Professional Documents
Culture Documents
/z
NASA ICASE
Contractor Report
194978
S
ON THE DYNAMICS SHOCK-BUBBLE OF A INTERACTION
James J. Quirk
Smadar Karni
N95-22325
Unclas
G3/64
0039171
Contract September
NAS
1-19480
1994
Institute for Computer Applications NASA Langley Research Center Hampton, VA 23681-0001
in Science
and
Engineering
Operated
by Universities
Space
Research
Association
On
the
dynamics
of a shock-bubble
James J. Quirk 1 and
interaction
Institute
for
Computer NASA
Applications Langley
Engineering
Hampton,
Karni
1,2 Sciences
10012
Abstract
a detailed
study
of the interaction
of a weak
shock
wave
with
an isolated to mix-
study
which
by repeated
refractions
of acoustic performed
bubble
interface. a Ms
by Haas
impinges
bubble
contains
either Euler
compressible simulations
component
restricted
to single
of successful at material
single
component Here
schemes we avoid
interfaces. scheme.
In addition, extremely
we have utilized
a sophisto
adaptive
mesh relatively
enables
that
were observed
experimentally
to irregular
refraction, etc.),
multi-shock
structures,
description
of a shock-bubble
IThis research was supported by the National Aeronautics and Space Administration under NASA Contract No. NAS1-19480 while the authors were in residence at the Institute for Computer Applications in Science and Engineering (ICASE), NASA Langley Research Center, Hampton, VA 23681. 2Supported in part by an NSF Postdoctoral Fellowship, NSF grant #DMS92 03768, ONR grant #N0001492-J-1245, DOE _:ontract #DEFG0288ER25053 and a Packard Fellowship to Leslie Greengard.
Introduction
lucid
paper,
Haas with
and
Sturtevant
(1987)
presented that
study
of the
of weak
shock
waves
isolated that
of either
bubbles.
They
argued behaviour
to shed random
light
phenomenological
whereby
propagating their
media
were into
a resounding several
which
provide
processes, given
interface
the nature
inevitably
to control
imperfect
the bubble
interface
(Abd-El-Fattah does
Moreover,
membrane
as does the
support
structure
needed
to hold the
There shock
of the experiment
& Sturtevant
images in that
be measured limitations
nature
(e.g.
vorticity)
or because
of practical study
background,
modern
method
Sturtevant
(1987)
mechanisms
through gap
that
a study
between although
of geometrical ignores
understanding to account
experimental
observations, Similarly,
theory
while
yon Neumann
with regions
of non-uniform (Ben-Dot
applicable hand,
at curved 1958)
1985).
extensions but
(Catherasoo
1983, it does
interfaces, Moreover,
the theory
is approximate
this
approach
cannot provide
any details
refracted
shock
fronts.
Hence
numerical
to supplement refraction
by theory. already
et a1.(1991)
simulations
of reflection-refraction 1978a,b).
which to shed
& Henderson
refraction
curved interface. Haas _ Sturtevant's experiments have already inspired several numerical studies. For example,
both Picone
determining investigated
& Boris
(1988)
and
Yang
et a/.(1993,
1994)
aimed
at
of the bubble
inhomogeneities, process.
have the
gas rather
binary the
by the experiment. results. Admittedly, extent, jump cannot Indeed, the can be must be the
expedient,
accuracy
of the and,
to some
But
desired the
density bubble
across
interface,
a single
model the
equilibrium
on the
density
would
be 2.08 _imes
for studying
shock-bubble
efficiently
with errors
supersonic
systems could
et al. 1987).
Clearly,
circumstances,
be tolerated. and
to misinterpretation. of the above shortcomings. by the First, proper Euler upon account equations the is taken for a
flow is modelled
compressible depending
fluid
22 (R22)
Although
over the
single
popular
they
produce
oscillations a significant
interfaces the
artifacts
evolution
interface,
to be avoided. 1994a).
a somewhat allows
scheme
to avoid
numerical
difficulty
(Karni
In essence,
for a controlled
conservation While
so as to maintain of strict
between
different
this relaxation
perceived
wisdom
schemes
for flows with shock the shortcoming (Quirk simulations 1991). and of This so it
excellent
resolution
mesh
refinement
of performing
would
otherwise our
As will be shown simulations purpose in some The and detail which could
in this
paper,
a means
of producing is general it
agree
remarkably
well with
be profitably
we describe
(although
of this
do not
in this
of our
in 4 we detail by four
experiments First,
is followed
discussion.
comparison
against
experiment with
two quantitative
comparisons
experiment,
velocities
of certain
key
flowfeatures, the
the axis vorticity discussion resulting
other
deals
with These
of pressure
traces
at various
locations
by a discussion
the bubble
purpose
it is pertinent Finally,
at determining remarks
evolution study.
in 9 we close
concerning
our numerical
Multicomponent
multicomponent equations. just
Flows
flows using the compressible in this paper Euler we focus equations augmented by a requisite components components
We model number
of species
For clarity,
on flows with
a single species
equation;
the extension
of our discussion
straightforwardly. using Cartesian coordinates (z, y), the governing equations may be written
w, + r(w), + G(W) = 0
pv
pu
W
pu 2 + P
pvu
(1)
pv I E p pY
; r(w)=
puv
pu( S + p) puY
G(W)
pv 2 + P pv( E + P) prY
Note
that
these
equations
are written
form,
p is the
density two.
of a binary It is assumed
whose fluid
equilibrium
whose
components
are u and v in the z and y directions, only if the density Here, perfect E is the total gases, by variation energy between
slip is reasonable
as is generally Both
with ratios
of specific
the pressure,
p, is given
1 p=(7(Y)-I)(E-_pu where standard the effective 7 for the mixture reasoning depends to be
71CvlY
(2) from
on the species
thermodynamic
"4- 72Cv_(1
Y)
.fir) =
It is well known governing equation equations that solutions to (1) may valid in their
+ Cv2(1- Y)
develop discontinuous form. remains shock fronts, across which
(3)
the
are no longer
differential which
Using valid
Gauss's at a shock
divergence
theorem,
into an integral
form
-_
O//o
Wdzdy
Fdy - Gdz
= 0
(4)
jump
(Courant
& Friedrichs
1948). in
a so-called F and
scheme solution.
up around
the problem
of how best
1986).
