Computers & Fluids xxx (2013) xxx–xxx

Contents lists available at SciVerse ScienceDirect

Computers & Fluids

j o u r n a l h o m e p a g e : w w w . e l s e v i e r . c o m / l o c a t e / c o m p fl u i d

A pressure-based, compressible, two-phase flow finite volume method

for underwater explosions
S.T. Miller a,⇑, H. Jasak b, D.A. Boger a, E.G. Paterson a, A. Nedungadi c
a
Applied Research Laboratory, The Pennsylvania State University, P.O. Box 30, State College, PA 16804, United States
b
WIKKI Ltd., 31 Dolben Court, Montaigne Close, London SW1P 4BB, United Kingdom
c
The Johns Hopkins University Applied Physics Laboratory, 11100 Johns Hopkins Road, Laurel, MD 20723, United States

a r t i c l e i n f o a b s t r a c t

Article history: Underwater explosions present many challenges for numerical modeling. The density ratio between
Received 14 May 2012 explosion gases and the surrounding liquid are typically Oð103 Þ, and the pressure ratios can be just as
Received in revised form 28 February 2013 high. We develop a cell-centered finite volume method to solve the governing equations of a two-phase
Accepted 4 April 2013
homogeneous fluid. Isentropic and isothermal equations of state are used to relate densities and pressure,
Available online xxxx
in lieu of solving an energy equation. A volume fraction is used to distinguish between the disparate
phases. A pressure-based, segregated algebraic solution procedure is used to solve for the primitive vari-
Keywords:
ables representing volume fraction, velocity, and piezometric pressure. Numerical examples include ver-
Finite volume
Pressure-based
ification and validation for a number of canonical test cases. In particular, we examine an advecting
Segregated solution material interface and show the absence of pressure oscillations across the contact discontinuity. The
Multiphase flow non-conservative form of our equations does not guarantee exact mass balance, but numerical experi-
Underwater explosions ments indicate that this dissipative mechanism has a stabilizing affect on the method. Shock tube and
UNDEX axisymmetric underwater explosion problems are presented to demonstrate the robustness of the algo-
rithm when very large density and pressure discontinuities are present. We simulate a shallow water
explosion in three dimensions and examine the effects on the free surface and explosion bubble topology.
Ó 2013 Elsevier Ltd. All rights reserved.

1. Introduction [4–6,8]. As such, surface tension and viscous effects play very minor
roles in the solution of UNDEX problems.
Underwater explosions (UNDEX) refer to the detonation of Modeling the conversion of the explosive charge into an effective
explosive devices immersed in water. Fig. 1 is a schematic repre- pressure and density for the hydrodynamical UNDEX model is
senting a generic UNDEX event. The explosive charge is detonated presented in Cole [4], the reports from the Office of Naval Research
at some depth d relative to the free surface, possibly near a solid [1–3], and more recently in Chisum [14]. Most models assume the
body. The physics of the explosion itself are extremely compli- total mass of the explosive charge is converted to gas mass. The
cated, involving rapid chemical changes during the detonation pro- gas is presumed to behave isentropically, with c = 1.25 [4,15].
cess, and they occur on an extremely short timescale. The physics The initial bubble state is presumed to have zero momentum. There
of underwater explosions is discussed in great detail in [1–12]. are then only three remaining unknowns to be specified: the pres-
The hydrodynamical response immediately after detonation is sure, initial radius of the gas bubble, and the isentropic constant.
modeled by assuming the explosion converts all of the fuel into a By knowing any two of these, the third may be calculated.
high pressure, high density gas bubble [4]; any solid particles or Empirical relationships are utilized to determine the initial
other remnants of the explosive device are ignored [13]. The result- pressure and radius from knowledge of the initial explosion weight
ing model is similar to a three-dimensional, spherical shock prob- and depth. The pressure on the surface of the bubble is given in
lem; the initial conditions are discontinuous at the initial bubble [4,15] as
radius at time t = 0. Due to the large pressure and density jumps in  c
the initial conditions, UNDEX events are largely inertially controlled W
PR ¼ 7:8 þ r; ð1Þ
V

⇑ Corresponding author. Tel.: +1 814 867 3052. where W is the weight of the explosive in grams of TNT equivalent,
E-mail addresses: [email protected] (S.T. Miller), [email protected]
V is the bubble volume in cm3, and PR is in kilobars (1 kilobar=108
(H. Jasak), [email protected] (D.A. Boger), [email protected] (E.G. Paterson), Ashish. Pascals), and r is a surface tension term. Other sources (see, e.g.,
[email protected] (A. Nedungadi). [16]) have used

0045-7930/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compfluid.2013.04.002

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
2 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx

(BEM), showing good agreement with both the theoretical Ray-

leigh–Plesset equation (see Section 4.3) as well as the experimental
results from Swift and Decius.
Chan et al. [19] computationally examined bubble-structure
interaction. They used isentropic and isothermal state equations
for gas and water, respectively. They used a 4-field finite difference
method to solve for velocity, density, pressure, and phasic masses,
along with an interface tracking method. Numerical studies
investigate bubble jetting near solid structures, demonstrating that
potential flow methods for UNDEX are inappropriate when the
jetting and subsequent breakup occur, creating a non-simply con-
nected region; however, Zhang et al. [16] utilized some aspects of
potential theory to calculate the velocity potential, which is used to
calculate bubble position and shape. They supplemented this with
a BEM and an elastic mesh motion technique to solve for the three
dimensional dynamics of underwater explosion bubbles.
Kadioglu and Sussman [20] presented an interface tracking
method for underwater explosion problems that utilized adaptive
Fig. 1. Schematic of an underwater explosion event. refinement strategies. Their approach is unique in that it is not
two-phase/fluid, but represents the gas phase analytically inside
 c of the tracked interface, and it uses a Jones–Wilkins–Lee (JWL)
W
P0 ¼ 1:39  105 ; ð2Þ equation of state to calculate the bubble content properties. They
V0 examined cases relating to both UNDEX and implosion near solid
with W in kilograms, V0 in cubic meters, and P0 in Pascals. Setting obstacles. Kan et al. [21] also studied the phenomena of bubble col-
c = 1.25 and working through the unit conversions, we find that lapse under a flat plate through the use of a two-fluid CFD method.
these two expressions are essentially the same1. Zhang [16] also Popinet and Zaleski [22] study the viscosity effects present in the
provides an equation that allows us to calculate the initial radius: collapse and rebound of a bubble near a solid surface. Axisymmet-
!c " ric RANS calculations via finite volumes are done with marker
 3ðc1Þ #  3
1:39  105 1 3W R0 R0 points and cubic splines to track the interface. Underwater implo-
1 ¼  1: ð3Þ sions (the opposite problem of UNDEX, but with very similar phys-
DP c  1 4pR3m Rm Rm
ics) have been studied by Farhat et al. [23] using a two-phase ghost
In (3), the charge weight W is given in kilograms and the radii R0, Rm fluid method with an exact Riemann solution. Their nearly conser-
are in meters. The pressure term DP is taken to be the ambient pres- vative method was found to yield results in good agreement with
sure P1 (neglecting any contributions due to Pc, the constant vapor experimental data on implosions.
pressure inside the bubble). Zhang et al. [16] also provide the The literature is rich with methods that could, in theory, be ap-
empirical relationship for maximum bubble radius as plied to UNDEX problems. Compressible multi-fluids [24–31], free
 1=3 surface flows, volume of fluid, and level set methods [32–35], seg-
W regated/projection methods for multi-phase flows [36–39], mesh-
Rm ¼ 3:38 : ð4Þ
H þ 10 less methods including smoothed particle hydrodynamics [40,41],
and incompressible two-phase methods [42] (and all of the refer-
Using both (3) and (4), along with either (1) or (2), the full initial ences therein) are all such examples in the recent literature.
conditions for the gas bubble are known.
1.2. Outline
1.1. Numerical solutions for UNDEX problems
In this paper, we review the governing equations used herein to
Wardlaw and Mair [17] investigated spherical solutions of an model UNDEX problems in (2). These are the same governing equa-
underwater explosion bubble using multiple one-dimensional tions used by Zwart et al. [43–47], who introduced their work to
models, including both compressible and incompressible models. study the unsteady free surface flow around ships. We rearrange
Their most detailed model was an inviscid model that accounted the differential forms of the equations into a form suitable for
for the fluid motion inside of the gas bubble, which was solved implementation within a co-located, cell-centered finite volume
using an Arbitrary Lagrangian–Eulerian (ALE) framework and a method (see 3). Our formulation differs from the majority of com-
Godunov method. Most notably, they found that modeling the pressible multi-phase or multi-fluid methods in that we use a pres-
water as incompressible increased the bubble period by about sure-based, segregated solution procedure. We present numerous
25%. Brett [18] used the finite element software DYNA2D to exam- numerical results in 4 to demonstrate the performance of our
ine shock waves and pressure generated by underwater explosions. method in capturing the solution features present in UNDEX.
He found that the FEA model was able to capture the bubble peri-
od, maximum diameter, and the pressure impulses; however, peak 2. Governing equations
magnitudes of the shock waves were poorly predicted, probably
due to poor spatial resolution. The macro-scale physics of underwater explosions are essen-
Chisum [14] used MSC/DYTRAN to perform finite volume anal- tially inviscid with little to no mixing over sufficiently short time
yses of UNDEX to investigate the dynamic behavior of the gas bub- intervals in the near field [48]. As such, our model equations are
ble. His validation cases present good qualitative agreement with those of a homogeneous two-phase fluid: immiscible with no slip
studies conducted by Swift and Decius [2]. Wilkerson [15] exam- between phases. This allows us to use a single momentum equa-
ined underwater explosions with the Boundary Element Method tion and velocity field to model both phases.
The derivation of the mixture transport equations for the homo-
1
The unit conversion actually yields a coefficient of 7.82 in (1). geneous treatment of multi-species fluids can take a number of

