A phenomenological analysis of droplet

shock-induced cavitation using a multiphase

modeling approach
Cite as: Phys. Fluids 35, 013312 (2023); https://doi.org/10.1063/5.0127105
Submitted: 19 September 2022 • Accepted: 11 December 2022 • Published Online: 06 January 2023

L. Biasiori-Poulanges and K. Schmidmayer

COLLECTIONS

Paper published as part of the special topic on Cavitation

This paper was selected as Featured

ARTICLES YOU MAY BE INTERESTED IN

Regulation of droplet size and flow regime by geometrical confinement in a microfluidic flow-
focusing device
Physics of Fluids 35, 012010 (2023); https://doi.org/10.1063/5.0130834

The stressful way of droplets along single-fiber strands: A computational analysis

Physics of Fluids 35, 012110 (2023); https://doi.org/10.1063/5.0131032

Collective transport of droplets through porous media

Physics of Fluids 35, 013304 (2023); https://doi.org/10.1063/5.0129477

Phys. Fluids 35, 013312 (2023); https://doi.org/10.1063/5.0127105 35, 013312

© 2023 Author(s).
 Physics of Fluids ARTICLE scitation.org/journal/phf

A phenomenological analysis of droplet

shock-induced cavitation using a multiphase
modeling approach
Cite as: Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105
Submitted: 19 September 2022 . Accepted: 11 December 2022 .
Published Online: 6 January 2023

L. Biasiori-Poulanges1,a) and K. Schmidmayer2,b)

AFFILIATIONS
1
ETH Zurich, Department of Mechanical and Process Engineering, Institute of Fluid Dynamics, Sonneggstrasse 3, Zurich 8092,
Switzerland
2
INRIA Bordeaux Sud-Ouest, project-team CAGIRE, Universite de Pau et des Pays de l’Adour, E2S UPPA, Laboratory of Mathematics
and Applied Mathematics (LMAP), Pau, France

Note: This paper is part of the special topic, Cavitation.

a)
Author to whom correspondence should be addressed: lbiasiori@ethz.ch
b)
URL: https://kevinschmidmayer.github.io/

ABSTRACT
Investigations of shock-induced cavitation within a droplet are highly challenged by the multiphase nature of the mechanisms involved.
Within the context of heterogeneous nucleation, we introduce a thermodynamically well-posed multiphase numerical model accounting for
phase compression and expansion, which relies on a finite pressure-relaxation rate formulation. We simulate (i) the spherical collapse of a
bubble in a free field, (ii) the interaction of a cylindrical water droplet with a planar shock wave, and (iii) the high-speed impact of a gelatin
droplet onto a solid surface. The determination of the finite pressure-relaxation rate is done by comparing the numerical results with
the Keller–Miksis model, and the corresponding experiments of Sembian et al. and Field et al., respectively. For the latter two, the pressure-
relaxation rate is found to be l ¼ 3:5 and l ¼ 0:5, respectively. Upon the validation of the determined pressure-relaxation rate, we run
parametric simulations to elucidate the critical Mach number from which cavitation is likely to occur. Complementing simulations with a
geometrical acoustic model, we provide a phenomenological description of the shock-induced cavitation within a droplet, as well as a discus-
sion on the bubble-cloud growth effect on the droplet flow field. The usual prediction of the bubble cloud center, given in the literature, is
eventually modified to account for the expansion wave magnitude.
C 2023 Author(s). All article content, except where otherwise noted, is licensed under a Creative Commons Attribution (CC BY) license (http://
V
creativecommons.org/licenses/by/4.0/). https://doi.org/10.1063/5.0127105

I. INTRODUCTION cavitation (i.e., isothermal inertia-driven phase change). Liquid

The interaction of a liquid droplet with a shock wave, or the rupture arises when subject to tension exceeding a threshold value,
impact of the droplet on a solid substrate, results in the transmis- which depends on the nature of the liquid and its purity. Pure
sion of a confined shock to the droplet. Because of the large acous- liquids cavitate when the random thermal motions of molecules
tic impedance ratio between the surrounding gas and the liquid, cause microscopic voids.5 This process is usually referred to as
and so a poor gas-to-liquid energy transfer,3 the droplet interface homogeneous cavitation. When liquids are not pure, that is, con-
acts as a nearly perfect mirror that traps the transmitted wave tain pre-existing nuclei/impurities, cavitation results from the
energy within the droplet. The confined shock therefore experien- expansion of submicroscopic gas pockets trapped on particles pre-
ces near total reflections, as well as focusing which results in the sent in the liquid. The process of bubble formation by this mecha-
amplification of the shock local interaction with the liquid.4 The nism is referred to in the literature as heterogeneous nucleation.
reflected wave of the transmitted shock being an expansion wave, Shock-induced cavitation within a droplet, upon impact or inter-
and some regions of the droplet are thus exposed to a tensile force action with a shock wave, occurs in wide range of applications, as a
which, under some conditions, may generate hydrodynamic desired or adverse effect, ranging from raindrop impact on aircraft,6 to

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-1

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

combustion and detonation of multiphase mixtures,7 through ink-jet computed from the numerics, and complemented with the geometri-
printing or liquid jet-based physical cleaning,8,9 to name but a few. cal acoustic model of Biasiori-Poulanges and El-Rabii. We finally
The comprehension of the bubble dynamics within the droplet is thus examine the sensitivity of l on the material properties of the liquid
of major importance to evaluate the erosion efficiency of the bubble- mixture constituting the droplet, by simulating the high-speed droplet
compounded droplet, related to the collapse and jetting processes of impact experiments of Field et al.
the cavitation bubbles. Given that the presence of cavities inside the
droplet alters its interfacial dynamics,10,11 under some conditions, II. PROBLEM DESCRIPTION
changes in the fragmentation process are to be expected. The interaction of a confined fluid volume with a shock wave is
The experimental characterization of shock-induced cavitation known to generate a complex time-dependent wave pattern. Taking
within a droplet is particularly challenging.11 By reducing the droplet into account for the compression and expansion effects in a two-phase
to a water column, Sembian et al. and Field et al. however successfully liquid–gas droplet, the canonical wave structure is modified. Based on
imaged the growth of a bubble cloud during the interaction of a cylin- the droplet internal wave structure, this section first gives a phenome-
drical droplet with a planar shock wave, and the high-speed impact of nological description of the shock-induced cavitation within a liquid
a cylindrical droplet with a solid substrate, respectively. To overcome droplet. It also introduces recent works on the analytical description of
the experimental limitations and address the shock-induced cavitation the wave pattern, which has been interpreted using the classical ray-
within a droplet under near-reality conditions, previous attempts to tracing approach to geometrical acoustics. Note that, in this section,
explicit the conditions for the bubbles to grow mostly relied on numer- the description is based on the interaction of a shock wave with a
ical simulations. As a first approach, past numerical studies used droplet, but the phenomenology is also valid for the high-speed drop-
numerical models not accounting for phase change or phase expan- let impact.
sion.1,4,12–14 The occurrence and intensity of the cavitation were evalu-
ated by probing the pressure field and comparing the low-pressure A. Phenomenology
region magnitude to the cavitation threshold given by the classical
The phenomenology of the shock-induced cavitation within a
nucleation theory for homogeneous cavitation (134 MPa at 300 K),15
liquid droplet is sketched in Fig. 1, where the wave pattern inside
or the Blake threshold pressure.16 Recently, Kyriazis et al. simulated
the droplet is drawn as time proceeds. The time origin, t ¼ 0, corre-
the experiment of Field et al., that is, high-speed droplet impact, using
sponds to the instant at which the shock wave interacts with the
a thermodynamically well-posed model incorporating phase change.
droplet. This interaction results in the transmission of a shock to
They successfully demonstrated that such models are well adapted to
the droplet [Fig. 1(a)], while part of the incident shock is diffracted
simulate the growth of bubbles and to examine its effect on the droplet
around the droplet. The transmitted shock is a compression wave
dynamics. However, a direct comparison of the numerical results with
the experimental observations revealed that the numerical model sig- that spherically propagates in the stream direction. When the
nificantly overestimates the size of the bubble cloud. This is because of transmitted shock meets the droplet boundary, and as a conse-
the thermodynamic equilibrium assumption, corresponding to an quence of the large water-to-air acoustic impedance ratio, the
instantaneous equilibrium of pressures, temperatures, and velocities, transmitted shock reflects at the interface as an expansion wave
in other words an analogous to infinite relaxation rates for the pres- [Fig. 1(a)], thereby forming low-pressure regions in the internal
sures, temperatures, and velocities, which enable the instantaneous flow field. At the early stage, the acoustic ray theory has shown this
expansion of the gas phase when subjected to a tensile wave. reflection to be a two-segment wavefront. On reaching the down-
In this work, we introduce a multiphase numerical model, in stream droplet surface, the transmitted shock is completely
velocity equilibrium, with a finite pressure-relaxation rate, l, to reflected back [Fig. 1(b)]. The only expansion wave remains and
address the over-expansion of the gas phase as previously reported for propagates upstream by converging and amplifying due to the
an infinite l. The finite pressure-relaxation rate is defined on the droplet curvature. The low-pressure region generated by the
0; 1 range. Shock-induced cavitation primary resulting from hetero- expansion wave thus locally exposes the liquid to a pulling force,
geneous cavitation, we do not account for phase change. The droplet which, under some conditions, results in the cavitation and growth
containing preexisting nuclei is modeled as a liquid–gas mixture. of bubbles [Figs. 1(a)–1(d)]. Once the convergence of the expan-
Considering the difference in the acoustic impedance between both sion wave is completed, it diverges by shaping a horseshoe
phases, such a modeling enables to simulate each phase response, [Fig. 1(d)]. Before this transition, portion of the expansion wave
within the mixture, to compression and expansion effects. We first crosses at the droplet axis where the bubbles eventually meet, and
simulate the spherical collapse of an air bubble in a free field, over the form a single bubble cloud. The bubble cloud collapses over time
l range, and compare the results with the solution of the [Figs. 1(d)–1(f)], while the expansion wave continues to propagate
Keller–Miksis equation to eliminate l values that do not agree with upstream. When the collapse is completed, a spherical shock wave
the theoretical bubble behavior. Second, we simulate the experiment of (CiS) originating from the cloud center is emitted [Fig. 1(f)]. Upon
Sembian et al. with a Mach 2.4 shock wave for which cavitation bub- reaching the droplet interface, the CiS similarly reflects as an
bles have been imaged. After investigating the influence of l on the expansion wave [Fig. 1(g)], which, under some conditions, may
shock-induced cavitation, we calibrate the finite pressure-relaxation also result in the cavitation and growth of bubbles. The wave pat-
rate against the experimental image and eventually validate the cali- tern drawn in Figs. 1(f) and 1(g) corresponds to some of the suc-
brated value by computing the experiment with a Mach 1.75 shock cessive reflections of the transmitted shock, that is, the reflections
wave. A phenomenological analysis of the shock-induced cavitation is of the TS reflection. Note that when the expansion wave reflects at
eventually proposed by interpreting the droplet internal wave pattern, the droplet interface, it transforms into a compression wave.

