-

A sharp interface Cartesian grid hydrocode

Sambasivan, Shiv Kumar
https://iro.uiowa.edu/esploro/outputs/doctoral/A-sharp-interface-Cartesian-grid-hydrocode/9983777170502771/filesAndLinks?index=0

Sambasivan, S. K. (2010). A sharp interface Cartesian grid hydrocode [University of Iowa].

https://doi.org/10.17077/etd.b55egfyg

https://iro.uiowa.edu
Free to read and download
Copyright © 2010 Shiv Kumar Sambasivan
Downloaded on 2025/03/12 05:30:25 -0500

-
A SHARP INTERFACE CARTESIAN GRID HYDROCODE

by

Shiv Kumar Sambasivan

An Abstract

Of a thesis submitted in partial fulfillment of the

requirements for the Doctor of Philosophy
degree in Mechanical Engineering
in the Graduate College of
The University of Iowa

May 2010

Thesis Supervisor: Professor H.S. Udaykumar

 1

ABSTRACT

Dynamic response of materials to high-speed and high-intensity loading condi-

tions is important in several applications including high-speed flows with droplets, bubbles

and particles, and hyper-velocity impact and penetration processes. In such high-pressure

physics problems, simulations encounter challenges associated with the treatment of mate-

rial interfaces, particularly when strong nonlinear waves like shock and detonation waves

impinge upon them. To simulate such complicated interfacial dynamics problems, a fixed

Cartesian grid approach in conjunction with levelset interface tracking is attractive. In this

regard, a sharp interface Cartesian grid-based, Ghost Fluid Method (GFM) is developed for

resolving embedded fluid, elasto-plastic solid and rigid (solid) objects in hyper-velocity im-

pact and high-intensity shock loaded environment. The embedded boundaries are tracked

and represented by virtue of the level set interface tracking technique. The evolving multi-

material interface and the flow are coupled by meticulously enforcing the boundary condi-

tions and jump relations at the interface. In addition, a tree-based Local Mesh Refinement

scheme is employed to efficiently resolve the desired physics. The framework developed is

generic and is applicable to interfaces separating a wide range of materials and for a broad

spectrum of speeds of interaction (O(km/s)). The wide repertoire of problems solved in this

work demonstrates the flexibility, stability and robustness of the method in accurately cap-

turing the dynamics of the embedded interface. Shocks interacting with large ensembles of

particles are also computed.

 2

Abstract Approved:
Thesis Supervisor

Title and Department

Date
A SHARP INTERFACE CARTESIAN GRID HYDROCODE

by

Shiv Kumar Sambasivan

A thesis submitted in partial fulfillment of the

requirements for the Doctor of Philosophy
degree in Mechanical Engineering
in the Graduate College of
The University of Iowa

May 2010

Thesis Supervisor: Professor H.S. Udaykumar

 Graduate College
The University of Iowa
Iowa City, Iowa

CERTIFICATE OF APPROVAL

PH.D. THESIS

This is to certify that the Ph.D. thesis of

Shiv Kumar Sambasivan

has been approved by the Examining Committee for the thesis re-
quirement for the Doctor of Philosophy degree in Mechanical Engi-
neering at the May 2010 graduation.
Thesis Committee:
H. S. Udaykumar, Thesis Supervisor

P. Barry Butler

Christoph Beckermann

James Buchholz

Pablo M. Carrica

George Constantinescu
 ACKNOWLEDGEMENTS

It is my pleasure to thank those who made this thesis possible. First and foremost,

I owe my deepest and sincere gratitude to my thesis supervisor, Prof. H.S. Udaykumar,

without whose guidance and constant encouragement this work would have been entirely

impossible. His endless patience and professional acumen are undeniably the reason for

instilling inspiration and reinforcing confidence in me. I have interacted with him both at

personal and professional levels, and I am grateful for the invaluable lessons that I have

learned from these interactions. I consider myself extremely fortunate to have him as my

thesis supervisor and I undeniably attribute the success of this thesis to his guidance.

I would also like to extend my sincere gratitude to all the members of my thesis

committee. Their painstaking effort in reading my thesis, the insightful suggestions and

the noteworthy advice are greatly and immensely appreciated.

Special thanks to my friend Haribalan Kumar for helping me compiling the figures

and movies presented in this thesis, and not to mention his excellent culinary skills that

made my stay supremely comfortable. I should also thank all the past and the present mem-

bers of my lab and friends who have made this journey a pleasant and an enjoyable one. I

would also like to thank the staff members of the Department of Mechanical Engineering,

for their timely help and support that kept me on track with the official responsibilities with

least encumbrance.

Finally words are not enough to thank my family, to whom this thesis is dedicated,

for their unconditional support that made this thesis feasible. Their continuous encourage-

ii
ment bolstered my confidence in overcoming every obstacle that I came across during the

course of this thesis.

The work was performed under the grants from the AFOSR computational mathe-

matics group and from the AFRL-MNAC (Eglin AFB).

iii
 TABLE OF CONTENTS

LIST OF TABLES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii

LIST OF FIGURES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . viii

CHAPTER

1 INTRODUCTION . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.1 Specific Objectives . . . . . . . . . . . . . . . . . . . . . . . . . . . 1

1.2 Highlights of the Thesis . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.3 Constitutive Relations and Governing Equations . . . . . . . . . . . . 9
1.3.1 Constitutive Relations . . . . . . . . . . . . . . . . . . . . . . 9
1.3.2 Governing Equations . . . . . . . . . . . . . . . . . . . . . . 11
1.3.3 Equation of State (E.O.S) . . . . . . . . . . . . . . . . . . . . 15
1.3.4 Radial Return Mapping Algorithm . . . . . . . . . . . . . . . 17
1.3.5 Material Models . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.4 Local Mesh Refinement . . . . . . . . . . . . . . . . . . . . . . . . . 20
1.5 Tracking of Embedded Interface . . . . . . . . . . . . . . . . . . . . 21
1.5.1 Implicit Interface Representation Using Level sets . . . . . . . 21
1.5.2 Classification of Grid Points . . . . . . . . . . . . . . . . . . 24
1.6 Ghost Fluid Method (GFM) . . . . . . . . . . . . . . . . . . . . . . . 24
1.6.1 The Basic Mechanism of the GFM . . . . . . . . . . . . . . . 25
1.6.2 Defining the Ghost States . . . . . . . . . . . . . . . . . . . . 26
1.7 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 34

2 LOCAL MESH REFINEMENT . . . . . . . . . . . . . . . . . . . . . . . . 45

2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 45
2.2 Automatic Grid Adaptation Schemes . . . . . . . . . . . . . . . . . . 47
2.3 Local Mesh Refinement (LMR) . . . . . . . . . . . . . . . . . . . . . 50
2.3.1 LMR Terminologies and Hierarchical Data Structure . . . . . 51
2.3.2 Coarsening and Refinement Operations . . . . . . . . . . . . . 52
2.3.3 Refinement Criteria . . . . . . . . . . . . . . . . . . . . . . . 54
2.3.4 Computing ENO Based Numerical Flux at the Mesh Interface 57
2.4 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
2.4.1 One-Dimensional Example . . . . . . . . . . . . . . . . . . . 63
2.4.2 Two-Dimensional Example . . . . . . . . . . . . . . . . . . . 66

3 SHARP INTERFACE TREATMENT FOR FLUID OBJECTS . . . . . . . . 79

iv
 3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 79
3.1.1 Survey of the Ghost Fluid Method (GFM) for Resolving Fluid-
Fluid Interfaces . . . . . . . . . . . . . . . . . . . . . . . . . 84
3.1.2 Riemann Problem at the Interface . . . . . . . . . . . . . . . . 85
3.1.3 Constructing the Local Riemann Problem at the Interface . . . 87
3.2 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 92
3.2.1 One-Dimensional Examples . . . . . . . . . . . . . . . . . . 92
3.2.2 Two-Dimensional Examples . . . . . . . . . . . . . . . . . . 104
3.3 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 123

4 SHARP INTERFACE TREATMENT FOR ELASTO-PLASTIC SOLIDS . . 172

4.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 172

4.2 Governing Equations and Constitutive Models . . . . . . . . . . . . . 174
4.3 Capturing the Response of the Material Interface . . . . . . . . . . . . 175
4.3.1 Populating the Ghost Points . . . . . . . . . . . . . . . . . . . 175
4.3.2 Collision Detection Algorithm . . . . . . . . . . . . . . . . . 178
4.4 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 181
4.4.1 Impact of a Copper Rod over a Rigid Substrate - Axisymmet-
ric Taylor Bar Experiment . . . . . . . . . . . . . . . . . . . 181
4.4.2 2D Axisymmetric Penetration of Steel Target by WHA Long
Rod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 183
4.4.3 Shock Wave Interaction with Hemispherical Groove . . . . . . 186
4.4.4 Void Collapse in a Copper Matrix . . . . . . . . . . . . . . . 187
4.4.5 Very High Speed Impact of Aluminum Sphere over Aluminum
Target . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 189
4.4.6 Perforation of Aluminum Plates by Conical Nose Projectile . . 191
4.4.7 High Velocity Impact Induced Explosion . . . . . . . . . . . . 196
4.4.8 Axisymmetric Dynamic-Tensile Large-Strain Impact-Extrusion
of Copper . . . . . . . . . . . . . . . . . . . . . . . . . . . . 198

5 SHARP INTERFACE TREATMENT FOR SOLID OBJECTS . . . . . . . . 246

5.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 246

5.2 Boundary Conditions . . . . . . . . . . . . . . . . . . . . . . . . . . 247
5.2.1 Boundary Condition Type I: Reflective Boundary Condition
(RBC) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 248
5.2.2 Boundary Condition Type II: Simple Boundary Condition (SBC)255
5.2.3 Boundary Condition Type III: Simple Boundary Condition us-
ing Coordinate Transformation (SBC-CT) . . . . . . . . . . . 257
5.2.4 Boundary Condition Type IV: Characteristic Based Matching
(CBM) [133, 135] . . . . . . . . . . . . . . . . . . . . . . . . 258
5.3 Numerical Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . 267
5.3.1 One Dimensional Examples . . . . . . . . . . . . . . . . . . . 267

v
 5.3.2 Two-Dimensional Examples . . . . . . . . . . . . . . . . . . 270
5.4 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 290

6 CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE WORK . . 341

6.1 Conclusions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 341

6.2 Significant Contributions of this Thesis . . . . . . . . . . . . . . . . . 341
6.3 Recommendations for Future Work . . . . . . . . . . . . . . . . . . . 342

REFERENCES . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 347

vi
 LIST OF TABLES

Table

1.1 Parameters for the Mie-Grüneisen E.O.S. for commonly used materials . . . . 36

1.2 Johnson-Cook material model parameters: A = Y0 , T0 = 294K and ǭ˙P0 = 1.0s−1 37

2.1 L2 errors computed with respect to the exact solution for frozen mesh condi-
tion, for Sod’s shock tube problem. . . . . . . . . . . . . . . . . . . . . . . . . 68

2.2 L2 (ρ) and OR for the two-dimensional Riemann problem . . . . . . . . . . . . 69

2.3 Initial conditions for the two-dimensional Riemann problem . . . . . . . . . . 70

3.1 L2 errors for the Gas-Gas Shocktube Problem and Air-Water Shocktube Prob-
lem I. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 125

4.1 Comparison of results for the axisymmetric impact of Copper rod at 227 m/s. . 200

4.2 Comparison of parameters for the axisymmetric impact of Copper rod for dif-
ferent impact velocities . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 201

4.3 Comparison with experimental and computational results for the jet velocity
and diameter. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 202

5.1 Timing analysis for a Mach 2.81 shock negotiating a stationary cylindrical ob-
1
stacle, for different levels of refinement with ∆xf = 800 . . . . . . . . . . . . . 292

5.2 CPU time taken for one complete (third-order) RK time iteration . . . . . . . . 293

vii
 LIST OF FIGURES

Figure

1.1 Spark-generated bubbles for modeling underwater explosion bubble dynam-

ics. This allows clear observation of re-entrant jet formation inside the bubble,
because the explosive detonation products generally occlude the view of the
interior of the bubble; Courtesy - Georges L. CHAHINE, Dynaflow, Fulton,
MD. Proc. 66th Shock and Vibration Symp., Biloxi, MS (1995). SAVIAC,
Arlington, VA (1995), vol. 2, p. 265 . . . . . . . . . . . . . . . . . . . . . . . 38

1.2 Snapshots of deformation of a bubble subject to a planar shock wave. . . . . . 39

1.3 Superposition of three flash radiographs showing a sequence of a 30-06 bul-

let in air penetrating a so-called Whipple shield - a configuration invented in
1946 by the astronomer Fred L. WHIPPLE. It is based on the principle that
small meteoroids and orbital debris explode when they strike a solid surface.
Courtesy - John P. BARBOUR and associates at Field Emission Corporation
in McMinnville, OR. See Proc. 7th Int. Congr. on High-Speed Photography,
Zurich (1965). Helwich, Darmstadt (1967), p. 292 . . . . . . . . . . . . . . . . 40

1.4 Crater formation details: Aerial view of Meteor Crater which is 1.2 km in
diameter and 200 m deep, officially called ”Barringer Crater” that is about
50,000 years old. Photo by David J. RODDY and Karl ZELLER, courtesy USGS 40

1.5 Snapshot of shock condensation effect around the jet fighter Boeing F-18 (Hor-
net) at transonic flight in humid air. It arises during acceleration when the air
flow at some parts of the fuselage reaches supersonic speeds. When the re-
sulting shock wave detaches, it builds up a sudden rarefaction that, lowering
the temperature, causes condensation of the ambient water vapor.[Courtesy: J.
GAY, U.S. Navy] A similar picture was recently taken from an F-4 (Phantom
II) during an air show at Point Magu Naval Air Station, CA. (The MilitaryAir-
craft Archive, http://www.milair.simplenet.com/) . . . . . . . . . . . . . . . . 41

1.6 Snapshot of the bow shock wave formed due to the supersonic flight of the
NASAs new Ares -1X rocket . . . . . . . . . . . . . . . . . . . . . . . . . . . 41

∗
1.7 Radial Return Algorithm: The predicted stress σn+1 lies outside the yield sur-
face. The predicted stress is brought back to the yield surface to obtain the
final corrected stress σn+1 , using a return mapping algorithm, along a direction
∂f
normal to the yield surface ∂σ ij
. . . . . . . . . . . . . . . . . . . . . . . . . . 42

viii
1.8 Classification of grid points - bulk points, interfacial ghost points and interior
ghost points. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 42

1.9 The GFM approach converts a two-material problem to two single-material

problem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43

1.10 The basic mechanism of the Ghost Fluid Method. . . . . . . . . . . . . . . . . 43

1.11 Discontinuity Ω0 (shock or material discontinuity) traveling inside a material

bounded by ∂. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44

2.1 Data Structures used in LMR: (a) Hierarchical (quad)tree structure; (b) hierar-
chical data structure . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 71

2.2 ENO-based flux construction at the fine-coarse mesh boundary: (a) adjoining
cells having one level difference along the diagonal direction (b) adjoining cells
having more than one level difference along the diagonal direction . . . . . . . 72

2.3 Sod’s shock tube problem with pressure ratio PP12 = 10: Plots of (a) density, (b)
pressure, (c) velocity (d) entropy. . . . . . . . . . . . . . . . . . . . . . . . . . 73

2.4 for Sod’s shock tube problem with pressure ratio PP21 = 10: Plots of (a) Density
and (b) entropy for different levels of mesh refinement, for frozen mesh condition. 74

2.5 Plots of errors: (a) Spatial variation of the absolute errors (kρ − ρEXACT k)
computed with respect to the exact solution for different refinement levels, for
frozen mesh condition (b) L2 (ρ) errors computed with respect to the exact
solution, for different refinement levels, for solution adaptive and frozen grid
conditions. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 75

2.6 Sod’s shock tube problem with pressure ratio PP21 = 1000: Plots of contours of
(a) density, (b) pressure, (c) velocity and (d) entropy. . . . . . . . . . . . . . . 76

2.7 Variation of |L2LM R − L2U nif orm | for different levels of mesh refinement for the
frozen mesh condition. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77

2.8 The two-dimensional Riemann Problem computed on a base mesh of size

∆xg = 0.02 with 5 levels of mesh refinement: Entropy contours (Figures (a)
and (e)), density contours (Figure(c)) and Mesh Topology (Figures (b), (d) and
(f)). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 78

3.1 The Riemann problem: (a) Approximate Riemann solver employed in MGFM;
(b) Procedure to construct the Riemann problem at the interface; (c) Typical
characteristic wave structure for Riemann problem. . . . . . . . . . . . . . . . 126

ix
3.2 Extension procedures: (a) One-dimensional version of the correction and ex-
tension procedure (b) Multi-dimensional correction and extension procedure. . 127

3.3 Example 1 - CASE(A): Plots of (a) density and (b) entropy for single fluid
Sod’s shock tube problem with pressure ratio PP21 = 10. . . . . . . . . . . . . . 128

3.4 Example 1 - CASE(A): Plots of (a) density and (b) entropy for single fluid
Sod’s shock tube problem with pressure ratio PP21 = 1000. . . . . . . . . . . . . 129

3.5 Example 1 - CASE(B): Plots of (a) density, (b) pressure, (c) velocity and (d)
entropy for Woodward Colella Bang Bang problem. . . . . . . . . . . . . . . . 130

3.6 Example 1 - CASE(B): Plots of total (a) mass, (b)energy and (c) momentum
conservation errors for Woodward Colella Bang Bang problem. . . . . . . . . . 131

3.7 Example 2 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
air-helium shock tube problem. The inserts correspond to the zoomed in view
of the variation close to the interface. . . . . . . . . . . . . . . . . . . . . . . . 132

3.8 Example 2 - Plots of total (a) mass, (b) energy and (c) momentum conservation
errors for air-helium shock tube problem. . . . . . . . . . . . . . . . . . . . . 133

3.9 Non-uniqueness in the numerical flux computed at the interface; Figure repro-
duced from “Abgrall, R. and Karni, S., Computations of Compressible Multi-
Fluids, Journal of Computational Physics, 169, 594-623(30), 20 May 2001”. . . 134