Irrespective
results
in a 'viscous' coincides
shock profile
discontinuity
(unless
although
smeared,
a nonconservative 1991).
discretization
inconsistent
solutions property,
1954,1972;
& Le Floch
a conservative
formulation and
is almost many
accepted schemes
as have
to date
a major
to multi-component fluid components 1991; in Figure interface, pressure. the solution computation given Ton
equilibrium
material 1994a;
(e.g.
Larrouturou illustrated the smeared the equilibrium contaminate conservative helium case
et al.
1994).
This
1. Even
though
might
erroneous
Figure
2 (a) shows a snapshot where bubble the start is marked has passed
of a shock-bubble
data
is identical
pressure
the shock
reflections sets
place.
Spurious
pressure
visible.
For other
trigger
significantly
and so might
Density
Momentum____
Energy PresslU'e_
___
Figure
1: Pressure
fluctuation
at a material
interface
due
to numerical
diffusion.
Numerical
problems
with
smeared
interfaces 4
can
be avoided
if fronts
are fitted
rather
than
1.6 l ' S_
1.6 mmmmam_
!.5
t 1.4
o 1.4
13
(a) Conservative
scheme
(Roe
1982)
Figure
2: Pressure
profiles
for a one
dimensional
'shock-bubble'
interaction.
(e.g.
1990),
but
front
fitting
introduces equations
Here form.
we The
an alternative system
(1) in primitive
U_ + AP(U)U_
= 0
It
V
v P Y
AP(U)=
! 000 l0 00/ /
0 0 0 0 u 0 0 u p-1 0 0 0 0 u 0 0 0 v 0 0 0 0 0 v
(5)
BP(U)=
0 0 0
p-1 v
7P 0
0 0
7P 0
advantages
of this _
consider
a planar both
material
interface
aligned
in the
interface, decoupled
It follows equations
to three constant.
completely Thus,
in p, v and Y and
however,
measure
inadequate. a set
proposed terms
developed errors
of high-order
which
to remove
leading This
so as to produce
conservative, captured
variable have
scheme.
'viscous'
profiles
which
a captured scheme
In essence, shock
present
primitive
variable scheme.
correction
so as to mimic
the 'viscous'
we outline
But first,
shock-bubble
it produces
free solutions,
Numerical
describe
Method
the major components These are: of our numerical (i) the primitive flows; method variable for investigating discretization adaptive whilst the - this mesh dynamics provides refinement low the a
We now
interaction.
of multicomponent in order
to resolve
intricate
maintaining of elucidating
flow visualization
- this facilitates
the process
investigation.
3.1
Following refinement
to the right
of each split
equation
as to control
conservation
we alternate
between
At = -_-Dx terms, Dx
U_ + BP(U)Uy depends
At = -_-Dy.
equation. operator
using
first-order
upwind
1982).
discretization system
(5) with
If Roe's second-order
approximation
to (1) but it is a
approximation
equation
=Y\At ('(IAIW.)_.
and mesh size used the leading terms viscous correct equation determine path
w.)
for the order integration terms and
(7)
A is
To leading
the viscous
numerical is produced
transition.
In this
case,
the numerical
is physicMly speeds
consistent
scheme scheme
shock
Similarly, approximation
equation
=T
In general, discretization, enforce the two viscous yields shocks forms that (7) and (8) are different. the Rankine-Hugoniot solution, the
v,,
.
arising the
(8)
from a conservative latter the does not. To
satisfy
consistent
shock
profiles
on the primitive
between
expressions primitive
(appropriately system
transformed)
has to be added
operator
(9)
and T is the conservative to primitive transformation 6 matrix oU o-'_V"
upwind
scheme,
to the order
of the primitive
Dx
T(T-1)=[AP[U= A
- T(T-1)_Ut.
(10)
Straightforward given by
algebra
shows
that
Euler
system
(5), the
correction
terms
are
D X =
(11)
where
observations
to replace
time
derivatives
by space Dx
derivatives, near
within
Dx
are scaled
vanishes
surfaces.
Consequently,
for first-order
upwinding, resolution
the latter's
superior
of contact
to first-order play.
accuracy
near shocks
is precisely
the correction
come into
correction
terms
using
asymptotic
arguments
based
while significantly
reduced,
are inherent
eliminated.
At 3. The correction terms depend on the ratio A = A-zx' and so some variation in their effect is to be expected with changes in Courant number (wave speed* A). It is our experience that the correction size of time terms step work best at Courant numbers process close to one, which is the upper bound on the
to be stable.
The correction
terms
operator
(6) may
be similarly
derived
and
are given
by
0
i
2p 1 Dy = _p clpyvv
-4p, u+
- 4ptvt
(12)
7--1
( c3pv_ + _c4p_v"
+ 2'v'pu_ 0
where
c1-c4
as those terms,
with u replaced
by v. is as follows.
Given
of the correction
method
The LHS for each split equation and-signal derivatives used form which (Roe are 1982). then The
are replaced
differenced, terms
values
take
by Roe's
scheme.
to cell-updates conservative
Thus
the x-sweep
of our 'nearly'
primitive
,_, U_+X = __ FRe(I'."_ j+ where _o, is the standard Roe evolution operator.
are alternated
so as to arrive
at a two-dimensional
3.2
The
An
(AMR) equations.
