You are on page 1of 11

ICMF 2004 International Conference on Multiphase Flow

Yokohama, Japan, May 30-June 3, 2004

Paper No.152

A Two-Phase Flow Model for Predicting Cavitation Dynamics

Philip J. Zwart1 , Andrew G. Gerber2 , Thabet Belamri3

1: ANSYS Canada; Waterloo, Ontario, Canada;
2: Dept. of Mechanical Engineering; University of New Brunswick; Fredericton, NB, Canada;
3: ANSYS Canada; Waterloo, Ontario, Canada;

Abstract A robust CFD methodology for predicting three-dimensional flows with extensive cavitation is presented.
The model is based on the multiphase flow equations, with mass transfer due to cavitation appearing as source and sink
terms in the liquid and vapour continuity equations. The mass transfer rate is derived from a simplified Rayleigh-Plesset
model. It is implemented into the CFX-5 software, which also features a control volume finite element discretization and a
solution methodology which implicitly couples the continuity and momentum equations together. The model is validated
on a range of applications including flow over a hydrofoil, an inducer, and transient cavitation in a venturi.


Cavitation is an important phenomenon which can have a profound effect on the performance of a number of
devices. Examples include pumps, inducers, propellors, and injectors. Computational Fluid Dynamics (CFD)
has been extensively used to predict the flow through these devices under non-cavitating conditions. However,
because of the physical and numerical challenges associated with cavitation, CFD has only recently started to
be used to predict cavitating flows.
Many of the CFD models developed for cavitation involve the use of a barotropic equation of state to
express the mixture density as a function of local pressure. This modelling approach is attractive because it can
be integrated into a basic-functionality CFD code without much effort. However, it must also be recognized that
these approaches are overly simplistic in their assumption of equilibrium thermodynamics. By this we mean
that as the flow conditions change, the two-phase fluid is assumed to instantaneously reach its equilibrium
thermodynamic state.
In reality, the finite rate effects involved in cavitation are important, and are based on complex physical
processes. Eisenburg [5] provides a summary of the transient dynamics of cavitation. Cavitation inception in
practical fluids is associated with the growth of nuclei having diameters ranging from 105 to 103 cm. These
nuclei contain a mixture of vapour and noncondensible gases. As the nuclei pass through regions where the
pressure drops below the vapour pressure, they typically grow explosively. (For the case of gaseous cavitation,
growth is limited by diffusion timescales and is therefore much slower.) The cavitation region is made up of a
large number of these bubbles. Further downstream, as the bubbles are swept into regions of higher pressure,
they collapse. The dynamics of collapse are complex and depend on a variety of factors including surface
tension, viscous effects, and noncondensible content. The large accelerations and pressures generated by the
final stage of collapse are responsible for much of the noise and damage generated by cavitating flows.
Not surprisingly, there are no cavitation models which attempt to account for all of these complexities.
There are models, however, which do account for nonequilibrium effects. Most of these models are based upon
the Rayleigh Plesset equation [3], which describes the growth and collapse of a single bubble subjected to a
far-field pressure disturbance. Examples of such models are given by Schnerr and Sauer [11], Gerber [6], and
Senocak and Shyy [12]. The details of the models developed by these and other researchers vary, but they share
the common feature of modelling the vapour composition by using a continuity equation having a source term
which can be traced back to the Rayleigh Plesset equation.
In this paper, a new Rayleigh-Plesset based cavitation model is described. The model has been implemented
in CFX-5, a general-purpose three-dimensional commercial CFD package. Recognizing that cavitation funda-
mentally involves phase change, the model has been implemented as an interphase mass transfer process within
the codes multiphase framework. This has some important numerical benefits, the most significant being that
the global continuity constraint is cast in volumetric form while remaining fully conservative.
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

The resulting model has been validated on a range of cavitating devices featuring a number of different
fluids. Three test cases will be discussed in this paper: a cavitating hydrofoil, cavitation in an inducer, and
transient cavitation in a venturi.