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx 3

forms, such as ensemble, temporal, or spatial averaging of micro- 2.2. Balance equations: non-conservative form
scale control volumes in order to derive the differential equations
for the macroscopic properties [49–52]. Section 2.1 reveals that we have two independent mass balance
Eq. (5), one momentum Eq. (8), two equations of state (9) and (11),
2.1. Balance equations: conservative form and an algebraic constraint relating the volume fractions (7); a
total of three differential equations and three algebraic equations.
Mass balance for each phase k is We also have six unknown field variables: {a1, a2, q1, q2, u, pd}. We
need to choose three fields to solve for via the differential equa-
@ tions, and we can obtain the other three fields through application
ðak qk Þ þ div ðak qk uÞ ¼ 0; k ¼ f1; 2g; ð5Þ
@t of the algebraic relationships.
with (ak, qk,u) being the volume fraction, density, and velocity of Since the equations of state are given in terms of pressure, we
will solve for pressure and obtain {q1, q2} algebraically. The
phase k, respectively. The subscript on velocity is often omitted
since the homogeneous model only has a single velocity field. Total momentum equation is a natural choice to solve for velocity. We
can solve for one volume fraction differentially and one algebrai-
mass balance is obtained by summing the phasic mass balances
over both phases k: cally, with the constraint that all volume fractions have values in
the interval [0, 1]. Hence, our unknown field variables will be
@ {a1, pd, u}, and our algebraically obtained fields are {a2, q1, q2}.
ðqÞ þ div ðquÞ ¼ 0; q ¼ a1 q1 þ a2 q2 ð6Þ
@t With this closure, we are effectively solving a simplified version
of the isentropic Baer–Nunziato equations [53].
The volume fractions obey the algebraic relationship
The momentum Eq. (8) will be solved for the velocity field. Eq.
a1 þ a2 ¼ 1: ð7Þ (5) with k = 1 and (6) need to be ‘rearranged’ into a form more
amenable to the solution procedure (see Section 3).
The single momentum equation for the homogeneous mixture We begin by reconsidering mass balance for phase k, regarding
model is the phase density as a function of pressure: qk = qk(p). We expand
@ (5) for k = 1 as
ðquÞ þ divðqu  uÞ  ðgrad uÞ grad leff  leff Du
@t ak ðq_ k þ u  grad qk Þ þ ak qk div u þ qk ða_ k þ u  grad ak Þ ¼ 0: ð13Þ
¼ grad pd  ðgrad qÞg  x; ð8Þ
Next, we apply the chain rule to the derivatives of density and rear-
where the piezometric pressure is pd = p  qg  x and the effective range terms to obtain
(composite) viscosity is volume averaged as leff = 4(a1l1 + a2l2)/3. ak @ qk _
The gaseous phase in an UNDEX event is typically modeled as ða_ k þ u  grad ak Þ ¼  ðp þ u  grad pÞ  ak div u ð14Þ
qk @p
an isentropic gas. Water is treated as either incompressible or
slightly compressible; we choose the later and adopt an isothermal Recall that (6) was obtained by summing (5) over all phases.
(constant wave speed) equation of state. The equations of state Similarly, we sum (14) over both phases to obtain the statement
used specify the phasic densities as functions of pressure. As such, of total mass balance
we need not solve an energy equation.  
a1 @ q1 a2 @ q2 _
þ ðp þ u  grad pÞ þ div u ¼ 0; ð15Þ
q1 @p q2 @p
2.1.1. Isentropic equation of state for gas phase
A fluid undergoing only isentropic processes is modeled with where we set p = pd + qg  x.
the equation of state The final form of our volume fraction Eq. (14) is derived by
using (15) to replace terms on the right side:
p !
¼ ac ¼ constant; ð9Þ
qc q1;p q2  q2;p q1
a_ 1 þ div a1 u  a1 div u ¼ a1 a2 div u ð16Þ
a1 w1 q2 þ a2 q2;p q1
where c is the ratio of (constant) specific heats and ac is the isentro-
pic constant. where qk,p = @ qk/@p.
For future use, note that the total derivative of q with respect to
pressure is given as
3. Pressure-based, segregated finite volume method
   1c
@ q  1 p c
¼ : ð10Þ The governing differential equations that we solve are (16), (8),
@p s ac c ac and (15), which we solve for a1, u, and pd, respectively. We use a
cell-centered, co-located finite volume method which is imple-
mented in OpenFOAM. Details of the numerics are provided in
2.1.2. Isothermal equation of state for liquid phase
[54–56]. In this section, we provide details of the linearization
The speed of sound in a fluid medium is given by the relation
and overall solution algorithm.
 
dq  1 The segregated projection algorithm used is reminiscent of the
¼ ; ð11Þ
dp s c2 PISO algorithm [57,58]; however, we solve a pressure equation
rather than a pressure corrector equation. Also, the ‘projection’ as-
where c is the speed of sound and the derivative is taken at constant pect of the scheme is not strictly a discrete projection in the case of
entropy. Assuming a constant sound speed, integration of (11) compressible flow. We do find that adopting and extending the
yields incompressible algorithm to two-phase compressible flow works
very well. We use the superscripts {n + 1, n, ⁄} to denote the current
1
q  q0 ¼ wðp  p0 Þ; w ¼ ; ð12Þ value, the value from the previous timestep, and the value from the
c2
previous iteration, respectively. Time derivatives will be approxi-
with q0 and p0 arising as constants of integration representing a ref- mated via backward Euler for conciseness of presentation; in prac-
erence density and pressure, respectively. tice, one can use any of a variety of time differencing schemes.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
4 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx

3.1. Momentum equation loops per time step to tightly couple each solution field with the
rest. We summarize our algorithm as:
The linearized momentum equation is
FOR EACH TIME STEP, DO
q unþ1  qn un
þ div ðq u  unþ1 Þ ðgrad u Þ grad leff FOR EACH OUTER CORRECTOR, DO:
Dt
1. Solve momentum predictor
 leff Du ~  via (20)
2. Set u
¼ grad pnþ1
d  ðgrad q Þg  x; ð17Þ 3. Solve (21) for pnþ1
d
n+1 4. Set p = pd + qg  x
which comes from (8) and is solved for the new velocity u . Using
5. Evaluate state equations
a projection step, the velocity is written as
6. Compute an+1 from (24)
unþ1 ¼ A1 Hðu Þ  A1 ðgrad q Þg  x  A1 grad pnþ1
d : ð18Þ 7. Set the composite density
END OUTER CORRECTOR LOOP
~  , we
Grouping the first two terms on the right side together as u END TIMESTEP
write the velocity as