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-2

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 1. (a)–(g) Sketch of the internal wave pattern and phenomenology of the shock-induced cavitation within a liquid droplet. Time is indicated with a red-to-blue colormap,
with t ¼ 0 the instant at which the incident shock reaches the droplet. Not all internal reflections are drawn for the sake of clarity and educational purposes.

B. Analytical wave description equations of the caustic surface associated with the kth reflected wave-
The wave pattern inside a spherical water droplet impacted front, which are given by
with a planar shock wave has recently been extensively analyzed. xcaustic ¼ Rd f ðaÞ cos ck þ Rd ½ f ðaÞ  1 cos ðck  2hÞ; (2a)
Within the context of ray theory, Biasiori-Poulanges and El-Rabii
described the time-dependent shape of the internal wavefront ycaustic ¼ Rd f ðaÞ sin ck þ Rd ½ f ðaÞ  1 sin ðck  2hÞ; (2b)
whose dominant feature has been shown to be the existence of where
cusp singularities, and examined in detail the focusing of the singly
reflected wavefront. Authors also derived the parametric equations 1 2n2 ðk  1Þ sin 2a  sin 2h
f ðaÞ ¼ : (3)
for the transmitted wavefront and its multiple internal reflections, 2 n2 ð2k  1Þ sin 2a  sin 2h
which read

xM ¼ ½cl t  nRd ð1  cos aÞ  2ðk  1ÞRd cos h cos ðck  hÞ

Rd cos ðck  2hÞ; (1a)

yM ¼ ½cl t  nRd ð1  cos aÞ  2ðk  1ÞRd cos h sin ðck  hÞ

Rd sin ðck  2hÞ; (1b)
where k ¼ 1 corresponds to the transmitted shock and k ¼ 2; 3; … to
the successive internal reflections, and n is the water-to-air sound
speed ratio cl =cg and ck ¼ 2kh  a  ðk  1Þp. Rd is the droplet
radius. We denote fk the wavefront associated with the kth reflection.
The incident and refraction angles, a and h, are related by the funda-
mental law of refraction, sin h ¼ n sin a.
As visible in Fig. 1, internal reflections are two-segment fronts,
which exhibit a singular point. These points trace out surfaces that are
the caustics of the associated kth reflection, which also have a singular FIG. 2. Caustic traced out by the singular point ðsÞ of the singly reflected wave-
point (Fig. 2). From the singularities of the energy flux density of the front. The caustic exhibits a cusp ðcÞ. The red-to-blue colored segments are the
refracted wave, Biasiori-Poulanges and El-Rabii derived the parametric singly reflected wavefronts as time proceeds.

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-3

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

The cuspidal point of the caustic is located on the droplet axis A. Governing equations
yc ¼ 0, while the x-coordinate has been found to be
The thermodynamically well-posed, pressure- and temperature-
ð1Þk n disequilibrium, multi-component flow model conserves mass,
xc ¼ Rd : (4) momentum, and total energy. It reads for N phases
ð2k  1Þn  1
@ak
þ u  rak ¼ dpk ;
@t
C. Problem dimensions @ak qk
þ r  ðak qk uÞ ¼ 0;
The Mach number M of the shock wave, the Weber number We, @t (7)
and the Reynolds number Re are defined as @qu
þ r  ðqu  u þ pIÞ ¼ 0;
@t
Us qU 2 d0 qUd0 @ak qk ek
M¼ ; We ¼ ; and Re ¼ : (5) þ r  ðak qk ek uÞ þ ak pk r  u ¼ pI dpk ;
c r l @t
In the configuration of the shock–droplet interaction (high-speed where ak, qk, pk, and ek are the volume fraction, density, pressure, and
droplet impact, respectively), Us is the incident shock wave velocity internal energy of each phase, respectively, for which k indicates the
(impact velocity, respectively), c is the gas sound speed in the pre- phase index. The mixture density and pressure are
shocked state (sound speed in the liquid, respectively), q is the density X
N X
N
of the post-shocked gas (droplet density, respectively), U is the post- q¼ ak qk and p¼ ak pk ; (8)
shocked gas velocity (impact velocity, respectively), l is the dynamic k¼1 k¼1
viscosity of the gas (dynamic viscosity of the liquid, respectively), and
r is the surface tension coefficient and d0 is the diameter of the cylin- while the mixture total energy is
drical droplet. 1
Table I reports high values of We and Re, indicating that, in both E ¼ e þ jjujj2 ; (9)
2
experiments, the inertial forces dominate the flow over the surface ten-
sion and the viscous forces, respectively. where e is the mixture-specific internal energy
In addition to the shock–droplet interaction and high-speed X
N
droplet impact configurations, we herein also simulate the spheri- e¼ Yk ek ðqk ; pk Þ: (10)
cal collapse of an air bubble in a free field. Viscous and capillary k¼1
effects are trivially shown to be also negligible by computing the
Rayleigh–Plesset equation. In this work, inviscid flows are there- In Eq. (10), ek ðqk ; pk Þ is defined via an equation of state (EOS) and Yk
fore modeled and capillary effects are not accounted for. are the mass fractions
The phenomenology of the shock-induced cavitation within a ak qk
Yk ¼ : (11)
droplet is described using dimensionless parameters. Unless otherwise q
specified, non-dimensionalization of the space and time variables, L
and T, is done using the initial droplet diameter d0 and the sound Herein, we consider two-phase mixtures of gas (g) and liquid (l) for
speed in water cl which the gas is modeled by the ideal-gas EOS

~¼ L
L ~ ¼ T cl ;
and T (6) pg ¼ qg ðcg  1Þðeg  eg;ref Þ; (12)
d0 d0
and the liquid is modeled by the stiffened-gas (SG) EOS
where ð~Þ denotes a non-dimensional quantity.
pl ¼ ql ðcl  1Þðel  el;ref Þ  cl p1 ; (13)
III. NUMERICAL MODELLING
19
We use herein a slightly modified version of the modeling where c, eref , and p1 are model parameters. The interfacial pressure
proposed by Schmidmayer et al.18 to simulate the compression and is defined as
expansion of each phase within the liquid–gas mixture, while 0 1
ignoring phase change. The modification is only related to the X
N X
N
@pk zj A
form of the pressure-relaxation terms (right-hand side) and is
k j6¼k
detailed in Sec. III B. pI ¼ ; (14)
P
N
ðN  1Þ zk
TABLE I. Weber and Reynolds numbers associated with the experiments of k
Sembian et al. and Field et al.
where zk ¼ qk ck and ck are the acoustic impedance and speed of
Configuration We Re sound of the phase k, respectively.
Since pressures are in disequilibrium here, the total energy equa-
Shock–droplet1 103 106 tion of the mixture is replaced by the internal-energy equation for
High-speed impact2 106 106 each phase. Nevertheless, the conservation of the mixture total energy
can be written in its usual form