3.10 Example 3 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-water shock tube problem. . . . . . . . . . . . . . . . . . . . . . . . . . . 135

3.11 Example 4 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-water shock tube problem. . . . . . . . . . . . . . . . . . . . . . . . . . . 136

3.12 Example 5 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
water-air shock tube problem. . . . . . . . . . . . . . . . . . . . . . . . . . . 137

3.13 Example 6 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-gas shock tube problem at time T = 0.03 units. . . . . . . . . . . . . . . . 138

3.14 Example 7 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
air-water shock tube problem (I) at time T = 0.0015 units. . . . . . . . . . . . . 139

3.15 Example 8 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-water shock tube problem (II) at time T = 0.001 units. . . . . . . . . . . . 140

x
3.16 Initial configuration for single phase spherical Riemann problem. . . . . . . . . 141

3.17 Snapshots of numerical Schlieren image obtained with and without GFM treat-
ment for the initially spherical contact discontinuity. . . . . . . . . . . . . . . . 142

3.18 Spherical Riemann problem: Comparison of (a) density and (b) pressure vari-
ation at y = 0.6 at T = 0.5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 143

3.19 Initial configuration for a Mach 1.22 shock impinging on a cylindrical helium
bubble. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 144

3.20 Snapshots of numerical Schlieren image at different instants in time for a Mach
1.22 shock impinging on a cylindrical He bubble: (a) T = 85 µs, (b) T = 188
µs, (c) T = 427 µs, and (d) T = 803 µs. . . . . . . . . . . . . . . . . . . . . . 145

3.21 Snapshots of density contours at different instants in time for a Mach 1.22
shock impinging on a cylindrical He bubble: (a) T = 85 µs, (b) T = 188 µs, (c)
T = 427 µs, and (d) T = 803 µs. . . . . . . . . . . . . . . . . . . . . . . . . . 146

3.22 Plots of density (Figure 3.22(a)) and pressure (Figure 3.22(b)) along the hor-
izontal line of symmetry, and pressure as a function of time at 3mm down-
stream of the initial bubble location (Figure 3.22(c)). The insert displayed in
Figure 3.22(c) corresponds to the plot obtained by Marquina et al[2003]. . . . . 147

3.23 Initial configuration for a Mach 6 shock interacting with a cylindrical helium
bubble. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 148

3.24 Snapshots of numerical Schlieren image for a Mach 6 shock interaction with a
cylindrical He bubble at different instants in time. . . . . . . . . . . . . . . . . 149

3.25 Topology of the interface (zero level set field) at different instants in time. . . . 150

3.26 Initial configuration for underwater explosion near a free surface. . . . . . . . . 151

3.27 Snapshots of numerical Schlieren image and the corresponding density con-
tours at different instants in time. . . . . . . . . . . . . . . . . . . . . . . . . . 152

3.28 Initial configuration for underwater explosion near a free surface with im-
mersed structure. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 153

3.29 Snapshots of numerical Schlieren image (left) and corresponding pressure con-
tours (right) at different instants in time. . . . . . . . . . . . . . . . . . . . . . 154

xi
3.30 Plots of non-dimensionalized force components (Figure 3.30(a)) and moment
(Figure 3.30(b)) exerted on on the submerged square structure. . . . . . . . . . 155

3.31 Comparison of non-dimensionalized horizontal (Figure 3.31(a)) and vertical

(Figure 3.31(b)) force components exerted on the submerged square structure. . 156

3.32 Richtmyer-Meshkov Instability (RMI): (a) Comparison of density contours for

LMR calculations, with 3 levels of mesh refinement, with the corresponding
uniform fine mesh calculation (b) Topology of the interface at different instants
in time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 157

3.33 Snapshots of mesh evolution and density contours for the RMI at different
instants in time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 158

3.34 Snapshots of mesh evolution and numerical Schlieren image for a planar shock
(Mach 1.22) interacting with R22 cylindrical bubble at different instants in time. 159

3.35 Mach 1.22 shock interacting with R22 cylindrical bubble: Enlarged view of the
(a) interface evolution at different instants in time and (b) instabilities occurring
at the interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 160

3.36 Mach 1.47 shock interacting with spherical water droplet: (a), (b) and (c) Snap-
shots of mesh evolution and numerical Schlieren image computed on a base
1
mesh ∆xg = 200 with 5 levels of mesh refinement (d) Pressure distribution
along the line of symmetry at T = 0.015µs. . . . . . . . . . . . . . . . . . . . 161

3.37 Comparison of unsteady drag coefficient computed on the surface of a cylin-

drical droplet . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 162

3.38 Mach 3 shock interacting with spherical water droplet: (a), (b) and (c) Snap-
shots of mesh evolution and numerical Schlieren image computed on a base
1
mesh ∆xg = 200 with 4 levels of mesh refinement (d) Comparison with cylin-
1
drical water column computed on a base mesh ∆xg = 200 with 4 levels of
mesh refinement. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 163

3.39 Mach 6 shock interacting with spherical water droplet: (a) and (b) Snapshots
of mesh evolution and numerical Schlieren image computed on a base mesh
1
∆xg = 200 with 4 levels of refinement (c) Comparison with cylindrical bubble
1
computed on a base mesh ∆xg = 200 with 4 levels of refinement; (d) Pressure
distribution along the line of symmetry at T = 0.0031µs. . . . . . . . . . . . . 164

3.40 Numerical Schlieren image at T = 0.0316, for the collapse of cylindrical air
cavity under the impact of a strong shock wave in water . . . . . . . . . . . . . 165

xii
3.41 One dimensional water-air shock tube problem for the collapse of cylindrical
air cavity in water. Inserts shown in the figures correspond to the variation
close to the interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 166

3.42 Numerical Schlieren image for the solution with 4 levels of mesh refinement
on a base of size ∆xg = 0.16. . . . . . . . . . . . . . . . . . . . . . . . . . . . 167

3.43 Collapse of cylindrical air cavity:(a) Close in view of the numerical Schlieren
image at time T = 0.03 units; (b) Enlarged view of the Mach contours at time
T = 0.03 units; (c) Velocity of the tail (rear end) of the bubble as a function of
time; (d) Evolution of the interface (0-levelset). . . . . . . . . . . . . . . . . . 168

3.44 Initial configuration for high speed droplet impacting a flat substrate. . . . . . . 169

3.45 Snapshots of numerical Schlieren image illustrating the wave patterns gener-
ated from the impact of a high velocity droplet on a flat substrate at different
instants in time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 170

3.46 Snapshots of numerical Schlieren image illustrating the wave patterns gener-
ated from the impact of a high velocity droplet on a flat substrate at different
instants in time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 171

4.1 Embedding the boundary conditions with the interpolation procedure. . . . . . 203

4.2 Collision detection algorithm: (a) An elasto-plastic solid object immersed in

a free surface. The surrounding free surface is flagged as bulk material (b)
Algorithm to detect collision between any two level sets; φl indicate the value
of the level set function corresponding to the lth material interface and δ is the
distance between the approaching level sets at point P. . . . . . . . . . . . . . . 204

4.3 Initial configuration for two-dimensional axisymmetric Taylor test on a Copper

rod. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 205

4.4 Snapshots of pressure and effective plastic strain (ǫP ) contours at different in-
stants in time for the axisymmetric impact of Copper rod at 227 m/s. . . . . . . 206

4.5 Snapshots of pressure and effective plastic strain (ǫP ) contours at different in-
stants in time for the axisymmetric impact of Copper rod at 400 m/s . . . . . . 207

4.6 Snapshots of pressure and effective plastic strain (ǫP ) contours at different in-
stants in time for the axisymmetric impact of Copper rod at 600 m/s . . . . . . 208

xiii
4.7 Axisymmetric Taylor bar experiment: Figures (a), (b), and (c) show the evo-
lution and the topology of the interface at different instants in time for three
different impact velocities (d) Final configuration of the interface at 80µs. . . . 209

4.8 Initial configuration for the two-dimensional axisymmetric penetration of Steel

target by WHA long rod . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 210

4.9 Contours of equivalent plastic strain (ǫp ) and velocity of a Tungsten rod pene-
trating a steel plate at 1250 m/s. . . . . . . . . . . . . . . . . . . . . . . . . . 211

4.10 Contours of equivalent plastic strain (ǫp ) and mesh evolution of a Tungsten rod
penetrating a steel plate at 1250 m/s. . . . . . . . . . . . . . . . . . . . . . . . 212

4.11 Snapshots of the interface topology of a Tungsten rod penetrating a steel plate
at 1250 m/s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 213

4.12 Tungsten rod penetrating steel plate at 1250 m/s: (a) Trajectories and (b) ve-
locities of the nose and tail of the projectile. . . . . . . . . . . . . . . . . . . . 214

4.13 Contours of equivalent plastic strain (ǫp ) and velocity of a Tungsten rod pene-
trating a steel plate at 1700 m/s. . . . . . . . . . . . . . . . . . . . . . . . . . 215

4.14 Contours of equivalent plastic strain (ǫp ) and mesh evolution of a Tungsten rod
penetrating a steel plate at 1700 m/s. . . . . . . . . . . . . . . . . . . . . . . . 216

4.15 Snapshots of the interface topology of a Tungsten rod penetrating a steel plate
at 1700 m/s. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 217

4.16 Tungsten rod penetrating steel plate at 1700 m/s: (a) Trajectories and (b) ve-
locities of the nose and tail of the projectile. . . . . . . . . . . . . . . . . . . . 218

4.17 Snapshots of velocity contours and mesh evolution at different instants in time
for the response of a hemispherical groove to a shock wave . . . . . . . . . . . 219

4.18 Topology and evolution of the interface at different instants in time for the
response of a hemispherical groove to a shock wave . . . . . . . . . . . . . . . 220

4.19 Snapshots of velocity and equivalent plastic strain (ǫp ) contours at different
instants in time for the collapse of a cylindrical void in a Copper matrix . . . . 221

4.20 Topology and evolution of the interface for the collapse of a cylindrical void in
a Copper matrix . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 222

xiv
4.21 Initial configuration for the impact of a metallic sphere on a semi-infinite metal-
lic target . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 223

4.22 High velocity impact of metal sphere on a metal Surface: (a), (c), (e), and (g)
correspond to configuration 1 - impact of a heavy alloy Tungsten sphere on an
infinitely thick Steel target and (b), (d), (f), and (h) correspond to configuration
2 - impact of a Copper sphere on an infinitely thick Copper target . . . . . . . . 224

4.23 Toplogy and location of the interface at different instants in time: (a) configu-
ration 1 - impact of a heavy alloy Tungsten sphere on an infinitely thick Steel
target and (b) corresponds to configuration 2 - impact of a Copper sphere on an
infinitely thick Copper target . . . . . . . . . . . . . . . . . . . . . . . . . . . 225

4.24 Snapshots of effective plastic strain (ǫP ) and velocity contours at different in-
stants in time for the dynamic tensile extrusion of Copper . . . . . . . . . . . . 226

4.25 Snapshots of effective plastic strain (ǫP ) and mesh topology at different instants
in time for the 7000 m/s impact of Aluminum sphere on Aluminum target. . . . 227

4.26 Topology and evolution of the interface at different instants in time for the 7000
m/s impact of Aluminum sphere on Aluminum target. . . . . . . . . . . . . . . 228

4.27 Initial configuration for the perforation of Aluminum plate by a conical nose
projectile. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 229

4.28 Snapshots of effective plastic strain (ǫP ) contours and mesh evolution at differ-
ent instants in time for the perforation of 12.7 mm thick Aluminum plate at an
incident velocity of 1195 m/s. . . . . . . . . . . . . . . . . . . . . . . . . . . 230

4.29 Snapshots of effective plastic strain (ǫP ) contours and mesh evolution at differ-
ent instants in time for the perforation of 50.8 mm thick Aluminum plate at an
incident velocity of 1176 m/s. . . . . . . . . . . . . . . . . . . . . . . . . . . 231

4.30 Initial configuration for the explosion inside a Copper tube. . . . . . . . . . . . 232

4.31 Snapshots of numerical Schlieren image, at different instants in time, for the
response of a Copper tube to an explosion in water. . . . . . . . . . . . . . . . 233

4.32 Initial configuration for a shock wave interacting with metal particle. . . . . . . 234

4.33 Snapshots of pressure contours for a Tungsten (left) and an Aluminum (right)
particle subject to a shock wave in water. . . . . . . . . . . . . . . . . . . . . . 235

xv
4.34 Aluminum particle subjected to a planar shock wave: (a)Topology and location
of the interface at different instants in time; (b) locus of leading and trailing
edge as function of time. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 236

4.35 Snapshots of pressure contours for a cluster of Aluminum particles subject to

a planar shock wave in water. . . . . . . . . . . . . . . . . . . . . . . . . . . . 237

4.36 Initial configuration for the explosion initiated due to high velocity impact. . . . 238

4.37 Snapshots of effective plastic strain (ǫP ) and velocity contours at different in-
stants in time for the impact driven explosion problem. . . . . . . . . . . . . . 239

4.38 Snapshots of effective plastic strain (ǫP ) contours and mesh evolution at differ-
ent instants in time for the impact driven explosion problem. . . . . . . . . . . 240

4.39 Topology and evolution of the interface at different instants in time for the
impact driven explosion problem. . . . . . . . . . . . . . . . . . . . . . . . . . 241

4.40 Initial configuration for the axisymmetric dynamic-tensile impact-extrusion of

Copper. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 242

4.41 Snapshots of effective plastic strain (ǫP ) and velocity contours at different in-
stants in time for the dynamic tensile extrusion of Copper. . . . . . . . . . . . 243

4.42 Snapshots of velocity contours and mesh evolution at different instants in time
for the dynamic tensile extrusion of Copper. . . . . . . . . . . . . . . . . . . . 244

4.43 Topology and evolution of the interface at different instants in time for the
dynamic tensile extrusion of Copper. . . . . . . . . . . . . . . . . . . . . . . . 245

5.1 Illustration of grid points used in RBC . . . . . . . . . . . . . . . . . . . . . . 294

5.2 Isobaric fix and the SBC procedure: (a) Simple entropy extrapolation to correct
the density field for the interfacial points; (b) Evaluating the normal derivatives
at the interface. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 295

5.3 Configuration of a rotating shock tube embedded in the computational domain . 296

5.4 Plots of density along the centerline of the shock tube placed at different ori-
entations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 297

5.5 Configuration of a rotating shock tube embedded in the computational domain . 298

xvi
5.6 Woodward Colella “Bang Bang” Problem: Plots of Density (Figures 5.6(a)
and 5.6(b)), Pressure (Figures 5.6(c) and 5.6(d)) and Entropy (Figures 5.6(e)
and 5.6(f)) along the centerline for different orientations (α = 300 on the left
and α = 600 on the right) of the shock tube . . . . . . . . . . . . . . . . . . . 299

5.7 Plots of density and temperature for a Mach 10 shock reflecting off a stationary
wall . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 300

5.8 Mach 20 shock reflecting off a stationary wall: Plots of density (Figure 5.8(a))
and temperature (Figure 5.8(b)) . . . . . . . . . . . . . . . . . . . . . . . . . . 301

5.9 Snapshots of numerical Schlieren image of a Mach 1.3 shock diffracting a

wedge: Results from RBC-RS are shown in Figures 5.9(a), 5.9(c), 5.9(e) and 5.9(g)
and the the corresponding plots obtained from CBM are displayed in Fig-
ures 5.9(b), 5.9(d), 5.9(f) and 5.9(h). . . . . . . . . . . . . . . . . . . . . . . . 302

5.10 Snapshots of density contours of a Mach 1.3 shock diffracting a wedge: Results
from RBC-RS are shown in Figures 5.9(a), 5.9(c), 5.9(e) and 5.9(g) and the the
corresponding plots obtained from CBM are displayed in Figures 5.9(b), 5.9(d), 5.9(f)
and 5.9(h). . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 303

5.11 Comparison of primary shock detachment distance over time between CBM
and RBC-RS . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 304

5.12 Contours of density and corresponding numerical Schlieren image at time T

= 0.11: Results from RBC-RS are shown in figures 5.12(a) and 5.12(b) and
CBM in 5.12(c) and 5.12(d); Figure 5.12(e) corresponds to the experimental
Schlieren image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 305

5.13 Contours of density and corresponding Schlieren image at time T = 0.2: Re-
sults from RBC-RS are shown in Figures 5.13(a) and 5.13(b) and CBM in 5.13(c)
and 5.13(d); Figure 5.13(e) corresponds to the experimental Schlieren image. . 306

5.14 Contours of density and corresponding Schlieren image at time T = 0.5: Re-
sults from RBC-RS are shown in Figures 5.14(a) and 5.14(b) and CBM in 5.14(c)
and 5.14(d) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 307

5.15 Comparison between CBM and the RBC-RS : Shock detachment distance vs
time (a) and locus of upper triple point (b) . . . . . . . . . . . . . . . . . . . . 308

xvii
5.16 Comparison of drag coefficient (CD ) with experimental and numerical pre-
dictions, for Mach 1.3 and 2.6 shock waves negotiating stationary cylindri-
cal obstacles. The experimental image was obtained from “Takayama, K. and
Itoh, K.,Unsteady drag over cylinders and aerofoils in transonic shock tube
ows, Proceedings of the 15th International Symposium on Shock Waves and
Shock Tubes, Stanford, California, U.S.A., PP 439 485, 1985”. The corre-
sponding numerical image was obtained from ”D. Drikakis and D. Ofengeim
and E. Timofeev and P. Voionovich, Computation Of Non-Stationary Shock-
Wave/Cylinder Ineraction Using Adaptive-Grid Methods, Journal of Fluids and
Structures,11(6),665 - 692,1997“. . . . . . . . . . . . . . . . . . . . . . . . . 309

5.17 Numerical Schlieren image for the solution with 4 levels of mesh refinement
1
on a base of size ∆xg = 300 at different instants in time: (a) T = 0.314 µs; (b)
T = 0.612 µs; (c) T = 1.1 µs; (d) T = 2.4 µs; . . . . . . . . . . . . . . . . . . . 310