Mathematical Model

Conservation Equations
The equations governing multiphase flow include conservation of mass for each phase :

(r ) (r ui )
+ = S (1)
t xi
and conservation of momentum for the mixture (assuming no interphase slip):

(m ui ) (m uj ui ) P i ( ji )
+ = + m r g + , (2)
t xj xi xj

where r , ui , , S respectively represent the volume fraction, Cartesian velocity components, density, and
mass generation rate of phase ; g i represents acceleration due to gravity; P is the pressure; and ji is the stress
tensor, related to the deformation rates using Stokes law:

ji = m + . (3)
xj xi

m and m are the volume-weighted mixture density and viscosity. We have assumed that the mass sources
arise from interphase mass transfer, and therefore satisfy the constraint
S = 0. (4)

We also have the constraint that the phases must fill up the available volume:
r = 1. (5)

Eqs. (1), (2), and (5) together form a closed system involving (N + 4) equations and (N + 4) unknowns,
N being the number of phases. For convenience, we choose to replace one of the phasic continuity equations
with the sum of all continuity equations divided by their respective densities:
1 (r ui )
S = 0. (6)
t xi

When all phases are incompressible, this equation may be interpreted as requiring the velocity divergence to
balance the volume generation due to phase change. For example, cavitation involves a vapour and liquid phase,
with sources related by:
S v = S l = S lv (7)
Then Eq. (6) reduces to
1 1
= S lv . (8)
xi v l
When the flow is turbulent, the velocities in the above equations represent statistically averaged velocities,
and additional Reynolds Stress terms appear in the momentum equation. These stresses are modelled using an
eddy viscosity approach such as the k or Shear Stress Transport (SST) models [7].
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

Cavitation Model
The Rayleigh-Plesset equation describes the growth of a vapour bubble in a liquid:
d 2 RB 3 dRB 2
2 Pv P
RB 2
+ + = (9)
dt 2 dt RB l
where RB represents the bubble radius, represents the surface tension coefficient, and Pv represents the
vapour pressure. Neglecting the second order terms and the surface tension yields the simplified expression
dRB 2 Pv P
= . (10)
dt 3 l
The rate of change of mass of a single bubble follows as
dmB 2 2 Pv P
= 4RB v . (11)
dt 3 l
If there are NB bubbles per unit volume, we may express the vapour volume fraction as
4 3
rv = VB NB = RB NB (12)
and the total interphase mass transfer rate due to cavitation per unit volume is
3r v v 2 Pv P
S lv = . (13)
RB 3 l
This model has been derived assuming bubble growth (vaporization). It can be generalized to include
condensation as follows: s
3rv v 2 |Pv P |
S lv = F sign(Pv P ) (14)
RB 3 l
where F is an empirical calibration coefficient.
This model works well for condensation. However, it is physically incorrect (and numerically unstable)
if applied to vaporization. One of the key assumptions in its derivation is that the cavitation bubbles do not
interact with each other. This is plausible only during the earliest stages of cavitation, when the cavitation
bubble grows from the nucleation site. As the vapour volume fraction increases, the nucleation site density
must decrease accordingly. With this in mind, we replace rv by rnuc (1 rv ) during vaporization, where rnuc
is the nucleation site volume fraction. RB is interpreted as the radius of a nucleation site. The final form of the
cavitation model is: q
Fvap 3rnuc (1rv )v 2 Pv P ifP < Pv
RB 3 l
S lv = q (15)
Fcond 3rv v 2 P Pv ifP > P v
RB 3 l

The following model parameters have found to work well for a variety of fluids and devices: RB = 106 m,
rnuc = 5 104 , Fvap = 50, and Fcond = 0.01.
For unsteady cavitating flows, it has been observed in the literature that standard turbulence models fail to
properly predict the oscillating behaviour of the flow. This has also been observed with the cavitation model
we have proposed. We follow the example of [4] and use a modified formulation for turbulent viscosity. In the
standard k model, the eddy viscosity for the mixture is
tm = m C (16)

The modified expression effectively reduces the eddy viscosity in the cavitating regions by using
tm = f ()C (17)

where  n
v m
f () = v + (l v ). (18)
v l
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152


Figure 1: Element-based finite volume discretization of the spatial domain. Solid lines define element bound-
aries and dashed lines divide elements into sectors. Solution unknowns are colocated at the nodal points (),
and surface fluxes are evaluated at integration points (). Control volumes are constructed as unions of element
sectors (shaded region).