~   A1 grad pnþ1

unþ1 ¼ u d ; ð19Þ
The number of outer correctors take per time step is a user-de-
where fined parameter. A typical choice is 4 or 6 outer correctors, but the
user needs to ensure that residual values are sufficiently reduced/
~  ¼ A1 Hðu Þ  A1 ðgrad q Þg  x:
u ð20Þ small. In future work, automatic convergence checking will be
implemented.
3.2. Pressure equation
3.5. Numerical solution properties
The pressure Eq. (15) is linearized and discretized as
The numerical results in 4 all use the same discretization
  nþ1 
a1 w1 a2 w2 pd  pnd   schemes. The implicit backward Euler scheme is used for temporal
þ  þ div pnþ1
d u

 pnþ1
d div u þ k
q1 q2 Dt derivatives. A second-order upwind scheme is used for the convec-
tive terms in the momentum equation. The convective terms in the
þ div unþ1
volume fraction equation are computed using the van Leer limiter
¼ 0; ð21Þ [55]. Pressure equation convective terms are discretized with a
first-order upwind flux; we find this aids in stabilization of the pres-
where wi = @ qi/@p and the source term is given by
sure in the vicinity of strong shocks. Due to the strong discontinuities
ðqg  xÞ  ðqg  xÞn present in our example problems, we only expect to see first order
k¼ þ div ððqg  xÞ u Þ  ðqg  xÞ div u ð22Þ results in the neighborhood of shock waves and contact discontinu-
Dt
ities. Since we are only interested in problems involving many com-
The un+1 term is replaced with the expression in (19). That is, we
plex, discontinuous solution structures, we have omitted any sort of
use
grid study to establish ‘higher-order’ convergence rates. That is, we
~   div ðA1 grad pnþ1
div unþ1 ¼ div u resign ourselves to using a fast, stable, simple to implement algo-
d Þ; ð23Þ
rithm with higher mesh resolution to compute the complex material
which allows us to solve implicitly for the pressure pnþ1
d . and shock wave interactions for our problems of interest.
Neither total mass nor phasic mass are balanced exactly by our
3.3. Volume fraction equation numerical method. While this is somewhat disconcerting, in prac-
tice we obtain good results with a reasonable amount of mass er-
The continuum form of the volume fraction equation is given as ror. We postulate, but have not proven, that the mass error is a
(16). If we solve the volume fraction equation after solving the direct consequence of strongly enforcing the equations of state
momentum and pressure equations, we have the un+1, pn+1 values rather than solving a weak statement of an energy equation. Typ-
to use. Therefore, the linearized discrete version is ically, a slight energy dissipation lends stability to numerical meth-
ods; since we do not have this, our dissipation comes in the form of
anþ1  an1  
1
þ div anþ1
1 u
nþ1
 anþ1
1 div unþ1 mass error. The non-conservative form that we solve also has
Dt implications with respect to shock speeds and Rankine-Hugoniot
 
¼ anþ1
1 1  anþ1
1 ðasrc Þnþ1 ; ð24Þ conditions, as we shall see in 4.2.
which is solved implicitly for an+1. The source term (asrc)n+1 is given
4. Numerical results
as
!
wnþ1 wnþ1 All of our numerical results are solved in dimensional form,
ðasrc Þnþ1 ¼ 2
nþ1
 1
nþ1
ðpnþ1 þ div ðpnþ1 unþ1 Þ with units being the standard metric units unless stated otherwise.
q2 q 1 The isentropic equation of state is used for species represented
nþ1
p div unþ1 Þ: ð25Þ with a = 1, and isothermal equation of state (EOS) is used for
a = 0. The specified time steps are all chosen such that the CFL
number based on compressible wave speeds is less than 1.
3.4. Solution algorithm
4.1. Advecting interface
Eqs. (17), (21), and (24) are integrated over each control volume
in a computational mesh according to finite volume numerics (see Pressure oscillations in the vicinity of contact discontinuities
[54–56]). The linearization procedure allows us to solve each equa- are a known issue with many methods (see [24,26,27] and therein).
tion sequentially, with the possibility of multiple outer corrector We demonstrate the absence of these discontinuities in our solu-

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx 5

tion method by considering an advecting interface. Our domain is

one-dimensional (or solved in a quasi-1D setting), with 1000 com-
putational cells, x 2 [0, 1]. A fixed time step of Dt = 104 is used. Ini-
tial conditions are specified as

0 x < 0:5
a¼ ; u ¼ 1; p ¼ 1: ð26Þ
1 x > 0:5
Dirichlet boundary conditions are specified as
uðx ¼ 0Þ ¼ 1; aðx ¼ 0Þ ¼ 0; pðx ¼ 1Þ ¼ 1; ð27Þ
with all non-specified quantities being unconstrained (i.e. zero gra-
dient). Fluid parameters for the gaseous phase (i.e. a = 1) are
c ¼ 1:25; ac ¼ 1; ð28Þ
Fig. 3. The volume fraction clearly marks the location of the contact discontinuity
and the fluid phase parameters are at tf = 0.25.

q0 ¼ 1000; p0 ¼ 1; w ¼ 1: ð29Þ
Gravity is neglected for this simulation.
Fig. 2 shows the pressure and volume fraction fields at time
t = 0.4. We see that there are no pressure oscillations across the
contact discontinuity located at x = 0.9. Indeed, all of the cell values
are p = 1 to within machine precision.

4.2. Shock tube

We investigate a two-phase shock tube problem [59] to demon-

strate the ability to resolve rarefaction waves, contact discontinu-
ities, and strong shock waves. We use the same computational
domain as in 4.1, with a fixed time step of Dt = 104. Initial condi-
tions are specified as
 
0 x < 0:5 1:0 x < 0:5; Fig. 4. Pressure field at tf = 0.25 for the shock tube problem.
a¼ ; u ¼ 0; p¼ ð30Þ
1 x > 0:5 0:1 x > 0:5;
The final time of our simulation is tf = 0.25, which is before any propa-
gating information reaches the domain boundaries; as such, boundary
conditions are inconsequential. Viscous terms and gravity are neglected.