5.18 Enlarged view of the wake region of the numerical Schlieren image at two
1
different instants in time: (a) Uniform mesh calculation with ∆x = 800 at 1.1
µs; (b) LMR calculation at T = 1.1 µs; (c) Mesh topology in the wake region
at T = 1.1 µs; (d) LMR calculation at T = 2.4 µs; . . . . . . . . . . . . . . . . 311

5.19 Shock-cylinder interaction validation: (a) Comparison of locus of upper and

lower triple point (b) A parabolic fit for the locus of the upper and lower triple
point . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 312

5.20 Comparison of density contours ((a), (c) and (e)) and numerical Schlieren im-
age ((b), (d) and (f)) for LMR-based calculation with 6 levels of mesh refine-
1
ment, with the uniform fine mesh computations (with ∆x = 800 ) at different
instants in time: (a) and (b) at T = 0.314 µs, (c) and (d) at T = 0.47 µs and (e)
and (f) at T = 0.565 µs . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 313

5.21 Comparison of pressure distribution on the surface of the cylinder for different
levels of mesh refinement at time T = 5.65 µs . . . . . . . . . . . . . . . . . . 314

5.22 Sensitivity of the solution to the refinement criteria: (a) L2 (S) for entropy (b)
% Mesh occupancy ratio . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 315

5.23 Snapshots of numerical Schlieren image for a Mach 3 shock interacting with
multiple stationary cylinders at different instants in time: Results from RBC-
RS are shown in Figures 5.23(a), 5.23(c) and 5.23(e) and the corresponding
plots for the CBM approach are displayed in Figures 5.23(b), 5.23(d) and 5.23(f).316

5.24 Mach 2 shock interaction with a square inclined at 60o with the horizontal:
Figures (a) and (c) correspond to numerical Schlieren image and figures (d)
and (e) correspond to the density contours . . . . . . . . . . . . . . . . . . . . 317

xviii
5.25 Mach 2 shock interaction with a square inclined at 60o with the horizontal:
Figures (a) and (c) correspond to numerical Schlieren image and figures (d)
and (e) correspond to the density contours . . . . . . . . . . . . . . . . . . . . 318

5.26 Cylinder subject to Mach 3 shock: Numerical Schlieren image (RBC-RS in

Figure 5.26(a) and CBM in Figure 5.26(b)) and locus of center of mass (Fig-
ure 5.26(c)) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 319

5.27 Mach 10 shock interacting with a 30o ramp at T = 0.22: Contours of (a) density
and (b) entropy. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 320

5.28 Grid refinement study - (a) Density, (b) Pressure and (c) Entropy plots along
the ramp at T = 0.22 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 321

5.29 Mach 3 wind tunnel with a backward facing step: Density contours ((a), (c)
and (e)) and numerical Schlieren image ((b), (d) and (f)) . . . . . . . . . . . . 322

5.30 Mach 1.3 shock diffracting a 90o corner: (a) Density contours, (b) numerical
Schlieren image and (c) experimental Schlieren image . . . . . . . . . . . . . 323

5.31 Mach 2.4 shock diffracting a 90o corner: (a) Density contours, (b) numerical
Schlieren image and (c) experimental Schlieren image . . . . . . . . . . . . . 324

5.32 Mach 20 shock diffracting a 90o corner: (a) Density contours and (b) numerical
Schlieren image . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 325

5.33 Mach 2.1 shock negotiating a curved channel without obstacles: Density con-
tours ((a) and (c)) and numerical Schlieren image ((b) and (d)) at two different
instants in time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 326

5.34 Mach 2.1 shock negotiating a curved channel with stationary obstacles: Den-
sity contours ((a) and (c)) and numerical Schlieren image ((b) and (d)) at two
different instants in time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 327

5.35 Mach 2.1 shock negotiating a curved channel with a moving obstacle: Density
contours ((a) and (c)) and numerical Schlieren image ((b) and (d)) at different
instants in time . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 328

5.36 Series of numerical Schlieren image for a Mach 3 shock traversing through 120
randomly oriented stationary particles . . . . . . . . . . . . . . . . . . . . . . 329

5.37 Series of numerical Schlieren image (Right) and vorticity contours (Left) for
Mach 8 spherical shock interacting with 12 particles; Series correspond to the
shock wave dispersing initially stationary particles . . . . . . . . . . . . . . . 330

xix
5.38 Series of numerical Schlieren image (Right) and vorticity contours (Left) for
Mach 8 spherical shock interacting with 12 particles; Series correspond to the
shock wave reflecting off the domain walls. . . . . . . . . . . . . . . . . . . . 331

5.39 Configuration of a projectile travelling at Mach 1.22 . . . . . . . . . . . . . . . 332

5.40 Numerical Schlieren image and the mesh topology for a supersonic projectile
fired at Mach 1.22 at different instants in time. . . . . . . . . . . . . . . . . . . 333

5.41 Bluff body projectile fired at Mach 1.22: (a) Numerical Schlieren image at time
T = 7.03 ms; (b) Shock stand off as a function of time . . . . . . . . . . . . . . 334

5.42 Problem configuration for the dynamics of multiple particles introduced in high
speed flow . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 335

5.43 Dynamics of multiple particles introduced in high speed flow: (a) Locus of
horizontal coordinate of the center of mass; (b) Locus of vertical coordinate of
the center of mass; (c) Drag coefficient (d) Lift coefficient . . . . . . . . . . . . 336

5.44 Numerical Schlieren image and the mesh topology of cylindrical particles in-
troduced in high speed flows (a) T = 1.79 ms; (b) T = 5.9 ms; (c) T = 10.0
ms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 337

5.45 Initial configuration of the system for DNS of shock wave traversing through a
dusty layer of gas . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 338

5.46 Mach 1.67 shock wave traversing through a dusty layer of gas at time T = 76.15
µ s: (a) Numerical Schlieren image and (b) vorticity contours . . . . . . . . . . 339

5.47 Numerical Schlieren image for a Mach 3 shock wave traversing through a dusty
layer of gas at time T = 41.17 µ s . . . . . . . . . . . . . . . . . . . . . . . . . 340

6.1 Explosively Formed Projectile: (a) Components of EFP and (b) Radiograph
images of the jet formation process . . . . . . . . . . . . . . . . . . . . . . . . 345

6.2 The jet formation process: (a) Current calculations (b) Theoretical prediction . 346

xx
 1

CHAPTER 1
INTRODUCTION

The dynamic response of materials to high speed flows and/or high intensity loading

conditions due to shock and detonation waves is important in several applications including

high-speed flows with droplets [78], bubbles and particles [139,147], hypervelocity impact

and penetration [190], and detonation diffraction [158, 204]. In such high pressure physics

problems, there are computational challenges associated with the treatment of material in-

terfaces, particularly when the interface interacts with strong nonlinear waves like shock

and detonation waves. The result of such shock-interface interactions may be partial re-

flection, refraction or transmission of the impinging waves and rapid and severe distortion

of the material interfaces under the influence of the shocked fields [73]. To simulate such

complicated interfacial dynamics problems, a fixed Cartesian grid approach in conjunction

with level set interface tracking is attractive. In this Eulerian framework, the grid remains

indifferent to the fate of the interfaces, with the level sets carrying the onus of tracking the

interfaces through the likely topological changes. However, accurate treatment and repre-

sentation of these phenomena by retaining the interface as a sharp entity is a key challenge,

and this is addressed in this thesis.

1.1 Specific Objectives

In this work, a general framework is developed to solve multi-material interaction

problems with interfaces separating a wide range of materials (liquids, elasto-plastic solids

and rigid solids) and for a broad spectrum of speeds of interaction (for high Mach num-
 2

bers). Based on the material-strength/material-stiffness of the constituent media, materials

can range from negligible strength (fluid) to moderate strength (solids that undergo elas-

tic deformation followed by plastic flow) and infinite strength (i.e. rigid solids). These

categories of materials cover a plethora of applications ranging from shock-droplet inter-

actions at the small scale to geophysical impact at the large scale. In particular, the class of

problems that can be solved can be categorized as follows:

1. Resolution of Fluid Objects:

The framework can be applied to resolve fluid objects such as droplets and bubbles

immersed in high speed compressible flows. Fluids, by definition, are characterized

as materials with negligible strength as they offer no resistance to external shear or

torsional loads. Fluid objects, such as droplets and bubbles, embedded in the flow

interact with shock and detonation waves that may lead to reflection, refraction or

transmission of the impinging waves. Capturing the interface conditions in such

situations are important in applications such as shock wave lithotripsy [92], high

speed coating technologies [40, 78], under-water detonation/explosions [73] etc. For

instance, see Figures 1.1 & 1.2.

2. Resolution of Elasto-Plastic Solids:

The framework can be employed to study the deformation of elasto-plastic solids.

Elasto-plastic solids are materials with moderate to large material-strength. These

materials undergo very severe deformation when subject to intense shock and im-

pact loads. Elasto-plastic solids, unlike fluids, behave like a shape-memory material

for loads within the elastic regime [183]. When the applied load increases in strength
 3

and beyond the elastic limit, the material tends to flow (plastic deformation). Because

of the high speeds involved in the interaction process, these problems are character-

ized by nonlinear wave propagation phenomena in the material. The elasto-plastic

response of materials typically consist of a precursor elastic wave followed by a

plastic wave. These phenomena are common in high speed multi-material dynamics

problems such as munition-target interactions [17, 212], geological impact dynam-

ics [8, 150], shock-processing of powders [21, 22], vehicular collision crashworthi-

ness [23,214], formation of shaped charges [195,199] etc..Some of these applications

are displayed in Figures 1.3 & 1.4.

3. Resolution of Rigid (Solid) Objects:

Materials with infinite material-strength are classified as rigid (solid) objects. With

infinite material-strength, rigid objects (embedded in high speed flows) do not de-

form to external loads. Shock waves interacting with stationary or moving rigid solid

objects give rise to complex diffraction patterns that are important in applications

such as supersonic aerodynamics (Figures 1.5 & 1.6) of projectiles [96, 97], explo-

sive/shock dispersal of particles [114,154], shock waves traversing through dusty gas

layers [56] etc.. Thus, the framework developed in this work can be applied to study

and capture such shock diffraction patterns arising from shock-interface interactions.

1.2 Highlights of the Thesis

The unifying feature in the high-speed/high-strain rate phenomena is the hyperbolic

character of the system of conservation laws that govern the dynamics and the large mo-
 4

tions/deformations of the embedded interface in such systems. Furthermore, despite the

fact that the response of materials to such high speed multi-material interaction phenomena

has similar signature in different media, the numerical techniques developed thus far are ei-

ther directed towards specific applications [56, 78, 114, 129] or limited to resolve particular

material enclosed within the interface [58, 136, 180, 190, 192]. The reason for this dearth

in generality can be attributed to the modeling complexities and intricacies that one has to

pay close attention while solving such high speed interaction problems. In contrast, the

framework developed in this thesis is generic and has the simulation capabilities designed

to handle the modeling issues such as:

• Representation of nonlinear wave propagation phenomena in the material:

The response of materials to high speed and high intensity loading conditions are

dictated by the propagation of nonlinear stress waves in the interacting media. The

nature of the waves propagating in the material could be a combination of (longitu-

dinal) dilatational and (shear or transverse) distortional waves, which depend on the

material being subjected to the loading conditions. For instance, if the interacting

materials are made of fluids, then the waves generated are primarily dilatational (lon-

gitudinal pressure waves) in nature. On the other hand, if the constituent materials

are elastic or elasto-plastic, then the material can sustain elastic wave or elastic wave

followed by a slower, but intense plastic (distortional) wave. For very high speeds of

interaction, the nonlinear waves steepen to form much stronger (compressive shock

or tensile rarefaction) waves in the material. Thus the physical characteristics that

are common to these problems are dictated by the nature of stress waves propagating
 5

in the material. For instance, the interaction of these stress waves with the embed-

ded interface often lead to spallation, fracture or complete failure of the interacting

materials. Hence, special care must be exercised in developing numerical techniques

for solving such problems, so that the nonlinear wave propagation phenomena and

the associated material response are captured accurately. It will be shown later that

the methodology developed in this work is robust in resolving the wave-patterns and

wave-interactions for a wide class of materials.

• Resolution of the embedded interface undergoing large deformation:

Because of the very high speeds involved in the interaction process, the embedded

interface is often subject to very large deformations occurring at high strain-rate con-

ditions. Traditionally, numerical techniques developed to handle high speed multi-

material interaction phenomena treat the moving material boundaries by either allow-

ing the boundaries to flow through a fixed mesh while computing the flow field on

the fixed mesh (Eulerian framework), or by allowing the mesh to follow the material

points in the deforming materials (Lagrangian framework) [192]. An intermediate

approach, ALE (Arbitrary Lagrangian Eulerian), allows the mesh to move so as to

conform to the contours of the deforming object, but the mesh is not necessarily at-

tached to the material points [192]. In the Lagrangian moving mesh methods, consid-

erable complexity is enjoined by the need for mesh management, to accommodate

the large distortion of the embedded boundary. Therefore in such methods, peri-

odical re-meshing operations are required, so that an adequately refined mesh with

good mesh quality is maintained. In some cases, it is advantageous to use meshless

 6

methods such as the Smooth-Particle Hydrodynamics (SPH) to cope with severe de-

formations [214]. However such methods are not suitable for Eulerian-based flow

calculations. For very severe deformations, the use of embedded boundary meth-

ods wherein a fixed global mesh with embedded boundary tracking (sharp interface

treatment) and local mesh enrichment have emerged as promising alternatives.

In recent years, the use of Cartesian grid-based sharp interface approaches have be-

come immensely popular because such methods do not suffer from the complex grid

generation and grid management requirements inherent to other methods. In addi-

tion, Cartesian grid methods come with numerous advantages including simpler flux

construction and formulation, simplifications of the data structure, incidental cancel-

lation of truncation errors etc [198,207]. Cartesian grid methods retain the simplicity

associated with discretization of governing equations while shifting the onus to the

numerical treatment at the embedded interface. Hence in this thesis, a Cartesian

grid based Eulerian framework is developed for resolving and capturing the interface

conditions for embedded multi-material interface.

• Capturing interface conditions involving collision, fragmentation, and merging of the

multi-material interface:

The high speed interaction problems are often characterized by complex phenomena

such as collision, fragmentation, and merging of multi-material interfaces. Capturing

interface conditions in such situations pose significant modelling challenges partic-

ularly when the mesh does not conform and adhere to the shape of the embedded

interface. To relieve this shortcoming, in the present framework, level sets are used
 7

to track and represent the arbitrarily oriented interface [137]. With the level set tech-

nique, collision, fragmentation and merging of embedded interfaces can be naturally

incorporated in the interface tracking procedure.

• Sharp treatment of the multi-material interface:

Since the zero level set contour implicitly determines the boundary that separates

different media, mixing of materials inherent in techniques that smear the material

interface [5, 25, 98, 102] is prohibited. Hence, the exact location of the interface

and the materials enclosed by the interface are known at all times [190]. Thus, the

boundary conditions and the jump relations can be directly imposed on the interface

by suitably communicating with the flow and the tracked interface. In the present

framework, this task is accomplished via the Ghost fluid Method (GFM) developed

in [58].

The pivotal theme in the GFM approach lies in the definition of a band of ghost points

corresponding to each phase of the interacting media. The ghost band when supplied

with appropriate flow conditions, together with the respective real fluid, constitutes

a single flow field. Hence higher-order numerical schemes such as ENO [163] and

WENO [74] developed for single component flows, can be readily employed at the

interface. The success of the GFM approach largely depends on the accuracy with

which the ghost states are predicted. The ghost states in turn are derived based on the

material enclosed by the embedded object. Thus in the GFM framework, the treat-

ment of embedded interfaces essentially boils down to suitably defining the ghost

states such that the material properties and the interface conditions are depicted ac-
 8

curately.

• Resolution of disparate lengths:

While the GFM-based Cartesian grid approach significantly alleviates the complex-

ity associated with mesh management, the method lacks flexibility in effective and

automatic grid point placement in regions with rich structures in the flow field. This

is particularly crucial for problems involving disparate length scales and steep gradi-

ents in the flow field, and large deformations of the embedded interface [108, 191].

Typically in compressible flows, such regions are concentrated spatially, near shock

waves, contact discontinuities, rarefactions waves, detonation waves or in the re-

gions of baroclinic vorticity concentration, that occupy a small fraction of the com-

putational domain. To adequately capture such intricate flow features arising from

the diffraction patterns of shock or detonation waves, one must resort to numerical

schemes of inherently higher-order with extremely fine mesh resolution throughout

the computational domain [146]. An alternative approach is to supplement the so-

lution with adaptive grid technology that automatically refines and follows discon-

tinuities with the evolution of the flow field. Hence, the GFM approach developed

in this work, is augmented with the quadtree- (octree- in three dimensions) based

Local Mesh Refinement (LMR) technique developed in [108] to efficiently resolve

the desired physics.

In this thesis, the framework has been applied to capture the interface conditions for

embedded fluid, elasto-plastic solids and rigid solid objects in high speed multi-material

impact and shock loaded environment. The broad range of results presented in this work
 9

demonstrates the flexibility and robustness of the current approach. Shocks interacting with

stationary and moving boundaries have been computed.

1.3 Constitutive Relations and Governing Equations

The constitutive relations and the governing equations solved in this work have been

discussed in [190, 192]. For the sake of completeness, a brief review of the equations are

presented here.

