You are on page 1of 9

On the modied dispersion-controlled dissipative (DCD) scheme for computation

of ow supercavitation
Z.M. Hu
a
, H.S. Dou
a
, B.C. Khoo
b,
a
Temasek Laboratories, National University of Singapore, Singapore 117411, Singapore
b
Department of Mechanical Engineering, National University of Singapore, Singapore 119260, Singapore
a r t i c l e i n f o
Article history:
Received 4 January 2010
Received in revised form 1 July 2010
Accepted 1 October 2010
Available online 8 October 2010
Keywords:
Flow supercavitation
DCD scheme
Flux splitting
Interface
Cavitation model
a b s t r a c t
In the present study, a robust scheme initially proposed for capturing shock waves in gasdynamics, the
dispersion-controlled dissipative (DCD) scheme is modied to simulate supercavitating ows in the
framework of one-uid model. Due to the stiffness of the equations of state (EOS), the Steger-Warming
method becomes inapplicable and the LaxFriedrichs method should be used for the numerical ux split-
ting. Following the formulation, the updated scheme is validated by a series of comparisons with theo-
retical and experimental data. The validation shows that the modied DCD scheme performs very well
for supercavitating ows in the presence of steep gradients of both density and EOS. Several one-uid
cavitation models and speed of sound (SoS) models are compared for the numerical investigation of
supercavitating ows.
2010 Elsevier Ltd. All rights reserved.
1. Introduction
Bubbles or cavities, which are lled with vapor, may emerge in
the initially homogeneous liquid if the surrounding pressure
reduces to its saturated vapor pressure (SVP). This kind of phase
transition process is referred to as cavitation. The driving mecha-
nism for cavitation is not usually temperature related as for boiling
but a pressure drop generally controlled by the ow dynamics.
Cavitation often happens at an extremely high speed where the
liquid pressure can rapidly drop below the corresponding SVP
although it can occur at any other speeds. In hydrodynamic applica-
tions, cavitation is usually an unintended and undesirable phenom-
enon: the cavitation bubbles typically do not sustain and may
implode as the surrounding ow decelerates. The constant implo-
sionof these small-scalecavitationbubbles causes material damage.
As such, a key subject of decades of research on cavitation has been
to prevent its inception or to ensure control of its evolution. On the
other hand, the cavitation effects can be harnessed for useful
applications. The induced cavity over an underwater object moving
at a sufcient speed can extend as a single large bubble of vapor to
envelop the entire object, hence leading to ow supercavitation.
Because the viscosity and density of vapor respectively are two
and ve orders of magnitude lower than those of water, ow
supercavitation can lead to drastic reduction of the hydrodynamic
drag.
Research attention on ow supercavitation has been largely
drawn towards issues covering the unsteadiness of supercavitating
bubbles, the cavity instability due to vehicle planning or tail slap-
ping, and the interaction between cavity and ventilation or propul-
sion exhaust. Besides experimental and theoretical studies,
numerical analysis has become a viable and promising approach
for understanding the fundamentals of supercavitating ows. The
main numerical difculty in treating supercavitating ows, among
other multiphasic ows, arises in dealing with the dynamic inter-
faces. The presence of large discontinuities in the thermodynamic
properties and equations of state (EOS) at material interfaces may
result innumerical instabilities andnonphysical oscillations. Gener-
ally, there have been developed two classes of methods to overcome
these difculties. One of them is Sharp Interface Method (SIM) in
which the interface is numerically treated as a sharp discontinuity.
The other is Diffuse Interface Method (DIM) in which the interface
is considered as a diffuse zone, like a contact discontinuity in gasdy-
namic ows [1].
The popular class of SIM is based on the well known level-set
function [2] to locate the interface in the framework of the front
capturing methods (FCM) which have been successfully applied
in gasdynamic ows. To reduce numerical smearing and to prevent
nonphysical oscillations across the material interfaces, Fedkiw
et al. proposed the original ghost uid method (GFM) [3]. However,
the original GFM was found to admit inaccuracies such as non-
physical oscillations when especially applied to multi-uid
0045-7930/$ - see front matter 2010 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compuid.2010.10.001