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-4

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

@qE for a given mixture and flow regime, only one value within this range
þ r  ½ðqE þ pÞu ¼ 0: (15)
@t accurately reproduces the physics. This value changes from one con-
We note that (15) is redundant when the internal energy equations are figuration to another and must be determined by comparison with
also computed. However, in practice, we include it in our computa- appropriated experimental data.
tions to ensure that the total energy is numerically conserved, and thus
preserve a correct treatment of shock waves. C. Numerical method
Based on the hyperbolic study, the mixture speed of sound, also We numerically solve Eq. (7) using a splitting procedure between
called frozen speed of sound, is derived as the left-hand-side terms associated with the flow and the right-hand-
side terms associated with our relaxation procedure.
X
N
The left-hand-side terms are solved by an explicit finite-volume
c2 ¼ Yk c2k ; (16)
k¼1
Godunov scheme where, to ensure the conservation of total energy, a
procedure correcting the non-conservative terms of the internal-
which is found to be in agreement with previously reported energy equations is required and it uses the mixture total-energy rela-
expression.20 tion (15). The method corrects the total energy before the relaxation
We also recall that the model is in velocity equilibrium, respects procedure, during the flux computation of the hyperbolic step, and
the second law of thermodynamics, and is hyperbolic with eigenvalues therefore allows finite or infinite relaxations.18
either equal to u or u 6 c, where u is the velocity in the x direction. The relaxation terms (system of ordinary differential equations)
are integrated with a first-order, explicit, Euler scheme with time step
B. Expression of dpk subdivisions.18 The number of subdivisions is adapted at each time
For the pressure-relaxation terms between the phases, dpk reads step to verify the volume-fraction and pressure constraints. During
under its general form this procedure, if the pressures are completely relaxed, that is, a unique
pressure for all phases, we terminate the Euler scheme and we perform
X
N
from the initial state an infinite-relaxation procedure20 to guarantee a
dpk ¼ lk;j ðpk  pj Þ; (17) unique pressure and better estimate the solution. This also assures a
j6¼k
faster computation.
where the relaxation coefficients lk;j , related to the k–j interactions (j As a side note, after applying an infinite pressure relaxation
are components different from k) and appearing in the original form (l ¼ 1), the model converges to the mechanical-equilibrium model
of the complete disequilibrium model,21 can be expressed under differ- of Kapila et al.24 and the effective mixture speed of sound matches
ent forms. Wood’s
In most if not all the literature, for example, Schmidmayer
1 X
N
ak
et al.,18 Saurel et al.,20 Baer and Nunziato,22 Saurel and Abgrall,23 the ¼ : (19)
relaxation coefficients are taken as unique and constant for all interac- qcw k¼1 qk c2k
2

tions, that is, lk;j ¼ l.

Herein, we propose to use a different approach and to express A second-order-accurate MUSCL scheme with two-step time
them as lk;j ¼ ak aj l, where l is a constant parameter. This leads to integration is used,25 where the first step is a predictor step for the sec-
ond and the usual piece-wise linear MUSCL reconstruction26 with the
X
N monotonized central (MC)27 slope limiter is used for the primitive
dpk ¼ lak aj ðpk  pj Þ: (18) variables.
j6¼k In order to resolve the wide range of spatial and temporal scales
P of wavefronts and interfaces, an adaptive mesh refinement technique
First, this expression is consistent for N phases, meaning k dpk ¼ 0,
and it does not alter the model properties (first and second law of ther- is employed.28 The cell i is refined when the following criterion is
modynamics and hyperbolicity). Second, the combination of the vol- fulfilled
ume fractions allows specific behaviors: jXNbði;jÞ  Xi j
> e; (20)
• a dilute phase takes time to reach equilibrium with the carrier minðXNbði;jÞ  Xi Þ
fluid,
• two phases with approximately the same volume fraction (close where X is a given flow variable. The criterion is tested for all neigh-
to 0.5), for example, interfaces, or bubbles or droplets of approxi- boring cells, denoted by the subscript Nb(i, j), where the jth cell is the
corresponding neighbor of the ith cell. The threshold is conservatively
mately the size of the computational cells, quickly reach equilib-
set to e ¼ 0:02. The above refinement criterion is tested for density,
rium, and
• two dilute phases within a carrier fluid hardly communicate and velocity, pressure, and volume fraction and refines the cell if the crite-
rion is fulfilled for any of these variables. In addition, neighboring cells
therefore take a significantly long time to reach equilibrium
of refined cells are also refined to prevent oscillations as well as loss of
through their own interactions.
precision.
Hence, the local relaxation rate adapts over time to the volume- This modeling is implemented in ECOGEN,25 which has been
fraction configuration within the computational cell. Note that l is a validated, verified, and tested for finite-relaxation rate in various setups
finite parameter, which can be selected in the 0; 1 range. However, such as gas bubble dynamics problems, including free-space and

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-5

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

near-wall bubble collapses, and liquid–gas shock tubes. Using infinite-

relaxation rate, it has also been validated for surface-tension problems
as well as column and droplet breakup due to high-speed flow (see,
e.g., Refs. 18 and 29–33).

D. Computational setup
The determination of the pressure-relaxation rate l is a two-step
approach. The first step consists in reducing the l-range by simulating
the spherical collapse of a bubble in a free field and determined l val-
ues that agree with the predicted bubble dynamics given by the
Keller–Miksis equation.34 The second step consists in calibrating l
against the experiment of Sembian et al., where a Mach 2.4 planar
shock wave interacts with a cylindrical water droplet. In this experi-
ment, the growth of a bubble cloud has been imaged. The calibrated l
is eventually validated against a second experiment of Sembian et al.
with M ¼ 1.75, for which no bubble cloud has been recorded. This
determination procedure is done for a given fluid, that is, water drop-
let. To evaluate the sensitivity of l on the material properties of the
mixture constituting the droplet, the experiment of Field et al., also
showing shock-induced cavitation, is simulated. It consists in a spheri-
cal gelatin droplet, which is impacted by a metallic slider at 110 m/s.

1. Spherical bubble collapse in a free field

This test case aims to present the behavior of the relaxation rate
during the spherical collapse of a bubble in a free field [Fig. 3(a)]. To
reduce the computational cost, a one-dimensional (1D) domain of
3 mm long is used with spherical axisymmetry to mimic a three-
dimensional bubble.31 The domain consists in high-pressure water at
p1 ¼ 50 atm with density ql ¼ 1000 kg=m3 . An air bubble is located
at the origin of the computational domain. The initial bubble radius is
R0 ¼ 0:1 mm. The bubble pressure and density are pb ¼ 3550 Pa (i.e.,
vapor pressure) and qb ¼ 0:027 kg=m3 , respectively. Initial velocities FIG. 3. (a) Volumetric representation of the spherical collapse of a bubble in a free
are nulls. One could note that this configuration enforces an initial field. (b) and (c) Computational setups corresponding to the experiments of
interface disequilibrium. A nonreflecting boundary condition is used Sembian et al. and Field et al., respectively.
at the far field limit, while a symmetry boundary condition is used at
the origin of the domain. The mesh contains 150 cells from 0 to initialized inside the domain and travels from left to right in air at
0:3 mm, which corresponds to 100 cells per bubble diameter, and atmospheric conditions. For the incident shock Mach number M, the
then, the grid is stretched non-uniformly to accommodate the large initial flow field is determined from the Rankine–Hugoniot jump rela-
computational domain. tions using a downstream density of 1:204 kg=m3 and a 1 atm pres-
sure. The water has a density of 1028 kg=m3 and is modeled using the
2. Cylindrical droplet interaction with a planar shock SG EOS [the water density is calculated to agree with the sound speed
wave in water calculated from the experimental observations of Sembian
et al. ( 1512 m/s), when using the Eq. (13)],19 with c ¼ 2:35 and
The two-dimensional (2D) computational setup, corresponding p1 ¼ 109 . The initial air volume fraction in water is 106 . This corre-
to Sembian et al. experiments, is shown in Fig. 3(b), where the x axis is sponds to the preexisting nuclei in non-purified water. We recall that
the axis of symmetry on which the center of the cylindrical droplet of considering the difference in the acoustic impedance between both
radius Rd is located. Simulations are performed in a ½12Rd  6Rd  rect- phases, the modeling enables to simulate each phase response, within
angular computational domain. A symmetric boundary condition is the mixture, to compression and expansion effects, that is, heteroge-
applied to the bottom side of the computational domain, and non- neous cavitation (without phase change).
reflective boundary conditions are imposed to the remaining bound-
aries. The droplet is initially located at the center and is assumed to be
3. High-speed cylindrical droplet impact
in mechanical equilibrium with the surrounding air. The initial droplet
is resolved by 100 cells per diameter. Adaptive mesh refinement The computational setup corresponding to Field et al. experi-
(AMR) composed out of three grid levels and adapted to follow the ments is shown in Fig. 3(c). The spherical droplet is modeled using a
flow discontinuities is used. The AMR level is selected based on the 2D formulation, where the y axis is the axis of symmetry on which the
analysis of the grid sensitivity (see Sec. III E). The shock wave is center of the droplet of radius Rd is located. Simulations are performed