1.3.1 Constitutive Relations

The response of materials to high intensity (shock/impact) loading conditions are

modeled by assuming the additive decomposition of strain rule,

E P
Dij = Dij + Dij (1.1)

where Dij is the total strain-rate tensor given as,

 
1 ∂ui ∂uj
Dij = + (1.2)
2 ∂xj ∂xi

E P
and Dij & Dij are the elastic and plastic strain-rate components respectively, and ui is the

velocity component. The validity of additive strain rule can be justified for the relatively

small elastic strain-rate experienced in the high speeds considered in this work. Assuming

P


incompressibility (isochoric) of the plastic flow (tr Dij = 0), the volumetric or dilata-

tional response is governed by an equation of state while the deviatoric response obeys a

conventional flow theory of plasticity [190]. Hence, the total stress in the material can be

expressed as

σij = Sij − P δij (1.3)

 10

where σij is the Cauchy stress tensor, Sij is the deviatoric component and P is the hydro-

static pressure taken to be positive in compression. Using Eq 1.1, the rate of change of

deviatoric stress component can be modeled using the hypo-elastic stress-strain relation

(Hooke’s law):
▽
P
S ij = 2G(D̄ij − Dij ) (1.4)

▽
where G is the shear or rigidity modulus, S ij is the Jaumann derivative [201]

▽
S ij = Ṡij + Sik Ωkj − Ωik Skj (1.5)

and Ωij is the spin tensor. The Jaumann derivative is used to ensure objectivity of the stress

tensor with respect to rotation. The spin tensor used in Eq 1.5 is given by:

 
1 ∂ui ∂uj
Ωij = − (1.6)
2 ∂xj ∂xi

The deviatoric strain-rate component, D̄ij , in Eq 1.4 is given by:

1
D̄ij = Dij − Dkk δij (1.7)
3

P P
The isochoric plastic strain-rate component (Dij = D̄ij ) in Eq 1.2 is modeled assuming a

coaxial flow theory (Druckers’ postulate) for strain hardening materials [105]:

p
Dij = ΛNij (1.8)

Sij
where Nij = √
Skl Skl
is the unit outward normal to the yield surface and Λ is a proportional

positive scalar factor called the consistency parameter [190]. The consistency parameter Λ

is determined using J2 Von Mises yield condition. The effective plastic stress (Se ) and the
 11

effective plastic strain-rate (ǭ˙p ) are given by:

3

Se2 = 2
tr (Sij Sji ) = 3 Sxx 2
+ Syy 2
+ Szz (1.9)
2
2 2 p p

2 2
(ǭ˙p ) = tr Dij Dji = Λ (1.10)
3 3

The evolution of temperature due to heat conduction and thermal energy produced by work

done during elasto-plastic deformation is written as

ρC Ṫ = k∇2 T − α (3λ + 2µ) To ǫ̇ekk + β Ẇp (1.11)

where T is the temperature, C the specific heat, k thermal conductivity, α thermal expan-

sion coefficient, λ & µ are the Lamé constants, Ẇp is the stress power due to plastic work

and β is the Taylor-Quinney parameter [186]. The Taylor-Quinney parameter implies the

fraction of mechanical power converted to thermal power and is taken as 0.9. The stress

power due to plastic work is given by

Ẇp = ǭ˙P Se (1.12)

where Se is the Von-Mises effective stress (Eq 1.9). For the applications considered in this

work, the conduction and elastic work terms are small in comparison with the plastic stress

power term Ẇp .

1.3.2 Governing Equations

To solve high strain-rate deformation of materials, the traditional operator splitting

algorithm is employed [164]. The integration of the mass, momentum and energy balance

P
laws are performed assuming pure elastic deformation (with Dij = 0) to obtain the elastic
 12

predictor step, followed by a radial return mapping to bring the predicted stress back to the

yield surface (Kuhn-Tucker consistency conditions) [165].

Because of the high speeds involved in the interaction process, the governing equa-

tions comprise a set of hyperbolic conservation laws. Cast in Cartesian coordinates, the

governing equations take the following form:

~
∂U ∂ F~ ~
∂G ~
∂H
+ + + Φ3D ~
=S (1.13)
∂t ∂x ∂y ∂z

For the elastic predictor step, in addition to the mass, momentum and energy equations, the

constitutive models for deviatoric stress terms are evolved. Thus the conservative variable

and the fluxes in Eq 1.13 take the form given below:

       
 ρ   ρu   ρv   ρw 
       
       
 ρu   ρu2 + p   ρuv   ρuw 
       
       
       
 ρv   ρuv   ρv 2 + p   ρvw 
       
       
       
 ρw   ρuw   ρvw   ρw2 + p 
       
       
       
 ρE   u (ρE + p)   v (ρE + p)   w (ρE + p) 
       
       
       
 ρǭp   ρuǭp   ρvǭp   ρwǭp 
       
~
U =  , F~ =  ~ =
,G ~ =
,H .
       
 ρT   ρuT   ρvT   ρwT 
       
       
       
       
 ρSxx   ρuSxx   ρvSxx   ρwSxx 
       
       
       
 ρSxy   ρuSxy   ρvSxy   ρwSxy 
       
       
       
 ρSyy   ρuSyy   ρvSyy   ρwSyy 
       
       
       
 ρSxz   ρuSxz   ρvSxz   ρwSxz 
       
       
ρSyz ρuSyz ρvSyz ρwSyz
 13

The source term in Eq 1.13 is given as:

 
 −ΦA ρu
x 
   
 
∂Sxx ∂Sxy ∂Sxz ρu2

 ∂x
+ ∂y
+ Φ3D ∂z
− Φ A x
− S xx + S zz


 
   
 ∂Sxy ∂Syy ∂Syz ρuv Sxy 
 ∂x
+ ∂y
+ Φ 3D ∂z
− ΦA x
− x 
 
   

 Φ3D ∂S∂xxz + ∂S∂yyz + ∂S∂zzz 

 
 
 SE 
 
 
 
 ρuǭp 
 −ΦA x 
~=
S .
 
 ρuT 
 −ΦA x 
 
 
 
 SSxx 
 
 
 
 SSxy 
 
 
 
 SSyy 
 
 
 
 SSxz 
 
 
SSyz

where


 ∂


 ∂x
(uSxx + vSxy + Φ3D wSxz ) +




 ∂

∂y
(uSxy + vSyy + Φ3D wSyz ) +
SE = (1.14)

 ∂


 Φ3D ∂z (uSxz + vSyz + wSzz ) −



  
 ΦA u(ρE+P ) − uSxx +vSxy

x x



 2ρΩ S + Φ 2ρΩ S +
xy xy 3D xz xz
SSxx = (1.15)


 2ρGD̄xx − ΦA ρuSx xx
 14






 ρΩxy (Syy − Sxx ) +


SSxy = Φ3D (ρΩxz Szy − Ωzy Sxz ) + (1.16)






 2ρGD̄xy − ΦA ρuSx xy



 2ρΩ S + Φ 2ρΩ S +
yx xy 3D yz yz
SSyy = (1.17)


 2ρGD̄yy − ΦA ρuSx yy



 ρΩxz (Szz − Sxx ) +
SSxz = Φ3D (1.18)


 (ρΩxy Syz − Ωyz Sxy ) + 2ρGD̄xz



 ρΩyz (Szz − Syy ) +
SSyz = Φ3D (1.19)


 (ρΩyx Sxz − Ωxz Sxy ) + 2ρGD̄yz

where the switch functions Φ3D & ΦA take the value 1 for three-dimensional and ax-

isymmetric problems respectively. For the two-dimensional problems of interest in this

work, the switch function Φ3D is set to zero. The deviatoric strain-rate components used in

Eqs 1.15 - 1.19 are given as follows:

  u
∂u 1 ∂u ∂v ∂w
D̄xx = − + + Φ3D − ΦA (1.20)
∂x 3 ∂x ∂y ∂z x
  
∂v 1 ∂u ∂v ∂w u
D̄yy = − + + Φ3D − ΦA (1.21)
∂y 3 ∂x ∂y ∂z x

 
1 ∂u ∂v
D̄xy = D̄yx = Dxy = + (1.22)
2 ∂y ∂x
 
1 ∂u ∂w
D̄xz = D̄zx = Dxz = + (1.23)
2 ∂z ∂x
 
1 ∂v ∂w
D̄yz = D̄zy = Dyz = + (1.24)
2 ∂z ∂y

The evolution of effective plastic strain (ǭp ) and temperature (T ) (in conservative form)
 15

included in the governing equations are given by:

∂ρǭp ~ ρuǭp
+ ∇• (ρ~uǭp ) = −ΦA (1.25)
∂t x
∂ρT ~ k 2 e
 ρuT
+ ∇ (ρ~uT ) =
• ∇ T − α (3λ + 2µ) To ǫ̇kk + β Ẇp − ΦA (1.26)
∂t c x
P
where in Eq 1.25 the plastic deformation (Dij ) is neglected in the elastic predictor step.

Eigenvalue analysis for the (one-dimensional) system of equations conducted in [194], re-

veal that the Eigenvalues are real for the problems of interest in this work. In the one dimen-

sional set up, the Eigenvalues of the system of equations are λ1 = λ2 = λ3 = u, λ4 = u + c

& λ5 = u − c, where c is the speed of sound.

1.3.3 Equation of State (E.O.S)

Closure for the set of governing equations is obtained by modelling the dilatational

(pressure) response of the material using a suitable equation of state. For this purpose, the

pressure P, specific internal energy e and specific volume (V = ρ1 ) are related through a

relation of the form:

(e − ec (V )) e
P (e, V ) ≈ Γ(V ) + Pc (V ) = Γ + f (V ) (1.27)
V V

where ec and Pc denote the reference specific internal energy and pressure at 0 K. The

E.O.S shown in Eq 1.27 is the incomplete Mie-Grüneisen formulation [190]. Eq 1.27 can

also be viewed as the first-order approximation of the state surface in the neighborhood

of the measured Hugoniot curve along an isochoric path [83]. Γ(V ) in Eq 1.27 is the

Grüneisen parameter defined as

∂P Γ0 ρ 0
Γ(V ) = V ( )|v = (1.28)
∂e ρ
 16

where ρ0 is the density of the unstressed material. As pointed out in [83], it is important to

note that the Mie-Grüneisen formulation is not applicable for problems with phase change.

The overall pressure is expressed as the sum of thermal excitation and attractive/repulsive

forces in the lattice atoms [190]. Accommodating for negative pressure (tension) and pre-

serving the positivity of sound speed-squared, the function f (V ) in Eq 1.27 is written as



 ρ0 c20 Φ Γ

 (1−sΦ2 )2
[1 − 2V
(V0 − V )] if V ≤ V0
f (V ) = (1.29)


c20 ( 1 −
 1
) if V > V0
V V0

In the above expression, the constants c0 is the bulk sound speed and s is related to the

isentropic pressure derivative of the isentropic bulk modulus [190]. The c0 & s coefficients

relate the shock speed Us and the particle velocity Up . Experiments on solids provide a

relation between Us and Up . A first approximation consists of a linear relation given as

Us = co + sUp (1.30)

The expression for the speed of sound in the material is then given by

∂P P ∂P P
c2 = |e + 2 ( )|ρ = Γe + f ′ (V ) + Γ (1.31)
∂ρ ρ ∂e ρ

The parameters for the Mie-Grüneisen E.O.S. for some of the commonly used materials

are given in Table 1.1. For additional details regarding the equation of state the reader may

refer to [12, 190] and the references cited therein.

 17

1.3.4 Radial Return Mapping Algorithm

The plastic deformation of material is governed by the yield function that constrains

the stress to remain on or within the elastic domain [143]:

f (Sij , αk ) ≤ 0 ⇒ admissible stress state (1.32)

f (Sij , αk ) > 0 ⇒ inadmissible stress state (1.33)

where f is a generic yield function and αk is the scalar or tensor hardening parameter [105].

In the traditional operator splitting algorithms, if the predicted “trial” elastic state (deter-

mined by freezing the plastic flow) falls within the yield surface, i.e. f < 0, then the

deformation is purely elastic and the final stress state is indeed the predicted trial state.

The yield and the subsequent plastic flow is said to occur when f = 0. The inadmissible

trial state for f > 0 is corrected by bringing the stress back to the yield surface, i.e. by

enforcing the consistency condition f = 0, along a direction normal to the yield surface
∂f
( ∂σ ij
, Figure 1.7). Conventionally, the return mapping to obtain the final corrected stress

state is effected by classical radial return algorithms [130, 142, 143, 164, 165]. In this work,

the algorithm due to Ponthot et al [142, 143] is adopted and is explained below.

The radial return algorithm due to Ponthot et al [142, 143] is based on the J2 Von-

Mises flow theory which assumes the existence of a yield function (for isotropic materials)

of the form

f (Sij , σY ) = Se − σY = 0 (1.34)

with linear hardening law given by:

r
2
σ̇Y = hΛ (1.35)
3
 18

where σY is the current yield stress (determined from material models discussed in sec-

tion 1.3.5) and h (also called plastic modulus) is the slope of the effective stress versus

effective plastic strain curve under uniaxial loading. Using Eq 1.10, the yield stress can be

written as

σ̇Y = hǫ̇¯Pij (1.36)

When elastic deformation occurs, f < 0 and Λ = 0. Plastic deformation is said to occur

when the consistency condition holds true, f˙ (Sij , σY ) = 0. Thus, for elastic and plastic

deformation, f and Λ can be obtained from the Kuhn-Tucker conditions of optimization

theory [143, 164, 165]:

Λf = 0, Λ ≥ 0, f ≤ 0 (1.37)

In conjunction with the operator splitting algorithm, deviatoric stress update

P


Ṡij + Sik Ωkj − Ωik Skj = 2G D̄ij − D̄ij (1.38)

is split into two parts - predictor and corrector steps. The predicted trial elastic state “tr” is

P
obtained by freezing the plastic flow (D̄ij = 0),

Ṡij,tr + Sik,tr Ωkj − Ωik Skj,tr = 2GD̄ij (1.39)

where Sij,tr is the trial elastic stress tensor. The trial elastic state is determined by numer-

ically integrating the governing equations (Eq 1.13). The plastic corrector step is enforced

to bring the computed trial stress back to the yield surface:

P
Ṡij,cor = −2GDij = −2GΛNij (1.40)
 19

where Sij,cor is the corrected stress update and Nij is the normal direction in which the

return mapping is effected:

Sij,tr
Nij = p (1.41)
Skl,tr Skl,tr

In discrete form, the plastic corrector step can be obtained by integrating Eq 1.40 (by hold-

ing the global configuration fixed) to obtain

Sij,cor = Sij,tr − 2GNij,tr ζ (1.42)

R t1
where ζ = t0
Λdt, with t0 and t1 denoting the beginning and end of the time interval

of integration. The parameter ζ is determined by enforcing the generalized consistency

condition, f = 0, at time t = t1 .
r
3
f= [(Sij,tr − 2GNij,tr ζ) (Sij,tr − 2GNij,tr ζ)] = 0 (1.43)
2

Integrating Eqs 1.10 & 1.36 in time, we get

r
2
ǭP1 = ǭP0 ζ + (1.44)
3
r
2
σY1 0
= σY + hζ (1.45)
3

where “0” and “1” denote the values at t0 and t1 , respectively. Substituting for σY1 , Eq 1.43

is simplified
  r !  
2 4 p 4 2 2 02
4G − ζ2 − 4G Sij,tr Sij,tr + h ζ + Sij,tr Sij,tr − σY = 0 (1.46)
9 3 3 3

to obtain
p
Sij,tr Sij,tr − 32 σY0
ζ= h

(1.47)
2G 1 + 3G

Thus, once ζ is obtained, the correction of the predicted deviatoric stresses is performed

using Eq 1.40 and the consistency condition is enforced.

 20

1.3.5 Material Models

Material models are required to determine the flow (yield) stress to enforce the

consistency conditions in the return mapping algorithm. The two main models used in

this work for strain hardening materials are the rate independent Prandtl-Ruess material

model [105] (Eq 1.48) and thermal softening based Johnson-Cook material model [99]

(Eq 1.49), given as:

σY = A + B(ǭP )n (1.48)
  P 
P n

ǭ˙
σY = A + B(ǭ ) 1 + Cln P (1 − θm ) (1.49)
ǭ˙0

T −T0
where A, B, C, n, m, ǭ˙P0 are model constants and θ = Tm −T0
, where Tm and T0 are mate-

rial melting and the reference room temperatures respectively. The specific values of the

parameters used in the Johnson-Cook model are given in Table 1.2. In the Johnson-Cook

material model (Eq 1.49), the flow stress σY , increases with an increase in effective plastic

strain and effective plastic strain-rate and decreases with an increase in temperature [190].

The yield stress in fact approaches zero as the temperature approaches the melting temper-

ature of the material.

1.4 Local Mesh Refinement

A crucial aspect of this thesis is the resolution of dominant structures and disparate

length scales present in the computational domain. As mentioned before, a tree-based Local

Mesh Refinement (LMR) scheme is used for grid adaptation. In contrast to traditional

grid adaptation approaches [26, 82, 196], the LMR scheme sub-divides each cell that is

tagged for refinement to form four (quadtree in two dimension) or eight (octree in three
 21

dimensions) child cells resulting in highly unstructured mesh. Since each cell is created

and destroyed individually, the LMR scheme does not require constant re-meshing and

update of the global mesh. As the resulting mesh is unstructured, the hierarchical data

structure associated with LMR scheme contains neighbor and parent-child connectivity

information stored in the cell structure. With hierarchical data structure the grid refinement

and coarsening operations are trivial to accomplish. Furthermore, as the LMR scheme

does not require optimized rectangular patches of mesh, fewer mesh points are used in the

computation resulting in significant savings in computational memory and on a Cartesian

mesh, features that are misaligned with the mesh can be captured by mesh refinement

tangent to the feature. Unlike the AMR approach, the flow field is evolved only on the

finest (undivided) cells (termed leaf cells in LMR terminology). Thus the solution for

every time step is achieved in a single sweep of solution step making the LMR scheme

more attractive than its counterpart. Since the flow field is evolved only on the leafcells,

no special treatment is required for points near the embedded interface and the numerical

scheme can be uniformly integrated throughout the computational domain. The coupling

algorithm, refinement criterion to trigger refinement and coarsening operations, along with

the effectiveness of the LMR approach are discussed at length in chapter 2.

1.5 Tracking of Embedded Interface

1.5.1 Implicit Interface Representation Using Level sets

Sharp interface treatment requires tracking and representation of material interfaces

as the underlying global mesh does not conform to the shape of the interface. In this
 22

work, level sets introduced by Osher and Sethian [137] are used to represent the embedded

interface. The value of the level set field, φl , at any point is the signed normal distance

from the lth embedded interface with φl ≤ 0 inside the immersed boundary and φl > 0

outside. The interface is implicitly determined by the zero level set field, i.e., the φl = 0

contours represent the lth immersed boundary. The levelset field is defined only in a small

neighborhood of the interface (φl = 0 contours) called the narrow band. For interfaces

representing bubbles, droplets and elasto-plastic solids that undergo large deformations,

the level set field is advected using the traditional level set advection equation [161, 162].