Corresponding author.
E-mail addresses: tslhz@nus.edu.sg (Z.M. Hu), mpekbc@nus.edu.sg (B.C. Khoo).
Computers & Fluids 40 (2011) 315323
Contents lists available at ScienceDirect
Computers & Fluids
j our nal homepage: www. el sevi er. com/ l ocat e/ compui d
problems, e.g., the impacting of a strong shock wave on a gas
water interface. As such, a modied ghost uid method (MGFM)
has been proposed by Liu et al. to deal with such limitations of
GFM with a series of successful applications [4,5]. The MGFM
was reported to satisfy the entropy condition across the material
interfaces. A different SIM approach, front-tracking method
(FTM), has been proposed to use explicit marker points to capture
waves and discontinuities [6,7]. However, this method also suf-
fered from numerical diffusion or spurious oscillations at the inter-
faces under some extreme conditions. In addition, as the uid
interfaces are explicitly tracked by connected marker points, FTM
becomes very complicated for reconstructing complex interfaces
and discontinuities. Unfortunately, none of these SIM methods is
deemed able to satisfactorily resolve the dynamically emerging
interfaces separating multiple uids [1].
The Diffuse Interface Method (DIM) considers an interface as a
numerically diffused zone. This method has been proved to be able
to solve the dynamical appearance of multi-uid interfaces such as
the cavity boundary which is absent in an initially homogeneous
liquid [1]. The challenge for this class of methods arises from deriv-
ing a physically and mathematically consistent EOS for the multi-
phasic uids in the numerically diffused zone. Representative
studies using DIM are performed by Abgrall [8], Saurel and Abgrall
[9], Murrone and Guillard [10], Coutier-Delgosha et al. [11], Liu
et al. [12], and Goncalves and Patella [13] among others. For
cavitating ows, the two-phase methods can be further divided
into three categories: one-uid model or homogenous model,
two-uid model, and hybrid model.
The two-uid model [14,15] assumes that both phases coexist
everywhere in the oweld and each phase ow is governed by
its own set of partial difference equations. Because the exchange
of mass, momentum and energy is explicitly treated in this
approach, the two-uid model can easily take into account the
physical details occurring at the interface. However, the two-uid
model is less often employed in unsteady cavitation ows because
of the dearth of knowledge on the parameters associated with
phase transition rates. A one-uid model, on the contrary, treats
the cavitating ow medium as a mixture of different uids, how-
ever, behaving as one. These models are based on the assumption
that the phases are in locally kinetic and thermodynamic equilib-
rium; i.e. unique velocity, temperature and pressure for both
phases. Therefore, one set of differential equations similar to the
single-phase ow is used to govern the whole multi-uid ow.
Models of this kind have been extensively used to simulate cavitat-
ing ows (for instance see [1113,1618] among others) because
they are relatively easy to treat the dynamic inception and collapse
of cavitating bubbles. However, physical phenomena relating to
vaporization, condensation or other phase transition processes
are assumed to complete instantaneously. Therefore, the one-uid
model cannot reect any strong thermodynamic or kinetic non-
equilibrium effects if present. Hybrid models are intermediate
models between the one-uid and two-uid models. In these mod-
els, each phase is governed by its own continuity equation with a
source term accounting for the phase exchange rate. Other local
kinetic and thermodynamic properties are treated the same way
as in one-uid models. The hybrid models are employed in simu-
lating cavitating ows (see [1921] among others). Some difcul-
ties lie in the determination of the source terms for phase
transition.
In the present study, a numerical scheme, dispersion-controlled
dissipative (DCD), is modied to simulate supercavitating ows in
the framework of the one-uid model and Diffuse Interface Meth-
od (DIM). The DCD scheme was proposed by Jiang et al. intending
for shock wave capturing [22,23]. The primary principle of the
scheme aims at removing non-physical oscillation across strong
discontinuities by making use of the dispersion characteristics of
the modied equation instead of adding articial viscosity. The
DCD scheme has also been applied successfully to chemically
reacting ows for compressible multi-component mixtures [24]
and ows with strong shock wave interactions [25,26]. Such men-
tioned applications have shown that the DCD scheme is fairly
robust, computationally efcient, and capable of resolving strong
discontinuities. The paper is organized as follows. In Section 2,
we have the numerical models and the modied DCD scheme. This
is followed by Section 3 on validation of the assembled code via
comparing with theoretical and experimental results. Several cav-
itation models are compared for modelling supercavitating ows.
Finally, a brief summary is given in Section 4.
2. Numerical models and DCD scheme formulation
2.1. Governing equations and numerical models
In the present study, the supercavitating ow is assumed to be
isothermal, along with the locally homogeneous and equilibrium
assumption. Therefore, the vapor and liquid phases share the same
pressure, temperature and velocity while the energy equation can
be avoided in the governing equations. In two-dimensional Carte-
sian coordinates, the compressible laminar NavierStokes (NS) sys-
tem reads:
@U
@t

@F
@x

@G
@y

i
y
S
@Fv
@x

@Gv
@y

i
y
Sv 1
For planar and axisymmetric ows, the control parameter i takes
value of 0 and 1, respectively. Here, the convective vectors U, F, G
and the geometric source term S are
U q; qu; qv
T
2
F qu; qu
2
p; quv
T
3
G qv; quv; qv
2
p
T
4
S qv; quv; qv
2

T
5
In the above NS equations, the viscous vectors, Fv, Gv, and Sv, are
not listed for conciseness. The viscosity coefcient for the two-
phase medium is l = al
v
+ (1 a)l
w
, where l
v
and l
w
are the vis-
cosity coefcients for pure vapor and water at a given temperature.
The void fraction, a
qq
sw
q
sv
q
sw
, varies from zero to unity between pure
liquid phase to pure vapor condition. Here q
sv
and q
sw
denote the
vapor and water densities in the mixture and will be explained
later. In the governing equations, q, p, u and v denote the average
density, pressure, velocities in x and y directions, respectively. To
close system Eq. (1), an equation of state (EOS) is necessary to
determine pressure based on other ow variables. One of the
challenges in one-uid methods is to dene an appropriate EOS
applicable for the vapor, liquid phases and the mixture condition
in the numerical diffused zone. For instance, there are different ver-
sions of barotropic EOS that directly link pressure p to density q for
modelling cavitating or supercavitating ows. Examples are the cut-
off model, Schmidts model [16], isentropic one-uid model [12],
sinusoidal model [13], and so on. Here, they are plotted in Fig. 1
for an isothermal cavitation process in initially homogeneous water
at T
0
= 295 K and p
0
= 101,325 Pa. The Schmidts model [16] gives
negative pressure under this condition and is not represented on
the gure. Here, the saturation line is obtained using the formulae
provided by IAPWS [27]. The abbreviations sw and sv correspond
to the saturated water and saturated vapor conditions, respectively.
Several of the one-uid models which will be calibrated for ow
supercavitation in the present study are briey documented as
follows.
The cut-off model of EOS, as shown by the solid line in Fig. 1,
xes the pressure of the two-phase mixture at the SVP value once
316 Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323
the water ow reaches the saturated condition. If the Taits EOS is
employed for the pure water, the cut-off EOS can be written as
p
B
0
q
q
sw
_ _
N
B
0
A
0
; q > q
sw
p
s
; q
sw
Pq Pq
sv
p
s
q
q
sv
_ _
c
; q < q
sv
_

