Professional Documents
Culture Documents
FEDSM2002-31011
EVALUATION OF CAVITATION MODELS FOR
NAVIER-STOKES COMPUTATIONS
Inanc Senocak and Wei Shyy
Department of Aerospace Engineering, Mechanics and Engineering Science
University of Florida
Gainesville, FL 32611
inanc@aero.ufl.edu, wss@aero.ufl.edu
ABSTRACT
The predictive capability of three transport equation-based
cavitation models is evaluated for attached turbulent, cavitating
flows. To help shed light on the theoretical justification of these
models, an analysis of the mass and normal-momentum
conservation at a liquid-vapor interface is presented. The test
problems include flows over an axisymmetric cylindrical body
and a planar hydrofoil at different cavitation and Reynolds
numbers. Proper grid distribution for high Reynolds number
cavitating flows is emphasized. Although all three models give
satisfactory predictions in overall pressure distributions,
differences are observed in the closure region of the cavity,
resulting from the differences in compressibility characteristics
handled by each model.
Keywords: CFD, cavitation, turbulence
INTRODUCTION
Navier-Stokes computations of turbulent cavitating flows
have received a growing attention due to advances in
computational capabilities and physical understanding for these
problems. The cavitating flow problem is complicated due to
high Reynolds number and multiphase nature of the flow with
large disparity between fluid properties of each phase. The
inception and concurrent development of a cavity into other
stages of cavitation, namely sheet, cloud, vortex and
supercavitation, is driven by the phase change due to
hydrodynamic pressure drop and bubble dynamics [1]. To date,
no computational model can offer comprehensive capabilities of
all these flow regimes based on first principles.
Due to changes in fluid properties and physical
mechanisms across the liquid-vapor boundaries, conventional
computational algorithms of single-phase incompressible flow
experience severe convergence and stability problems for
cavitating flows. To remedy this situation, improved numerical
methods have been proposed. In the context of density-based
methods, the artificial compressibility method has been applied
with special attention given to the preconditioning technique [25]. The preconditioning technique is guided by examining the
eigenvalues of system and may not be unique depending on the
forms of the governing equations [6]. Following the spirit of the
well-established SIMPLE algorithm [7], a pressure-based
method for turbulent cavitating flows has been developed [8]. It
is shown that the pressure correction equation for turbulent
cavitating flows shares common features of high speed flows
and a pressure-density coupling scheme is developed that
results in a unified incompressible-compressible formulation.
Upwinded density interpolation is adopted to improve mass and
momentum conservation in cavitating regions [8]. Highresolution non-oscillatory convection schemes have been
applied in both pressure-based and density-based methods [2-5,
8]. The use of such convection schemes is very important
especially in the vicinity of sharp density gradients. Both
methods, pressure-based and density-based, have been
successful to compute turbulent cavitating flows around
axisymmetric bodies and hydrofoils with comparable accuracy.
A common approach in cavitation modeling is to use the
homogeneous flow theory. In this theory, the mixture density
concept is introduced and a single set of mass and momentum
equations are solved. Different ideas have been proposed to
generate the variable density field; a review of such studies is
given in Ref. [9]. Some of the existing studies solve the energy
equation and determine the density through suitable equations
of state [3]. Since most cavitating flows are isothermal,
arbitrary barotropic equations have been proposed to
supplement the energy consideration [10]. Another popular
approach is the transport equation-based model (TEM) [2, 4-6,
8]. In TEM, a transport equation for either mass or volume
fraction, with appropriate source terms to regulate the mass
transfer between phases, is solved. Different source terms have
been proposed by different researchers, which will be discussed
in the next section. A recent experimental finding [11] helps
assess the adequacy of the above-mentioned physical models. It
C prod L2 (1 L )
L t
The empirical factors have the following values, which have
been determined previously [8]. (Cdest=9.0x105, Cprod=3.0x104)
Model-3 (from Ref. [13])
The vapor mass fraction is the dependent variable in the
transport equation. Both evaporation and condensation terms
are functions of pressure. The model equations that are
presented here are slightly different from the ones in the original
paper [13], which considers only a non-condensable gas. No
such a limitation is imposed here.
( m fV )
r
(4)
+ ( m fV u) = ( m
& +m
& +)
t
m& = Cdest
U
2 P P
L V V
3 L
if : P < PV
The final form reads the following after further arranging the
terms.
(5)
U
2 P PV
LV
if : P > PV
3 L
1
f
(1 fV )
(6)
= V +
m V
L
The empirical factors have the following values (Cdest/=1225,
Cprod/=36750).
m& + = C prod
V
V
1
1
PV PL = + + 2V V ,n 2 L L,n
(8)
n
n
R1 R2
+ L (V L ,n V I ,n ) 2 V (VV ,n V I, n ) 2
Figure 1 illustrates a typical representation of a liquid-vapor
interface based on the homogeneous flow theory. The mixture
density is defined as follows based on the liquid volume
fraction [18].
(9)
m = L L + V (1 L )
Assuming that a hypothetical interface lies in the liquid-vapor
mixture region, as illustrated in Fig. 1, and neglecting the surface
tension and viscosity effects, the mass and normal momentum
conservation conditions reduces to the following forms.