∂φl ~ ~
+ Vl · ∇φl = 0, (1.50)
∂t

where V~l denotes the level set velocity field for the lth embedded surface. A fourth-order

ENO scheme for spatial discretization and a fourth-order Runge-Kutta time integration are

used in solving the level set advection equation. The velocity of the levelset field V~l , is

defined only on the embedded interface (the zero-levelset contours). The value of velocity

field at the grid points that lie in the narrow band around the zero-levelset is determined by

solving the extension equation [15] to steady state as given below:

∂Ψ ~ =0
+ H(φl )~n · ∇Ψ (1.51)
∂τ

where Ψ is the variable extended across the interface, τ is the pseudo-time and H(φl ) is the

unit Heaviside function defined as




0 if φl ≤ 0.0,

H(φl ) =



1 if φl > 0.0.
 23

A reinitialization procedure [179] is initiated at regular intervals to enforce the condition

~ l | = 1. The reinitialization of the levelset field is effected by solving the following

|∇φ

equation to steady state:

∂φl
+w
~ · ∇φl = sign(φl ) (1.52)
∂τ

where
∇(φl )0
w
~ = sign((φl )0 )
|∇(φl )0 |
(φl )0
sign((φl )0 ) = q
(φl )0 2 + ∆x2
with initial condition φl (~x, 0) = (φl )0 (~x) For the case of embedded (rigid) solid objects

considered in this work, the level set field is reconstructed by advancing the center of mass

based on the net (pressure) force acting on the body:

Z
d2~xcm
ρs ∀s = P n̂l da (1.53)
dt2 s

where ρs , ∀s & ~xcm are the density, volume and the coordinates of the center of mass of the

embedded rigid object. The normal vector (n̂l ) in equation (1.53) and the curvature (κl ) at

the interface can be computed from the level set field according to

~ l
∇φ
n̂l = , (1.54)
~ lk
k∇φ

~ · n̂l ,
κl = −∇ (1.55)

For other relevant details regarding the implementation of level set methods, the reader is

referred to the literature [137, 160].

 24

1.5.2 Classification of Grid Points

The grid points on the Cartesian mesh can be classified as bulk points and interfacial

points. The points which lie immediately adjacent to the interface are tagged as interfacial

points as shown in Figure 1.8. Identification of interfacial points is straightforward with

the level set field. If φcurr φnbr ≤ 0 , where the subscript curr denotes the current point and

nbr denotes the neighboring point, then the current and the neighboring point are tagged as

interfacial points. All the other points are classified as bulk points. A band of ghost points

are defined for each phase of the interacting media as shown in Figure 1.8. The ghost point

band typically extends up to 4 max(∆x,∆y) distance from the interface. Again, the level

set field can be used to define the band of ghost points. The set of ghost points which are

immediately adjacent to the interface are tagged as interfacial ghost points similar to the

regular interfacial points.

1.6 Ghost Fluid Method (GFM)

One of the main highlights of this thesis is the development a general framework

for treating embedded interfaces subject to intense loading conditions. For this purpose,

the Ghost Fluid Method (GFM) due to Fedkiw et al [58] is employed to capture the in-

terface conditions. The Ghost Fluid Method was first proposed by Glimm et al [67], later

adopted by Fedkiw et al [58], and is frequently employed to treat compressible flow prob-

lems with embedded interfaces. The GFM formulated by Fedkiw and co-workers provides

a simple framework to inject the boundary conditions for sharp representation of discon-

tinuities [133]. The central idea is the definition of a band of ghost points corresponding
 25

to each phase of the interacting materials. For instance, consider the case of two materials

separated by an interface as shown in Figure 1.9. Once the set of ghost field is identi-

fied and defined with flow conditions, the two-material problem can be converted to two,

single-material problem consisting of the respective real field and the corresponding ghost

field. Thus, effective communication with the flow solver can be achieved by suitably pop-

ulating these ghost points with flow properties such that the real field together with the

corresponding ghost field depict the interface properties and conditions precisely.

1.6.1 The Basic Mechanism of the GFM

The first step in the GFM approach is to identify the band of ghost points corre-

sponding to each phase of the interacting materials. The band of ghost points can be easily

identified based on the sign of the level set field. For instance, in the one-dimensional set-

up shown in Figure 1.10, the ghost points for the two interacting fluids can be identified

based on the change in sign of the level set field across the interface. Once the band of ghost

points are determined, the next step is to infuse these band of ghost points with appropriate

flow conditions. For the case of the interface that separates two-fluids (Figure 1.10), the

following approach was adopted in [58]. In [58], the ghost cells were populated based on

the fact that the pressure and velocity are continuous across the contact discontinuity. This

enables the pressure and velocity to be directly copied from the real field onto the ghost

fluid. Since the entropy advects with material velocity (corresponding to the characteristic

wave traveling with fluid/particle velocity), there is no entropy exchange between the two

fluids. This allows the entropy to be extended to the ghost points (isobaric fix) which can
 26

then be used to compute the density for the ghost points (Figure 1.10). Thus with the in-

fused ghost-field and the corresponding real field the two-fluid problem can now be solved

as two single-fluid problem.

1.6.2 Defining the Ghost States

In the original GFM approach [58] described in the previous section, a methodol-

ogy was formulated to populate ghost points that apply particularly for interfaces separating

fluid media. In this work, the GFM approach is extended to resolve interfaces separating

materials with varied stiffness. In general, for any surface of discontinuity (nonlinear shock

wave or linear contact discontinuity), such as Ω0 in Figure 1.11, the following jump rela-

tions in mass, momentum and energy (Eqs 1.56 - 1.58) hold true (for any point on Ω0 ):

[ρ(~u − ~uI )•n̂] = 0 (1.56)

[ρ~u(~u − ~uI )•n̂ − n̂•σ] = 0 (1.57)

[ρE(~u − ~uI )•n̂ − n̂•σ •(~u − ~uI ) + qn ] = 0 (1.58)

where ~u and ~uI are the flow velocity and the velocity of the discontinuity and qn is the

heat flux crossing the interface. Denoting the values immediately ahead of and behind the

discontinuity as (R) and (L), the jump relation for any quantity Ψ, is given as

[Ψ] = ΨR − ΨL (1.59)

When [Ψ] = 0, the quantity Ψ is conserved (continuous), otherwise the quantity is discon-

tinuous. It is desirable to populate the ghost points by solving the mass, momentum and

energy jump relations across the embedded material interface. As noted in [58], due to
 27

the linear degeneracy associated with material interface, the solution for the jump relations

(Eqs 1.56 - 1.58) is not straightforward. In fact for a material interface the jump relations

are trivially satisfied [58]. However, depending on the type and strength of the materi-

als enclosed/separated by the interface, the jump relations can be simplified such that the

resulting set of equations can be solved to define the ghost states.

1.6.2.1 Conditions Enforced at the Interface Separating Fluid Media

By definition, fluids are materials that offer negligible resistance to external shear or

torsional loading. Neglecting the material-strength associated with shear/torsional loading,

the conservative variables and fluxes in the governing equations (Eq 1.13) reduce to
       
 ρ   ρu   ρv   ρw 
       
       
 ρu   ρu2 + p   ρuv   ρuw 
       
       
       
~ =
U ~  ~ = ~ =
,F = 
ρv  ρuv ,G ρv 2 + p ,H ρvw .
     
       
       
ρw2 + p
       
 ρw   ρuw   ρvw   
       
       
ρE u (ρE + p) v (ρE + p) w (ρE + p)
In the above equations, the terms corresponding to the constitutive equations for modelling

the deviatoric response of the material are neglected. However, the dilatational response

of the material is retained and is governed by an E.O.S. of the form given in Eq 1.27.

Simplifying Eq 1.27, we get

P = ρe(γ − 1) − γP∞ , (1.60)

where γ = Γ + 1 is the specific heat ratio and P∞ is a material dependent constant [43]. For
cp
the case of ideal gas we have γ = cv
and P∞ = 0. For fluids like water, the Grüneisen expo-

nent and the material dependent constant take the values of 4.5 and 0.613 GPa respectively.
 28

From the definition of sound speed,

s
γ(P + P∞ )
c= . (1.61)
ρ

For embedded fluid objects such as droplets, bubbles etc. immersed in strongly shocked

flows, it has been well established that the original approach due to Fedkiw et al [58] for

the definition of ghost states failed to maintain a non-oscillatory pressure field, particularly

when strong shock and detonation waves reside in close proximity to the interface [118].

During the few time steps when the shock and the interface are coincident, the interface

treatment technique should temporarily reduce to a shock tracking method. Therefore,

in this regard, the Rankine-Hugoniot (R-H) jump conditions must be solved across the

interface to define the ghost states. Thus, with the modified governing equations and the

E.O.S., the jump relations (Eqs 1.56 - 1.58) are simplified to obtain:

[ρũ] = 0 ⇒ ρL ũL = ρR ũR (1.62)

 2 
ρũ + P = 0 ⇒ ρL ũ2L + PL = ρR ũ2R + PR (1.63)

[ũ (ρE + P )] = 0 ⇒ ũL (ρL EL + PL ) = ũR (ρR ER + PR ) (1.64)

ũL|R = ~uL|R •n̂ − Us (1.65)

The R-H jump equations given in Eqs 1.62 - 1.64, are cast in a reference frame attached

to the material discontinuity, and the jump relations are solved assuming that the shock

and the material discontinuity are coincident. The nature of the reflected and the refracted

waves could be shock waves or rarefaction waves or both, and can be determined from the

solution of the R-H jump conditions at the interface. The existence and the uniqueness

of the solution for the set of (hydrodynamic) jump relations are well established [189].
 29

Exact and approximate solutions (Riemann solvers) for the system of algebraic equations

are available and have been well documented [189]. The ghost states defined with the

solution obtained from such Riemann solvers satisfy the jumps in mass, momentum and

energy conservation laws and also respect the nature of the wave interactions occurring at

the material interface. The approach developed to construct the Riemann problem at the

interface is discussed in detail in chapter 3 of this thesis.

1.6.2.2 Conditions Enforced at the Interface Separating Elasto-Plastic Material from other

media

For materials with moderate to large material-strength, such as elasto-plastic solids,

the governing equations (Eq 1.13) along with the constitutive relations are solved in its

entirety. Hence, unlike the case of embedded fluid objects, a Riemann solver based solution

is not readily available for the jump relations (Eqs 1.56 - 1.58). Obtaining an exact solution

for the system of equations (Eqs 1.56 - 1.58) is far fetched particularly for the complex

E.O.S such as the Mie-Grüneisen E.O.S. employed in the present work.

Alternatively, since the high speed multi-material interaction phenomena are dic-

tated by the propagation of nonlinear waves and their interaction with the material interface,

the ghost states can be defined such that they are in line with the wave-interface interaction

phenomena. For instance, a compressive wave incident on a free surface is reflected as a

tensile wave and vice-versa. In the hydrodynamic case, the wave interaction phenomena

are resolved by solving the R-H jump conditions at the interface. For elasto-plastic mate-

rial, the wave-reflection and refraction phenomena are captured by enforcing one or more
 30

of the following conditions at the material interface:

• Continuity of normal velocity component:

[~u•n̂] = 0 (1.66)

• Discontinuity in density and tangential velocity components:

[ρ] 6= 0 (1.67)
 
~u•t̂1 6= 0 (1.68)
 
~u•t̂2 6= 0 (1.69)

• Discontinuity in tangential stress components:

[σ̃t1 t1 ] 6= 0 (1.70)

[σ̃t1 t2 ] 6= 0 (1.71)

[σ̃t2 t2 ] 6= 0 (1.72)

• Traction-based condition for normal stress components:

1. For material-void interface i.e. for interfaces that separate elasto-plastic mate-

rials from voids (free-surfaces), zero-traction condition (or alternatively wave

reflection condition) is enforced at the interface:

σ̃nn = 0 (1.73)

σ̃nt1 = 0 (1.74)

σ̃nt2 = 0 (1.75)

[P ] 6= 0 (1.76)
 31

2. For (elasto-plastic) material - (elasto-plastic) material interface, continuity of

normal stress is enforced:

[σ̃nn ] = 0 (1.77)

[σ̃nt1 ] = 0 (1.78)

[σ̃nt2 ] = 0 (1.79)

[P ] = 0 (1.80)

3. For FSI coupling i.e. for interfaces separating elasto-plastic solids from fluids,

continuity of normal stress is enforced in the following manner:

EP
σ̃nn + P EP = P L (1.81)

EP
σ̃nt1
= 0 (1.82)

EP
σ̃nt2
= 0 (1.83)

[P ] 6= 0 (1.84)

where the subscripts EP and L denote the elasto-plastic and the liquid material

respectively.

4. For interfaces separating elasto-plastic material and rigid solids, the discontinu-

ity in normal stress components are enforced:

[σ̃nn ] 6= 0 (1.85)

[σ̃nt1 ] 6= 0 (1.86)

[σ̃nt2 ] 6= 0 (1.87)

[P ] 6= 0 (1.88)
 32

In addition to the above conditions, the normal velocity at the interface is mod-

ified as

~u•n̂ = Un (1.89)

where Un is the normal velocity of the rigid solid object.

The total stress tensor in the normal and tangential coordinates (σ̃) is given by

σ̃ = JσJT (1.90)

where  
 nx ny nz 
 
 
J= t t t
 1x 1y 1z 
 (1.91)
 
 
t2x t2y t2z

is the Jacobian matrix, and n̂, t̂1 & t̂2 are the local normal and tangential vectors defined

at the interface. The ghost points are populated with flow properties such that the condi-

tions hold true exactly at the embedded interface. The approach developed to enforce the

aforementioned conditions at the interface is discussed in detail in chapter 4 of this thesis.

1.6.2.3 Conditions Enforced at the Interface Separating Rigid Objects

Solids with infinite material-strength, that register negligible deformation under in-

tense loading conditions, are classified as rigid objects. There are significant differences

between the treatment of fluid/elasto-plastic materials and rigid solid objects in the ghost

fluid framework. For instance, in the case of elasto-plastic materials the flow variables such

as velocity and pressure that are continuous across the interface are directly injected into
 33

the ghost field. In the latter case, due to the absence of flow fields on one side of the in-

terface (the solid side), it is not immediately clear how to populate the ghost cells with the

virtual flow properties. At the interface of a solid body, the following boundary conditions

are to be applied for velocity, temperature and pressure fields

i. No-penetration condition for the normal velocity :

vn = Un , (1.92)

where Un is the velocity of the center of mass.

ii. Slip condition for the tangential velocity:

∂vt1
= 0, (1.93)
∂n
∂vt2
= 0, (1.94)
∂n

iii. Adiabatic condition for temperature :

∂T
= 0, (1.95)
∂n

iv. Normal force balance on pressure :

∂p ρs vt1 2
= − ρs an , (1.96)
∂n R

where,

vn = V~ • n̂, (1.97)

vt1 = V~ • tˆ1 , (1.98)

 34

vt2 = V~ • tˆ2 (1.99)

are the normal and (contravariant) tangential velocities in the body fitted curvilinear coor-

dinate, V~ is the velocity vector in the global Cartesian coordinate, n̂, tˆ1 , tˆ2 are the normal

and tangential vectors, R is the radius of curvature and an is the acceleration of the inter-

face. The set of boundary conditions given in Eqs (1.92) - (1.96) govern the behavior of

the flow near the embedded solid body and must be enforced on the real fluid by suitably

populating the corresponding ghost cells. There are several ways in which this task can be

accomplished which is the subject of discussion in chapter 5 of this thesis.

1.7 Numerical Examples

The illustrative numerical examples presented in this work were simulated by solv-

ing the governing equations using a third-order TVD-based Runge-Kutta scheme for time

integration and third-order convex ENO scheme [120] for spatial discretization. Since the

numerical schemes implemented in this work are well established, the implementation de-

tails are not presented here. Interested readers may refer to the original articles [120, 163]

for details. The purely inviscid Euler equations were non-dimensionalized based on free
q
stream density ρ∞ , pressure p∞ and the sonic velocity a∞ = ρp∞ ∞
. Unless stated to the

contrary, the standard atmospheric conditions are chosen to non-dimensionalize the gov-

erning equations. A suitable length scale is chosen depending on the dimensions of the

immersed object. The CFL number was set to 0.6 for all of the simulations presented in

this work unless stated to the contrary. Numerical Schlieren images shown in this work
 35

were generated using the method outlined in [147]:

k∇ρk
υ = e−κ k∇ρkmax (1.100)

where κ is a problem-dependent coloring constant and υ is the shade function.

 36

W J
Material ρ0 ν c m−K
K kg−K
Γ0 c0 ms S

Copper 8930 0.35 383.5 401 2.0 3940 1.49

Tungsten heavy alloy 17600 0.29 477 38 1.43 4030 1.24

High-hard steel 7850 0.30 134 75 1.16 4570 1.49

Aluminum 2700 0.30 896 166.9 1.99 5386 1.339

HMX 1891 - 1000 0.4 1.1 2058 2.38

Polymer Binder 1100 - 1.00 5.1 1.55 2350 1.7

Table 1.1: Parameters for the Mie-Grüneisen E.O.S. for commonly used materials
 37

Material Y0 (GP a) B (GP a) N C m G (GP a) Tm (K)

Copper 0.4 0.177 1.0 0.025 1.09 43.33 1358

Tungsten 1.51 0.177 0.12 0.016 1.0 124.0 1777

High-hard steel 1.50 0.569 0.22 0.003 1.17 77.3 1723

Aluminum 0.324 0.114 0.42 0.002 1.34 26.0 925.0

HMX 0.26 NA NA NA NA NA NA

Polymer Binder 89.0 NA NA NA NA NA NA

Table 1.2: Johnson-Cook material model parameters: A = Y0 , T0 = 294K and ǭ˙P0 = 1.0s−1
 38

Figure 1.1: Spark-generated bubbles for modeling underwater explosion bubble dynamics.
This allows clear observation of re-entrant jet formation inside the bubble, because the
explosive detonation products generally occlude the view of the interior of the bubble;
Courtesy - Georges L. CHAHINE, Dynaflow, Fulton, MD. Proc. 66th Shock and Vibration
Symp., Biloxi, MS (1995). SAVIAC, Arlington, VA (1995), vol. 2, p. 265
 39

Figure 1.2: Snapshots of deformation of a bubble subject to a planar shock wave.