It attempts
grid to the local requirements flows, a fine mesh is generally elsewhere a relatively coarse
in simulations
flow discontinuities, which savings (Quirk & Oliger employs rectangular accrue of more
savings
refinement
are totally
significant; phenomena
of detonation
foundations outlined
with the work of Berger The AMR to a single patches. domain. grid mesh G1. algorithm
a hierarchical patch
'mesh'
topologically
mesh
patches
Go and This
they
such
domain
be refined form
by embedding grid
Go. These
embedded
the next
is effectively
formed ratio
by subdividing is arbitrary,
choice Thus,
for the
refinement
by construction,
see Figure
a restriction
on the size of time step the smaller More that the allowable but smaller at
on fine grids
than
on coarse (the
in a fashion
the rate
move
relative
Courant grids
number)
is comparable
This
undesirable step
situation
coarse
numbers
the time
grid's
stability
(e.g.
Lax-Wendroff)
numbers. is retained of integration in both time even in regions is always and space different of grid coarse overlap and so all grid levels
from
integrations any
at the
span
temporal to project
take
Periodically, underlying
on to its
projections
for a three
refinement
integration of dummy
is extremely
simple the
in concept. dummy
Each
mesh
to integrating is consistent
a grid,
boundary
have to
is then
integrated
dependent, cell-centred
never
actually
for a single
topologically
In general solution
it is necessary
to adapt
in the evolving
is dynamic where
outlines main
go to make diffracted
to the apex
features corner
the flow,
namely
vortex shown
of the
Although between
many
adaptions patch
might
appear
actually
several dovetails
hundred with
A large
of small
grid movements 4.
occurs that
integrations
adaption
so as to ensure
is never A grid
of more essentially
coarse with
adaption
produces before
data
from
partially shovelled
of the same
In regions coarse
the
required
solution
is found
by interpolation
the underlying
grid solution. In a typical patch application the finest grid of data will contain several hundred The Multiple mesh parallel Data patches. AMR Thus, algorithm model. the mesh (Quirk Each
is a sufficiently 1993)
fine unit
parallelism. Program
Hanebutte
is implemented
(SPMD)
F I F F I l
A
I # I f I / 1 I / l/ I I / / I I
/ I I_';_'
/ / / I I I I I I I I ! I_11 gltf Illl Ill| IJ IIII Ig| I | I I
;
II
iiiiiiiii[iiiii_lllll
, \\ \
!
1/1111_'
l I / I I I I I I I I I I
I
I I
I till
////////
employs a hierarchical
PROJECTION
i\\\\\
grid system.
ADAPTION
GRID
INTEGRATION
TIME
STEP
Go G1 4xG2
At A_/2
4 x At8 G_ _ G_
62
GI 4 x G_
Atl2 4 x At/8
G2 ---* G_
G1 _ Go
62
Ga
Figure
4: Grid operations
are recursively
interleaved
(to
Figure
5: The AMR
algorithm
employs 10
a dynamic
grid system.
node
executes
the
basic
serial
(Quirk between
1991) the
in isolation nodes
from
nodes, that
messages
to supply during
to be missing, at which
integration the
a processor
is during
cells.
Whereas
in a serial memory
are from
some
are from
processor.
a schedule grid.
so as to prime to produce
If running off-processor
in parallel, fetches.
a schedule
necessitate fetch
point, every
processors
can exchange
subsets
schedules,
a schedule
of messages
that
it must
send
the priming
First,
case.
Second,
then waits for those data from arrive grid its own schedules is unimportant. levels also necessitate
process amounts
sizeable cells.
of communication,
balancing
the
AMR
algorithm before
rests the
the best
amongst
the processing
nodes
Currently, 'Bin
using
to classical account
algorithms costs
complication
transfer
3.3
Flow
Visualization to make
Images comparison (Liepmann between our numerics 1957). Such and experiment, are useful that we present for identifying the sensitivity is little to a
a qualitative images
of schlieren-type which
images
It should
be appreciated
images trying
which the
might from
Instead
to elicit simplicity,
maximum
amount
of information provide
possible a very
images refraction
effective
the magnitude
of the gradient
IVpl=
and hence gradient. following they The may density be viewed derivatives as idealized schlieren images; the darker the image the larger the
and the
nonlinear
shading
function,
where
k is a constant colour
that graphics
took the value 600 for the light fluid and 120 for the heavy system the grey shades outside the bubble were produced
fluid. using
Usthe
ing a 24 bit
< R, G, B > triplet < 204_, 204_, 255_ We also present useful for determining
within
the bubble
were produced
using
lit surface
plots,
8,10
are
the strengths
of certain
flow features.
lack of space
us from
describing
were produced.
Computational
Set-up
of the dynamics performed wave, of a shock-bubble by interaction (1987). we have reproduced Namely, bubble the numerically of a or
interactions helium
shock
moving
of either
(CHCIF2). acoustic
Whereas lens,
than
the surrounding
air and
as a divergent
is heavier
therefore
acts as a convergent
In the experiments
made from a very thin membrane of the bubble and the resultant
was exercised
can be expected
of the bubble
sets of results
counterparts, spark
a compilation be taken
from each run, and so the complete delay times to the exposure the accuracy
by repeating While
this method
produced
images, is derived
necessarily
suffered:
the repeatability
of the experiment.
