (deg)
10
Number of Blades
3

whole computation domain.

(3) Pressure of liquid water is equal to the pressure of gas in
bubbles locally.
(4) The pressure of gas is sum of air partial pressure and vapor
partial pressure.
(5) Phase change is enough fast to maintain bubbles saturated
with vapor.
From these assumptions the state equation for a locally
homogeneous gas -liquid two-phase media can be expressed as
follows.
=

Impeller shroud is also transparent and flat as seen in Fig.2.

Cavitation in the blade passage can be clearly observed from
axial direction. Similar setup for a centrifugal pump impeller is
also used by Kleinert et al. [5]. Cavitation is photographed by a
high speed video camera, a digital still camera. Eighteen absolute
pressure transducers and linear DC amplifiers are instrumented
inside blades. Pressure taps of 0.5 mm diameter are drilled at
midspan of the pressure and suction sides. Natural frequency of
pressure measurement system is 2.5 KHz and its characteristic is
linear up to 0.5 KHz. The transducer mounting is similar to Drei?
and Kosyna [2]. Pressure signal from eighteen transducers is
transferred through slip ring connectors to stationary frame and
stored simultaneously in multi channel digital data acquisition
system. For more detailed information about experimental rig see
Nohmi et al. [6].
NUMERICAL METHODS FOR CAVITATION FLOW
Mathematical Modeling
A locally homogeneous model of compressible gas -liquid
two-phase medium by authors has already been applied to
cavitation flow (Iga et al. [4]). In the present study its modeling is
slightly modified taking account of phase change. In addition to
the original modeling, following assumptions are applied.
(1) Bubbles in each control volumes are made of air and vapor.
(2) Mass fraction of air in mixture media i.e. Y is constant in

( p + p c )( p pv )
(1 Y )( p pv ) K (T + T0 ) Y v K (T + T 0 ) Ra T + Y ( p + p c )R aT
(1)

Also the equations of void fraction and mass ratio of

vaporization Y are expressed.
= 1Y =

Y
(1 Y v Y ) =
=
w
a
a w a v

(2)

Y v Ra T
<1
(1 Y )( p p v )

(3)

If the pressure is enough decreased (p < p v saturated) to vaporize

whole liquid water the mixture media is regarded as single phase
perfect gas. The following equations are applied.
p p a + pv = (YRa + (1 Y ) Rv )T

(4)

pa = a Ra T = YR aT
pv = v Rv T = (1 Y )R vT

(5)
(6)

Y =1

(7)

=1

(8)

This model is named TE model from the first letters of Tohoku

university and Ebara research company.

Rotation

Volute Casing

Pressure
Taps

Outlet Collector

Window

Impeller
Pressure Transducers

Shroud
Guide
Vane
Fig.1 Pump Impeller

Fig.2 Pump Cross Section

2

Numerical Method
In this study steady cavitation flows are simulated solving
the governing equations by using finite volume method. A
commercial CFD code STAR-CD is applied. Compressible NS
equations are solved with above mentioned state equation in
users subroutine. In STAR-CD the governing equations are
discretized by using SIMPLE scheme. Prior to cavitaion CFD
ordinary incompressible computation is carried out and its result
is used as initial condition of cavitation CFD.
Computation by CFX-TASCflow
CFX-TASCflow uses enthalpy based modeling of cavitation
named Constant Enthalpy Vaporization (CEV) model (Bouziad et
al. [7]). In CEV model enthalpy of water vapor mixture is
assumed constant along streamlines. There is not slip velocity
between water phase and vapor phase. Flow is assumed steady
and standard k- model is used. Incompressible CFD result is
used as initial condition of cavitation CFD in the same way of
STAR-CD.
Computation Domain and Boundary Conditions
The flow in the pump stage including inlet volute casing,
inlet guide vane, impeller and exit collector casing can be
computed simultaneously using cavitation CFD solvers. It will
be, however, very time consuming. In this study impeller part is
only computed with periodic boundaries. For the discretization of
the computation domain, multi blocked structured grids are
adopted. Its meridional view is shown in Fig. 3. The total cell
number is 96336. Same computational grids are used for STARCD and CFX-TASCflow. After preliminary study it was found that
the grid is coarse for stable computation in the case of the high