 40

Figure 1.3: Superposition of three flash radiographs showing a sequence of a 30-06 bullet
in air penetrating a so-called Whipple shield - a configuration invented in 1946 by the as-
tronomer Fred L. WHIPPLE. It is based on the principle that small meteoroids and orbital
debris explode when they strike a solid surface. Courtesy - John P. BARBOUR and asso-
ciates at Field Emission Corporation in McMinnville, OR. See Proc. 7th Int. Congr. on
High-Speed Photography, Zurich (1965). Helwich, Darmstadt (1967), p. 292

Figure 1.4: Crater formation details: Aerial view of Meteor Crater which is 1.2 km in
diameter and 200 m deep, officially called ”Barringer Crater” that is about 50,000 years
old. Photo by David J. RODDY and Karl ZELLER, courtesy USGS
 41

Figure 1.5: Snapshot of shock condensation effect around the jet fighter Boeing F-18 (Hor-
net) at transonic flight in humid air. It arises during acceleration when the air flow at some
parts of the fuselage reaches supersonic speeds. When the resulting shock wave detaches,
it builds up a sudden rarefaction that, lowering the temperature, causes condensation of the
ambient water vapor.[Courtesy: J. GAY, U.S. Navy] A similar picture was recently taken
from an F-4 (Phantom II) during an air show at Point Magu Naval Air Station, CA. (The
MilitaryAircraft Archive, http://www.milair.simplenet.com/)

Figure 1.6: Snapshot of the bow shock wave formed due to the supersonic flight of the
NASAs new Ares -1X rocket
 42

∗
Figure 1.7: Radial Return Algorithm: The predicted stress σn+1 lies outside the yield sur-
face. The predicted stress is brought back to the yield surface to obtain the final corrected
stress σn+1 , using a return mapping algorithm, along a direction normal to the yield surface
∂f
∂σij
.

Figure 1.8: Classification of grid points - bulk points, interfacial ghost points and interior
ghost points.
 43

Figure 1.9: The GFM approach converts a two-material problem to two single-material
problem

Figure 1.10: The basic mechanism of the Ghost Fluid Method.

 44

Figure 1.11: Discontinuity Ω0 (shock or material discontinuity) traveling inside a material

bounded by ∂.
 45

CHAPTER 2
LOCAL MESH REFINEMENT

2.1 Introduction

Grid adaptation has gained renewed interest in the last decade since the pioneer-

ing work by Berger and co-workers [26, 27]. Adaptivity requires automatic and optimized

grid point placement in the domain of interest such that the resulting grid topology sig-

nificantly reduces computational time and effort. There are several approaches that have

been used to accomplish grid adaptation both in compressible [19, 26, 82, 136, 146, 196]

and incompressible flows [72, 108, 121, 122, 178]. All of these methods have the same un-

derlying principle of operation. In general, the process of grid adaptation can be described

as follows: A fixed base mesh that is fine enough to capture the global flow features is

employed as a starting point. Then a suitable refinement criteria (usually first or second

gradient gradient-based [108, 191] or in some instances, wavelet-based [32]) is invoked to

identify flow structures (such as shock waves, strong rarefaction waves, vorticity, etc.) and

embedded objects that are under-resolved. Once such regions are identified, either indi-

vidual cells encompassing the regions (LMR [108, 191]) or patches of rectangular mesh

(AMR [26, 136, 146]) are recursively subdivided until the desired level of refinement is

achieved. In essence, the grid adaptation technique judiciously identifies and quarantines

fine mesh cells to those regions with active structures and enables the use of coarse grid

elsewhere. This results in dramatic savings in memory and processor time.

In this chapter, the GFM-based approach is augmented with the quadtree- (octree-
 46

in three dimensions) based Local Mesh Refinement (LMR) technique developed in [191],

for computationally efficient calculations involving strong shock interactions with embed-

ded solid and fluid objects(s) in high speed compressible multiphase flows. It was demon-

strated by Berger et al [26], and later by Jameson [93], that a combination of a higher-order

scheme and adaptive grid refinement technique is the best strategy, if a solution of rea-

sonable accuracy with least computational overhead is sought [146]. The tree-based mesh

adaptation schemes for compressible flows developed in the past used lower-order linear

reconstruction technique such as the minimum energy [45] or linear Green-Gauss [45,207]

flux reconstruction technique. In this work, the higher-order aspect for temporal and spatial

resolutions are retained by means of a third order TVD-based Runge-Kutta scheme [163]

and third order convex ENO scheme [120] respectively. Higher-order schemes such as

ENO use wider support cells that demand a conservative approach to compute numeri-

cal fluxes at mesh interfaces. Application of third and higher-order ENO/WENO schemes

have been proven computationally effective only for uniform and smoothly varying curvi-

linear meshes [159]. Particularly, limited progress has been made in this direction to couple

higher-order schemes with tree-based grid adaptive schemes. Recently, Cecil et al [36] have

developed generalized ENO schemes on octree-based unstructured non uniform meshes.

However, the applications reported in their work are limited to one-dimensional weak

shocks traversing in single fluid medium. The current work focuses on extending such

an approach for strong shock applications and to shock-interface interactions. In essence,

the following efforts are reported in this chapter:

i. A simple, conservative formulation for accurate computation of ENO-based numer-

 47

ical fluxes at the fine-coarse mesh boundaries in tree-based LMR schemes. The

methodology adopted in this work provides a general framework to uniformly in-

tegrate the solution throughout the computational domain. Thus, the GFM approach

and other interface treatment techniques can be directly employed in the current

framerwork.

ii. A refinement criteria is developed for identifying and tagging cells for refinement.

Obtaining suitable criteria that simultaneously track strong and weak shocks, rar-

efaction waves, sliplines and contact discontinuities (material interfaces) is a tricky

proposition. The criteria proposed in this work is simple and effective in identifying

under-resolved regions.

The broad range of results presented in this work demonstrates the flexibility and robustness

of the current method in efficiently resolving dominant features and disparate length scales

for a wide spectrum of Mach numbers. Shocks interacting with multiple moving boundaries

are also computed.

2.2 Automatic Grid Adaptation Schemes

Among the various mesh enrichment techniques that have been used in the past [26,

82, 121, 196], Adaptive Mesh Refinement (AMR) schemes due to Berger et al [26, 27] have

been extensively used for compressible [26, 136, 138, 145], incompressible [72, 126, 151,

178] and astrophysical [30, 41, 104] flow computations. AMR generates blocks of struc-

tured, non-overlapping, Cartesian mesh patches over the underlying coarse mesh, and the

flow field is evolved on all mesh patches. The two main operations involved in the AMR
 48

technique are the prolongation and restriction operations [30]. The flow field correspond-

ing to the newly formed mesh at the finer levels are interpolated from the corresponding

mesh at coarser levels based on the flow solution from the preceding time step (prolonga-

tion operation). Thus in the prolongation operation, information propagates from coarse

mesh to fine mesh patches. On the other hand, the evolution of the flow field follows from

the finest to the coarsest level (restriction operation); i.e. the flow field is first evolved on

the fine mesh patch, and this solution from the fine mesh patch is projected on the coarse

mesh patch before computing the flow field on the coarse mesh patch. Thus the AMR

scheme requires repeated interpolation back and forth between different grid levels at each

time step [191]. As rectangular blocks of structured mesh are formed at every level, the

implementation of AMR scheme is less complicated and more straightforward than other

methods, and demands no major changes in the data structure. As the flow field is evolved

on all grid levels, different time steps could be used for grids at different levels. Hence the

smallest time steps are local to the finest levels and is not enforced for the grids at coarser

levels.

Although the AMR scheme is attractive and has been proven to be effective for a

wide gamut of problems, their applicability for sharp interface calculations involving fluids

with high impedance mismatch is not straightforward. The problem lies with the restric-

tion operation which requires interpolation from fine mesh to the coarse mesh patch, and a

conservative interpolation at the interface results in mixing of fields and numerical fluxes

across the interface in a manner that is detrimental to the overall numerical computation.

Nourgaliev et al [136], have identified and demonstrated this potential pitfall of the AMR
 49

approach for interfaces separating fluids of high impedance mismatch. To extend its ap-

plicability to sharp interface multiphase flow computations, the AMR approach described

in [136] uses a non-conservative Natural Neighbor Interpolation (NNI) [29, 173] method.

The NNI approach carefully avoids interpolation across the interface of the embedded ob-

ject, but results in complications in the implementation that detract from the inherently

simple AMR scheme. Alternatively, one could use a generic grid adaptation technique

such as the hierarchical tree-based Local Mesh Refinement (LMR) [191] scheme, that does

not require the evolution of the flow field on all grid levels. This latter approach is the one

elaborated in the present work.

In contrast to the AMR approach, the LMR scheme sub-divides each cell that is

tagged for refinement to form four (quadtree in two dimension) or eight (octree in three

dimensions) child cells resulting in highly unstructured mesh rather than blocks of rectan-

gular grid patches. Since each cell is created and destroyed individually, the LMR scheme

does not require constant re-meshing and update of the global mesh. As the mesh is now

unstructured, the hierarchical data structure associated with LMR scheme contains neigh-

bor and parent-child connectivity information stored in the cell structure. With hierarchical

data structure the grid refinement and coarsening operations are trivial to accomplish. Fur-

thermore, as the LMR scheme does not require optimized rectangular patches of mesh,

fewer mesh points are used in the computation resulting in significant savings in computa-

tional memory and on a Cartesian mesh, features that are misaligned with the mesh can be

captured by mesh refinement tangent to the feature. Unlike the AMR approach, the flow

field is evolved only on the finest (undivided) cells (termed leaf cells in LMR terminology).
 50

Thus the solution for every time step is achieved in a single sweep of solution step making

the LMR scheme more attractive than its counterpart. Since the flow field is evolved only

on the leafcells, no special treatment is required for points near the embedded interface and

the numerical scheme can be uniformly integrated throughout the computational domain.

2.3 Local Mesh Refinement (LMR)

As stated in the previous section, the octree-based mesh refinement schemes are

more general but come with an added price of complexity associated with the data structure,

and pose significant challenges for algorithmic implementation [121, 122]. However the

implementational intricacies associated with octree and quadtree-based LMR schemes have

been clearly identified and outlined by Greaves et al [68–71]. The application of octree-

/quadtree-based LMR schemes for compressible flows dates back to the work of Zeeuw et

al [205–207]. Although LMR-based adaptive schemes have found limited applications for

compressible flow computations [36, 44, 45, 50], one may find a multitude of applications

pertaining to incompressible flow simulations [7,108,121,122,128,144,169–172,198]. For

compressible flow simulations, AMR schemes have been preferred over LMR approach

despite of the potential savings in computational resources. Although the reason for this

trend is not apparent, it was pointed out in [121, 122] that spurious and artificial reflected

waves were generated with the use of LMR schemes, when waves of large magnitude were

allowed to pass through the fine-coarse mesh boundaries [41]. One possible reason for the

generation of such spurious waves could be the fact that the low frequency errors on a given

fine grid becomes high frequency errors when passed on to coarse grids [207]. As will be
 51

evident from the current work, LMR schemes can be employed effectively by judiciously

enforcing refinement criteria that prohibit waves of large magnitude from passing the fine-

coarse mesh boundaries. The impact of the wave passage through the fine-coarse mesh

interfaces have been carefully examined and further discussion on this topic is deferred to

section 2.4.1.1

2.3.1 LMR Terminologies and Hierarchical Data Structure

Since computational cells are created and destroyed locally and individually, im-

plementation of LMR scheme requires compatible data structures for easy cell addition,

cell removal, access of neighbor cells and mesh traversal [191]. To facilitate such oper-

ations, the most obvious choice is to use a hierarchical (quadtree in two dimensions and

octree in three dimensions) data structure that stores cell-connectivity and cell-hierarchy

information. When a cell is marked for refinement, it is divided to form four cells in two

dimensions or eight cells in three dimensions. The four (eight) new cells formed during the

refinement operation are called child cells and the cell which is refined to form these child

cells is called the parent cell. The undivided cells at the bottom of the hierarchy/tree level

(that has no children) are called the leafcells. It is in these undivided leaf cells in which the

flow calculations are evolved. A parent cell can also be coarsened by aggregating its child

cells and pruning their branches.

Whenever a cell is coarsened, the child cells are destroyed and the parent cell be-

comes an undivided leaf cell. Typical representation of the tree structure is displayed in

Figure 2.1(a). The data structure used to describe the cell is displayed in Figure 2.1(b).
 52

As shown in Figure 2.1(b), in addition to the the regular cell attributes pertaining to the

flow variables, the data structure for individual cell stores pointer information pointing to

its parent cell, child cells and to its neighbors. The pointer information for each cell can

be updated independently, every time a cell is refined or coarsened. Thus, with the use of

a hierarchical data structure, each time a cell is refined or coarsened no global re-meshing

is required and the grid adaptation operations are local. Since the flow field is evolved

only on the undivided cells, the attributes corresponding to the flow variables need to be

stored only on the leaf cells resulting in savings in memory. The hierarchical data struc-

ture makes it feasible for easy traversal through the tree to access the parent and child cell

information. For ease of computation and to obtain balanced tree distribution, the refined

grid is smoothed to ensure that no two neighboring cells have more than one grid-level

difference. As pointed out in [167], balancing enables easy navigation through the tree

structure to efficiently access neighbor cell information, besides providing equal load dis-

tribution [36] and a gradually varying mesh distribution. For additional details pertaining

to the implementation of the tree-based LMR scheme in the current context, the reader may

refer to [192].

2.3.2 Coarsening and Refinement Operations

Cells are refined and coarsened based on the presence of structures and dominant

features in the flow field. As the refinement process is solution adaptive, a base mesh

that is fine enough to capture the global flow structure is selected, and the initial level set

field is constructed on this base mesh. Once the base mesh is defined, all those cells that
 53

exceed a preset threshold limit of the refinement criteria are tagged, and the refinement

operation is carried out on these cells until the desired level of refinement is achieved.

The computational cells straddling the interface of the embedded object(s), that lie within

the narrow band of the level set field are refined to the maximum level. This eliminates

interpolation errors and associated instabilities from being introduced at the interface [167].

Thus initial discontinuities (such as shock and detonation waves) and the cells in the narrow

band of the level set field are uniformly refined to the finest level, and are maintained at the

finest level throughout the computation. Other flow structures that develop with evolution

of the flow field are refined by repeated application of the refinement criteria at regular

intervals in time (the interval at which LMR is invoked is determined by the fastest traveling

(characteristic) wave in the domain (CFL condition)). In addition, adequate number of

buffer cells are provided in refinement zones around strong shock and rarefaction waves.

As a band of fine (buffer) cells always surround these flow structures, any new structure

formed with the evolution of the flow field will remain within these band of fine cells. This

further ensures that interpolations are not performed near discontinuities in the flow field.

Since each time a cell is divided four (eight) new child cells are formed, a constant

refinement factor of 2 is maintained i.e. a cell which is recursively refined k times (for

k > 1) forms 2k child cells. Therefore, a (child) cell at k th level formed by k − 1 recursive

1
divisions of a parent cell will have sides of length ∆x = 2k−1
∆xg , where ∆xg is the size

of the base mesh with level 1 denoting the global base mesh (Figure 2.1(a)).

Divided cells that are not tagged for refinement are coarsened. Whenever a cell is

coarsened, the flow properties for the parent cell are determined from aggregating the flow
 54

properties from its child cells.

NX
child
~ = 1 ~ ich
U U (2.1)
Nchild ich=1

~ corresponds to the primitive variables and Nchild = 4 for quadtree and Nchild = 8
where U

for octree data structures. The flow attributes for the newly formed child cells are de-

rived from interpolating on the leafcells in the preceding time step. One of the most com-

monly used approach for child cell interpolation in both LMR and AMR framework is the

MUSCL-based interpolation scheme [145, 146]. Cecil et al [36] have used higher-order

ENO-based interpolations for defining the flow attributes for the newly created child cells.

The ENO-based interpolations for child cells were carried out by choosing stencils along

the diagonal and grid directions. In this work, a simple conservative bilinear interpolation

due to Berger et al [26] is used to define the flow attributes for the newly created child cells.

Noting that the child cells are created and destroyed only at the edges of the buffer cells

that surrounds the fine mesh cells encompassing dominant flow features, a simple bilinear

interpolation scheme appears to be an obvious choice.

2.3.3 Refinement Criteria

As mentioned before, the refinement criteria identify and tag cells in the regions

with distinct structures and steep gradients in the flow field. The refinement criteria is

critical for tree-based mesh adaptive schemes particularly when applied to compressible

flow calculations, as interpolations between grid levels near discontinuities are not permis-

sible. Such interpolations can be averted if the fine-coarse mesh interface is sufficiently

far away from discontinuities and the flow field varies smoothly across the mesh interface.
 55

Berger et al [26,27] employed Richardson extrapolation based estimate of the local trunca-

tion error of the solution, to identify regions with discontinuities. A similar approach was

used in [104] for astrophysical flow computations. In [45, 198, 207] a statistical description

based on the cell-size-weighted curl and divergence of the velocity field were employed

to identify regions that are excessively compressed and sheared. Flow field gradient-based

(density, velocity and speed of sound) estimates were used in [211] for refining and coars-

ening grid patches. Recently, Karni et al [103] developed a smoothness indicator (SI)

based on weak local truncation error estimate to identify roughness and discontinuities in

the solution. The approach, appears promising, may not be computationally efficient as

wider support stencil and time history (up to three consecutive time levels) are required to

compute the SI.