_
6
where B
0
, A
0
and Nare model constants of the Taits EOS while c is the
specic heat ratio of vapor. p
s
, q
sw
and q
sv
denote the saturated
pressure, saturatedwater density andsaturatedvapor density at a gi-
ven temperature, respectively. At T
0
= 295 K, B
0
= 3.07 10
8
Pa,
A
0
= p
s
= 2621 Pa, N = 7.15,c = 1.31, q
sw
= 997.76 kg/m
3
, q
sv
=
0.0193 kg/m
3
.
In the isentropic one-uid EOS [12], the two-phase mixture EOS
can be calculated via Newtons iteration algorithm from
q
kq
cav
v
q
cav
w
pB
0
A
0
p
cav
B
0
A
0
_ _
1=N
k
p
p
cav
_ _
1=c
7
where q
cav
v
and q
cav
w
respectively denote the associated vapor and
water densities at the cavitation pressure p
cav
. k = a
0
/(1 a
0
),
where a
0
is the known void fraction of the mixture at p
cav
. It should
be noted that the original isentropic model was developed assum-
ing that the cavity boundary is a strong discontinuity, where a
jumps from zero to a
0
across the cavity interface. The ow is judged
as pure water when a < a
0
. However, the natural cavitation process
is considered continuous for the DIM method as used in the present
work, i.e., cavitation is assumed to start immediately when a
exceeds zero. Otherwise, a negative pressure will occur if the Taits
EOS as given in Eq. (6) is used when 0 < a < a
0
. On the other hand, an
innite pressure will be obtained if the isentropic one-uid model is
employed in this region. Therefore, unlike the original one, the
currently used isentropic model as plotted in Fig. 1 switches to
the cut-off model when a < a
0
. Details on the original isentropic
one-uid model can be found in the literature [12].
The sinusoidal EOS was further developed by Goncalves and
Patella [13] from its original version to avoid the innite SoS. This
model in conjunction with the Taits EOS for the pure water can be
described as
p
B
0
q
q
sw
_ _
N
B
0
A
0
; p > p
s
Dp
p
s

q
sw
q
sv
2
_ _
c
2
min
sin
1
A1 2a; p
s
Dp Pp Pp
s
Dp
_
_
_
8
where the parameters A and c
min
can be determined by the continu-
ity relation between the pure water and the mixture for a given sat-
uration condition. The quality of the problem-related parameter,
DP, has to be smaller than p
s
to avoid negative pressure.
One of the purposes of cavitation model is to obtain a positive
speed of sound (SoS) a via ensuring that
dp
dq
> 0 inside the two-
phase domain. The Walliss SoS [28] for the isentropic one-uid
EOS can be written as
a q
a
q
sv
a
2
v

1 a
q
sw
a
2
w
_ _ _ _
1=2
9
where the density of the two-phase mixture is expressed by linearly
combining the vapor phase density and the water phase density as
q aq
sv
1 aq
sw
10
Here, a
v
and a
w
are the speeds of sound of the pure vapor and water
at the given temperature, respectively. The SoS for the sinusoidal
EOS can be computed as
a c
min
A

1 A
2
1 2a
2
_
_

_
_

_
1=2
11
From Fig.1 and Eq. (6), it is clear that the cut-off model does not en-
able a positive SoS and an articial model of SoS should be strictly
provided to get a hyperbolic system of governing equations.
The well known Wallis [28] and sinusoidal [13] models of SoS
are depicted in Fig. 2 by the solid and dashed lines, respectively.
Both models are widely used in the modelling of low-speed
cavitating ows in hydrodynamics. However, these two SoS mod-
els exhibit a non-monotonic variation with the void fraction
which likely poses serious computational challenges. For exam-
ple, to model high-speed supercavitating ows especially in the
light of applying the class of DIM for interface resolution, these
models can lead to numerical inconsistency; i.e., the ow inside
the numerically diffused zone may be hypersonic due to the very
low value of SoS. This can thus lead to a delay in the wave trans-
mission through the interfaces [1]. It may be noted that both
Wallis [28] and sinusoidal [13] models may not present any dif-
culty when one employs the SIM method for interface resolu-
tion. The SoS can alternatively switch between the water and
vapor speeds of sound across the cavity boundaries because there
is a clear and sharp interface between phases in a SIM solution.
To circumvent this difculty associated with the employment of
p
(
P
a
)
10
-2
10
-1
10
0
10
1
10
2
10
3
10
-4
10
-2
10
0
10
2
10
4
10
6
10
8
0
Cut-off cavitation EOS
Taits water EOS
Saturation line (IAPWS)
Barotropic vapor EOS
Sinusoidal cavitation EOS
Isentropic cavitation EOS

p
s
v
3
(kg/m )
s
w
Fig. 1. Homogeneous EOS models for the simulation of cavitating ows.
(
m
/
s
)
10
0
10
1
10
2
10
3