The general
measurements
is thought
of a few instances
set-up
is shown
in Figure
6. We have assumed
the axis of the shock tube and so only the top half of the flow field (ABCD) boundary conditions a standard were applied reflecting to the flow domain: procedure sides BC and the
The following
boundary
(Quirk
1991);
inflow along side CD was specified zeroth-order downstream relevance profile extrapolation boundary
behind neither
the incident
was used
that
the upstream
treatment
is critical
since no physical
'start-up'
are generated
discontinuity
(Hillier
was placed
some distance
of the bubble
themselves
waves moving
on the passive
characteristics
12
to interfere modelled
as perfect constant
specific
capacity
shock
shock
wave
and'pressure thermal
quiescent mechanical
to be unity.
equilibrium bubble
R_ir/Rb_,b_te.
contamination
by Haas
(1987).
in the
experiment
Therefore,
initialised
with
one of two
depending
on whether
lay inside
or outside
_mm
222J mm
Figure
6: A schematic
of the
computational
domain
(not
to scale).
R
kJ/kg K
cv
kJ/kg 0.72 0.365 3.11 2.44 K
Table
1: Gas
properties
for the
simulations.
computational
domain
was discretized
using
20 coarse
mesh
patches
formed
Additionally,
a refinement to
to resolve
grid mm.
of 16,000
by 800
a spatial cluster
computations each
on a small
of workstations we make
(8 Sun no claims
Sparcl0
51s)
paper,
as to the that
excellent R22
computational bubble
efficiency
of our
to consider require
for the
computation
the equivalent
13
(16 x I,592 iterations a mesh 16,000 by 800 cells). on For our flow solver, singleprocessorof a a CKAY Y-MP might manage one cellupdate every 10ps in which case it would need 905 hours to
run the simulation.Brute forcecomputations on super computers do not representa sensibleoption for investigationsf shock wave phenomena. o
Results
and
Discussion:
_Flow
Visualization
images which reveal the experiment certain subtleties of the
we present interactions
from either
or previous
numerical
5.1 Figure
R22
Bubble
Convergent of schlieren-type
Case images from the simulation images of the R22 bubble is also shown. In order case, by Pleasingly, out the describe recall
7 shows
a sequence
sequence
of experimental features
to bring now
we shall which
the images
follow,
from
position
by a light support
circle
in the
numerical images.
by what
like a dark
(a) of Figure
some 55ps after it is first hit by the incident has already vertical undergone a slight near deformation. the top and the for
as two short
which
shock.
strengths (Figure
two waves
gained from
and pressure
shock
the incident
inside
outside
shock From
thickened plots
as to why
the surface
the pressure
chamfered. between
the pressure
jumps
the weak
moves
in sound (Figure
speeds 7 (b))
between
the
more
apparent,
now run roughly fields for this In essence, time for the
density
each side
the bubble,
air-R22 the
interface
a series
of compression
are required
to turn
degrees:
horizontal
flow is vertical,
the induced
flow behind
14
portion have
shock
is largely the
Note half
that
the
shock interface
to diffract
around roll-ups
of the
shows shock by
where
generated
of the incident
improve
fine mesh
be required
scales accurately. of compression and waves which turns the flow around in the surface each of the two bends density with and the wave
By 135ps in the refracted pressure side shows images, curvature the the limbs
is clearly
visible
plots
for the
fields (Figure
beyond
by LShner
thickening
refracted
up much it should
shadowgraphs
it does in the numerical represents tube an integration used will the within to make
schlieren of the
of the
in the
is one such
experimental that
artifact.
Because,
limbs
appear
in experimental image.
it is just
possible
out a line
the front
shown
Other
artifacts
undue
confusion.
gave rise to a number from the tube's the influence walls present such
As did the walls of the shock waves also appear to perform dynamics
so these
particular
in the a series
numerical
beyond what
it would effects
to determine
of interaction
been
focused
down
to a point. plots behind shock and have shock started bubble along surface
increase
pressure
in the
(Figure an Ms
larger than
segments
following
diffraction
downstream separate
discontinuities
contacts
of fluid that
shocks bubble.
tube
through
these
counterparts fluids.
speeds
between
roll-ups in the
bubble
much they
more
and
prominent Note
plot for the pressure of the number top and bottom of cylindrical front
appear through
(Figure bubble
8 (d)).