An ideal refinement criteria must not only tag cells that straddle the discontinuity,

but also tag cells in the vicinity of the discontinuity. This is required to ensure that enough

(buffer) cells are provided to confine discontinuities within the fine grid For any differen-

tiable, Lipschitz continuous function f (x), there exists a unique derivative at the point xi ;

i.e
df (x) df (x) df (x)
lim− = |xi = lim+ (2.2)
x→xi dx dx x→xi dx

In such regions where the function is continuously differentiable, a second order central-

difference estimate will converge with the corresponding second-order one-sided estimates

for the first derivative. On the contrary, for those regions where the function f (x) is dis-

continuous, the central difference scheme based estimate will significantly deviate from

the corresponding one sided estimates of the first derivative. Thus cells enclosing smooth
 56

regions can be effectively segregated from those cells straddling discontinuities by compar-

ing the first derivative estimates. In essence the following steps are performed to identify

potential cells for refinement:

• At each cell center, second order one sided and second order central difference based

derivatives are evaluated as follows:

df (x) − f (xi−2 ) − 4f (xi−1 ) + 3f (xi ) 1 2 d3 f (x)

= + ∆x ; ∀x ≤ xi (2.3)
dx 2∆x 3 dx3
df (x) + −3f (xi ) + 4f (xi+1 ) − f (xi+2 ) 1 2 d3 f (x)
= + ∆x ; ∀x ≥ xi (2.4)
dx 2∆x 3 dx3
df (x) f (xi+1 ) − f (xi−1 ) 1 2 d3 f (x)
= − ∆x (2.5)
dx 2∆x 6 dx3

• Since in smooth regions, the one sided estimates approach the central difference

scheme based value, the one sided derivatives are compared with the central differ-

ence scheme based estimate to obtain

df (x) df (x) −
ξ1 = | − | (2.6)
dx dx
df (x) df (x) +
ξ2 = | − | (2.7)
dx dx
df (x) − df (x) df (x) +
ξ3 = M AX(| |, | |, | |) (2.8)
dx dx dx

where ξ1 , ξ2 & ξ3 are the estimates of error.

df (x) − df (x) df (x) +

• As the derivatives dx
, dx & dx
are comparable in smooth regions, the error

estimates ξ1 & ξ2 are of O(∆x2 ). For those regions with discontinuities, the error

estimates max(ξ1 , ξ2 ) ≥ δ1 or ξ3 ≥ δ2 , where δ1 > 0 & δ2 > 0 are user-defined

tolerance limits. The criterion δ1 is primarily used to identify dominant flow struc-

tures such as shocks and contact discontinuities, while δ2 is used for resolving weak
 57

shocks and other weak structures such as expansion waves and sliplines, that may

evolve in the flow field. Since shocks and contact discontinuities are O(1) disconti-

nuities, the threshold for the criterion δ1 ≈ O(∆x). As will be shown in the results

section, based on the numerical experiments conducted in this work, δ1 is set in the

range δ1 ∆xg ≈ (∆xg − 1.0) where ∆xg is the size of the base mesh. The value of

the δ2 depends on the features and gradients that the user desires to resolve.

The above procedure is repeated along all grid directions. Since the refinement

criteria for a point xi uses points {xi−2 , xi−1 , xi , xi+i , xi+2 } in the neighborhood of xi , the

procedure is highly effective in flagging cells that fall in the vicinity of the discontinuity.

Furthermore, in addition to the above refinement criteria, three layers of buffer cells are

provided at the fine-coarse mesh boundary to ensure that the discontinuities are confined

well within the fine mesh level. Thus, in contrast to the refinement criteria developed

in [26, 27], the criterion adopted in this work neither requires data storage at different time

levels nor involves complex interpolation at different grid patches/levels. The function f

chosen in the current work correspond to the density, pressure and entropy fields that reside

at the centers of the computational cells.

2.3.4 Computing ENO Based Numerical Flux at the Mesh Interface

One of the crucial aspects of this work concerns the extension of the higher-order

ENO-based flux discretization schemes developed for uniform meshes, to the fine-coarse

mesh boundary of the unstructured, adaptive, tree-based mesh refinement schemes. Several

unstructured mesh-based ENO schemes have been developed in the past [4, 34] but such
 58

schemes vary in complexity depending on the grid structure [36]. Cecil et al [36] have

developed an efficient scheme to construct ENO-based fluxes by identifying ghost points at

the mesh interface. The ghost points in turn derived their flow properties from higher-order

ENO-based interpolation. The success of this approach is largely dependent on a stringent

restriction imposed on the the grid level difference between neighbor cells. In their ap-

proach, the neighbor cells and the neighbor of neighbor cells, both in the grid and in the

diagonal directions, were prohibited from having more than one level difference. One may

have to extend the restriction to many cells for higher-order schemes. In contrast, in the

current approach only the neighbor cells along the grid directions are not allowed to have

more than one level difference in grid level, and thereby provides easy extension to any

desirable order of accuracy. The approach adopted in this work also uses ghost points to

construct fluxes at the mesh interface. As mentioned before, since the fine-coarse mesh

transitions are designed to lie away from significant gradients in the flow and occupy a

lower-dimensional space in the computational domain, a (conservative) bilinear interpola-

tion scheme is found to be adequate to obtain the values at these ghost points. Moreover,

with bilinear interpolation, extension to three-dimensions is straightforward. Details per-

taining to the ENO-based flux construction approach for a uniform mesh distribution are

discussed in [190, 192]; the generic procedure for constructing the ENO-based numerical

fluxes at the fine-coarse mesh interface is outlined below.

 59

2.3.4.1 Flux Construction for a Cell that is finer than its Neighbor

Consider the typical case of the coarse-fine mesh boundary as shown in Figure 2.2(a).

In Figure 2.2(a), cell P is at a level higher than its immediate neighbor cells W and N. In

this section, the flux construction approach for the west face of cell P is illustrated:

i. The ghost points required for completing the ENO stencils are identified. The ghost

points can be identified by simply comparing the status (divided or undivided) and re-

finement level of the neighboring cells. For cell P, three ghost points (GW1 , GW2 , GW3 )

on the west side, and two ghost points (GE2 , GE3 ) on the east side are required to

complete the support stencil.

ii. Once the ghost points are identified, the next step is to derive the fluxes and the flow

properties for these ghost points. As mentioned before, several interpolation schemes

such as the ENO-based [36] or MUSCL-based [145, 146] schemes have been used

in the past. Berger et al [26] employed a conservative bilinear interpolation proce-

dure to determine the fluxes and the flow field at the ghost points. Quirk [145, 146]

used a MUSCL-based interpolation procedure to define the ghost points. Losasso et

al [122] used a direct injection method. In their approach, the flow properties and

fluxes from the nearest undivided leafcell were directly injected to the ghost points.

Recently, Choi et al [41] proposed a quadratic ghost point filling technique. Rec-

ognizing that a linear interpolation procedure at the ghost points results in spurious

reflected waves, a quadratic interpolation procedure along with explicit gradient con-

tinuity were enforced at the mesh interface for determining the flow properties at the

ghost points. The method was shown to be effective in suppressing the spurious os-
 60

cillations arising at the mesh interface, but was found to be complicated for two- and

three-dimensional applications. Cecil et al [36] modified their linear interpolation

based ghost point filling approach [35] with an efficient and higher-order ENO based

interpolation procedure.

In this work, a simple conservative bilinear interpolation approach is adopted, as

the bilinear interpolation procedure is used to determine the flow properties for the

newly formed child and parent cells. Moreover, the refinement criteria developed

in this work ensures that the flow field spanning the fine-coarse boundary is smooth

and that the discontinuities are contained well within the fine mesh cells. Hence a

conservative, bilinear interpolation approach, is the most straightforward choice.

In order to construct the attributes for the ghost point GW 1 , a non-conservative in-

terpolation technique can either use points P, W, NW and N or points P, PS, W and

NW for interpolation as shown in Figure 2.2(a). Here the choice for interpolation

points is not unique. On the other hand, a conservative interpolation scheme uses a

unique set of points for interpolation such as points W, NW, PI and N (Figure 2.2(a)).

The point PI corresponds to the parent of cell P, and the flow attributes for point PI

can be obtained by aggregating from its child cells i.e from cells P, PE, PS and PSE.

Essentially, a conservative interpolation scheme uses points on the same level for in-

terpolation. To clarify this procedure, a complex scenario involving cells at different

levels of refinement is presented in Figure 2.2(b). As shown in Figure 2.2(b), the

fine-coarse grid interface involves grids with more than one level difference (along

the diagonal direction). In such cases, points NWI, NI, PI and WI on the same level
 61

are used for interpolation. Figure 2.2(b) shows how the flow properties and fluxes are

aggregated from child cells to determine the flow properties at point NI. A similar

procedure can be adopted to determine the flow properties and fluxes at points PI and

WI. Identification of points of interpolation on the same grid level is straightforward.

The procedure described above is applied repeatedly to define the flow attributes and

fluxes for the remaining ghost points.

iii. Once the attributes for the ghost points are determined, the construction of flux fpW

follows from the uniform mesh approach. Thus the method outlined in this section

does not deviate much from the uniform mesh-based flux construction approach, and

retains the simplicity of the ENO scheme flux construction for uniform meshes. The

approach for constructing fluxes for other faces of cell P follow directly the procedure

described above.

2.3.4.2 Flux Construction at the Fine-Coarse Interface for Cells Abutting Divided Neigh-

bors

If the cell under consideration is at a level lower than its immediate neighbor, the

numerical flux corresponding to that face is not explicitly computed. With reference to

Figure 2.2(b), cell W that abuts the fine mesh cells (P and PS) is at a level lower than its

potential neighbors P and PS. Hence the flux crossing the mesh interface at the east face of

cell W is not explicitly computed. Instead, the west fluxes for cells P and PS are summed

to determine the net flux crossing the east face of cell W. Thus for the flux crossing the east
 62

face of cell W:

fE = fPW + fP S W

where fPW and fP S W correspond to the fluxes crossing the west face of cells P and PS

respectively. The fluxes fPW and fP S W are computed based on the procedure outlined

above. Thus, explicit conservation of numerical fluxes particularly at the interface of fine-

coarse boundaries is achieved.

2.3.4.3 Flux Construction Near the Embedded Interface

Since all cells in the narrow band of the level set field corresponding to the em-

bedded interface are refined uniformly to the same level, no special treatment is required

for constructing the fluxes near the interface of the embedded object(s). Therefore, the

Riemann solver based GFM technique developed for uniform mesh can be directly applied

near the interface of the embedded object(s).

2.4 Numerical Examples

As a representative measure of the computational memory usage, the percentage

Occupancy Ratio (OR) (Eq (2.9)) is computed. The OR is defined as the ratio of the max-

imum number of cells used in the solution adaptive LMR calculation with the number of

cells if a uniform fine mesh (corresponding to size of the finest mesh in the LMR calcula-

tion) was used.

M AX(N CELLSLM R (t))

OR = × 100 (2.9)
N CELLSU N IF ORM F IN E M ESH

For problems involving moderate to strong shocks interacting with gas-gas or gas-
 63

solid interfaces, the threshold for the refinement criteria were set in the range δ1 = 1.0 −

100.0 and δ2 = 1.0 − 100000.0 (arbitrarily large value). The threshhold limit for δ2 is set

to arbitrarily large value if the resultant shock diffraction pattern does not contain weak

structures and gradients that are under-resolved. For gas-water interface problems, due to

very large difference in the stiffness of the material strength (impedance mismatch ρa), the

threshhold limits were set in the range δ1 = 100.0 − 10000.0 and δ2 to an arbitrarily large

value.

2.4.1 One-Dimensional Example

For the one-dimensional test case considered in this section, a computational do-

main corresponding to a shock tube of unit length and unit height was chosen. A base

mesh of size ∆xg = 0.02 was used for the simulations and the solutions were evolved

with three different (3, 4 and 5 levels of mesh refinement corresponding to fine mesh size

1 1 1
∆xf = 200
, ∆xf = 400
& ∆xf = 800
respectively) levels of mesh refinement.

2.4.1.1 Sod’s Shock Tube Problem

A simple, canonical, Sod’s shock tube problem is considered first. The initial con-

ditions for the problem are as follows:




(1.0, 10.0, 0.0, 1.4) for x < 0.5,

(ρ, P, u, γ) =



(0.125, 1.0, 0.0, 1.4) for x ≥ 0.5.

P2
Pressure ratio across the discontinuity is fairly low, P1
= 10. The initial discontinuity

results in a rightward moving shock wave and contact discontinuity, and a leftward moving

rarefaction wave. In this example no interface treatment technique was employed to resolve
 64

the contact discontinuity. The simulations were evolved to time T = 0.167 units. Since

the strengths of the shock wave and the expansion wave are low, the threshold for the

refinement criteria were set at δ1 = 1.0 and δ2 = 100000.0. The solution obtained from

the current simulations along with the exact solution are displayed in Figure 2.3. As can

be seen from the plots, the numerical solutions from the current calculations closely follow

the exact solution.

In order to investigate the impact of the interpolation procedures at the fine-coarse

mesh boundary, the mesh refinement was frozen after the first time step, prohibiting the

mesh from adapting to the solution. This allows the shock wave, the contact discontinu-

ity and the rarefaction wave to pass through the fine-coarse mesh boundary and hence the

interpolation errors generated at the fine-coarse mesh interface can be examined. The mo-

tivation for this analysis comes from the fact that spurious reflected waves were observed

in [41] when waves were made to pass through coarse-fine mesh interface. Although the

refinement criteria developed in this work prohibits the passage of strong discontinuities

across mesh interface, the intent of this analysis is to examine the errors generated in the

limiting case of a shock wave passing through the mesh interface.

The density and the entropy plots for the frozen mesh condition are displayed in

Figure 2.4. Once the shock and the contact discontinuity move into the coarse base mesh

region, they assume a more diffusive profile (corresponding to the size of the coarse base

mesh ∆xg ). The plots are synchronous with the uniform mesh solutions. The spatial vari-

ation of the absolute errors (ǫ = |ρN umerical − ρExact |) and the L2 (ρ) errors (also computed

with respected to the exact solution), for different levels of mesh refinement are plotted in
 65

Figure 2.5. As can be seen from Figure 2.5(a), no spurious reflected waves from the mesh

interface are observed. The errors near the region occupied by the shock and the contact

discontinuity are consistent with the prediction made by the uniform base mesh solution.

But the errors near the head and the tail of the rarefaction waves are significantly smaller

than the errors generated by the uniform base mesh solution. This is because of the pres-

ence of the frozen patch of fine mesh around the initial discontinuity (i.e. at x = 0.5) that

partly spans the region occupied by the tail of the rarefaction wave. Furthermore, in the

region occupied by the shock and the contact discontinuity, the error distribution decreases

with increase in mesh refinement level. This is counter-intuitive, particularly for higher

levels of mesh refinement which involve interpolations across different grid levels. The

reason for this can be attributed to the nature with which the solution is evolved in the

LMR framework. As mentioned before, unlike AMR schemes, the flow field is evolved

in a single sweep for any given time step. As a result, the smallest time step determined

by the size of the fine mesh cell is enforced for the global time update. With the increase

in mesh refinement levels, smaller and smaller time steps are enforced globally and hence

make the interpolation errors more benign. The L2 (ρ) errors computed for both the frozen

mesh condition and the LMR solution are plotted in Figure 2.5(b) clearly confirm this fact

(the errors decrease with increase in mesh refinement levels).

P2
Next a stronger shock was simulated with pressure ratio P1
= 1000. The solution

obtained from the current simulations for different levels of mesh refinement are displayed

in Figure 2.6. The threshold for the refinement criteria are the same as in the previous case,

and the solutions are evolved to time T = 0.021 units. The numerical solutions tend towards
 66

the analytical solutions with increase in mesh refinement level. Similar to the previous case,

the mesh evolution was frozen after the first time step, and L2 errors computed with respect

to the exact solution are displayed in Table 2.1. The errors incurred, although very large

for some variables, are comparable for different levels of mesh refinement. The difference

in the L2 errors computed for the uniform base mesh solution and the LMR solution with

frozen mesh condition, are plotted in Figure 2.7. In contrast to the previous case, the L2

errors increase with mesh refinement level. Thus, as one would expect, it can be inferred

from the above two examples that the errors incurred due to interpolation across shock

waves increase with shock strength. Such errors can be averted if shocks and other strong

discontinuities are contained well within the fine mesh level, by inhibiting their movement

across fine-coarse mesh boundaries.

2.4.2 Two-Dimensional Example

2.4.2.1 Two-Dimensional Riemann Problem (TDRP)

The two-dimensional Riemann problems (TDRP) consist of a combination of one-

dimensional wave patterns, giving rise to 19 different configurations [110]. As one can

construct numerous elementary wave patterns, it is not possible to derive an analytical so-

lution for the TDRPs. However, the solution for each configuration of the TDRP is well es-

tablished and has been widely used to test several high resolution numerical schemes [103,

110]. In this work, three different configurations (6,8 & 19) of the TDRP are considered, to

test the accuracy and performance of the LMR scheme in the two-dimensional set up. The

initial conditions for each configuration correspond to a constant state in each quadrant of a
 67

computational domain of unit length and unit height. The boundaries of the computational

domain are imposed with Neumann conditions. The initial conditions corresponding to

each configuration are listed in Table 2.3.

Simulations were performed with the traditional shock capturing approach (ENO

scheme) with no explicit interface treatment technique. Shown in the figure are the density

contours and the mesh topology for 5 levels of mesh refinement, on a base mesh of size

∆xg = 0.02. The details of the numerical solutions are displayed in Table 2.2. The L2

errors displayed in the table are computed by comparing the solution with the correspond-
∆xg
ing uniform fine mesh calculations (with ∆x = 2k−1
;k = 3, 4 & 5.). The L2 errors remain

almost constant for different levels of mesh refinement, indicating that the solutions ob-

tained from the LMR calculations are consistent, and concur with the corresponding fine

mesh calculations. In addition, the OR shown in Table 2.2, decreases with increase in mesh

refinement level. In particular, the OR for configuration 19 is 5 % (only 5 % of the com-

putational domain is occupied by fine mesh cells). Thus with the current formulation of

the tree-based LMR scheme, substantial savings in computational memory can be attained.