The material parameters for the gaseous phase (i.e. a = 1) are

c ¼ 1:25; ac ¼ 1:34543; ð31Þ
and the isothermal parameters are
q0 ¼ 0; p0 ¼ 0; w ¼ 1: ð32Þ
Thus, at t = 0, we have

1:0 x < 0:5
q¼ : ð33Þ
0:125 x > 0:5

Fig. 5. Velocity field at tf = 0.25 for the shock tube problem.

The exact solution for this problem can be obtained only semi-
analytically. The procedure in [60,61] is followed. We satisfy the
Rankine-Hugoniot conditions across the shock wave, and we set
the isothermal Riemann invariants equal across the rarefaction.
The isothermal equation of state holds to the left of the contact dis-
continuity, and the isentropic to the right. Noting that both veloc-
ity and pressure are continuous across the contact, we are able to
finally solve for the location of the right edge of the rarefaction and
the shock speed in an iterative fashion.
Fig. 3 shows that the computed volume fraction ah matches the
exact solution, with no spurious oscillations around the disconti-
nuity. The contact discontinuity is resolved across eight cells,
Fig. 2. The advecting interface test case shows that there are no pressure which is consistent with a first-order method in the presence of
oscillations across the contact discontinuity. discontinuous solution features.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
6 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx

0.45
h = 0.0100
h = 0.0050
h = 0.0033
0.4 h = 0.0025
h = 0.0020
Rayleigh-Plesset
0.35 Experiment

0.3
Radius (m)

0.25

0.2

0.15

0.1

0.05
0 0.005 0.01 0.015 0.02 0.025
Time (s)

Fig. 6. Deep UNDEX on a Cartesian mesh of various sizes.

0.45
h = 0.0100
h = 0.0050
h = 0.0033
0.4 h = 0.0025
h = 0.0020
Rayleigh-Plesset
0.35 Experiment

0.3
Radius (m)

0.25

0.2

0.15

0.1

0.05
0 0.005 0.01 0.015 0.02 0.025
Time (s)

Fig. 7. Deep UNDEX on a radially symmetric mesh of various sizes.

0.45
Radial: h = 0.0020
Cartesian: h = 0.0020
Rayleigh-Plesset
0.4 Experiment

0.35

0.3
Radius (m)

0.25

0.2

0.15

0.1

0.05
0 0.005 0.01 0.015 0.02 0.025
Time (s)

Fig. 8. Comparison of Cartesian versus radially symmetric meshes; the smallest grid spacing is compared.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx 7

Fig. 9. Depiction of a deep water UNDEX event under a flat plate.

Fig. 10. Flat plate results at t = 0.4 ms. The evolving pressure field is on the left, volume fraction on the right.

The pressure solution ph also shows qualitatively good agree- 4.3. Deep water UNDEX
ment with the exact solution, as shown in Fig. 4. The edges of the
rarefaction wave are slightly rounded due to numerical dissipation. Deep water UNDEX problems are, essentially, the three-dimen-
The shock propagates at what appears to be the correct speed; how- sional analog of the shock tube problem examined in 4.2. As the
ever we see a slight error in the wake of the shock. This error is on explosion bubble expands, a shock wave is sent out through the
the order of two percent. We see a similar solution discrepancy, of surrounding water medium, and an expansion wave travels to-
the same order, in the shock wake region of the velocity; see Fig. 5. wards the center of the explosion gases. Due to the detonation
We postulate that these errors are due to our method not being to- depth, the effects of gravity are negligible and thus ignored in
tally conservative. In particular, the non-conservative form that we the numerical model.
solve for does not satisfy the same Rankine-Hugoniot conditions The size of an oscillating free bubble can be characterized ana-
that a conservative form of the equations does. The mass error ef- lytically. The Rayleigh–Plesset equation for bubble dynamics with-
fect combined with incorrect jump conditions are a likely candidate out thermal effects is (cf. Brennen [62])
for the errors we see in the shock wake region. Further research is  3c
pB ðT 1 Þ  p1 ðtÞ pG0R0 € þ 3 R_ 2 þ 4mL R_ þ 2S ;
needed to quantify these errors and possibly augment our non-con- þ ¼ RðtÞR ð34Þ
servative equations with terms to correct the deficiencies. qL qL R 2 R qL R

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
8 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx

Fig. 11. Flat plate results at t = 0.8 ms. The pressure field is on the left, volume fraction on the right.

Fig. 12. Flat plate results at t = 1.0 ms. The pressure field is on the left, volume fraction on the right.

where the bubble radius R = R(t). The initial gas pressure is pG0 . A The initial gas bubble with R0 = 0.0667815 is centered at the
subscript ‘L’ refers to the surrounding liquid medium. ‘S’ is the sur- coordinate origin. The explosion depth of the 0.30 kg TNT charge
face tension, which we shall neglect for this study.2 While this or- is 167.64 m. Other relevant initial conditions are:
dinary differential equation cannot be solved analytically, standard
numerical integrators such as a Runge–Kutta method are able to pro- pG0 ¼ 1:31  108 ; p1 ¼ patm þ qL gh
vide numerical solutions. ¼ 1; 786; 987:1; qbubble ¼ 240:472: ð35Þ
Our test case will be the same as Wilkerson [15], Fig. 5. We
chose this case because we can compare to both the Rayleigh–Ples- The isentropic parameters for the explosion gas are
set solution as well as the experimental results from Swift and Dec- c ¼ 1:25; ac ¼ 138999:59; ð36Þ
ius [2] (as quoted in [15]). In the current work, we have converted
and the surrounding water uses the values of
all of Wilkerson’s parameters to standard metric units.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx 9

Fig. 13. Initial condition for shallow water explosion. The gray iso-surface represents the undisturbed free surface. The red iso-surface is the initial explosion bubble. (For
interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.)