(10)
V (VV ,n VI ) = m (Vm ,n V I )
PV PL = m (Vm ,n V I )2 V (VV ,n V I )2
(11)
(V m,n VI ) =
V (VV ,n V I )
m
(12)
(13)
PV PL = V (VV ,n V I ) 2 V 1
m
At this point the definition of mixture density, given in Eq. (9), is
incorporated into the above equation that leads to the following
forms:
(14)
PV PL = V (VV ,n V I ) 2
1
(
1
)
L L
V
L
( V L ) L =
L =
( PV PL ) L L + ( PV PL ) V (1 L )
V (VV ,n V I ) 2
L ( PL PV ) L
( PL PV )(1 L ) (16)
+
2
V (VV ,n V I ) ( L V ) (VV ,n V I ) 2 ( L V )
L ( PL PV ) L
S& = L =
t
(VV ,n V I ) 2 ( L V ) t
1V44
44
4244444
3
(17)
( PL PV )(1 L )
+
2
(VV ,n V I ) ( L V ) t
1444424444
3
L
r
L ( PL PV ) L
+ ( L u ) =
t
V (VV ,n V I ) 2 ( L V ) t
(18)
( PL PV )(1 L )
+
(VV ,n V I ) 2 ( L V ) t
No empirical constants appear in the above equation. The first
term on the right hand side, compared to the second term, is
scaled naturally by a factor of the nominal density ratio ( L/ V).
To utilize the above equation, the vapor phase velocity normal
to the interface (VV,n) and the velocity of the interface (VI) are
needed.
The derivation of the model is based on an existing
interface; hence, conditional statements are required on the
pressure terms in order to couple the model to the flow
computation. Cavitation inception condition suggests that
cavitation incepts once the hydrodynamic pressure drops below
the vapor-pressure value of the corresponding liquid. As seen
from Eq. (17), in the pure liquid phase (L=1) the second term
will be zero. Hence the inception condition is imposed as
minimum (MIN) function on pressure difference term of the first
term of Eq. (17). Then the desinent cavitation condition is
imposed as a maximum (MAX) function on the pressure
difference term of the second term. As a result the first term of
Eq. (17) is responsible for conversion of liquid phase to vapor
phase (evaporation) and the second term of Eq. (17) is
responsible for conversion of vapor phase back to liquid phase
(condensation). With these inclusions, the model equation to be
solved along with Navier- Stokes equations is the following:
(15)
Eq. (19) forms a fundamental cavitation model for NavierStokes computations of cavitating flows. Unlike the existing
transport equation-based models in the literature, the proposed
model does not require empirical constants to regulate the mass
transfer process. It is general to handle the time-dependency
since the interface velocity (VI) is taken into account in the
model. The model requires that an interface be constructed in
order to compute the interface velocity for time-dependent
computations, as well as the normal velocity of the vapor phase.
However in sheet type of attached cavitation the cavity is often
modeled in a steady-state computation and the assumption can
be valid and does produce satisfactory results [4, 5, 8]. Hence
the interface velocity (VI) is zero for steady-state computations
and the normal velocity of the vapor phase can be computed by
taking the gradient of the liquid volume fraction [19, 20]. The
vapor phase normal velocity is the dot product of the velocity
and the normal vector.
L
(20)
VV ,n = u n
n=
L
RESULTS
Two flow configurations have been considered, namely, (i)
an axisymmetric cylindrical object with a hemispherical
forehead (referred to as hemispherical object) at a Reynolds
number of 1.36x105 and a cavitation number of 0.30, and (ii)
the NACA66MOD hydrofoil at an angle of attack of 4 with
Reynolds number of 2.0x106. It is experimentally observed that
sheet (attached) cavitation occurs for both of the geometries
under given conditions and time averaged experimental data of
pressure distribution along the surface is available [21, 22].
Hence steady-state computations are carried out in this study.
Figure 2 compares the performance of the cavitation
models for the hemispherical object at a cavitation number of
0.30.
All three models match the experimental data
satisfactorily. Differences in the performance are more
noticeable in the closure region, where the vapor phase
condenses. Figure 3 shows the corresponding density
distribution along the surface. As seen from density plots, the
liquid phase first expands and vapor phase appears uniformly
inside the cavity, then the vapor phase compresses, in a shock
like fashion, back to the liquid phase. The differences among
the three models in density profiles are significant. This implies
that each model generates different compressibility
characteristics, although they produce very similar steady-state
pressure distributions. This issue has important implications in
time dependent problems; which model produces the correct
compressibility is an open question and needs further
investigation. Figure 4 shows the distribution of density
throughout the cavity and the spanwise vorticity distribution.
The interface is captured sharply with Model-2 and Model-3
compared to Model-1, especially in the downstream region of
the cavity. As seen from the spanwise vorticity distribution, also
given in Fig. 4, there is additional generation of vorticity at the
[22]
[23]
[24]
Model-1
Model-2
Model-3
0.8
density
[21]
0.6
0.4
0.2
0.5
1.5
2
s/D
2.5
3.5
Model-1
Model-2
Model-3
1.2
Model-1
Model-2
Model-3
Exp. data
1
0.8
Cp
0.6
0.04
0.04
GRID-A
0.4
GRID-B
0.035
0.035
0.03
0.03
0
-0.2
-0.4
0
0.2
0.025
0.025
Cavity edge
0.5
1.5
2.5
3.5
s/D
0.02
-0.23
-0.225
-0.22
-0.215
-0.21
0.02
-0.23
-0.225
-0.22
-0.215
-0.21
1.5
Grid-A
Grid-B
Exp. data
Model-1
Model-2
Model-3
Exp. data
0.8
0.6
0.5
-Cp
-Cp
0.4
0.2
0
0
-0.2
-0.5
-0.4
-1
-0.6
0.2
0.4
0.6
0.8
-0.8
X/C
10
0.2
0.4
0.6
0.8
X/C
Grid-A
Grid-B
defect layer
10
y+=400.0
2
Y+
10
log layer
y+=11.63
10
viscous sublayer
0
10
0.2
0.4
0.6
0.8
X/C
Model-1
Model-2
Model-3
Exp. data
1
0.8
-Cp
0.6
0.4
0.2
0
-0.2
-0.4
-0.6
0.2
0.4
0.6
0.8
X/C