Model of speed of sound


Wallis
Sinusoidal
Frozen
a
0 0.2 0.4 0.6 0.8 1
Fig. 2. Models for the speed of sound (SoS): Wallis [28], Sinusoidal [13] and frozen
SoS.
Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323 317
DIM, a frozen speed of sound is introduced. After substituting
the isentropic vapor EOS and the Taits EOS into Eq. (10) one
can get
q aq
sv
p
p
s
_ _
1=c
1 aq
sw
p B
0
A
0
B
0
_ _
1=N
12
Under the assumption that the composition of the two-phase mix-
ture, i.e., a, does not vary with a innitesimal change of pressure,
the frozen SoS then can be written as
a
dp
dq
_ _
1=2

aq
sv
cp

1 aq
sw
Np B
0

_ _
1=2
13
It may be noted that the present frozen model of speed of
sound is different from the frozen SoS dened for the non-
equilibrium 6-equation model [1]. As plotted by a dasheddotted
line in Fig. 2, the frozen SoS continuously and monotonously
varies from SoS of water at about 1483 m/s to that of vapor at
about 425 m/s as a increases from zero to unity. Other detailed
properties of the various one-uid cavitation models and the
related discussions on the physical characteristics can be found
in [12].
One more technical difculty in the numerical implementa-
tion of compressible one-uid model arises from the stiffness
(or incompressibility) of the EOS for the pure water. From the
initial condition where T
0
= 295 K and p
0
= 101,325 Pa to the
saturated water state where p
s
= 2621 Pa, the change of water
density is extremely small from q
0
= 997.81 kg/m
3
to q
sw
=
997.76 kg/m
3
. As a consequence, the pressure eld is subjected
to round-off errors during the simulation of the ow cavitation.
For numerical computations of low-speed cavitating ows like
the work of [13], the stiffness can be articially eliminated via
employing a very small value of SoS. However, this approach is
not feasible for high-speed cavitating ows where the compress-
ibility is not negligible. In the present study, a lter function is
introduced to re-initialize the density of the water phase by q
0
whenever the density change is less than a specied value as
following
q
q
0
; jq q
0
j < 0:02q
0
q
sw

q; else
_
14
Obviously, the re-initialization only works for the water phase in
the vicinity of the initial density.
2.2. Modication of DCD scheme for ow supercavitation
The DCD scheme [22,23] for the simulation of high-speed gas-
dynamic ows is different from other conventional shock-captur-
ing schemes. It satises the dispersion conditions to ensure the
solution is free of nonphysical oscillations across shock waves or
contact surfaces inside the complex ow-elds. The attractive fea-
ture of the scheme is that it can systematically regulate the scheme
dissipation effect according to the discontinuity intensity. Thus,
over- or under-dissipated solutions can be avoided without intro-
ducing any free parameters which are generally problem depen-
dant. More details on the DCD scheme can be found in the
review work [23].
The second-order DCD scheme for the convective part of Eq. (1)
in two-dimensional curvilinear coordinates (n, g) can be written as
follows:
@

U
@s
_ _
conv
i;j

1
Dn
F
i
1
2
F
i
1
2
_ _

1
Dg
G
j
1
2
G
j
1
2
_ _
15
where
F
i
1
2

i
1
2
;j

i
1
2
;j
G
j
1
2

i;j
1
2

i;j
1
2

i
1
2
;j


F

i;j

1
2
minmod

F

i;j

i1;j
;

i1;j

i;j
_ _

i;j
1
2

i;j

1
2
minmod

G

i;j

i;j1
;

G

i;j1

i;j
_ _

i
1
2
;j


F

i;j

1
2
minmod

F

i2;j

i1;j
;

i1;j

i;j
_ _

i;j
1
2

i;j

1
2
minmod

G

i;j2

i;j1
;

G

i;j1

i;j
_ _
The symbol indicates the respective vectors in the computa-
tional space. It can be seen from the above equations that the
DCD scheme belongs to the scheme family based on the ux split-
ting method. In the original DCD scheme [22,23], the Steger-Warm-
ing (SW) method was suggested to obtain the split uxes

F

or

G

.
For the gasdynamic ows with an ideal gas EOS, the method works
well and very efciently [2426]. Based on the Steger-Warming
method, the split ux for the ux vector,

F n
x
F n
y
G, in the com-
putational space can be written as


q
2
~
k

2

~
k

3
u k
x
a
~
k

2
u k
x
a
~
k

3
v k
y
a
~
k

2
v k
y
a
~
k

3
_

_
_

_: 16
Here, k

i

1
2
k
i

k
2
i
e
_
_ _
, (i = 1, 2, 3, 0 < e1). The three eigen-
values of the Jacobian matrix