flow rate by STAR-CD. The refined grid of the total cell number
187344 is adopted only for the high flow rate case of STAR-CD.
More detailed grid dependency is not evaluated in this study.
In the case of STAR-CD mass flow rate and flow angle are
fixed at the inlet boundary and static pressure is fixed at the
outlet. In the case of CFX-TASCflow total pressure and flow
angle are fixed at the inlet boundary and mass flow rate is fixed at
the outlet. In the objective pump inlet swirling flow is generated
by the inlet volute casing. Inlet flow angle is estimated by using
CFD applied for the flow in volute casing in the previous study
(Nohmi et al. [6]). The estimated angle is slightly tuned by
comparison to the measured pressure distribution over blade
surfaces (see results section). Skin friction of hub and tip wall is
calculated using a wall function approach.
RESULTS AND DISCUSSIONS
Pump Characteristics
Measured head, efficiency and volume flow rate
characteristics are shown in Fig.4. Best efficiency point is found
at about Q=0.95 m3/min. Results at 0.95, 1.25 m3/min are used in
the following comparisons.
Results of Incompressible CFD
In Fig.5 measured and computed pressure distributions over
blade surface are shown. Measured results are time averaged of
133 impeller rotations. By same slight tuning of inlet flow angle
incompressible CFD of both STAR-CD and CFX-TASCflow show
good correspondence to experimental results. From the
stagnation in CFD results, it is found that the attack angle at
Q=0.95 m3/min, the best efficiency point, is almost zero and the
attack angle at Q=1.25 m3/min is negative.
Measured and Computed Head Drop Characteristics
In Fig.6 measured head drop characteristic curves are
shown. Computed head drop curves are also shown in Fig.6.
Note that computed impeller head of CFX-TASCflow is
underestimated because loss at the inlet volute casing and the
outlet collector are not included in cavitation CFD. NPSH of CFD
is estimated at the sum of NPSH of cavitation CFD inlet and the
loss head of inlet volute casing. The loss head of inlet volute
12

45

Head

10

Efficiency

Head m

30

6
4

15

Q=0.95

0
0

0.5

1.5

Q m /min

Fig.3 Computational Grids for Impeller

Fig.4 Pump Head and Efficiency Curves

Efficiency %

Governing Equations
The governing equations for two-phase media are ordinary
3D compressible Navier-Stokes equations and the mass
conservation equations. In this study temperature change is
neglected and the energy conservation equation is not applied
i,e, fluid is assumed barotropic. Standard k- model is used for
the turbulence closure.

160000

160000

Absolute Static Pressure Pa

Absolute Static Pressure Pa

180000

CFD PS

140000
120000

CFD SS

100000
80000
0

0.2

0.4

0.6

0.8

Normalized Meridional Length

CFD PS

140000
120000
100000

CFD SS

80000
60000
40000
0

(a) STAR-CD Q=0.95m3 /min

(b) STAR-CD Q=1.25m3 /min

180000

160000

160000

Absolute Static Pressure Pa

Absolute Static Pressure Pa

0.2
0.4
0.6
0.8
Normalized Meridional Length

CFD PS

140000
120000

CFD SS

100000
80000
0

0.2

0.4

0.6

0.8

Normalized Meridional Length

CFD PS

140000
120000
100000
80000

CFD SS

60000
40000
0

(c) CFX-TASCFlow Q=0.95m3 /min

0.2
0.4
0.6
0.8
Normalized Meridional Length

(d) CFX-TASCFlow Q=1.25m3/min

Fig.5 Measured and Computed Pressure Distribution over Blade Surfaces

casing is calculated from the CFD in previous study (Nohmi et al.
[6]).
As seen in experimental results in Fig.6 the curve at Q= 0.95
m3/min shows very sharp head drop. The curve at high flow rate
of 1.25 m3/min shows creeping head drop. CFD results of both
TE model and CEV model at Q=0.95 m3/min show sharp head
10

E Q=0.95m3/min
CEV Q=0.95m3/min

Head m

8
7

Exp. Q=0.95m3/min

E Q=1.25m3/min
CEV Q=1.25m3/min
Exp. Q=1.25m3/min

6
5
4
3
2
1
0
0

10

11

12

NPSH m

of Q= 1.25 m3/min, the result of TE model looks starting creep,

however very high void fraction in computation domain makes
computation unstable at less NPSH condition. Computed NPSH
of creep starting is a little bit higher than that of experiment. By
using CEV model stable computation can be carried out until the
head totally breaks down at Q= 1.25 m3/min. However its head
increases prior to head breakdown. For the case of Q= 1.25
m3/min further modification are necessary for both codes to
achieve stable and accurate results. In stead of SIMPLE method
some explicit scheme may be preferable for TE model. More
detailed discussions about head drop characteristics in CFD are
written in the next section.
Flow Characteristic at Cavitation Breakdown
Absolute static pressure distribution over the blade surface
changes as the NPSH decreases as shown in Fig.7 and Fig.8. In
Fig.7 measured pressure value is time averaged of 133 impeller
rotations. Cavitation photographs at breakdown are shown in
Fig.9 and the isosurfaces of computed void fraction =30% at
breakdown are shown in Fig.10. These photographs and void
fraction figures correspond to circles and squares shown in
Fig.6.