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-6

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

in a ½6Rd  6Rd  square computational domain. A wall boundary con- One can observe the convergence of the bubble-cloud radius. We
dition and a symmetry boundary condition are applied to the bottom consider the solution with three levels of refinement to be sufficiently
and left sides of the computational domain, respectively. Non- converged for the purpose of the paper.
reflective boundary conditions are imposed to the two remaining The rate of convergence is also presented in Fig. 5 in terms of the
boundaries. The droplet moves downward with a velocity discrete L2-error e as
ui ¼ 110 m=s. The initial droplet is resolved by 100 cells per diameter,
and the three-grid level AMR is used. In the experiments, the droplet 1 X Nt
jjRðti Þ  Rref ðti Þjj
e¼ ; (22)
is made of a 12 wt. % gelatin. It is modeled using the SG EOS13,35 with Nt i¼0 Rref ðti Þ
c ¼ 6:72 and p1 ¼ 3:70  108 . Similarly to the previous computa-
tional setup, the initial air volume fraction within the droplet is 106 . where Nt is the number of time steps in the temporal window ~t
As a simplification of the experiments, the present setup does not 2 ½0; 3:25; Rðti Þ is the bubble radius at time ti of our simulations, and
account for the material properties of the metallic slider, used in the Rref is the reference solution, here chosen as the solution of the simula-
experiments, and simulates a droplet impacting a wall at velocity tion AMR level 3. We see that the method converges at 1.7 order,
110 m=s.17 matching the expected rate for AMR simulations of flows mainly gov-
erned by tension waves and exhibiting shocks and interfaces.
E. Grid convergence for finite pressure relaxation
IV. RESULTS AND DISCUSSION
To consider the spatial convergence of the numerical method,
a grid resolution study is performed by simulating the interaction In this section, we discuss the results from the three-step proce-
of the cylindrical water droplet with a Mach 2.4 plane shock wave dure for the determination of the pressure-relaxation rate.
at four different resolutions (Fig. 4). Keeping constant the initial
mesh size to 100 cells per diameter, four AMR grid levels are used
A. Spherical bubble collapse in a free field
which eventually result in 100, 200, 400, and 800 cells per diame-
ter. For this study, the pressure-relaxation rate is chosen equal to The finite pressure-relaxation-based method allows for an infinite
l ¼ 10. As time proceeds, the grid sensitivity is examined by com- range of pressure-relaxation rate. As a first approach, it is instructive to
paring the growth and collapse of the shock-induced bubble evaluate the sensitivity of l by simulating the spherical collapse of a bub-
cloud, within the droplet. Note that in this work, the “shock- ble in free field [Fig. 3(a)] for various orders of magnitude of l and com-
induced bubble cloud” is a liquid–gas mixture. Figure 4 plots the pare the bubble response to the analytical solution of the Keller–Miksis
equivalent radius of the gas phase within the droplet, denoted R equation,34 and the compressible form of the Rayleigh–Plesset equation.
and defined as Assuming a spherical collapse, the Keller–Miksis equation is based on
an asymptotic expansion in the Mach number of the Bernoulli equation.
 ð 1=2
1 The use of the Keller–Miksis equation in the present work is predicated
R¼ ag dV ; (21) on larger measured relative errors than errors related to the asymptotic
p Vd
expansion and the inherent presumption of sphericity. This assumption
where Vd is the droplet volume. is borne out by the results displayed in Fig. 6, which discloses the com-
parison between numerical simulations and the Keller–Miksis solution

FIG. 4. Growth and collapse of the gas phase within the droplet at four grid resolu-
tions. The pressure-relaxation rate is l ¼ 10. The droplet is initially resolved by 100 FIG. 5. Convergence order of the numerical method. The discrete L2-error, e, repre-
cells per diameter. sentative of the convergence rate is given by Eq. (22).

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-7

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 6. (a) Radial bubble-wall evolution for a spherical bubble collapse with p1 =pb ¼ 1427 and N ¼ 100 cells, with (b) a magnified view in the rebound region.

by plotting the radial bubble-wall evolution. In agreement with the 1D the minimum radius up to 19%. The best agreement with theory is
formulation, the effective bubble radius, Rb, is defined as given for l ¼ 1, where the relative error er on the minimum bubble
radius, with respect to the Keller–Miksis solution, is 10%. Note that
X
N
Rb ¼ ag;i Vc;i ; (23) minðRb =R0 Þjl¼10  minðRb =R0 Þjl¼1 . One should also note that
i¼1 decreasing errors is expected for refined grids and for three-
dimensional simulations, for which the non-conservative terms of the
where N is the number of grid cells, and ag;i and Vc;i are the gas vol- axisymmetry are absent. Major discrepancies are however reported for
ume fraction and the volume of the ith cell, respectively. The dimen- l ⱗ 0:1, which significantly overestimate the minimum bubble radius.
sionless time is given by the ratio of the dimensional time t with the From this first sensitivity analysis on l, the range of pressure-
Rayleigh collapse time relaxation rate can be restricted to 0:1; 1. Note that this range
rffiffiffiffiffiffi can be extended when refining the mesh size, so that l 2 a; 1 with
ql
tr ¼ 0:915R0 ; (24) a ! 0 for an infinitely small mesh size. The l-range 0:1; 1 is here
p1
valid for a reasonable resolution of 100 cells/diameter. In the following,
which is the nominal total collapse time, that is, the time required for we calibrate l upon experimental observations, which is therefore
the bubble to complete its collapse.36 Note that the solutions are only expected to be in the l-domain here determined.
displayed until t ¼ 1:05tc , right after the minimum bubble radius is
reached, since the subsequent rebounds for large pressure ratios for
B. Cylindrical droplet interaction with a planar shock
the Keller–Miksis equation are well-known to be physically inaccu-
wave
rate.37 When comparing the Keller–Miksis equation with our simula-
tions, our results are expected to converge toward the analytical In this section, we simulate the experiments of Sembian et al. to
solution up to the first rebound because of the reducing diffusion. first investigate the influence of l on the droplet internal flow field,
Beyond the collapse, the solution of the Keller–Miksis equation is here and then to calibrate and validate the corresponding pressure-
not accurate as the Mach of the interface is high at the time of the col- relaxation rate. Upon validation, parametric simulations are eventually
lapse, and so the compressibility effects are important. These condi- run to evaluate the critical Mach number from which, bubbles are
tions appear to be out of the validity domain on which the likely to grow.
Keller–Miksis equation has been derived (i.e., low Mach number). The
convergence should thus not occur. 1. Influence of the pressure-relaxation rate
Figure 6 and Table II show that pressure-relaxation rates 1ⱗ l
ⱗ 1 are in satisfying agreement with theory, while underestimating The effect of the pressure-relaxation rate is investigated by com-
paring numerical results at four different rates, 1, 10, 100, and 1, and
by using the computational setup dedicated to the shock–droplet
TABLE II. Relative error, er , between the numerics and the Keller–Miksis solution on interaction [Fig. 3(b)]. Figure 7 shows that larger values of R are
the minimum bubble radius.
reached for higher values of l. The maximum radius asymptotically
increases with l, so that maxðR=R0 Þjl¼100  2maxðR=R0 Þjl¼10 and
M 0.01 0.1 1 10 1
Rðl ¼ 100Þ  Rðl ¼ 1Þ. The simulation thus converges to an infi-
er (%) 735 275 10 19 18 nite pressure-relaxation rate modeling as l increases. Conversely,
smaller R are reported for lower values of l. Numerical results