A @

F=@

U are
~
k
1
n
x
u n
y
v;
~
k
2

n
x
u n
y
v ha, and
~
k
3
n
x
u n
y
v ha, respectively. Here,
h

n
2
x
n
2
y
_
, k
x
= n
x
/h and k
y
= n
y
/h where n
x
and n
y
are components
of transformation matrix between the physical and computational
coordinate systems, i.e., (x, y) and (n, g). The speed of sound a can
be one of the forms as listed in the previous subsection. For cavitat-
ing ows however, the above eigenvalue splitting method results in
mathematical inconsistency, i.e., the ux

F cannot be recovered
simply via

F

. The reason of the failure of SW method may


be associate with the stiffness of the barotropic EOS. On the other
hand, the LaxFriedrichs ux splitting method is applicable and
the numerical ux can be written as follows,


q
2
~ u
~
k
m
~ u~ u
~
k
m
n
x
p
~ v~ u
~
k
m
n
y
p
_

_
_

_;

G


q
2
~ v
~
k
n
~ u~ v
~
k
n
g
x
p
~ v~ v
~
k
n
g
y
p
_

_
_

_: 17
Here,
~
k
m
and
~
k
n
are the maximum absolute values of the eigenvalues
of the Jacobian

A @

F=@

U and

B @

G=@

U, respectively. It is easy
to note that the ux

F can be recovered by

F

. The reason that


LaxFriedrichs ux splitting is applicable as opposed to the Steger-
Warming approach for computing cavitating ows is that the split-
ting operation is not required for the pressure term for the former.
In general, the attractive feature of DCD scheme for computational
gasdynamics is that it can achieve non-oscillatory solutions without
the need of adding articial viscosity or otherwise. This feature is
also very critical to capture the cavity boundaries accurately in
the presence of extremely sharp gradients of ow density or other
thermodynamic properties.
3. Numerical results and discussion
3.1. 1-D natural cavitation
A one meter-long tube is initially lled with pure water at the
atmospheric pressure with density q
1
= 997.81 kg/m
3
. An initial
318 Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323
velocity discontinuity is located at x = 0.5 m. The velocity is set as
u = 100 m/s on the left and u = 100 m/s on the right. The cut-off
cavitation model and the frozen SoS model are used in this
computation. The results obtained on a uniform grid with 1000
nodes are shown in Fig. 3 at time t = 1.85 ms. As strong rarefaction
waves propagate in the tube, evaporation occurs resulting in a
cavitation pocket. This indicates that the present algorithm is able
to handle the dynamic appearance of two interfaces that are absent
initially.
3.2. Flow supercavitation and validation
In the following computations for validation, the ow is
assumed laminar and axisymmetric. The freestream temperature
of water is 295 K, and the cavitation process is assumed isother-
mal. For the convective part of Eq. (1), the DCD scheme as
described above is used for all the computations. A second-order
central difference scheme is applied to discretize the viscous terms,
and a third-order RungerKutta algorithm is used for the time-
marching procedure. In addition, because of the presence of extre-
mely steep gradients of density and EOS across the supercavity
boundaries, a small CFL number is used, i.e., CFL = 0.1.
The rst case for validation follows the experimental work of
Hrubes [29] for high-speed supercavitating underwater projectiles.
The experiment was conducted in a test range submerged at 4 m
depth in a fresh water tank. A disc cavitator of r = 0.71 mm in ra-
dius was mounted at the head of the conical shaped projectile.
The projectile travelled at 970 m/s corresponding to a Mach num-
ber of 0.65 and a very small cavitation number r = 2 10
4
. A
fairly uniform and transparent cavity was obtained in the experi-
ments [29] under such conditions. Here, the cavitation number is
dened as
r
p
1
p
c
1
2
q
w
V
2
1
18
where p
1
, p
c
, q
w
and V
1
are the ambient water pressure, cavity
vapor pressure, density of the water and the cruising speed of the
cavitator, respectively. A multi-block axisymmetric grid is used
with 137,000 total nodes over a computational domain of
360r 140r. The mesh is locally rened near the projectile surfaces.
Fig. 4a and b shows a visual comparison between the computational
shadowgraph and experimental shadowgraph from Hrubes [29].
The cut-off cavitation model supplemented with the frozen SoS
(see details on Figs. 1 and 2) is used in the simulation to obtain
the computational shadowgraph. The MunzerReichardt [30] theo-
retical solution, Hrubess experimental data (obtained from the
shadowgraph) [29] along with the numerical density contour are
shown together in Fig. 4c. The coordinates are non-dimensionalized
by the radius of the cavitator, r. In general, the computational result
agrees well with experimental data and theory. It can be seen from
(a) that a slight angle of attack in the test results in a slight asym-
metry of cavity shape; this reects one of the deciencies in exper-
imental research on the supercavitation of underwater projectiles.
Therefore, the data in (c) which are obtained from the shadowgraph
x (m)
-0.4 -0.2 0 0.2 0.4
0
0.2
0.4
0.6
0.8
1