that
shocks waves
interface
has given
which
then
to form
a shock
in a manner
reminiscent
of Huygen's Once
reconstruction. shock which has been focussed is cylindrical (Figure it emerges 7 (e)). from the downstream The downstream interface to become bubble
a transmitted
interface
of the
15
necessarily alignsitselfith the resultantvelocityfieldwhich isalmost radialand so it takes on w a wedge-likeshape. Note that the cylindricalransmitted wave is in the stages of catchingup the t two diffracted segments of the incidentshock front. Although the agreement between experiment and computation is poor at thismoment in time, it isworth remembering that each shadowgraph
was produced from a separate experimental run. Therefore,the fact that we are generallyable to match our numerical schlierens closely the shadowgraphs istestimony to the repeatability f so to o the experiment. In thisone instance, itwould appear that the experimental run was relativelyoor p and that the grossfeaturesof the computation are correctly positioned. If there is any criticism the simulation,itshould be directedat a few subtle shortcomings of on the small scale.For example, the two-pronged featureemanating from the left-handsideof the bubble (Figure 7 (e)onwards), seems unduly exaggerated in our simulation.This featureis caused by a narrow jet of fluidwhich isshot forward during the focusingof the refractedwave. As yet, we cannot categoricallytate the cause of thisexaggeration. It is probably due to the lack of real s viscosity our flow model. In the experiment viscosity in causes the jet to spread thus reducing its range of influence. the simulation, In which isinviscid, any spreading of the jet issimply clown to residualnumerical diffusion. iven the resolution our computation, thisresidualdiffusion svery G of i small and so the spreading of the jetwillbe underdone giving itan exaggerated range of influence. However, itisconceivablethat the exaggerationisyet another obscure numerical failing the type of catalogued by Quirk (1994a). By 342/_s the bubble has moved appreciably from itsoriginalpositionand it has started to elongate (Figure 7 (g)).Inside the bubble there is a backward moving shock which was born from the internal reflection ofthe refracted shock from the downstream interface. n the numerical image a I number of weaker waves are alsoapparent,these are caused by waves which pass through the bubble because of reflections from the wailsof the shock tube and which subsequentlylead to other internal reflections from the bubble interface. Outside the bubble, the transmitted wave has reflected rom f the wallsof the shock tube. Interestingly, can be seenfrom the surfaceplotsfor thistime (Figure as 7 (g)),spikes in the pressure and density fields still persist where the transmitted wave intersects the bubble interface. The apparent featheringof the transmitted shock isdue to itspassage over what isnow a corrugated surfacegiven the many roll-upsalong the bubble interface. The internally back-reflected shock wave eventually emerges from the upstream interfaceto
become a backscatteredwave (Figure 7 (h)). While the waves resulting from the reflectionf the o transmitted shock from the top and bottom walls of the shock tube in their turn start to pass
through the bubble, furtherpromoting the generation of vorticity along the interface. The bubble continues to elongateand by much latertimes itevolvesintoa largevortex pair(Figure 7 (h)).For these late times,when viscous effectsight be expected to dominate proceedings,itisremarkable m that an inviscid simulation givessuch qualitatively ood agreement with experiment. g
16
_-
_L
_i_:i _
schlieren
images
and
(!Iaas right
interaction bubble.
1.22 (b)
cylindrical
5.S ys,
11.5 ys,
17
Numerical
schlieren
images of an Ms
and
experimental
(llaa,s
&
= 1.22 shock
from right to
Times:
(f) 318 #s, (g) 342 tts, (h) 417 #s, (i) 1020 its. Press 1987. Reprinted with permission of
QCambridge Press.
University
18
of nn Ms
= 1.22
19
2O
5.2
Helium
Bubble
- Divergent
Case images from the simulation of the Helium bubble process. shock case,
interaction
Figure 9 (a) shows a view of the helium bubble 32 ps after it is first hit by the incident wave. As before, there is a curved refracted shock which lies inside the bubble, helium has a higher sound speed than the surrounding moves ahead of the incident shock. Outside shock nor a simple expansion wave. A one-dimensional across this wave is 19% of the density jump between air). Indeed, the surface plots for the pressure air (aair/aHe = 0.35), the refracted
the bubble, the curved reflected wave is neither a simple analysis for the precise moment the incident (the density jump the undisturbed bubble and the surrounding is
shock hits the bubble suggests that the reflected wave should be a weak expansion
and density
10 (a)). However, away from this axis there is very little acts as a solid surface giving rise to a reflected system which accounts for the lower pressure to be
found behind the rest of the reflected wave due to the collapse of the bubble. The difference in sound speeds between parent by 52 ps (Figure A four shock configuration reflection-refraction the relative intersection (TRR). has formed A schematic the bubble and the surrounding air becomes more apwave.
which Henderson et al. (1991) have termed twin regular for this shock configuration contact is shown in Figure 11. Given emanates from their discontinuity
point as would be expected in the general case; although one does become visible by 72 Around 62 ps (Figure 9 (c)) the refracted wave emerges from the left-hand internally side reflected wave appears as 10 (d)). By
ps (Figure 9 (d)).
two cusps. As can be seen from Figure 9 (d), this reflected wave is convergent and is being focused along the axis of the bubble but the local increase in pressure is quite small (Figure 82 ps (Figure 9 (e)) the internally reflected waves have crossed and are now diverging, here they
appear as a small loop. The two branches of the transmitted (Figure 9 (f)), along the axis of flow symmetry merged. Meanwhile, the walls of the shock tube. Interestingly, can lead to large increases wall and eventually has just started interacts the shock tube's walls, is reinforced to diffract around reflected
shock have also now crossed. At 102/_s shock have almost shock have reflected from shock, where it meets of the incident shock in
both the original reflected wave and the transmitted in local pressure. with the bubble. interface
as can be seen from Figure 10 (f), such spurious reflections Here the foot of the incident At this time, what remains side of the bubble, as a weak back scattered This spike then proceeds to move away from the and the internally reflected
wave, so weak in fact that it does not appear in the experimental and spreads laterally in the process When this than structures
images. As time moves on, the bubble becomes kidney shaped (Figure 10 (g)). This change in shape is driven by vorticity jet impinges (Figure on the air at the downstream it spreads laterally
generated
to the passage of the shock which induces a jet of air along the axis of flow symmetry. edge of the bubble, the lighter helium, 10 (i)). 21 and the bubble forms a pair of distinct vortical
schlieren
images
(ltaa.s
of ;LZlMS Times:
from right
cyliml['h'a.l
(a)32
Figure9: (Contd.) Numericalschlieren imagesandexl)erimental hadowgraphs s (llaas & Sturtev_tnt 987_ 1 from the intera.ction an Ms = 1.22 shock wave moving from right to of
left over a tlelium Experimental Cambridge cylindrical bubble. Times: University (f) 102 #s, (g) 245 #s, (h) 427 tts, (i) 674 #s. Press 1987. Reprinted with permission of images University @Cambridge Press.
23
= 1.22
24
id
(d)
l"igur('
10: (('onld.)
Surface
plots
of the density
and
pressure Times:
of
all M._' = 1.22 shl,('k w_lve wilh azL lie cylindrical 2,1Y_ its.
bubb[(,.