Fig. 14. Shallow water explosion at first maximum bubble radius.

w ¼ 4:54  107 ; q0 ¼ 1025; p0 ¼ p1 ð37Þ To compute the average bubble radius in our simulations, we
simply convert the total volume of gas into the radius of an equiv-
alent sphere. Figs. 6 and 7 compare our numerical solution with the
for the isothermal EOS. Rayleigh–Plesset solution as well as the experimental data (pulled
The computational domain is modeled as axisymmetric with an from [15]). For both mesh types, decreasing the grid spacing
additional symmetry plane introduced such that we are only solving slightly decreases the maximum bubble radius and shortens the
in the first quadrant, i.e. (x P 0,y P 0). The domain is created with a time for the first bubble collapse. Both the maximum radius and
maximum radius of 50; this allows any boundary reflections (e.g. shock collapse time are shorter than that predicted by the Rayleigh–Ples-
waves, etc.) to not affect the bubble radius over our time of interest. set equation; however, our predicted collapse time agrees fairly
Two grid topologies are studied: a standard Cartesian grid, and a well with the experimental data. Possible reasons for the discrep-
radially symmetric grid. For both topologies, we use equally spaced ancy with the Rayleigh–Plesset solution include: isothermal com-
cells in the r 2 (0, 0.55) range, and gradually increase cell size to the pressible water instead of incompressible (RP), non-uniform
far field from there. The explosion bubble remains inside of the pressure inside of the bubble, viscous effects, etc. The mass errors
equally spaced cells for all time. Five grid sizes are used in this re- introduced by our proposed numerical method could also be a
gion: h ¼ f0:01; 0:005; 0:003;  0:0025; 0:002g. In all cases, the initial source of deviation.
time step is chosen to be Dt = 107. Adaptive time stepping is used, Fig. 8 compares the bubble radii computed on the different
with a CFL limitation of 0.25. mesh topologies. The predicted bubble radius and collapse times

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
10 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx

Fig. 15. Shallow water explosion at second bubble rebound.

show good agreement for the same mesh spacing parameter. Plots and density. We will continue to use our isentropic EOS for the
of the volume fraction solution (not shown) confirm that the radi- explosive gases, with c = 1.25.
ally symmetric mesh preserves the known symmetry of the prob- Fig. 10 shows the pressure field (left) as the initial shock wave ap-
lem to a much higher degree. proaches the edge of the flat circular plate. The corresponding vol-
ume fraction profile is at right. Figs. 11 and 12 show the pressure
4.4. Deep explosion near a plate and volume fraction fields at t = 0.8 ms and t = 1.0 ms, respectively.
Upon reaching the edge of the plate, the shock wave curls around
Kadioglu and Sussman [20] considered a deep explosion under- and propagates towards the centerline while still radiating out-
neath a circular flate plate, as depicted in Fig. 9. We use the same wards. The bubble radius remains unaffected by the outwardly prop-
problem setup, including the Tait equation of state for water agating pressure field, and is growing at a rate much slower than the
(ignoring cavitation pressure) shock speed. These results are in very good qualitative agreement
 1 with those presented in [20].
p þ B  A ct
q ¼ q ; ð38Þ
B
4.5. Shallow water explosion near a free surface
with B = 3.31  108 Pa, A = 105 Pa, ct ¼ 7:15, q ¼ 103 kg=m3 . The
JWL equation of state for water is used in [20]; they are able to Shallow water underwater explosions exhibit very different
use this because they track the contact location between the phases characteristics from deep UNDEX. Gravity (buoyancy) affects the
and use the JWL equation to compute a uniform bubble pressure bubble collapse event, resulting in a reentrant jet that pierces the

Fig. 16. Shallow water explosion at third bubble rebound.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx 11