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-8

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

As a result of this scattering process, only f2;a remains and continues

to propagate while expanding the gas phase within the droplet [see
Fig. 8 at ~t ¼ 1:21 for l ¼ 10 and l ¼ 100, and the schematics in
Fig. 8(c)]. In this context and assuming an infinite pressure-relaxation
rate, the volume of the bubble cloud corresponds to volume through
which segment f2;a propagates. The complete singly reflected wave-
front (f2;a and f2;b ) is known to be a spherically converging wave-
front whose intensity is amplified due to the focusing. The maximum
intensity is reached at the cuspidal point of the caustic (c). When the
wavefront meets c, it then spherically diverges by propagating
upstream (Fig. 8 at ~t ¼ 1:56  1:90 for l ¼ 1). However, due to the
scattering of f2;b , the energy collected at c decreases as l increases.
This is qualitatively visible in Fig. 8 at time ~t ¼ 1:56, where a slight
diverging wave is observed for l ¼ 10 (light blue box), and no discon-
tinuity is detected for l ¼ 100. Note that the Schlieren in the light blue
box has been post-processed with a contrast enhancement algorithm
to highlight the wavefront. In the case of l ¼ 10, the diverging wave-
front propagating upstream weakens as traveling and reflects again at
the droplet boundary (Fig. 8 at ~t ¼ 1:90). The resulting multiple wave-
FIG. 7. Effect of the pressure-relaxation rate on the growth and collapse of the gas fronts are finally invisible on the numerical Schlieren and other com-
phase within the cylindrical droplet. puted flow fields. From time 1.56 to 2.56 (2.93, respectively), the
simulation with l ¼ 10 (l ¼ 100, respectively) does not exhibit any
computed for l ¼ 1 seem to approximate simulations that do not flow discontinuity related to the internal wave reflections. Between
account for cavitation. time 2.56 and 2.76 (2.93 and 3.28, respectively), the bubble cloud col-
The image sequences of Fig. 8 show the internal flow field of the lapse has been completed in the case l ¼ 10 (l ¼ 100, respectively)
droplet and illuminate the various discontinuities using a numerical and the collapse-induced shock (CiS) has been emitted (Fig. 8 at ~t
Schlieren (i.e., exponential of the negative, normalized density gradi- ¼ 2:76 for l ¼ 10). Upon collapse, the bubble cloud does not experi-
ent32,38). Image sequences are displayed for simulations with l in the ence an additional growth phase, which allows the reflection of the CiS
[1, 10, 100] range. For l ¼ 1, the numerical Schlieren images are over- at the rear side of the droplet to propagate upstream without being
laid with the predicted wavefronts computed from Eq. (1) (solid red scattered (Fig. 8 at ~t ¼ 2:93 for l ¼ 10 and ~t ¼ 3:28 for l ¼ 100). The
lines), as well as the caustic associated with the singly reflected wave- CiS is a compression wave, which implies that its reflection is an
front given by Eq. (2) (dashed red line). This superposition reveals an expansion wave. This expansion is strong enough to generate a new
excellent agreement between the simulation and the theory. The para- bubble cloud as clearly visible with the third discontinuities at time
metric equations of the transmitted shock and the kth reflections being 3.28 for l ¼ 100. This last phenomenon is consistent with Figs. 4 and
derived by assuming they propagate in an homogeneous phase, and 7, where a rebound in the time-dependent R=R0 -ratio is observed
this agreement indicates the absence of bubble growth within the between ~t ¼ 3:00 and ~t ¼ 3:25.
droplet. This is consistent with Fig. 7, where the dimensionless ratio
R=R0 does not increase as time proceeds. Note that both the theoreti- 2. Pressure-relaxation rate calibration with experiment
cal and numerical singular points ðsÞ well trace out the caustic associ-
ated with the first reflection (k ¼ 2). The pressure field within the droplet has been previously shown
Now considering the simulation results for l ¼ 10 and l ¼ 100, to be dependent on the pressure-relaxation rate (Sec. IV B 1). We now
Fig. 8 shows a very similar phenomenology. It however displays strong determine the pressure-relaxation rate by calibrating the simulation
discrepancies with the simulation with l ¼ 1 that the comparison of against the experimental data of Sembian et al. We first compare and
the theoretical wave pattern with the numerical Schlieren, along with analyze three different orders of magnitude of l. For l equals to 1, 10,
Fig. 7, evidences to be related to the growth of the gas phase within the and 100, Fig. 9 displays a comparison of the numerical solutions with
droplet. First, the transmitted shock propagates downstream. Upon an experimental Schlieren photograph at same initial conditions and
interaction with the droplet boundary, it reflects as an expansion wave. time. The numerics plot the colored volume fraction of air (yellow-to-
As previously described, this wavefront consists in two segments, f2;a black colormap) overlaid with the grayscale Schlieren. Note that the
and f2;b [see Fig. 8 at ~t ¼ 0:87 for l ¼ 1 and l ¼ 10, and the sche- numerical Schlieren can only contour the interface between the gas
matics in Fig. 8(a)]. This expansion wave generates a low-pressure and the liquid phase, while an experimental Schlieren photograph,
region, which, under some conditions, is likely to expand the gas phase sensitive to the first derivative in density, images the line-path inte-
within the droplet. When the expansion wave is strong enough to gen- grated volume of the cloud (dark area).
erate a bubble cloud, it modifies the canonical wave pattern observed in Test case l ¼ 1 shows a very good agreement in both shape and
the simulation with l ¼ 1. In this configuration, the forehead segment location of the internal wavefronts, but does not disclose the existence
f2;a initiates the growth of the gas phase, which scatters the incoming of a bubble cloud as evidenced by the volume fraction mapping and
subsequent segment f2;b [see Fig. 8 from ~t ¼ 0:87 for l ¼ 100 and the lack of collapse-induced shock (CiS). This suggests that l ¼ 1
from ~t ¼ 1:21 for l ¼ 10, and the schematics in Figs. 8(a) and 8(b)]. underestimates the experimental relaxation rate. The internal wave

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-9

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 8. Influence of l on the internal flow field of a 22-mm-diameter cylindrical droplet interacting with a planar shock wave propagating at M ¼ 2.4. (a)–(c) Schematics of the
bubble formation and the scattering process.

structure reported for l ¼ 10 does not agree neither with the experi- Discrepancies are also observed by comparing the center of the experi-
ments nor the theory, but however reveals a CiS generated by the pre- mental bubble cloud and the origin of the CiS. These observations
vious growth and collapse of a bubble cloud. The absence of the indicate that l ¼ 10 overestimates the experimental relaxation rate.
wavefront as predicted, and seen for l ¼ 1, results from the early The pressure-relaxation rate l ¼ 100 is eventually simulated. As
development of the cloud that scattered the segment f2;b . expected, it also overestimates the experimental relaxation rate. The

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-10

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 9. Comparison of the internal droplet structure, at ~t ¼ 2:72, between (top) numerical simulations at different l and (bottom) experiments of Sembian et al. The upper-
halves display the volume fraction of air (yellow-to-black colormap) overlaid with numerical Schlieren images (white). The lower-halves disclose experimental Schlieren images,
which, for l ¼ 1, is compared with the theoretical predictions given by Eq. (1) for k 2 ½1; 7. Reproduced with the permission from Sembian, et al., “Plane shock wave interac-
tion with a cylindrical water column,” Phys. Fluids, 28, 056102 (2016). Copyright 2016 AIP Publishing.

internal wave structure does not agree with the experiments due to the the numerical Schlieren. Figure 13(a) reports an excellent agreement
obvious growth of the gas phase, which has not yet completed its col- both on the internal wave structure (dashed red line), and the size and
lapse. Finally, comparing simulation results for l equals to 1, 10, and location of the bubble cloud (solid red line). Within the region r
100 indicates that the experimental relaxation rate should be between delimited by the dashed red line, one can note the growth of a second
1 and 10. bubble cloud. It results from the interaction of the third reflected
To identify the pressure-relaxation rate exhibiting the better transmitted shock (f3 , red dashed line) with the CiS reflection. It is
agreement with the experiments of Sembian et al., we compared the
center of the bubble cloud and the collapse time, denoted xc and tc ,
respectively, between the experiments and the numerical simulations.
The experimental center of the cloud is determined by detecting the
cloud contour on the lower halves of Fig. 9 using an edge detection
algorithm, and computing the center-of-mass. The collapse time is
determined based on the image sequence available in Fig. 8 in
Sembian et al. from which we assume the collapse to occur between
frame (e) and (f). Using the position of the internal and external wave-
fronts (see Fig. 10), and knowing the size of the droplet as well as the
shock wave Mach number, the dimensional time of frames (e) and (f)
has been determined. We denote Dtc the time interval between frames
(e) and (f).
Figure 11 shows the functional dependency of xc and tc on the
pressure-relaxation rate, which is estimated using non-linear least
squares fits of the form alb þ c. The two plots exhibit an asymptotic
behavior as l ! 1, which is consistent with the analysis of Fig. 9.
Figure 12 plots the ðxc ; tc Þ-coordinates for various l in the ½3; 1
FIG. 10. Interaction of a planar shock wave at M ¼ 2.4 with a 22-mm-diameter
range. The dark solid line corresponds to the experimental xc -coordi-
cylindrical water droplet. The upper half is a numerical Schlieren and the lower-half
nate, and the gray filled area to Dtc . It appears that the dimensionless is a Schlieren photograph from Reproduced with permission from Sembian et al.
cloud center coordinate ~x c linearly depends on the collapse time ~t c . “Plane shock wave interaction with a cylindrical water column,” Phys. Fluids 28,
The linear interpolation of the ðxc ; tc Þ-coordinates intersects the 056102 (2016). Copyright 2016 AIP Publishing. The agreement in the internal and
experimental xc coordinate on l ¼ 3:5, and within the Dtc window. external wave locations enables to calibrate the dimensional time of the experi-
Figure 13 shows the numerical results for l ¼ 3:5 compared ments. Both the numerics and the experiment image the irregular Mach reflection
over the droplet, which consists in the reflected shock wave, the Mach stem, the
with the experimental Schlieren image. The upper half of Fig. 13(a) slip surface, and the incident shock wave (not visible here) all connected by the tri-
plots the volume fraction of air with a numerical Schlieren, and the ple point. They also image the third internal reflection, f3 , of the transmitted
upper half of Fig. 13(b) maps the mixture pressure field overlaid with shock.