Density at saturated condition


x (m)
-0.4 -0.2 0 0.2 0.4
0
200
400
600
800
1000
1200
1400
a
(
m
/
s
)
x (m)
-0.4 -0.2 0 0.2 0.4
-1
-0.8
-0.6
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
u
/
u

x (m)
-0.4 -0.2 0 0.2 0.4
0
0.2
0.4
0.6
0.8
1

Fig. 3. One-dimensional natural cavitation showing the proles of density, speed of sound, velocity and void fraction (Cavitation model: cut-off and frozen SoS).
Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323 319
[29] entail experimental uncertainties. On the other hand, the def-
inition of the cavity shape by the MunzerReichardts model [30]
does not include any information regarding the diffused zone as
for the present DIM computations. As such, the comparison is
broadly qualitative rather than quantitative. Nevertheless, the com-
putational cavity boundary as determined by the maximum density
gradient conforms better with the experiment than the theory. The
MunzerReichardts model for an axisymmetric cavity shape is an
early model based on the potential ow assumption and can be de-
scribed as
R
c
x R
max
4x
L
max
1
4x
Lmax
_ _
_
_
_
_
1=2:4
19
Here R
c
, is the cavity radius at location x along the centerline of cav-
ity originating from the cavitator nose. The maximum radius R
max
and the length L
max
[31] of the cavity are respectively given as
R
max
r

C
d
r=r
_
L
max
2r

C
d
r=r
2
ln1=r
_
20
where r, C
d
and r are the cavitator radius, the cavitator drag coef-
cient and the cavitation number, respectively. The hydrodynamic
drag coefcient for a disc-shaped cavitator at a zero angle of attack
is given by C
d
(r) = C
d0
(1 + r), where C
d0
= 0.815.
Fig. 5 shows the cavity shape over a high-speed projectile
travelling at a much higher speed of 1530 m/s corresponding to a
transonic Mach number of 1.03 in the water. The cavitation model
used in the computation is again the cut-off model. The numerical
ow structure looks fairly similar to that shadowgraph given by
Hrubess experiment [29]. The bow shock wave is well captured
in the simulation. Due to deformation of the transonic projectile
and the lack of higher quality images, the experimental result
shown in (a) is of relatively low delity.
Fig. 6 shows the comparison of cavity shape for a cylinder pro-
jectile travelling at a relatively low speed of 100 m/s. Here, the ow
condition corresponds to a small Mach number of about 0.07 and a
cavitation number r = 0.02. The computational results agree rea-
sonably with the theoretical solution obtained via the Munzer
Reichardts model over a disc cavitator.
3.3. Cavitation model effects
Different EOS models are tested and compared for the supercav-
itating ow over a high subsonic underwater projectile at V
1
=
970 m/s. First of all, it may be mentioned that an attempt to apply
the sinusoidal model fails in which a severe density oscillation oc-
curred. This failure indicates that the sinusoidal model may be not
Fig. 4. Supercavitating ow over a high-speed underwater projectile, V
1
= 970 m/s:
(a) Experimental shadowgraph [29]; (b) numerical shadowgraph; (c) comparison of
cavity shape among theory, experiment, and the computation with cut-off model.
Fig. 5. (a) Experimental shadowgraph [21,29] and (b) numerical density contour for
a transonic projectile travelling at Mach 1.03 (cavitation model: cut-off).
Fig. 6. Supercavity over a cylinder projectile travelling at 100 m/s (cavitation
model: cut-off).
320 Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323
suitable or appropriate for high-speed supercavitating ows
although it has been proved to perform well in computations of
low-speed cavitating ows [13]. On the other hand, both the com-
putational cavitation proles obtained via the isentropic [12] and
cut-off models in conjunction with the frozen SoS are closer to
the experiment data than the MunzerReichardt theory as shown
in Fig. 7. In this gure, each strip corresponds to the numerical dif-
fused zone covering the range of void fraction 0.1 6 a 6 0.9,
respectively. Moreover, the cut-off model results in a relatively
narrower diffused zone and shows better performance than the
isentropic cavitation model.
To investigate the effects of SoS model on the cavity shape, for
the same ow condition as in Fig. 7, the cavity boundary obtained
via Walliss SoS is compared to the frozen SoS model solution as
presented in Fig. 8; the computation is based on the cut-off cavita-
tion model. It is clear that the former cannot reach the correct cav-
ity shape. In addition, the ow Mach number nonphysically
exceeds 200 inside the numerical diffused zone for the Walliss
model solution due to the presence of extremely small speed of
sound. Fig. 8) shows the total pressure proles at x/r = 100 for both
computations. The Walliss SoS simulation results in high-
frequency oscillations in the water region. In the cavity, there
appears to be no oscillation. Saurel et al. pointed out that the
numerical diffused zone created by using an equilibrium SoS mod-
el like the Walliss may have serious consequences regarding wave
propagation [1]. The speed of sound in the cavity which was care-
fully measured in an experiment [32] was found to be between
those of the water and vapor thereby implying the frozen SoS is
applicable. The obstructed exchange of information via wave
propagation/interaction between the cavity and the ambient water
ow appears to be the primary cause to the under-predicted cavity
shape as shown in Fig. 8a. As such, the Walliss model of SoS is
numerically inapplicable to the present DIM computations of
high-speed supercavitating ows under the framework of the
one-uid cavitation model.
3.4. Flow regime effects
Next, the interest is to examine the effects of ow regime on the
supercavity geometry in a two-dimensional planar ow versus axi-
symmetric ow. The computation is based on the same geometri-
cal setup and ow conditions as for Fig. 7. The planar ow
simulation indicates a considerably larger cavity as shown on the
upper half for a comparison to the axisymmetric ow solution de-
picted on the lower half in Fig. 9. This great difference implies that
the cavity shape is highly sensitive to the different ow regimes,
i.e., whether planar or axisymmetric ow conguration. Our future
work will involve a full three-dimensional conguration where the
underwater projectile is mounted at different angles of attack; it is
surmised that the effect on the cavity geometry is likely to be pro-
found judging from the above preliminary study of planar versus
axisymmetric ow regime at the same zero angle of attack.
3.5. Partial cavitation
It should be pointed out that the laminar ow assumption will
breakdown at the rear portion of the cavity. In the cavity closure
region, very complicated ow phenomena associating with diverse
uid physics occur, e.g., collapse of vapor bubbles as accompanied
by small scale implosions, collision and coalescence of droplets,
condensation, and turbulence, etc. As such, the computational
domain used in the above cases has to be appropriately selected
to exclude the closure region of a supercavity. However, a partial
cavitation will occur over the projectile if its travelling speed is
not sufciently high resulting in a cavity closure on the surface
of the projectile. In such a circumstance, laminar ow simulation
may not be applicable. To demonstrate the limitation of the pres-
ent numerical techniques, the computational result for a partial
axisymmetric cavitation ow at r = 0.3 is shown in Fig. 10. The
case is water owover a cylinder geometry with experimental data
from Rouse and McNown [33]. Details of other ow conditions
x
y
0 100 200
0
10
20
/
r
/r
Projectile
Munzer Reichardt Theory
Experiments (J.D. Hrubes)
Cut-off model
Isentropic model
0.1 0.9 }
Fig. 7. Cavitation model comparison for the simulation of the supercavitating ow
over a high-speed underwater projectile, V
1
= 970 m/s (Experimental data [29];
Cavitation model: cut-off v.s. isentropic model [12]; The radial coordinate is
magnied for clarity).
p /
y
/
r
0.9 0.92 0.94 0.96 0.98 1 1.02
10
20
30
40
50
60