Numerical Model

Discretization Scheme
The conservation equations described above are discretized using an element-based finite volume method [10].
The mesh may consist of tetrahedral, prismatic, pyramid, and hexahedral elements. A control volume is con-
structed around each nodal point of the mesh, as illustrated in Figure 1. The subface between two control
volumes within a particular element is called an integration point (ip); it is at integration points that the fluxes
are discretized. Integration point quantities such as pressure and velocity gradients are obtained from nodal
values using finite element shape functions, with the exception of advected variables which are obtained using
an upwind-biased discretization.
We now consider the discretization of the conservation equations at each control volume. The discretization
is fully conservative and time-implicit. The conservation equations are integrated over each control volume,
volume integrals are converted to surface integrals using Gauss divergence theorem, and surface fluxes are
evaluated in exactly the same manner for the two control volumes adjacent to an integration point. In the
following discussion V represents the volume of a control volume, Aiip the area vector of an integration point,
t the time step, and the superscripts n + 1 and n mean that the quantity is evaluated at the new and old time
step, respectively.
The discrete conservation equations for the phasic continuity may be viewed as evolution equations for the
volume fractions:
V  X
( r )n+1 ( r )n + ( ui Ai )n+1
ip (r,ip )
= 0, (19)

The advection scheme used to evaluate r,ip in terms of neighbouring vertex values must give solutions which
are both bounded and accurate. We write it in the form
r,ip = r,up + r R, (20)

where r,up is the upwind vertex value and R ~ is the vector from the upwind vertex to the integration point. If
= 0, this scheme recovers the first-order upwind scheme, which is bounded but excessively diffusive. If =
1, this scheme is a second-order upwind-biased scheme, but unbounded. A bounded high-resolution scheme
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

can be obtained by making as close to 1 as possible, but reducing where necessary to prevent overshoots and
undershoots from occurring. We use a method similar to that described by Barth and Jesperson [2].
The mass flows must be discretized in a careful manner to avoid pressure-velocity decoupling. This is
performed by generalizing the interpolation scheme proposed by Rhie and Chow [9], such that the advecting
velocity is evaluated as follows:  
i i
uip = u ip + dip , (21)
xi xi ip

dip V /a (22)
a m V /t + b (23)

and b represents the sum of advection and viscous coefficients in the discretized momentum equation. The
overbar denotes the average of the control volume values adjacent to the integration point.
The discretized phasic momentum equations may be viewed as an evolution equation for the phasic velocity
V  X X
n+1 i
(m ui )n+1 (m ui )n + (m uj Aj )n+1 (ui )n+1 = Pip A + n+1 i
m g V + (( ji )n+1 Aj )ip .
ip ip ip
A standard second-order or high-resolution scheme may be used for the advected velocity in this equation, and
finite-element shape functions are used to evaluate the gradients for the pressure and viscous forces.
Finally it remains to derive a discrete equation for pressure. This is obtained by integrating Eq. (6) over the
control volume:
X 1 V n+1 X n+1
n + r ui Ai ip S V = 0,
=1 ip

which yields a diagonally-dominant equation for pressure because of the special interpolation used for uiip and
through the derivative S /P .

Solution Strategy
The set of algebraic equations (19), (24), (25), and (5) represent equations for the volume fraction, velocity, and
pressure fields. With two phases, these equations form a 66 coupled system of equations at each nodal point.
Equation (5) is an algebraic equation which may be decoupled from the active set. In addition, equation (19)
is currently also decoupled from the pressure-velocity system and is treated in a segregated manner. The linear
system of equations is solved using the coupled algebraic multigrid technique developed by Raw [8].


Hydrofoil Cavitation
This test case involves a cavitating hydrofoil [13] at two angles of attack. At a one-degree angle, cavitation is
induced along the midchord. At a four-degree angle, cavitation is induced at the leading edge. The mesh is
solved as a three-dimensional problem with two layers of nodes in the depth direction and symmetry boundaries
applied to these two planes. There are 15,288 nodes on each of the two planes, and symmetry conditions are ap-
plied to these two planes. A close-up of the mesh in the vicinity of the hydrofoil is shown in Figure 2. Along the
hydrofoil surface a no-slip condition is applied. Free-slip conditions are used for the far-field boundaries above
and below the hydrofoil. A velocity-specified boundary is applied to the inlet to yield a specified Reynolds
number. At the outlet, the pressure is imposed to yield a specified cavitation number. The fluid properties are
taken to be saturated water conditions at 25 C.
For the one-degree angle of attack, cavitation numbers of 0.43, 0.38, and 0.34 are considered. The Reynolds
number based on chord length is 3 106 . A comparison between experimental and computed pressure coeffi-
cients for these situations is plotted in Figure 3. The plot shows excellent agreement with the data.
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

Figure 2: Hydrofoil mesh

For the four-degree angle of attack, cavitation numbers of 1.00, 0.91, and 0.84 are considered. The Reynolds
number based on chord length is 2 106 . The calcalated and experimental pressure coefficients for these cases
are plotted in Figure 3. In this case, the trends compare well with experiment but but the length of the cavitation
zone is somewhat underpredicted at the lower cavitation numbers.