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-11

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 11. Dependence of ~x c and ~t c coordinates on l. Both fits are non-linear least
squares fits of the form alb þ c with (a) a ¼ 0:15; b ¼ 0:90, and c ¼ 0.36,
and (b) a ¼ 2:61; b ¼ 0:73, and c ¼ 3.10.

evidenced by Fig. 13(b) which shows two regions r1 and r2 with

opposite curvatures and propagation directions. Region r1 is driven
by the downstream propagation of the f3 wavefront, while region r2
is induced by the upstream propagation of the CiS reflection. The FIG. 13. Simulation results for l ¼ 3:5 compared with the experimental Schlieren
interplay between f3 and the CiS reflection results in a low-pressure image of Reproduced with the permission from Sembian et al., “Plane shock wave
region at the origin of the growth of a second bubble cloud. A closer interaction with a cylindrical water column,” Phys. Fluids 28, 056102 (2016).
look at the experimental Schlieren image [see the magnified view in Copyright 2016 AIP Publishing. The upper half shows (a) the volume fraction of air
Fig. 13(b)] shows discontinuities, highlighted with dashed-dotted with numerical Schlieren and (b) the mixture pressure field superposed with the
numerical Schlieren.

white lines, which align with the upstream contour of region r2

(white dashed line). We infer these discontinuities to be the CiS reflec-
tion. In light of the comparison displays in Fig. 13, the numerical sim-
ulation very well agrees with the experiment of Sembian et al. (note
that Fig. 10 is plotted for l ¼ 3:5), thus validating the pressure-
relaxation rate l ¼ 3:5.
A descriptive comparison of the numerical simulations with the
experimental observations for l 2 ½1; 3:5; 10, and different times, is
given in Appendix A.

3. Validation and critical Mach number

Sembian et al. carried out experiments at M ¼ 2.4 and M ¼ 1.75.
They observed the growth of a bubble cloud for M ¼ 2.4, but they did
not capture it for M ¼ 1.75. We thus simulate the experiments at
M ¼ 1.75 with l ¼ 3:5 to assess the determined value of the pressure-
relaxation rate. Figure 14 plots the dimensionless equivalent radius of
the cloud for various shock wave numbers. It shows that the gas phase
FIG. 12. Identification of the pressure-relaxation rate corresponding to the experi- inside the droplet does not expand for M ¼ 1.75, which agrees with
ment of Sembian et al. and based on the time and x-location of the cloud center. the experimental observations [maxðRÞ ~  1010 ]. Running the

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-12

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

parametric simulations for M 2 ½1:75; 2:4 to evaluate the critical

Mach number, Mc, from which cavitation occurs. From Fig. 14, we
estimate Mc to be in the [1.9,2.0] range where maxðRÞ ~ varies from
109 to 108 . For M ¼ 1.9 (M ¼ 2.0, respectively), the minimum
pressure behind the expansion wave f2 , measured from its total
reflection to complete focusing, ranges from –0.37 to –1.9 MPa (from
–0.5 to –2.4 MPa, respectively). This is consistent with the pressure
thresholds for the heterogeneous cavitation reported in the literature
and found to be between –0.1 and –1 MPa.39

C. High-speed droplet impact

We previously determined and validated the pressure-relaxation
rate for a water droplet impacted by a planar shock wave. To evaluate
the dependence of l on the material properties, we now simulate the
high-speed impact of a cylindrical gelatin droplet on a solid substrate
[Fig. 3(c)] and compare the results to the experiment of Field et al.
After parametric investigation on the pressure-relaxation rate
and calibration on the experimental observations, the simulation,
which agrees best with the experiment, uses l ¼ 0:5. A comparison
FIG. 14. Parametric simulations of the shock–droplet interaction for various shock between the numerics and the experiment is shown in Fig. 15. The
wave Mach numbers ranging from 1.75 to 2.4. Labels indicate the ðM; lÞ combina-
tion used in the simulations. left-hand side of Fig. 15(a) plots the volume fraction of air, superposed
with a numerical Schlieren. The only numerical Schlieren is plotted in
Fig. 15(b). Figure 15(c) sheds light on common features reported on
simulation for M ¼ 1.75, but with l ¼ 1, however, shows the both numerical and experimental images, used to validate the simula-
significant expansion of the gas phase, which is completely off the tion. The dashed gray line corresponds to the contour of the volume
experimental observations. Hence, we validate the value l ¼ 3:5. Note fraction of air as drawn in Fig. 15(a) with the dashed white line. The
that similar results than for ðM; lÞ ¼ ð1:75; 1Þ would also be red dashed lines are discontinuities extracted from the photograph,
expected for thermodynamically consistent, mechanical-equilibrium, while the white dashed lines have been plotted from the numerical dis-
or thermodynamical-equilibrium models. In addition, we also run continuities. While discrepancies in the shape and the location of all

FIG. 15. Comparison of the numerical simulation (left) with the experiments of Field et al.2 (right). Reproduced with the permission from Field et al., “The effects of target com-
pliance on liquid drop impact,” J. Appl. Phys. 65, 533–540 (1989). Copyright 1989 AIP Publishing. The numerical image shows (a) the colored volume fraction of air overlaid
with a Schlieren image. (b) the numerical Schlieren image alone. (c) is a magnified view of the bubble cloud observed on (b). (d) is a second magnified view of the bubble
cloud showing, in white dashed line, the bubble cloud contour computed from the numerics.

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-13

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

these discontinuities are observed, numerics and experiments present while an asymptotic behavior of the relative error between xc;n and xc
a very similar pattern. Among others, a very good agreement is is reported when l ! 1.
reported on the shape and location of the denser region of the bubble Three regimes of droplet shock-induced cavitation can be
cloud, that is, the centered dark region on the Schlieren images. defined: (i) no-cavitation regime, (ii) convergence-driven cavitation
Finally, we plotted the contour of the bubble cloud computed from the regime, and (iii) the immediate cavitation regime. When the transmit-
numerical simulation [white dashed line in Fig. 15(d)]. We remember ted shock is not strong enough, the droplet never experiences cavita-
that the numerical Schlieren only contours the cloud. The palm-like tion and bubble growth. This is the no-cavitation regime. We note i1
shape of the bubble cloud relatively well agrees with the experimental the intensity of the transmitted shock in this regime. In the other
observation. The location and the curvature of the stem perfectly extreme, the strongest transmitted shock instantaneously results in
match the darker region in the experiments, while the top of the palm cavitation upon the reflection of the transmitted shock as an expansion
connects the streamers experimentally observed. wave. This corresponds to the immediate cavitation. We note i2 the
intensity of the transmitted shock in this regime. However, for trans-
mitted shock with intensity i1 < i < i2 , cavitation does not imme-
V. LOCATION OF THE BUBBLE CLOUD
diately occur upon reflection, but during the convergence of the
The location of the caustic’s cuspidal point ðxc ; yc Þ has been expansion wave, which induces its amplification. This corresponds to
proven to be the highest density of shock-induced cavitation.40 This is the transitional convergence-driven cavitation regime.
because the density of rays tangent to the caustic, which gives a relative We denote, xt, the location of the expansion wave f2 , during its
measure of the focusing strength over the caustic, is maximum at the convergence, where the critical amplitude for the expansion wave to
intersection of the caustic and the droplet axis (y ¼ 0).4 Consequently, sufficiently expand the gas is reached. Locating the droplet center at
previous research efforts on shock-induced cavitation within a cylin- the domain origin so that ðx; yÞ ¼ ð0; 0Þ, the xt-coordinate is bounded
drical droplet assumed the cavitation bubble cloud to appear at the as xc < xt < Rd . Note that xt ¼ Rd is the condition for the immediate
focus of the reflected wavefront ðxc ; yc Þ.2,13 However, a close exami- cavitation regime. The diameter of the bubble cloud measured on the
nation of the bubble cloud center as seen in the experimental observa- x axis is then given by xt  xc . Assuming a volumetric collapse of the
tions shows discrepancies between the theoretical xc value and cloud, this implies that the cloud center Dx is given by
experimental measurements, errors of 42%610% and 23%65% for
xt  xc
the shock–droplet interaction and the high-speed droplet impact, Dx ¼ xc þ : (25)
respectively. As clearly visible in Fig. 8, the center of the cloud on the x 2
axis does not agree with the cuspidal point of the caustic ðcÞ. In the The bottom graph in Fig. 16 plots the xc;n =Dx ratio over the pressure-
following, we denote xc;n the location of the cloud determined from relaxation rate l and shows a very good agreement between the bubble
the numerical simulations. The top graph in Fig. 16 plots the xc;n =xc cloud center measured on the numerical simulation and the computed
ratio over the pressure-relaxation rate l. It is obvious that the theoreti- Dx. Note that the second term in the right-hand side of the Eq. (25),
cal value xc does not agree with the numerical value xc;n . Note that, in xt  xc =2, is a corrective term to Obreschkow et al.’s theory to esti-
the absence of the bubble cloud growth (l ¼ 1), a very good match has mate the location of the cavitation region in the convergence-driven
nevertheless been reported between xc and the focal point of the cavitation regime. Equation (25) agrees to locate the highest density
reflected wave simulated. As l decreases, the xc;n converges toward xc , of shock-induced cavitation at the caustic’s cuspidal point
(limxt !xc Dx ¼ xc ). In the immediate cavitation regime, Dx ¼ xc
þ ðRd  xc Þ=2. In agreement with the numerical simulations and the
experimental observations, and using Eq. (4) for k ¼ 2, the location of the
bubble cloud center during droplet shock-induced cavitation thus obeys
 