25
_Ur
._
_i..._-
...
--,, ",
Figure 6 Results and 11: Schematic Discussion: for twin regular reflection-refraction Velocities correct, Here, images. Whereas For each (TRR).
The results from the previous section clearly indicate that our simulations are qualitatively however, any serious numerical this included simulation, a quantitative investigation should contain some form of validation exercise. of several prominent flow features. check on the velocities
the positions of certain features were digitized from a sequence of schlieren-type x-t diagrams were then constructed velocities had an estimated so as to find the velocities. measured
uncertainty of 11%, here the uncertainty given the resolution of This equates to an to within +0.17 mm.
uncertainty of less that 1% in the worst case velocity measurement. to conservation errors is also small at less than 3%.
6.1
R22
Bubble
- Convergent
Case
The x - t diagramfortheshock interaction with theR22 cylindrical bubbleisshown in Figure12. Also shown in thisfigure isa schematicwhich identifies various the flowfeatures thathave been digitized.comparison A ofour computed velocities withtheir xperimentally e measured counterparts (Haas & Sturtevant 987)isgiveninTable2. The agreementbetweenthetwo setsofresults well 1 lies within thegiven11% experimentalrror;heworstcase(Vr)is e t just5.8%. Note thatwe haveignored thelarge discrepancy forVdisince theexperimentalalue v appearstohavebeen tabulated ncorrectly; i the experimental - t diagram indicates x that Vdiisclose 130 m/s which isin fairagreement to with the computation. Overall, thegeneral agreementbetween the two setsof velocities confirms theexperimentalists' thatthecontamination view ofR22 by airwas so small(theyestimated itat 3.4% by mass) as tobe negligible.
6.2
Helium
Bubble
- Divergent
is made in Table 3. As with the R22 case, the two sets of results are now significant. As detailed
in Section account
4, we have
assumed
that
the
contamination results
by
mass.
If no
is taken
similar. contamination
of 13.5%.
contamination would
necessarily the
air and
helium
are homogeneously
mixed.
experiment,
our correction
as accounting
of contamination. flow features move more near or less at constant velocity; in Figure to the point 13, at
shock
front
by the curved
transmitted
+ 4 + 4+
@ e 0 0
! HE E II
200
++
+ +
_ 4 0 ,1, 0 @ @ @ @ @ 111
M X> X X X X
Vui
150
100
+
4+ + + 4+
@ @ + 44+
X X X x X X
,I,X qK X@ X @ X @ X @ @ @ @ @ 4, @ e @ ! ,
50
+
+ +
; X + X t.X It< ! !
10
20
60
70
Figure
12: x-
cylinder
with
a schematic
the
points
used
V_
70 73 -4.1
y_ s
90 90 +0.0
vd_
116 78 N/A 82 78 +5.1
Table measured
2:
A comparison experimentally
of the by
computed and
velocities Sturtevant
for (1987);
the
R22
case 12.
with
those
Haas
for key,
27
250
vj
I00
5O
All '
IO 20
30 40 50 60 70 x (r_m)
Figure
showing shock,
the points
used
to construct shock,
V,,i - upstream
edge of bubble,
Vdi - downstream
edge of bubble,
V8 422 410
% Discrepancy +2.9 Table 3: A comparison of the computed experimentally by Haas and Sturtevant
Results
and
Discussion: shadowgraphs,
Pressure
In addition stations
to producing
along the axis of flow symmetry process. a smooth position, For example, pressure
disturbance
at a measuring disturbance
as might be expected
from a shock
wave. In fact, as was shown in 5, the diffracted of the bubble interface: ramps up gradually of the measured Although informative behind the diffracted
the surface plots in Figure 8 reveal that along the interface the pressure field wave and is not discontinuous,
disturbance. pressure traces only provide a local view of events and so are not as
the experimental
as to the
accuracy
the
traces
cannot
be
an accurate
benchmark endwall
placed the
the transducer
pressure
disturbances
reflecting
Consequently,
be expected
bubble indicated
suggests
disturbances a Ms
behind
provide setup.
we cannot pressure
make traces
a useful from
comparison
against case,
we present Figure 16 of
the heavy
bubble
& Sturtevant,
1987)
4.5 1
't
(a)
2.25
zp = 3mm 21
1.75
1.5
1.25
55O
(d)
Figure 14: Pressure histories for several stations downstream of the tt22 cylinder.
29
Results
and it takes
Vorticity purpose
Generation of this paper, recent we can use our studies numerical results mixing: flow that to the has
us beyond
to several
on shock-induced steady
to a three-dimensional,
proposed (Marble
to ensure Drummond
mixing
of air and fuel in supersonic Essentially, drives a mixing vorticity process Rupert which
combustions is impulsively
systems generated
by the passage
(Richtmyer
Meshkov
1970;
of vorticity
generated
of a shock-bubble produced
therefore
of predicting
conditions. the vorticity, w, is not in doubt. equation, Recall that the vorticity the baroclinic
produces
contains
term
This
equation 1
in the form
-N- -from which it can be seen that of the density misMignment largely and pressure occurs because of the have from vorticity fields (Shercliff
...
1977).