free surface. For our last numerical example, we simulate a 0.30 kg [5] Holt M. Underwater explosions. Ann Rev Fluid Mech 1977;9(1):187–214.
[6] Kaltwasser M. Hydra program-theoretical and experimental determination of
TNT charge (the same as in 4.3) at a detonation depth of 6 m below
energy partition of selected underwater explosives. Tech. rep., DTIC Document;
the free surface. At this depth, the initial bubble radius is computed 1963.
to be R0 = 0.60 m at a pressure of 72.7 MPa. [7] Goertner JF. Vacuum tank studies of gravity migration of underwater explosion
Physical properties of the both the atmospheric air and the bubbles. Tech. Rep. 3902, NAVORD; 1956.
[8] Pritchett J. An evaluation of various theoretical models for underwater
explosion gas are taken to be identical, and are specified the same explosion bubble pulsation. Tech. rep., DTIC Document; 1971.
as in 4.3. The only changes to the water parameters are [9] Kolodner I. Underwater explosion bubbles. 2. The effect of gravity and the
change of shape. Tech. rep., DTIC Document; 1953.
q0 ¼ 1025 kg=m3 ; p0 ¼ 101; 325 Pa: ð39Þ [10] Blake J, Gibson D. Growth and collapse of a vapour cavity near a free surface. J
Fluid Mech 1981;111:123–40.
2
Gravity is 9.81 m/s in the z-direction. We use a fully three-dimen- [11] Blake J, Gibson D. Cavitation bubbles near boundaries. Ann Rev Fluid Mech
1987;19(1):99–123.
sional grid consisting of 6.1 million Cartesian cells to discretize the
[12] Strahle W. Conventional weapons underwater explosions. Tech. rep., DTIC
domain [100, 100]2  [100, 10]. The calculation was done in par- Document; 1988.
allel using 128 processors. [13] Bryant E. Debris distribution in underwater explosions. Tech. rep., DTIC
Fig. 13 shows the initial conditions, with the red iso-surface Document; 1964.
[14] Chisum J. Simulation of the dynamic behavior of explosion gas bubbles in a
showing the high pressure explosion gas. The undisturbed free sur- compressible fluid medium. Ph.D. thesis, Naval Postgraduate School; 1996.
face is the gray iso-surface. Fig. 14 shows the explosion bubble at [15] Wilkerson S. Boundary integral technique for explosion bubble collapse
its maximum radius. Note that the water surface is not broken at analysis. Tech. Rep. ARL-TR-184, Army Research Laboratory; August 1993.
[16] Zhang AM, Yao XL, Yu XB. The dynamics of three-dimensional underwater
this time; however, the upward momentum of the fluid above explosion bubble. J Sound Vib 2008;311:1196–212.
the explosion will continue to carry fluid particles vertically. [17] Wardlaw Jr A, Mair H. Spherical solutions of an underwater explosion bubble.
Fig. 15 shows the explosion bubble at maximum radius after the Shock Vib 1998;5(2):89–102.
[18] Brett J. Numerical modelling of shock wave and pressure pulse generation by
first expansion-collapse event. Buoyancy has caused the bubble underwater explosions. Tech. rep., DTIC Document; 1998.
to roll into a toroidal shape during the first collapse, and it retains [19] Chan P, Kan K, Stuhmiller J. A computational study of bubble-structure
the shape in the resulting expansion. Fig. 16 depicts the ‘crowning’ interaction. J Fluids Eng 2000;122:783.
[20] Kadioglu SY, Sussman M. Adaptive solution techniques for simulating
of the free surface at the maximum radius of the third bubble
underwater explosions and implosions. J Comput Phys
expansion. All of these phenomena are qualitatively extremely 2008;227(3):2083–104. http://dx.doi.org/10.1016/j.jcp.2007.10.019.
similar to that reported in experiments [2,4,5]. [21] Kan K, Stuhmiller J, Chan P. Simulation of the collapse of an underwater
explosion bubble under a circular plate. Shock Vib 2005;12(3):217–26.
[22] Popinet S, Zaleski S. Bubble collapse near a solid boundary: a numerical study
5. Conclusions of the influence of viscosity. J Fluid Mech 2002;464(1):137–63.
[23] Farhat C, Rallu A, Shankaran S. A higher-order generalized ghost fluid method
for the poor for the three-dimensional two-phase flow computation of
We have introduced a novel finite volume formulation for underwater implosions. J Comput Phys 2008;227(16):7674–700.
homogeneous two-phase compressible flow. Our formulation uses [24] Abgrall R, Karni S. Computations of compressible multifluids. J Comput Phys
state equations to enforce pressure-density relationships. Equa- 2001;169(2):594–623.
[25] Allaire G, Clerc S, Kokh S. A five-equation model for the simulation of interfaces
tions are solved in a non-conservative form, and mass is not strictly between compressible fluids. J Comput Phys 2002;181(2):577–616. http://
conserved. However, the ease of implementation within a large dx.doi.org/10.1006/jcph.2002.7143.
open-source CFD package (OpenFOAM) combined with the small [26] van Brummelen E, Koren B. A pressure-invariant conservative godunov-type
method for barotropic two-fluid flows. J Comput Phys 2003(185):289–308.
linear systems used in a segregated solution procedure make our
[27] Fedkiw R, Liu X, Osher S. A general technique for eliminating spurious
FVM formulation attractive for many real-world solution proce- oscillations in conservative schemes for multiphase and multispecies Euler
dures. Our experience indicates that the method is extremely sta- equations. Int J Nonlinear Sci Numer Simulat 2002;3(2):99–105.
ble and robust, even for problems with very large pressure ratios [28] Miller GH, Puckett EG. A high-order Godunov method for multiple condensed
phases. J Comput Phys 1996;128(1):134–64.
present. Future work remains to investigate the full effects of the [29] Venkateswaran S, Lindau J, Kunz R, Merkle C. Computation of multiphase
non-conservative form and to quantify the resulting errors. mixture flows with compressibility effects. J Comput Phys 2002;180(1):54–77.
Numerical examples in Section 4 demonstrate the method’s [30] Saurel R, Petitpas F, Berry R. Simple and efficient relaxation methods for
interfaces separating compressible fluids, cavitating flows and shocks in
ability to handle shock waves, contact discontinuities, and rarefac- multiphase mixtures. J Comput Phys 2009;228(5):1678–712.
tions. In particular, we do not see any spurious pressure oscilla- [31] Kokh S, Lagoutière F. An anti-diffusive numerical scheme for the simulation of
tions across material interfaces. Deep water UNDEX results show interfaces between compressible fluids by means of a five-equation model. J
Comput Phys 2010;229(8):2773–809. http://dx.doi.org/10.1016/
quantitatively reasonable agreement with the experiment of Swift j.jcp.2009.12.003.
and Decius [2] as well as to the Rayleigh–Plesset equation [62]. Our [32] Apsley D, Hu W. CFD simulation of two-and three-dimensional free-surface
shallow water UNDEX shows that we are able to delineate between flow. Int J Numer Methods Fluids 2003;42(5):465–91.
[33] Gopala V, van Wachem B. Volume of fluid methods for immiscible-fluid and
explosion gases and atmospheric air for UNDEX phenomena where free-surface flows. Chem Eng J 2008;141(1–3):204–21.
the free water surface is broken. The qualitative phenomena of [34] de Niem D, Kührt E, Motschmann U. A volume-of-fluid method for simulation
bubble expansion, collapse, re-expansion and crowning are all cap- of compressible axisymmetric multi-material flow. Comput Phys Commun
2007;176(3):170–90.
tured by our model.
[35] Shukla RK, Pantano C, Freund JB. An interface capturing method for the
simulation of multi-phase compressible flows. J Comput Phys
Acknowledgement 2010:7411–39.
[36] Puckett EG, Almgren AS, Bell JB, Marcus DL, Rider WJ. A high-order projection
method for tracking fluid interfaces in variable density incompressible fluids. J
This research was funded under the The Johns Hopkins Univer- Comput Phys 1997;130:269–82.
sity Applied Physics Laboratory Internal Investment Program. [37] Moukalled F, Darwish M. A unified formulation of the segregated class of
algorithms for fluid flow at all speeds. Numer Heat Transfer, Part B: Fundam:
An Int J Comput Methodol 2000;37(1):103–39.
References [38] Darwish M, Moukalled F, Sekar B. A unified formulation of the segregated class
of algorithms for multifluid flow at all speeds. Numer Heat Transfer, Part B:
[1] Office of Naval Research. Underwater explosion research: vol. 1: the shock Fundam: An Int J Comput Methodol 2001;40(2):99–137.
wave, Tech. rep., Department of the Navy; 1950. [39] Senocak I, Shyy W. A pressure-based method for turbulent cavitating flow
[2] Office of Naval Research. Underwater explosion research: vol. 2: the gas globe, computations. J Comput Phys 2002;176:363–83.
Tech. rep., Department of the Navy; 1950. [40] Liu G-R, Liu M. Smoothed particle hydrodynamics: a meshfree particle
[3] Office of Naval Research. Underwater explosion research: vol. 3: the damage method. World Scientific Publishing Company Incorporated; 2003.
process, Tech. rep., Department of the Navy; 1950. [41] 6th International SPHERIC SPH Workshop. SPH simulation of gas explosion
[4] Cole RH. Underwater explosions. Princeton University Press; 1948. inside an underwater non-cohesive sediment deposit.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002
12 S.T. Miller et al. / Computers & Fluids xxx (2013) xxx–xxx