tot
Wallis SoS
Frozen SoS
0.5 V


2
(a)
(b)
Fig. 8. Density contour (a) and total pressure p
tot
prole at x/r = 100 (b) showing the
comparison between the Wallis (lower half) and frozen (upper half) SoS models
(cavitation model: cut-off).
Fig. 9. Flow regime effects: planar ow (upper half) v.s. axisymmetric ow (lower
half) (V
1
= 970 m/s; cavitation model: cut-off).
Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323 321
were not clearly specied in their report. Here, the water is
assumed to be under the standard condition, i.e., q
1
= 997.81 kg/
m
3
and p
1
= 1 atm, in the present simulation. In Fig. 10,
Cp
pp
1
0:5q
1
V
2
1
is the pressure coefcient, d is the diameter of the
cylinder, and s is the surface distance from the nose. The cavity is
not well predicted as compared to the experimental data [33] at
the same cavitation number. In particular, the ow phenomena
in the vicinity of the cavity closure have not been resolved due
the barotropic EOS used in the simulation. In previous studies on
partial cavitation where two-uid [15,20] or hybrid [19,21] cavita-
tion models were used, computations and experimental data agree
reasonably. This may be attributed to the tunable phase transition
rate which is problem dependent as employed in their cavitation
modelling.
4. Conclusions
Based on a robust scheme which was initially proposed for
shock-capturing in gasdynamics, the dispersion-controlled dissipa-
tive (DCD) scheme is further modied in the present study for the
purpose of solving steep discontinuities existing in supercavitating
ows. The numerical ux splitting algorithm of the original DCD
scheme is found inapplicable for the simulation of ow cavitation
and should be replaced with a LaxFriedrichs method. The modi-
ed DCD scheme is applied to several tests of supercavitating
ows. Comparisons with theoretical and experimental data help
verify the applicability of the revised DCD scheme for the numeri-
cal simulation of supercavitating ows.
In the one-uid cavitation models for the simulation of super-
cavitating ows, the cut-off model and the isentropic model are
able to reproduce the supercavity with acceptable accuracy as
compared to the experiments. On the contrary, the sinusoidal mod-
el which has been qualied for the simulation of low-speed cavi-
tating ows fails for the high-speed cavitating ows in the
present study. Moreover, a frozen model of speed of sound (SoS)
as used in the present simulation performs well for high-speed
cavitating ows while the well known Walliss model leads to
some numerical instabilities with an under-prediction of the cavity
shape. Finally, further improvement and modication of the pres-
ent numerical techniques are needed to accurately reproduce mul-
tiple physical phenomena at the closure region of a partial
cavitation ow; this is for the future work.
References
[1] Saurel R, Petitpas F, Berry RA. Simple and efcient relaxation methods for
interfaces separating compressible uids, cavitating ows and shocks in
multiphase mixtures. J Comput Phys 2009;28:1678712.
[2] Mulder W, Osher S, Sethian JA. Computing interface motion in compressible
gas dynamics. J Comput Phys 1992;100:20928.
[3] Fedkiw RP, Aslam T, Merriman B, Osher S. A non-oscillatory Eulerian approach
to interfaces in multimaterial ows (the ghost uid method). J Comput Phys
1999;152:45792.
[4] Liu TG, Khoo BC, Yeo KS. Ghost uid method for strong shock impacting on
material interface. J Comput Phys 2003;190:65181.
[5] Liu TG, Khoo BC, Wang CW. The ghost uid method for compressible gas
water simulation. J Comput Phys 2005;204:193221.
[6] Unverdi SO, Tryggvason G. A front-tracking method for viscous,
incompressible, multi-uid ows. J Comput Phys 1992;100:2537.
[7] Terashima H, Tryggvason G. A front-tracking/ghost-uid method for uid
interfaces in compressible ows. J Comput Phys 2009;228:401237.
[8] Abgrall R. How to prevent oscillation in multicomponent ow calculation. J
Comput Phys 1996;125:15060.
[9] Saurel R, Abgrall R. A multiphase Godunov method for compressible multiuid