+ p-vp vp,
whenever there is a misalignment in the gradients interaction, gradient such which a is In the case of a shock-bubble wave imposes a local pressure
is produced
shock
gradient analytic
imposed
inhomogeneity. to predict this the amount by making bubble Of the for two of
devised
Typically,
assumptions interaction
duration
Picone
& Boris
to provide which
the better
predictions. found
a vorticity
is within
the Picone
of 2. Yang
it retains
features Consider
indicate
case. interface resolution appeared of their correct. point a on point term, side of
the
moving more
shock detailed
simulations where
is essentially interface,
is produced amount
crosses
plots
of the
baroclinic
there
at the points
produces
any significant
and it eventually
becomes
30
Figure10(f). Figure15showswosnapshots t oftheaccumulated vorticity,cf. the surface plotsof thepressurenddensityields a f shown Figure in 10.Notethatverylittle vorticityisgenerated dueto plainshock curvature. Alsonotethatthedistribution vorticityis not symmetric. of Morevorticity is deposited onthewindward sideofthebubble thanontheleeward side. Theabove observations undermine theassumptions which upon mostvorticityprediction odels m arebased. y wayofexample, B taketheYangel al. (1994) model. Since vorticity is predominantly
generated the bubble. by the side Note that shock, the production incident But shock effectively shock has assumes the ceases only that bubble once the transmitted half the bubble wave by emerges the time from this traversed vorticity (i.e.
happens, time
9 (c).
is generated as long
continuously as is necessary), of 2.
it takes
incident
twice
it might
be expected
to over predict
amount on the
of vorticity windward
bubble
not one as in the model. torque remains shock. that across But
assumed
gradient
constant this
it is equal pressure
to the jump
case.
relevant while
jump
is significantly
at later
times that
(see Figure
10 (f)).
gradient which
always
bubble 8 (b)-(d).
is tangential
nature
it would
that
any set
of
analytic In
errors
some models
can provide
reasonable
vorticity
in a slightly
fortuitous
31
Concluding
interactions.
For example,
in the case of a weak shock interacting with a heavy bubble, that was observed experimentally. were marred Previous
of the refracted shock front within the bubble and thereby identify
for the first time the cause of the shock thickening computational efforts to reveal this internal structure artifacts the necessary
and by numerical
which interfered with the flow. Despite having only modest high resolution by using a sophisticated that this algorithm led to between Here, we estimate computation artifacts
resources, we were able to obtain mesh refinement algorithm. fold decrease in computational each simulation as a parallel
effort. The actual computation time was further reduced by running on a small cluster of workstations. Additionally, which generally plague multicomponent simulations are able to match simulations.
great care to avoid the numerical at material interfaces. parts both qualitatively experimental Simulations
essence, we elected to suffer a small, controlled conservation Thus the present and quantitatively, time-dependent
at least to the limits set by the physical model and by phenomena usually prove difficult to decipher, since numerical study we chose to present schlieren-type
uncertainties. of complex of data. Therefore, in part, the clarity of the present For this paper
flow since they enable us to compare our results directly against experiment. weak features which are often lost on contour the various mechanisms that anumber plots and so shock
Such images are useful for identifying they provide refraction phenomena.
a very effective means for assimilating We have also presented of individual analytic
comprise
of realistically lit, surface plots which are Here, for example, the amount they are used to
invaluable for gauging the strengths expose the weaknesses of current by a shock-bubble interaction.
flow features.
of vorticity produced
computational
excellent results, it is
only fair to point out that it represents a considerable hope to apply our machinery
of effort. However, the machinery Looking to the future, we that are not yet fully understood, study.
and so the large development costs can be defrayed. to other shock wave phenomena
Acknowledgements The authors hospitality would like to thank R.Abgrall, improve this paper. M.J. Berger and K.G. Powell for their perceptive comJJQ would like to thank the Courant Institute for their of this work. The authors also wish to thank
Prof. B. Sturtevant
and Cambridge University Press for granting permission from the Journal of Fluid Mechanics.
32
References
ABGRALL,
R. 1988 Generalisationof the Roe scheme for the computation of mixture of perfect
gases. La Recherche A_rospatiale 31-43. 6, ABD-EL-FATTAH, A.M. & HENDERSON, L.F. 1978a Shock waves at a fastslow gas interface. J.
Fluid Mech. 86, 15-32. ABD-EL-FATTAH, A.M. & HENDERSON, L.F. 1978b Shock waves at a slow fastgas interface. J.
Fluid Mech. 89, 79-95. BELL, J.,BERGER, M., SALTZMANN, J.& WELCOME, M. 1994 Three-dimensionM adaptive mesh J. Sci. Compul. 15, 127-138.
M.J. &
OLIOER,
C.J. &
STURTEVANT,
Mech. 127, 539-561. COLELLA P., HENDERSON L. & PUCKETT E.G. 1989 A numerical study ofshock wave refractions 9th CFD Conference, Buffalo,NY,
K.O.
York. DRUMMOND, J.P. & GIVI, P. 1994 Suppression and enhancement of mixing in high-speed reacting
flow fields. Combuslion in high speed,flows, In 191-229. Edited by Buckmaster, J.,Jackson, T.L. & Kumar,
GRAHAM,
416-429. GROVE, J.W. & MENIKOFF, 219,313-336. B. 1987 Interaction 181, of weak shock 41-76. E.G. 1-27. 1991 On the refraction of shock waves at a waves with cylindrical and spherical R. 1990 Anomalous reflectionf a shock wave at a fluidinterface. o
& STURTEVANT
gas inhomogeneities.
HENDERSON,
J. Fluid Mech.
P. &
L.F.,
COLELLA
PUCKETT 224,
slow-fast
HILLIER,
gas interface.
J. Fluid Mech.
R. 1991 Computation of shock wave diffractiont a ninety degrees convex edge. Shock a
Waves 1, 89-98. Hou, T.Y. & LE FLOCH, Error analysis.Comm. PH. 1991 Why nonconservativeschemes converge to wrong solutions:
33
KARNI,
S. 1992
Viscous
shock
profiles
and
primitive
formulations.
SIAM
Num.
Anal.