Inducer Cavitation
A second validation test case involves flow in an inducer tested at LEMFI [1]. The cavitating flow through
a single blade passage is modelled at a range of flow rates and cavitation numbers. The mesh, consisting of
250,000 hexahedral elements for the blade passage, was generated using CFX-TurboGrid. The total pressure
is specified at the inlet, and the mass flow rate at the outlet. A periodic boundary condition is used to connect
the sides of the domain together. No-slip walls are used for all other boundaries. Each calculation was declared
converged when the maximum normalized residual dropped below 104 .
The head drop-off curve was predicted for a range of flow rates. The flow rate is characterized by the
actual-to-nominal flow ratio, Q/Qn . For each curve, a noncavitating solution was first obtained. The cavitation
model was then activated and the inlet total pressure decreased by 10,000 Pa for each point on the curve. When
the head dropoff became significant, the inlet pressure was dropped in smaller increments of 1000 Pa.
The experimental and predicted dropoff curves (nondimensional pressure rise as a function of cavitation
number) are plotted for three different flow rates in Figure 5. At low flow rates (Q/Qn = 0.79), the pre-
dicted dropoff curve occurs smoothly and slightly before the experimental curve. Close to the design flow
rate (Q/Qn = 1.03), dropoff occurs more rapidly and simultaneously with the experimental measurements.
Agreement between the two results is very satisfactory. At high flow rates (Q/Qn = 1.15), the dropoff curve
occurs rapidly and slightly after the experimental measurements. At this flow rate, the pressure rise is slightly
underestimated, and cavitation-induced blockage through the blade passage is seen to be more extensive in the
experiments than the model predictions. Overall, the agreement is very encouraging.
For one of the flow rates (Q/Qn = 1.03), the predicted shape of the cavitation pockets was compared to
experimental visualizations. The comparison is illustrated in Figure 6. The general development of cavitation in
the inducer is illustrated at three operating points, corresponding to cavitation numbers of 0.09, 0.06, and 0.045.
The vapour first appears near the shroud at the leading edge of the suction side of the blade . As the cavitation
number decreases, the cavity remains attached to the blade but grows into the blade-to-blade channel and down
towards the hub. Finally, the vapour passes to the pressure side of the blade, the blockage becomes extensive,
and the performance breaks down. The plot also illustrates the predicted vapour bubbles corresponding to 10%
volume fraction in red. The hub is coloured in green and the blade in gray. The agreement is quite reasonable,
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152





Calculated, 0.43
Experiment, 0.43
Calculated, 0.38
0 Experiment, 0.38
Calculated, 0.34
Experiment, 0.34
0 0.2 0.4 0.6 0.8 1
Figure 3: Comparison of experimental and computed surface pressure coefficients for midchord cavitation on
hydrofoil for three cavitation numbers





0.2 Calculated, 1.00

Experiment, 1.00
Calculated, 0.91
0 Experiment, 0.91
Calculated, 0.84
Experiment, 0.84
0 0.2 0.4 0.6 0.8 1
Figure 4: Comparison of experimental and computed surface pressure coefficients for leading edge cavitation
on hydrofoil for three cavitation numbers
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

Figure 5: Predicted and experimental head dropoff curves at Q/Qn = 0.79, 1.03, 1.15.
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

Figure 6: Inducer flow visualizations at Q/Qn = 1.03 at cavitation numbers of 0.09, 0.06, and 0.045.

especially the location and evolution of the vapour region.

Transient Venturi Cavitation

The cavitation model was also tested for transient cavitation conditions. A venturi-type geometry was used for
the test approximating that used in the experiments of Stutz and Reboud [14]. In their experiments the venturi
was operated with unsteady shedding of vapour clouds just past the venturi throat. The shedding frequency is
characterized by a Strouhal number, defined as

f Lcav
St = (26)