[42] Majumder S, Chakraborty S. New physically based approach of mass [52] Hassanizadeh M, Gray WG. General conservation equations for multi-phase
conservation correction in level set formulation for incompressible two- systems: 1. Averaging procedure. Adv Water Resources 1979;2:131–44.
phase flows. J Fluids Eng 2005;127(3):554–63. http://dx.doi.org/10.1115/ [53] Baer M, Nunziato J. A two-phase mixture theory for the deflagration-to-
1.1899172. detonation transition (ddt) in reactive granular materials. Int J Multiphase
[43] Zwart P, Scheuerer M, Bogner M. Free surface flow modelling of an impinging Flow 1986;12(6):861–89.
jet. In: ASTAR international workshop on advanced numerical methods for [54] Jasak H. Error analysis and estimation for the finite volume method with
multidimensional simulation of two-phase flow. GRS Garching, Germany; applications to fluid flows. Ph.D. thesis, Imperial College, University of London;
2003. 1996.
[44] Zwart PJ. Industrial two-phase flow CFD. Tech. rep., von Karman Institute for [55] Rusche H. Computational fluid dynamics of dispersed two-phase flows at high
Fluid Dynamics; 2005. phase fractions. Ph.D. thesis, Imperial College of Science, Technology &
[45] Zwart P, Burns A, Galpin P. Coupled algebraic multigrid for free surface flow Medicine; 2002.
simulations. ASME; 2007. [56] Jasak H, Gosman A. Element residual error estimate for the finite volume
[46] 10th International symposium on practical designs of ships and other floating method. Comput Fluids 2003;32(2):223–48.
structures, ship hull simulations with a coupled solution algorithm. [57] Issa R. Solution of the implicitly discretised fluid flow equations by operator-
[47] Zwart P, Godin P, Penrose J, Rhee S. Simulation of unsteady free-surface flow splitting. J Comput Phys 1986;62(1):40–65.
around a ship hull using a fully coupled multi-phase flow method. J Mar Sci [58] Issa R, Gosman A, Watkins A. The computation of compressible and
Technol 2008;13(4):346–55. incompressible recirculating flows by a non-iterative implicit scheme. J
[48] Park J. A Runge Kutta discontinuous Galerkin-direct ghost fluid (RKDG-DGF) Comput Phys 1986;62(1):66–82.
method to near-field early-time underwater explosion (UNDEX) simulations. [59] Sod G. A survey of several finite difference methods for systems of nonlinear
Ph.D. thesis, Virginia Polytechnic Institute and State University; 2008. hyperbolic conservation laws. J Comput Phys 1978;27(1):1–31.
[49] Drew D. Mathematical modeling of two-phase flow. Ann Rev Fluid Mech [60] LeVeque R. Numerical methods for conservation laws. Birkhäuser; 1992.
1983;15(1):261–91. [61] LeVeque R. Finite volume methods for hyperbolic problems, vol.
[50] Drew D, Passman SL. Theory of multicomponent fluids. Springer-Verlag; 1999. 31. Cambridge Univ Pr; 2002.
[51] Gray WG. A derivation of the equations for multi-phase transport. Chem Eng [62] Brennen C. Cavitation and bubble dynamics, vol. 44. USA: Oxford University
Sci 1975;30:229–33. Press; 1995.

Please cite this article in press as: Miller ST et al. A pressure-based, compressible, two-phase flow finite volume method for underwater explosions. Com-
put Fluids (2013), http://dx.doi.org/10.1016/j.compfluid.2013.04.002