29,
1592-1609. KARNI, S. 1994a Multi-component 112, 31-43. instabilities in multicomponent the mass fractions 95, 59-84. hyperbolic equations and their numerical computation. fluids (in preparation). positive when computing compressible flow calculations by a consistent primitive algorithm. Y. Comp.
Phys. KARNI,
S. 1994b Interracial
LARROUTUROU,
multi-component LAX, P.D. Comm. LAX, P.D. waves. LIEPMANN, p. 157. LOHNER, R., through Aachen, MARBLE,
Phys.
of nonlinear 7, 159-163.
Hyperbolic Monograph
systems series.
of conservation
laws
theory
of shock
& ROSHKO,
A. 1957 Elements
of Gasdynamics
John Wiley
& Sons,
Chichester,
J.P.
produced Tubes
In Proc.
on Shock Germany,
Edited G.J.
Weinheim, E.E.
1988. toward
F. E.,
processes
Paper wave
of a shock
accelerated
between
two gases.
NASA
074 K.F. J.P. 189, 1988 Vorticity 23-51. An adaptive Institute mesh refinement U.K. Mesh Refinement Phys.). Debate. Intl. J. Num. Meth. algorithm. NASA CRalgorithm for computational shock generation by shock propagation through bubbles in
Thesis,
Cranfield
of Technology, Adaptive
(submitted Great
A Contribution
to the
18, 555-574. 1994b Dynamic load balanc!ng strategies for parallel adaptive mesh refinement (in
preparation). RICHTMYER, R..D. 1960 23, Taylor instability in shock acceleration of compressible fluids. Comm.
Math.
297-319. and signals dynamics. - A framework Edited for numerical K.W. evolution problems. M.J., 219-257, In
Fluctuations methods
for fluid
by Morton,
and Baines,
Press,
New York. schemes for the Euler Equations Ann. Rev. Fluid Mech. 18,
1986 Characteristic-based
34
RUPERT,
In Shock Waves, Proceedings of the 18th Intl.Symp. on Shock Waves, Held at Sendal,Japan 1991. Edited by Takayama, K., Springer-Ver!ag, Berlin,1992. 83-94. SCHWBNDEMAN, D.W. 1988 Numerical shock propagation in non-uniform media Y. Fluid Mech.
J. Num.
Anal. 5, 506-517. TON, V.T., KARAGOZIAN 101, Proc. Combustion An A.R., ENGQUIST Institute of fluid B.E. & OSHER, S.J. 1991 482-512. paper WSS/C191UCLA. Press, Stanford, Press. of blast 18, flowfields using a high California, p. 148.
album
Parabolic
J. 1963
Collected G.F.
resolution WHITHAM,
volume
scheme.
G.B.
approach 2, 145-171.
to problems
dynamics.
J. Fluid
of shock
waves through
regions
of non-uniform
area or
J. Fluid Mech.
1993 Applications
of shock-induced
mixing
to supersonic
J. 31,854-862. 1994 A model inhomogeneity. for characterization J. Fluid Mech. transformations of a vortex pair formed
passage
of nonlinear
J. Comp.
179-186
35
REPORT
DOCUMENTATION
PAGE
For,. Approved
OMB No 0704-0188
Public reporting burden for this collection of information ts estimated to average | hour gathering and maintaining the data needed, and completing and reviewing the collection collection of information, Including suggestions for reducing this burden to Washinglon Davis Highway, Su te 1204, Arlington, VA 22202-4302, and to the Office of Management
per response, including the time for reviewing instructions, searching ex st ng data sources, of information Send comments regard ng th s burden estimate or any other aspect of this Headquarters Serv ces, Directorate for Information Operations and Reports, 1215 Jefferson and Budget. Paperwork Reduction Pro)ect (0704-0188), Washington. DC ?0503.
L AGENCY
USE ONLY(Leave
b/an/<)
2. REPORT September
DATE 1994
3. REPORT Contractor
TYPE
AND
DATES COVERED
4. TITLE
AND
SUBTITLE
ON THE DYNAMICS
OF A SHOCK-BUBBLE
INTERACTION
C NAS1-19480 WU 505-90-52-01
ORGANIZATION Computer
NAME(S)
Applications
Hampton,
AGENCY Space
NAME(S)
AND
ADDRESS(ES)
Administration
10. SPONSORING/MONITORING AGENCY REPORT NUMBER NASA ICASE CR-194978 Report No. 94-75
NOTES Monitor: to the Journal Michael F. Card Mechanics. 12b. DISTRIBUTION CODE
of Fluid STATEMENT
13.
ABSTRACT
(Maximum
200
words)
We
present
a detailed Such shock waves the interaction interface. planar 22. fluid single avoid they such shock
numerical interactions propagating process Specifically, wave, flows or mostly by These are
study have
of
the
of
weak
shock in an
wave attempt
with to
an
isolated
gas
been
elucidate
Our study concentrates on and reflections of acoustic performed bubble compressible wave are which Euler multi-component generated scheme. at In material addition, by Haas contains equations are and either
the early fronts at Sturtevant: helium for now a two fairly of have
through
nphenomena
utilized a sophisticated to be performed relatively that were observed wave focusing, description for
which enables extremely to reproduce numerically to irregular etc.), refraction, and we formation,
high resolution all the intricate cusp can now formation present
experimentally
regular
structures, interaction.
15. NUMBER
OF PAGES
37
16. PRICE CODE
A03
17. SECURITY CLASSIFICATION OF REPORT Unclassified MSN 7540-01-280-5500 18. SECURITY CLASSIFICATION OF THIS PAGE Unclassified 19. SECURITY CLASSIFICATION OF ABSTRACT 20. LIMITATION OF ABSTRACT
Standard
Prescribed 298-I02
Focm 298(Rev.
by ANSI Std. Z39-18
249)