Fig.6
Measured
and
Computed
Suction
drop
and
head drop
p oint
is relatively
wellPerformance
predicted. InCurves
the case
4

160000

160000

NPSH=10.4
PS

NPSH=10.4
PS

140000
Static Pressure Pa abs.

Static Pressure Pa abs.

140000
120000

SS

100000

120000

NPSH= 1.86
1.17
1.09
1.081
1.079
1.075

80000
60000
40000

SS

100000

20000

NPSH= 4.86
3.21
2.67
2.34
2.01
1.83
1.61
1.53

80000
60000
40000
20000

0
0

0.2

0.4

0.6

0.8

0.2

Normalized Meridional Length

(a) Q =

0.4

0.6

0.8

Normalized Meridional Length

0.95m3/min

(b) Q = 1.25m3/min

Fig.7 Measured Blade Pressure Distribution at Head Drop

160000

100000

NPSH=10.2073.17

140000
Static Pressure Pa abs

Static Pressure Pa abs

NPSH=3.2871.505
80000

60000

40000

20000

120000
100000
80000
60000
40000
20000

0
0

0.2

0.4

0.6

0.8

0.2

0.6

0.8

Normalized Meridional Length

Normalized Meridional Length

(b) TE Model Q=1.25m 3/nin

(a) TE Model Q=0.95m3 /nin

100000

100000

NPSH=1.7161.095

80000

Static Pressure Pa abs.

Staic Pressure Pa abs.

0.4

60000
40000
20000

NPSH=3.3650.902

80000
60000
40000
20000
0

0
0

0.2

0.4

0.6

0.8

0.2

0.4

0.6

0.8

Normalized Meridional Length

Normalized Meridional Length

(d) CEV Model Q=1.25m3 /nin

(c) CEV Model Q=0.95m 3/nin

Fig.8 Computed Blade Pressure Distribution at Head Drop

5

(a) Q=0.95m3/min (NPSH=1.075m)

(b) Q=1.25m 3/min (NPSH=1.529m)

Fig.9 Cavitation Photographs at Head Drop

(a) TE Model Q=0.95m3/min (NPSH=1.505m) (b) TE Model Q=1.25m3/min (NPSH=3.17m)

(c) CEV Model Q=0.95m3 /min (NPSH=1.095m) (d) CEV Model Q=1.25m3/min (NPSH=0.902m)
Fig.10 Computed Void Fraction at Head Drop ( =30%)
In the case of Q=0.95 m3/min, cavitation bubbles appear at
mid chord on suction side where static pressure is minimum as
shown in Fig.5. While NPSH is further decreased, cavitation

bubbles are extended downstream and aft part of the pressure

distribution curve on suction surface becomes flat in Fig.7(a).
When bubbles reach throat area, wedge like cavity appears at the
throat on pressure side as seen in Fig.9(a). Fore part of the
6

pressure curve on pressure side is dented. Cavities on both sides

choke throat and head drops suddenly. In TE model pressure
jump appears at the suction surface in Fig.8(a). This pressure
jump cannot be observed in the measured pressure distribution.
Flow at the closure region of cavity must be refined in TE model.
The isosurface of void fraction by TE model looks very similar to
the experimental photograph. High void faction zone between
blades in Fig.10(a) corresponds to low static pressure induced by
blades and tip streamline curvature on a shroud. In experiments
cavitation bubble bursts are observed at the same zone. Similar
bubbles due to the curvature on a shroud of a centrifugal pump
were observed by Minami et al. [8] and Hofmann et al. [9].
Although pressure distribution over suction surface by CEV
model becomes flat at breakdown in Fig.8(c) high void fraction
zone on suction surface cannot be seen in Fig.10(c). Further
numerical investigation is necessary to clear discrepancies
mentioned above.
In the case of Q=1.25 m3/min, cavitation bubbles appear at
the leading edge on pressure side i.e. on throat where static
pressure is minimum as shown in Fig.5. Inception of visible
cavitation there causes almost 3% head drop. While decreasing
NPSH, cavitation bubbles grow gradually and pump head
decreases gradually as seen in Fig.6. When NPSH is further
decreased, bubbles at the leading edge become pure gas phase
like a wedge as seen in Fig.9(b). This gas wedge has very stable
structure. NPSH is decreased further, the cavitation bubbles start
appearing at the throat on suction side, opposite side of the gas
wedge. Along this flow change, fore part of the pressure curve
on pressure side is dented as shown in Fig.7(b). On suction side
aft part of the pressure curve decreases because throat velocity
is increased by the cavity of pressure surface. Finally aft part of
the pressure curve on suction side becomes less than fore part.
In both CFD high void fraction area on pressure surface is similar
to gas wedge in experiments . In TE model flow cavity at the
leading edge on pressure surface increases the loss and
decreases the pump head gradually. However in TE model flow in
high void fraction area is very unstable as mentioned in the
previous section. Some improvement was achieved by grid
refinement, however it was limited. In CEV model calculation is
more stable. As seen in Fig.8(d) pressure distribution change is
relatively well predicted by CEV model. In the case of
Q=1.25m3/min with high NPSH condition, loading at fore part of
the blade is negative due to the negative attack angle as seen in
Fig.5. By cavitation bubbles minimum pressure over blade
surface can be limited around saturation vapor pressure. At very
low NPSH condition minimum pressure at pressure surface is
limited and negative loading is reduced in CFD by CEV model.
This phenomena causes head increase numerically prior to head
drop in Fig.6.