n Dx 1 n
1þ : (26)
3n  1 Rd 2 3n  1
Note that the analysis on the cloud center, here shown for the configu-
ration of Sembian et al., is also valid for the high-speed droplet impact
configuration.
VI. CONCLUSION
In this paper, we introduce a multiphase numerical model, in
velocity equilibrium, using a finite pressure-relaxation rate. In the con-
text of heterogeneous cavitation, we demonstrate the finite formula-
tion to be more suitable and effective to simulate shock-induced
cavitation. Based on the shock–droplet interaction experiments of
Sembian et al., where cavitation bubbles within the droplet have been
reported, we calibrate and validate the pressure-relaxation rate, which
is found to be l ¼ 3:5. A parametric investigation on l shows the
FIG. 16. Comparison between the bubble cloud centers measured from the numeri- effect of the bubble cloud of the internal wave structure, which, for a
cal simulations and theoretical predictions. significant growth, scatters the incoming wavefronts resulting in their

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-14

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

annihilation. For l > 3:5, the size, the location, and the collapse time AUTHOR DECLARATIONS
of the bubble cloud are overestimated as l increases, and conversely, Conflict of Interest
these parameters are underestimated when l, below 3.5, decreases. We
The authors have no conflicts to disclose.
eventually determined the critical shock Mach number Mc from which
shock-induced cavitation is possible, to be 1:9 < Mc < 2:0. This is
consistent with Sembian et al. who observed cavitation at M ¼ 2.4, Author Contributions
while no bubbles has been imaged at M ¼ 1.75. Luc Biasiori-Poulanges: Conceptualization (equal); Data curation
Complementing the phenomenological analysis, based on the (equal); Formal analysis (equal); Investigation (equal); Methodology
numerical results, with ray theory from geometrical acoustics, we dis- (equal); Validation (equal); Writing – original draft (equal); Writing –
cussed the theoretical location of the bubble cloud center. Although review & editing (equal). Kevin Schmidmayer: Conceptualization
usually approximated to be located at the focal point of the singly (equal); Data curation (equal); Formal analysis (equal); Investigation
reflected wavefront, we actually show that it depends on the magnitude (equal); Methodology (equal); Validation (equal); Writing – original
of the continuously amplifying expansion wave, so that Dx
xc jk¼2 . draft (equal); Writing – review & editing (equal).
The correction of the Dx is done by adding the ðxt  xc Þ=2 term to the
f2 ’s focal point location xc . Future work should address the analytical DATA AVAILABILITY
determination of the xt location, which is here numerically determined. The data that support the findings of this study are available
To evaluate the sensitivity of the pressure-relaxation rate on the from the corresponding author upon reasonable request.
material properties where the cavitation occurs, we then simulated the
experiment of Field et al. consisting in the impact of a solid surface
onto a gelatin droplet. Field et al. observed the cavitation of bubbles APPENDIX A: TIME SEQUENCES OF SHOCK–DROPLET
under the internal reflection of the water hammer shock. Our calibra- INTERACTION
tion procedure identified l ¼ 0:5 to best match the experimental Figure 17 shows image sequences for the interaction of a 22-mm-
observations. The droplet dependency on l is consistent with the pre- diameter cylindrical droplet with a planar shock wave propagating at
vious observations in the simulation of Sembian et al.’s configuration. Mach 2.4. The upper halves are the colored volume fraction of air (yel-
Simulating the spherical collapse of a bubble in a free field and com- low-to-black colormap) overlaid with a numerical Schlieren in gray-
paring the bubble dynamics to the analytical solution of the scale. The lower-halves are the experimental Schlieren photographs
Keller–Miksis equation, the 0:1; 1 range has been found to very well from Sembian et al. The comparison is disclosed for l equals 1, 3.5,
agree with theory. Encouragingly, the two pressure-relaxation rates and 10. On frames (a) and (b), the three simulations are in excellent
determined fits in this l interval. Note that, in this work, the calibrated agreement with the experiments. The transmitted shock and the expan-
values of l are only valid for the two configurations we have been con- sion wave both agree on time and space. The simulations do not exhibit
cerned with and for an initial gas volume fraction ag ¼ 106 . Effects the expansion of the gas phase within the droplet, which is in accor-
of ag on the phenomenology of the shock-induced cavitation require dance with the absence of bubble cloud on the photographs. On frame
additional research efforts based on complementary experiments, (c), the three numerical wave patterns are still in line with the experi-
which are out of the scope of this paper. However, to shed light on the ments. However, discrepancies on the volume fraction of air arise for
possible influence of ag on the physics and the calibrated l, additional l ¼ 10 (red box), which shows the growth of the gas phase downstream
simulations have been run and discussed in Appendix B. the expansion wave, while no bubble cloud is imaged in the experiment.
Future works would be to integrate phase change in the numeri- On frame (d), only l ¼ 1 and l ¼ 3:5 well simulate the diverging
cal model, which would require to calibrate again the pressure- expansion wave, which is invisible in the simulations with l ¼ 10.
relaxation rate along with the chemical-potential relaxation rate. A Simulations with l ¼ 1 and l ¼ 3:5 also show a horseshoe-like shape
closer examination on the dependence of l on the material properties bubble cloud behind the expansion wave. It is difficult to discuss the
should also be considered. In this work, we infer the difference in the validity of this cloud against the experiment, as the Schlieren photo-
l values between the two configurations investigated to be related to graph exhibits an intricate structure inside the horseshoe shape, which
the changes in the properties of the material constituting the droplet, complicates the analysis of the image. However, it is obvious that the
that is, water vs gelatin. The present modeling and existing experimen- size of the cloud simulated with l ¼ 10 is significantly overestimated.
tal datasets do not allow for the identification of the governing param- On frame (e), experiments show a bubble cloud (contoured with the
eters, while the viscoelasticity, the spatial arrangement of molecules, red dashed line). In the simulation with l ¼ 1, no cloud is visible. In
the concentration and size of nuclei, or the equations of state should addition, the intensity of the numerical Schlieren is the same for the
be considered. Additional experiments, varying the material properties various internal reflections, while the f2 wavefront is significantly
and the regimes, should be conducted to independently investigate the stronger on the experimental Schlieren. Note that these observations
influence of these parameters on the cavitation dynamics. hold for frame (f). The wave pattern and the bubble cloud observed in
the simulation with l ¼ 3:5 are however in a very good agreement,
ACKNOWLEDGMENTS which is also true for frame (f). Although the simulation with l ¼ 10
significantly fails to reproduce the wave pattern, we note that a rela-
Authors thank Fabien Petitpas and Nicolas Favrie for their tively good match is reported on the bubble cloud. However, on frame
participation in the determination of the lk;j formulation. Authors (f) and still for l ¼ 10, the CiS reveals a late collapse of the bubble
acknowledge the financial support from the ETH Zurich when comparing with the discontinuities seen in the experiment [see
Postdoctoral Fellowship program. the magnified view in Fig. 13(b)].

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-15

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

FIG. 17. (a)–(f) Comparison of the internal droplet structure between (top) numerical simulations at different l and (bottom) the experiment of Sembian et al. The upper-halves
display the volume fraction of air (yellow-to-black colormap) overlaid with numerical Schlieren images (white). The lower-halves disclose experimental Schlieren images.
Reproduced with the permission from Sembian et al., “Plane shock wave interaction with a cylindrical water column,” Phys. Fluids 28, 056102 (2016). Copyright 2016 AIP
Publishing. The dashed red lines on frames (e) contour the bubble cloud as indicated in Sembian et al.

APPENDIX B: EFFECT OF THE INITIAL GAS VOLUME and Field et al., ag is an unknown parameter. We therefore esti-
FRACTION mated ag to be equal to 106 , which is assumed to be representa-
tive of purified water. The calibrated pressure-relaxation rates for
In this work, we are concerned with the growth and collapse the two configurations are thus only valid for simulations with
of the gas phase involved in the liquid–gas mixture initially consti- ag ¼ 106 . To assess the sensitivity of the pressure-relaxation rate
tuting the droplet. This requires to set the initial gas volume frac- on the initial gas volume fraction, we run additional simulations,
tion, ag within the droplet. In the experiments of Sembian et al.
corresponding to the experiment of Sembian et al., with ag ¼ 105
and ag ¼ 107 . Figure 18 plots the variation of the volume of the
gas phase within the droplet as time proceeds for various (ag, l)
combinations. Comparing the results for the calibrated l ¼ 3:50
with ag ¼ 106 , relative errors between 40% and 60% are reported
on the maximal volume, which remain relatively low when com-
pared to the errors resulting from an infinite pressure-relaxation
rate (650%). The relative errors on the numerical location of the
bubble cloud center with respect to the experiments are found to
be 10% for both ag ¼ 105 and ag ¼ 107 with l ¼ 3:50. We
then calibrated again the pressure-relaxation rates for ag ¼ 105
and ag ¼ 107 and found l ¼ 2:50 and l ¼ 4:60, respectively. It
appears that changing ag by one order of magnitude (lower or
higher) does not change the order of magnitude of the pressure-
relaxation rate.