and multiphase ows. J Comput Phys 1999;150:42567.
[10] Murrone A, Guillard H. A ve equation reduced model for compressible two
phase ow problems. J Comput Phys 2005;202:66498.
[11] Coutier-Delgosha O, Reboud JL, Delannoy Y. Numerical simulations in
unsteady cavitating ows. Int J Numer Methods Fluids 2003;42:52748.
[12] Liu TG, Khoo BC, Xie WF. Isentropic one-uid modeling of unsteady cavitating
ow. J Comput Phys 2004;201:80108.
[13] Goncalves E, Patella RF. Numerical simulation of cavitating ows with
homogenous models. Comput Fluids 2009;38:168296.
[14] Kunz RF, Boger DA, Stinebring DR, Chyczewski TS, Lindau JW, Gibeling HJ, et al.
A preconditioned NavierStokes method for two-phase ows with application
to cavitation prediction. Comput Fluids 2000;29:84975.
[15] Lindau JW, Kunz RF, Boger DA, Stinebring DR, Gibeling HJ. High Reynolds
number, unsteady, multiphase CFD modeling of cavitating ows. J Fluids Eng
2002;124:60716.
[16] Schmidt DP, Rutland CJ, Corradini ML. A fully compressible, two-dimensional
model of small, high speed, cavitating nozzles. Atomization Spray
1999;9:25576.
[17] Venkikos Y, Tzabiras G. A numerical method for the simulation of steady and
unsteady cavitating ows. Comput Fluids 2000;29:6388.
[18] Shin BR, Iwata Y, Ikohagi T. Numerical simulation of unsteady cavitating ows
using a homogenous equilibrium model. Comput Mech 2003;30:38895.
[19] Ahuja V, Hosangadi A, Arunajatesan s. Simulations of cavitating ows using
hybrid unstructured meshes. J Fluids Eng 2001;123:33140.
[20] Owis FM, Nayfeh AH. Computations of the compressible multiphase ow over
the cavitating high-speed torpedo. J Fluids Eng 2003;125:45968.
[21] Neaves MD, Edwards JR. All-speed time-accurate underwater projectile
calculations using a preconditioning algorithm. J Fluids Eng 2006;128:28496.
[22] Jiang ZL, Takayama K, Chen YS. Dispersion conditions for non-oscillatory
shock-capturing schemes and its applications. Comput Fluid Dynam J
1995;2:13750.
[23] Jiang ZL. On the dispersion-controlled principles for non-oscillatory shock-
capturing schemes. Acta Mech Sinica 2004;20:115.
[24] Hu ZM, Jiang ZL. Wave dynamic process in cellular detonation reection from
wedges. Acta Mech Sinica 2007;23:3341.
[25] Hu ZM, Myong RS, Kim MS, Cho TH. Downstreamow conditions effects on the
RR ?MR transition of asymmetric shock waves in steady ows. J Fluid Mech
2009;620:4362.
[26] Hu ZM, Wang C, Zhang Y, Myong RS. Computational conrmation of an
abnormal Mach reection wave conguration. Phys Fluids 2009;21:011701.
s/d
C
p
0 2 4 6 8 10
-0.4
-0.2
0
0.2
0.4
0.6
0.8
1
Rouse and McNown (1948)

Computation
= 0.3
(b)
(a)
Fig. 10. Computational partial cavitation compared to experimental data [33]: (a)
Surface pressure distribution; (b) Void fraction (V
1
= 25.8 m/s, r = 0.3; Cavitation
model: cut-off).
322 Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323
[27] Cooper JR, Dooley RB. The international associated for the properties of
water and steam: revised release on the IAPWS industrial formulation
1997 for the thermodynamic properties of water and steam. Switzerland:
Lucerne; 1997.
[28] Wallism GB. One-dimensional two-phase ow. New York: McGill-Hill; 1969.
[29] Hrubes JD. High-speed imaging of supercavitating underwater projectiles. Exp
Fluids 2001;30:5764.
[30] Munzer H, Reichardt H. Rotational symmetric source-sink bodies with
predominantly constant pressure distributions[sic]. ARE translation 1/
50. England: Aerospace Research Establishment; 1975 [as described in May,
1975].
[31] Garabedian PR. Calculation of axially symmetric cavities and jets. Pac J Math
1956;4:61184.
[32] Wu XJ, Chahine GL. Characterization of the content of the cavity behind a high-
speed supercavitating body. J Fluid Eng 2007;129:13645.
[33] Rouse H, McNown JS. Cavitation and pressure distribution: head forms at zero
angle of yaw. Studies in engineering, vol. 32. Bulletin: State University of Iowa;
1948.
Z.M. Hu et al. / Computers & Fluids 40 (2011) 315323 323

You might also like