where f is the frequency, Lcav is the average cavity length, and Vref is the inlet velocity. The Strouhal number
reported in the experiments is 0.27, based on a shedding frequency of 50 Hz, an average cavity length of 45 mm,
and a velocity of 8 m/s.
The current calculations use an inlet velocity of 7.9 m/s, an outlet pressure of 30 kPa, symmetry planes at
the front and back, and a no-slip condition on the venturi walls. A noncavitating solution was used to provide
initial conditions. The cavitation number (based on the average inlet total pressure) is 1.9. A time step of 104 s
is used. The predicted cavity length is 40 mm and average frequency is 55 Hz, yielding an average Strouhal
number of 0.278. The cavitation model parameters were set to Fvap = 0.4 and Fcond = 0.001. In addition, the
turbulence model modification described in Eq. (17) was used with n = 5.
Figure 7 shows a typical pressure response at a point 0.26 m downstream of the throat, indicating a nearly
regular perturbation of the flow field as vapour clouds are advected downstream. When a vapour cavity passes
the probe, the vapour pressure is plotted. The volume fraction variation with time at the same location is
also provided. Figure 8 shows a typical vapor distribution at a fixed time, along with a cross-hair indicating
the location where the readings for Figure 7 are taken. It should be noted that with the turbulence model
modifications required to induce the unsteadiness as well as the cavitation model constants, there is still a
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

Figure 7: Pressure and volume fraction response in venturi at a point 0.26 m downstream of the throat.

significant amount of tuning involved in the prediction of transient flow behavior of this kind. However, it is
encouraging to see that the general flow features and shedding frequency can be accurately predicted.


A new multiphase flow algorithm for predicting cavitation has been presented. The mass transfer rate be-
tween liquid and vapour phases is computed using a model based upon the Rayleigh Plesset equation. It has
been implemented in a conservative manner in an element-based finite volume method, and features careful
linearization and coupling behaviour in order to obtain good convergence behaviour.
Three validation examples have been provided. First, flow around a hydrofoil, with cavitation induced at
both the leading edge and midchord, has been successfully predicted. Second, the flow through an inducer has
been modelled, including the head-dropoff curves. Finally, the transient shedding of vapour bubbles through
a venturi has been modelled. This final case required some retuning of various model coefficients, including
a modification to the turbulent viscosity, but nonetheless indicates that transient cavitating flows can also be
simulated by this model.


[1] F. Bakir, R. Rey, A. G. Gerber, T. Belamri, and B. Hutchinson. Numerical and experimental investigations
of the cavitating behaviour of an inducer. International Journal for Rotating Machinery, 10:1525, 2004.

[2] T. J. Barth and D. C. Jesperson. The design and application of upwind schemes on unstructured meshes.
AIAA Paper 89-0366, 1989.

[3] C. E. Brennen. Cavitation and Bubble Dynamics. Oxford University Press, 1995.

[4] O. Coutier-Delgosha, R. Fortes-Patella, and J. L. Reboud. Evaluation of the turbulence model influence
on the numerical simulations of unsteady cavitation. Journal of Fluids Engineering, 125:3845, 2003.

[5] P. Eisenberg. Cavitation. In Handbook of Fluid Mechanics. McGraw Hill, 1961.

[6] A. G. Gerber. A CFD model for devices operating under extensive cavitation conditions. In International
Mechanical Engineering Congress and Exhibit, 2002.
ICMF 2004 International Conference on Multiphase Flow
Yokohama, Japan, May 30-June 3, 2004
Paper No.152

Figure 8: Snapshot of volume fraction in venturi. The crosshair indicates the probe location used for Figure 7.

[7] F. R. Menter and H. Grotjans. Application of advanced turbulence models to complex industrial flows. In
G. Tzabiras, editor, Advances in Fluid Mechanics: Calculation of Complex Turbulent Flows. WIT Press,

[8] M. Raw. Robustness of coupled algebraic multigrid for the Navier-Stokes equations. AIAA Paper 96-0297,

[9] C. M. Rhie and W. L. Chow. Numerical study of the turbulent flow past an airfoid with trailing edge
separation. AIAA Journal, 21:15251532, 1983.

[10] G. E. Schneider and M. J. Raw. Control volume finite-element method for heat transfer and fluid flow
using colocated variables 1. Computational procedure. Numerical Heat Transfer, 11:363390, 1987.

[11] G. H. Schnerr and J. Sauer. Physical and numerical modelling of unsteady cavitation dynamics. In 4th
International Conference on Multiphase Flow, 2001.

[12] I. Senocak and W. Shyy. Evaluation of cavitation models for Navier-Stokes computations. In ASME 2002
Fluids Engineering Division Summer Meeting, 2002.

[13] Y. T. Shen and P. E. Dimitakis. The influence of surface cavitation on hydrodynamic forces. In Proc. 22nd
ATTC, 1989.

[14] B. Stutz and J. L. Reboud. Experiments on unsteady cavitation. Experiments in Fluids, 22:191198, 1997.