REFERENCES
1
S. Sembian, M. Liverts, N. Tillmark, and N. Apazidis, “Plane shock wave inter-
action with a cylindrical water column,” Phys. Fluids 28, 056102 (2016).
2
J. E. Field, J. P. Dear, and J. E. Ogren, “The effects of target compliance on liq-
uid drop impact,” J. Appl. Phys. 65, 533–540 (1989).
3
FIG. 18. Variation of the volume of the gas phase within the droplet as time pro- H. Chen and S. M. Liang, “Flow visualization of shock/water column inter-
ceeds for various initial gas volume fraction ag and pressure-relaxation rate l. actions,” Shock Waves 17, 309–321 (2008).

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-16

C Author(s) 2023
V
 Physics of Fluids ARTICLE scitation.org/journal/phf

4 23
L. Biasiori-Poulanges and H. El-Rabii, “Shock-induced cavitation and wave- R. Saurel and R. Abgrall, “A multiphase Godunov method for compressible
front analysis inside a water droplet,” Phys. Fluids 33, 097104 (2021). multifluid and multiphase flows,” J. Comput. Phys. 150, 425–467 (1999).
5 24
S. Balibar and F. Caupin, “Metastable liquids,” J. Phys.: Condens. Matter 15, A. Kapila, R. Menikoff, J. Bdzil, S. Son, and D. Stewart, “Two-phase modeling
S75 (2003). of DDT in granular materials: Reduced equations,” Phys. Fluids 13, 3002–3024
6
K. Ando, “Effects of polydispersity in bubbly flows,” Ph.D. thesis (California (2001).
25
Institute of Technology, 2010). K. Schmidmayer, F. Petitpas, S. L. Martelot, and E. Daniel, “ECOGEN: An
7
J. C. Meng and T. Colonius, “Numerical simulations of the early stages of open-source tool for multiphase, compressible, multiphysics flows,” Comput.
high-speed droplet breakup,” Shock waves 25, 399–414 (2015). Phys. Com. 251, 107093 (2020).
8 26
H. F. Okorn-Schmidt, F. Holsteyns, A. Lippert, D. Mui, M. Kawaguchi, C. E. F. Toro, Riemann Solvers and Numerical Methods for Fluid Dynamics
Lechner, P. E. Frommhold, T. Nowak, F. Reuter, M. B. Pique, C. Cair os, and R. (Springer Verlag, Berlin, 1997).
27
Mettin, “Particle cleaning technologies to meet advanced semiconductor device B. Van Leer, “Towards the ultimate conservative difference scheme III.
process requirements,” ECS J. Solid State Sci. Technol. 3, N3069 (2014). Upstream-centered finite-difference schemes for ideal compressible flow,”
9
Y. Tatekura, T. Fujikawa, Y. Jinbo, T. Sanada, K. Kobayashi, and M. Watanabe, J. Comput. Phys. 23, 263–275 (1977).
28
“Observation of water-droplet impacts with velocities of o (10 m/s) and subse- K. Schmidmayer, F. Petitpas, and E. Daniel, “Adaptive mesh refinement algo-
quent flow field,” ECS J. Solid State Sci. Technol. 4, N117 (2015). rithm based on dual trees for cells and faces for multiphase compressible
10
S. R. G. Avila and C.-D. Ohl, “Fragmentation of acoustically levitating droplets flows,” J. Comput. Phys. 388, 252–278 (2019).
29
by laser-induced cavitation bubbles,” J. Fluid Mech. 805, 551–576 (2016). B. Dorschner, L. Biasiori-Poulanges, K. Schmidmayer, H. El-Rabii, and T.
11
L. Biasiori-Poulanges and H. El-Rabii, “Multimodal imaging for intra-droplet Colonius, “On the formation and recurrent shedding of ligaments in droplet
gas-cavity observation during droplet fragmentation,” Opt. Lett. 45, 3091–3094 aerobreakup,” J. Fluid Mech. 904, A20 (2020).
30
(2020). T. Trummler, S. H. Bryngelson, K. Schmidmayer, S. J. Schmidt, T. Colonius,
12
K. K. Haller, Y. Ventikos, D. Poulikakos, and P. Monkewitz, “Computational and N. A. Adams, “Near-surface dynamics of a gas bubble collapsing above a
study of high-speed liquid droplet impact,” J. Appl. Phys. 92, 2821–2828 crevice,” J. Fluid Mech. 899, A16 (2020).
31
(2002). K. Schmidmayer, S. H. Bryngelson, and T. Colonius, “An assessment of multi-
13
T. Kondo and K. Ando, “One-way-coupling simulation of cavitation accompa- component flow models and interface capturing schemes for spherical bubble
nied by high-speed droplet impact,” Phys. Fluids 28, 033303 (2016). dynamics,” J. Comput. Phys. 402, 109080 (2020).
14 32
L. Biasiori-Poulanges, G. T. Bokman, E. Baumann, and O. Supponen, Y. A. Pishchalnikov, W. M. Behnke-Parks, K. Schmidmayer, K. Maeda, T.
“Dynamics of a shocked bubble-encapsulated droplet,” Appl. Phys. Lett. 120, Colonius, T. W. Kenny, and D. J. Laser, “High-speed video microscopy and
260601 (2022). numerical modeling of bubble dynamics near a surface of urinary stone,”
15
J. C. Fisher, “The fracture of liquids,” J. Appl. Phys. 19, 1062–1067 (1948). J. Acoust. Soc. Am. 146, 516–531 (2019).
16 33
R. E. Apfel, “Acoustic cavitation prediction,” J. Acoust. Soc. Am. 69, 1624–1633 K. Schmidmayer, F. Petitpas, E. Daniel, N. Favrie, and S. L. Gavrilyuk, “A
(1981). model and numerical method for compressible flows with capillary effects,”
17
N. Kyriazis, P. Koukouvinis, and M. Gavaises, “Modelling cavitation during J. Comput. Phys. 334, 468–496 (2017).
34
drop impact on solid surfaces,” Adv. Colloid Interface Sci. 260, 46–64 (2018). J. B. Keller and M. Miksis, “Bubble oscillations of large amplitude,” J. Acoust.
18
K. Schmidmayer, J. Caze, F. Petitpas, E. Daniel, and N. Favrie, “Modelling Soc. Am. 68, 628–633 (1980).
35
interactions between waves and diffused interfaces,” Int. J. Numer. Methods V. Coralic and T. Colonius, “Finite-volume WENO scheme for viscous com-
Fluids (publised online, 2022). pressible multicomponent flows,” J. Comput. Phys. 274, 95–121 (2014).
19 36
O. L. Metayer, J. Massoni, and R. Saurel, “Elaborating equations of state of a C. E. Brennen, Cavitation and Bubble Dynamics (Oxford University Press, 1995).
37
liquid and its vapor for two-phase flow models,” Int. J. Therm. Sci. 43, 265–276 D. Fuster, C. Dopazo, and G. Hauke, “Liquid compressibility effects during the
(2004). collapse of a single cavitating bubble,” J. Acoust. Soc. Am. 129, 122–131
20
R. Saurel, F. Petitpas, and R. A. Berry, “Simple and efficient relaxation methods (2011).
38
for interfaces separating compressible fluids, cavitating flows and shocks in J. J. Quirk and S. Karni, “On the dynamics of a shock–bubble interaction,”
multiphase mixtures,” J. Comput. Phys 228(5), 1678–1712 (2009). J. Fluid Mech. 318, 129–163 (1996).
21 39
D. A. Drew and S. L. Passman, Theory of Multicomponent Fluids, Applied F. Caupin and E. Herbert, “Cavitation in water: A review,” C. R. Phys. 7,
Mathematical Sciences Vol. 135 (Springer, New York, 1998). 1000–1017 (2006).
22 40
M. R. Baer and J. W. Nunziato, “A two-phase mixture theory for the deflagra- D. Obreschkow, N. Dorsaz, P. Kobel, A. de Bosset, M. Tinguely, J. Field, and
tion-to-detonation transition (DDT) in reactive granular materials,” Int. J. M. Farhat, “Confined shocks inside isolated liquid volumes: A new path of
Multiphase Flow 12, 861–889 (1986). erosion?,” Phys. Fluids 23, 101702 (2011).

Phys. Fluids 35, 013312 (2023); doi: 10.1063/5.0127105 35, 013312-17

C Author(s) 2023
V