CONCLUDING REMARKS
The cavitation flow in a low specific speed centrifugal pump
is computed by STAR-CD with a locally homogeneous model of
compressible air-vapor-liquid two-phase medium (TE model) and
by CFX-TASCflow with Constant Enthalpy Vaporization model
(CEV model). At the cavitation breakdown of the BEP flow rate,

the throat is choked by cavities on both suction surface and

pressure surface and head drops steeply. The steep head drop is
well captured and the NPSH value at the breakdown point can be
predicted reasonably by both CFD codes. At the high flow rate
cavitation bubbles appear at the leading edge on pressure side
incipiently and the head drops gradually. Although gradual head
drop can be predicted, the computation is unstable by using TE
model. By using CEV model gradual head drop cannot be
predicted. Further modification is necessary for both codes to
achieve stable and accurate results.
ACKNOWLEDGEMENT
While planning the experimental facility, one of the authors
had opportunities to discuss cavitation experiments and CFD
with Professors. K. Kikuyama, Y. Hasegawa, Y. Tsujimoto, Y.
Yoshida, C.E. Brennen, S. Ceccio, B. Lakshminarayana, D.
Steinbring, C. Zierke, S. Deutsch, R.E.A. Arndt, and C.S. Song.
The discussions were very fruitful for this study. Professor A.
Furukawa s suggestion about loading improvement by cavitation
was very important to analyze CFD results. The authors are most
grateful to them. The authors thank Mr. T. Katsumata Mr. T. Sakai
and Mr. N. Kodama for their contribution to experimental and
CFD works.

REFERENCES
[1] Shen, Y. T., Dimotakis, P. E., 1989, The Influence of Surface
Cavitation on Hydrodynamic Forces, Proc. 22nd ATTC, St.
Johns, pp4-53
[2] Drei,? A., Kosyna, G., 1997, Experimental Investigation of
Cavitation-States in a Radial Pump Impeller, Proc. JSME
CENNTENIAL GRAND CONGRESS Intl.Conf. Fluid
Eng.,pp231-238
[3] Friedrichs, J., Kosyna, G., 2002, Rotating Cavitation in a
Centrifugal Pump imp eller of Low Specific Speed, Trans
ASME, J. Fluids Eng.Vol.124,pp356-361
[4] Iga, Y., Nohmi, M., Goto, A., Shin, B. R. and Ikohagi, T.,2001
Numerical Study of Sheet Cavitation Break-Off
Phenomenon n a Cascade Hydrofoil, CAV2001:B2.001
[5] Kleinert, H. J., Will, G., Finzel, R., 1979, Experimental and
Theoretical Investigations of the Flow in Radial Impellers of
Centrifugal Pumps at Cavitation,Proc. 6th Conf. Fluid
Machinery, pp575-585
[6] Nohmi, M., Goto, A., Iga, Y. and Ikohagi,
T. ,2003, Experimental and Numerical Study of Cavitation
Breakdown in a Centrifugal Pump, Proc FEDSM03,
FEDSM2003-45409
[7] Bouziad, Y. A., Guennoun, F., Farhat, M. and Avelan, F., 2003,
Numerical Simulation of Leading Edge Cavitation, Proc
FEDSM03, FEDSM2003-45312
[8] Minami, S., Kawaguchi, K. and Honma, T.,1959, Experiment
about Cavitation in a Volute Pump , J. JSME Vol. 62 No.485,
in Japanese
[9] Hofmann, M., Stoffel B., Coutier-Delgosha, O. Fortes -Patella,
R. and Reboud JL,2001, Experimental and Numerical Studies
on a Centrifugal Pump with 2D-Curved Blades in Cavitating
Condition, CAV2001:B7.005
7