Plots obtained from the current simulations are presented in Figure 2.8. The mesh topology

shown in the figure indicate that the refinement critetion developed in this work is effective

in identifying and following the important features such as shocks and entropy layers.
 68

Refinement Level L2 (ρ) L2 (P ) L2 (u) L2 (S)

P2
P1
= 10

1 4.07569 × 10−3 4.241 × 10−3 1.0865 × 10−2 1.2559 × 10−2

3 3.36566 × 10−3 3.503 × 10−3 1.049 × 10−2 1.08668 × 10−2

4 3.061 × 10−3 3.401 × 10−3 1.036 × 10−2 1.049 × 10−2

5 3.24515 × 10−3 3.379 × 10−3 1.031 × 10−2 1.038 × 10−2

P2
P1
= 1000

1 4.567 × 10−2 557.8874 0.958 1478.400

3 4.61 × 10−2 557.9053 0.9774 1478.465

4 4.613 × 10−2 557.9093 0.9808 1478.475

5 4.619 × 10−2 557.9098 0.9819 1478.484

Table 2.1: L2 errors computed with respect to the exact solution for frozen mesh condition,
for Sod’s shock tube problem.
 69

Configurations Time Levels δ1 δ2 L2 OR

6 0.3 3 25.0 1000.0 5.79 × 10−5 26.42 %

4 25.0 1000.0 5.33 × 10−5 13.03 %

5 25.0 1000.0 3.87 × 10−5 9.1 %

8 0.25 3 1.0 1000.0 9.077 × 10−6 31.58 %

4 1.0 1000.0 8.204 × 10−6 19.74 %

5 1.0 1000.0 6.278 × 10−6 11.78 %

19 0.3 3 20.0 1000.0 7.19 × 10−5 29.9 %

4 20.0 1000.0 4.13 × 10−5 16.1 %

5 20.0 1000.0 2.37 × 10−5 5.71 %

Table 2.2: L2 (ρ) and OR for the two-dimensional Riemann problem

 70

Configurations Initial Conditions

x < 0.5; y > 0.5 x > 0.5; y > 0.5

ρ2 = 2.0 P2 = 1.0 ρ1 = 1.0 P1 = 1.0

u2 = 0.75 v2 = 0.5 u1 = 0.75 v1 = −0.5

6 x < 0.5; y < 0.5 x < 0.5; y < 0.5

ρ3 = 1.0 P3 = 1.0 ρ4 = 3.0 P4 = 1.0

u3 = −0.75 v3 = 0.5 u4 = −0.75 v4 = −0.5

x < 0.5; y > 0.5 x > 0.5; y > 0.5

ρ2 = 1.0 P2 = 1.0 ρ1 = 0.5197 P1 = 0.4

u2 = −0.6259 v2 = 0.1 u1 = 0.1 v1 = 0.1

8 x < 0.5; y < 0.5 x < 0.5; y < 0.5

ρ3 = 0.8 P3 = 1.0 ρ4 = 1.0 P4 = 1.0

u3 = 0.1 v3 = 0.1 u4 = 0.1 v4 = −0.6259

x < 0.5; y > 0.5 x > 0.5; y > 0.5

ρ2 = 1.0 P2 = 1.0 ρ1 = 0.5197 P1 = 0.4

u2 = −0.6259 v2 = 0.1 u1 = 0.1 v1 = 0.1

19 x < 0.5; y < 0.5 x < 0.5; y < 0.5

ρ3 = 0.8 P3 = 1.0 ρ4 = 1.0 P4 = 1.0

u3 = 0.1 v3 = 0.1 u4 = 0.1 v4 = −0.6259

Table 2.3: Initial conditions for the two-dimensional Riemann problem

 71

(a) Hierarchical tree structure (b) Hierarchical data structure

Figure 2.1: Data Structures used in LMR: (a) Hierarchical (quad)tree structure; (b) hierar-
chical data structure
 72

(a) Adjoining cells with one level difference (b) Adjoining cells with more than one level
difference

Figure 2.2: ENO-based flux construction at the fine-coarse mesh boundary: (a) adjoining
cells having one level difference along the diagonal direction (b) adjoining cells having
more than one level difference along the diagonal direction
 73

(a) Density (b) Pressure

(c) Velocity (d) Entropy

P2
Figure 2.3: Sod’s shock tube problem with pressure ratio P1
= 10: Plots of (a) density, (b)
pressure, (c) velocity (d) entropy.
 74

(a) Density (b) Entropy

Figure 2.4: for Sod’s shock tube problem with pressure ratio PP12 = 10: Plots of (a) Density
and (b) entropy for different levels of mesh refinement, for frozen mesh condition.
 75

(a) Spatial variation of the absolute errors (kρ − (b) L2 (ρ) errors
ρEXACT k)

Figure 2.5: Plots of errors: (a) Spatial variation of the absolute errors (kρ − ρEXACT k)
computed with respect to the exact solution for different refinement levels, for frozen mesh
condition (b) L2 (ρ) errors computed with respect to the exact solution, for different refine-
ment levels, for solution adaptive and frozen grid conditions.
 76

(a) Density (b) Pressure

(c) Velocity (d) Entropy

P2
Figure 2.6: Sod’s shock tube problem with pressure ratio P1
= 1000: Plots of contours of
(a) density, (b) pressure, (c) velocity and (d) entropy.
 77

Figure 2.7: Variation of |L2LM R − L2U nif orm | for different levels of mesh refinement for the
frozen mesh condition.
 78

(a) Configuration-6: Entropy Contours (b) Configuration-6: Mesh Topology

(c) Configuration-8: Density Contours (d) Configuration-8: Mesh Topology

(e) Configuration-19: Entropy Con- (f) Configuration-19: Mesh Topology

tours

Figure 2.8: The two-dimensional Riemann Problem computed on a base mesh of size
∆xg = 0.02 with 5 levels of mesh refinement: Entropy contours (Figures (a) and (e)),
density contours (Figure(c)) and Mesh Topology (Figures (b), (d) and (f)).
 79

CHAPTER 3
SHARP INTERFACE TREATMENT FOR FLUID OBJECTS

3.1 Introduction

High-resolution and higher-order schemes such as ENO [163] and WENO [74]

perform very well for single fluid media with discontinuities [116]. However, when such

schemes are employed directly to solve multi-component flows, undesirable oscillations in

the form of pressure waves are prevalent near the material interface [98]. The difficulty

resides in maintaining pressure equilibrium between the fluid components at the material

interface, which results in computational inaccuracies and spurious oscillations [102]. Dis-

continuities such as shocks are nonlinear phenomena with converging characteristics and

so the numerical errors generated are confined within the (smeared) discontinuity [66]. On

the other hand, for contact discontinuities the characteristics run parallel to each other and

hence there is no steepening mechanism to counter the numerical dissipation of the errors

generated near the discontinuity [49]. Thus, errors in treating the presence of embedded

interfaces can permeate the solution away from the interface. To prevent these spurious os-

cillations when dealing with material discontinuities, care must be exercised when treating

the embedded interfaces in a sharp manner.

Several methods have been proposed in the past to extend the numerical schemes

developed for single component flows to multi-component flows [5, 6, 94, 98, 102]. These

methods are based on the fact that a strictly conservative scheme will not be able to maintain

pressure equilibrium across the material interface [6]. In Abgrall et al [5,6], a separate non-
 80

1
conservative transport equation was solved to update the specific function (Γ(γ) = (γ−1)
)

for the mixture. In Karni et al [102], a non-conservative pressure evolution equation was

used to compute the pressure near the material interface. In Jenny et al [94], a thermo-

dynamically consistent correction algorithm for the total energy was proposed. All of

these methods fall under the category of front-capturing methods. A good review of such

methods can be found in Abgrall et al [6]. Although these methods are at least quasi-

conservative, smearing of the interface is an inherent feature. Treating materials that are

separated by distinct sharp interfaces (as in droplet/bubble dynamics) by reformulating

the problem using a mixture model (with diffuse interfaces) and therefore as a single-

component inhomogeneous medium, casts the onus on the continuum formulation, while

relieving the numerical techniques. However, treatment of interfaces as non-sharp or dif-

fuse zones within a mixture formulation is not devoid of numerical problems. For example,

across the interface of multiphase compressible flow (say containing gas-liquid interfaces)

there are jumps in material properties. There are also sharp changes in material behavior

and constitutive laws, for example in the equation of the state. When a shock is transmitted

across an interface, failure to capture this discontinuous response of the contiguous ma-

terials results in severe numerical instabilities or unphysical flow fields. For example, if

the interface is not treated as a sharp entity, due to the presence of numerical diffusion at

the interface, a non-physical zone with artificially diffused density field will result [42]. In

such situations, it becomes necessary to use an ad-hoc equation of state (representing some

non-physical averaged material) to obtain a continuous pressure field [43]. This results in

inaccurate wave interactions and boundary conditions corresponding to this non-physical

 81

zone being enforced at the interface [118]. Nevertheless, it is important to point out that

despite these shortcomings, in some instances (like the shattering of droplets/bubble under

the impact of strong shocks [38, 125]), the diffuse interface approach may still be the most

judicious choice.

Alternatively, front tracking methods which preserve interfaces as sharp disconti-

nuities retain the two fluid regions as separate entities but must contend with the numerical

challenges of discretizing the governing equations to apply relevant boundary conditions

on the interface. In recent times the use of sharp interface methods for the represen-

tation of solid and fluid boundaries in incompressible [86, 124, 148, 177] and compress-

ible [57, 58, 118, 133, 135, 190] flows have become popular. These methods are particularly

attractive for the treatment of moving boundaries in the flow domain since grid generation

and management are dispensed with. In this regard, the Ghost Fluid Method (GFM) orig-

inally formulated by Fedkiw et al [58] provides a simple framework to implicitly transmit

the presence of an interface to the flow field. The GFM requires the definition of a band of

ghost fluid points corresponding to each phase of the interacting media. The band of ghost

fluid points (when populated with suitable flow properties) together with its respective real

fluid constitutes in effect a single flow field. Thus, discretization schemes with uniform or-

der of accuracy can be applied throughout the computational domain without requiring any

special treatment near the interface. Hence, the implementation of the numerical scheme

and the interface treatment are decoupled, and the onus is shifted to populating the ghost

points with the appropriate flow properties.

GFM has been widely used to treat the presence of embedded fluid-fluid and solid-
 82

fluid interfaces [57,58,60,118,133,135,190]. However, these variants in GFM differ in the

way in which the ghost points are populated. In Fekdiw et al [59], GFM was used in the

context of shock and detonation tracking. The Rankine-Hugoniot (R-H) jump conditions

were solved across the discontinuity to populate the ghost points. In a subsequent work,

Fedkiw et al [33, 57] extended the GFM approach for coupling stiff and non-stiff fluids

(gas-water interfaces). In this case, the ghost fluid states were defined by extrapolating the

velocity at the interface in the water medium and the pressure from the air medium [57].

This approach, although attractive for gas-water interfaces, was not suitable to represent

gas-gas interfaces. Aslam [13, 14] proposed to construct a Riemann problem across dis-

continuities (shocks and contacts) and populate the ghost points by solving a Riemann

problem normal to the interface. Liu et al [118], clearly demonstrating the failure of the

original GFM, proposed the Modified GFM (MGFM). Attributing the failure of original

GFM to inaccurate treatment of the wave interactions occurring at the interface, Liu and

co-workers solved the local Riemann problem by carrying out characteristic analysis on

the waves arriving at the interface. Later Hu et al [87], extended this approach by solving

two separate Riemann problems - one for the real fluids and second for the real and cor-

responding ghost fluid. By enforcing the condition that the two Riemann problems work

in conjunction such that the pressure and the velocity obtained from the real-ghost interac-

tions correspond to the real interactions [87], they were able to employ the isobaric fix to

compute the density field for the ghost points. This approach was called Interface Interac-

tion GFM (IGFM). Both IGFM and MGFM were successfully applied to solve a multitude

of problems involving strong shocks interacting with gas-gas and gas-water interfaces.
 83

In this work, a simple but efficient method to construct the Riemann problem has

been developed. The method developed provides a uniform formulation to treat both fluid-

fluid and solid-fluid interfaces. In the case of solid objects embedded in a compressible

flow, it is well known that numerical schemes suffer from acute over-/under-heating er-

rors [127,132]. Several corrective measures have been developed to suppress the growth of

such errors. A unified methodology should be able to suppress these over-/under-heating

errors, and also accurately capture the wave interactions occurring at the interface. Hence,

the focus of this work is on designing a simple yet robust method that can handle strong

shock interactions with gas-gas, gas-water and solid-fluid interfaces.

The approach adopted in this work has a multi-dimensional characteristic intrin-

sic to the construction procedure making it attractive for three-dimensional applications.

Briefly, a local Riemann problem is constructed at the interface which is then solved using

an exact Riemann solver. The resulting Riemann states obtained from solving the Rie-

mann problem are used to populate the respective ghost points. In chapter 5, this method

is extended to treat embedded solid objects in compressible flows, and has been shown to

be effective in suppressing the over-/under-heating errors. As demonstrated in section 3.2,

the results obtained from the current simulations clearly indicate that the proposed method

is consistent in generating satisfactory solutions for several complex configurations and

shocks interacting with interfaces; shocks interacting with droplets, bubbles and free sur-

face have been computed. The method is currently being applied to study the dynamics of

dense particulate compressible flows.

 84

3.1.1 Survey of the Ghost Fluid Method (GFM) for Resolving Fluid-Fluid Interfaces

In Fedkiw et al [58], the ghost points were populated based on the fact that the

pressure and velocity are continuous across the contact discontinuity. This enables the

pressure and velocity to be directly copied from the real fluid onto the ghost field. Since the

entropy advects with material velocity (corresponding to the characteristic wave traveling

with fluid/particle velocity), there is no entropy exchange between the two fluids. This

allows the entropy to be extended to the ghost points (a procedure called isobaric fix in

Fedkiw et al [58]) which can then be used to compute the density for the ghost points

(Figure 1.10).

This simple approach was found to be robust for weak shocks interacting with the

interface. However, when the shock strength and the material stiffness of the interacting

fluids were increased considerably, it was found that this method failed to maintain a non-

oscillatory pressure field. The reason for this failure is intuitive and is largely attributed

to the definition of the ghost states. When strong shocks reside on/near the interface, the

density, pressure and velocity fields cease to be continuous and hence cannot be directly

copied onto the ghost points. The problem is more severe with stiff fluids, as the pressure

and density fields are coupled with a highly sensitive stiffened equation of state. As a re-

sult, a small perturbation in density acutely amplifies the oscillations in the pressure field,

resulting in stalling of the computation. During the few time steps when the shock and the

interface are coincident, the interface treatment technique should temporarily reduce to a

shock tracking method. Therefore, in this regard, the Rankine-Hugoniot (R-H) jump con-

ditions must be solved across the interface to define the ghost states. Liu et al [118] pointed
 85

out the inability of the original GFM to accurately resolve the wave interactions at the in-

terface. According to them, GFM essentially solves two separate single medium Riemann

problems (real-ghost interaction Riemann problem) across the interface. The resolution of

the waves from the real-ghost interaction does not always concur with the Riemann states

generated from real fluid interactions [118]. Hence, with strong shocks impinging on in-

terfaces with high impedance mismatch, this discrepancy in wave representation results in

inaccurate shock and interface locations that result in unphysical oscillations in the flow

field [118].

3.1.2 Riemann Problem at the Interface

Carefully developed interface treatments can avoid the shortcomings of the original

GFM by decomposing the singularities in the flow field and material properties [118]. In-

terfacial states satisfying such conditions can be obtained by locally solving a Riemann

problem normal to the interface. Davis [49] solved an approximate Riemann problem

across the interface using the method of characteristics approach. The solution from this

approximate Riemann solver was then used to construct the second-order Godunov fluxes.

Cocchil et al [42] employed an exact Riemann solver to correct the numerical errors afflict-

ing the interfacial points. The interfacial points suffering from numerical diffusion were

treated using a combination of the solution obtained from the numerical scheme and Rie-

mann solution. The method was found to be promising for one-dimensional problems, but

was complicated when applied for multi-dimensional problems. A modified version of this

approach was incorporated in the GFM framework by Takahira et al [181]. This version
 86

of GFM was successfully used to simulate shocks interacting with gas bubbles and water

droplets. However, their method involves an iterative process in the correction procedure.

Fedkiw et al [59] obtained the numerical fluxes for the ghost points by directly solving the

R-H jump conditions across shocks and detonation fronts. Aslam [13,14] modified this ap-

proach by solving a local Riemann problem at the discontinuity to define the ghost states.

Liu et al [118] developed the Modified Ghost Fluid Method (MGFM); As shown in Fig-

ure 3.1(a) the Riemann problem formulated locally, in the normal direction to the interface,

was solved by carrying out a characteristic analysis of the nonlinear waves arriving at the

interface. Later Hu et al [87], extended this approach by solving two separate Riemann

problems - one for the real fluids and second for the real and corresponding ghost fluid.

By enforcing the condition that the two Riemann problems work in conjunction such that

the pressure and the velocity obtained from the real-ghost interactions correspond to the

real interactions [87], they were able to employ the isobaric fix to compute the density for

the ghost fluids. It is worth noting that, although explicit enforcement of isobaric fix [58]

(which is a simple extension of entropy from the real field to define the ghost field) substan-

tially reduced the over-/under-heating errors, such errors were not totally removed from the

solution.

Thus it is clear that the methods discussed above differed in the way in which the

Riemann problem was incorporated and solved at the interface. Along the same lines, the

methodology developed in this work focuses on designing a simple yet robust method for

constructing the Riemann problem along the local normal to the interface. The method

has multi-dimensional characteristics inbuilt in the construction procedure and is easy to

 87

implement. The method is an extension of Aslam [13, 14] and has been shown here to

successfully simulate strong shocks interacting with both gas-gas and gas-water interfaces.

It is worth noting that a similar method was independently developed by Wang et al [197].

As will be shown in section 3.2.1.2, the approach adopted in this work has been shown to

minimize and confine the associated conservation errors near the interface. In chapter 5,

this method is extended to mitigate over-/under-heating errors effectively and also to work

consistently for a wide range of problems involving strong and weak shock interactions

with embedded solid boundaries.

3.1.3 Constructing the Local Riemann Problem at the Interface

The aforementioned (GFM and non-GFM) sharp interface methods, require the

construction of a local Riemann problem normal to the interface. In this section, a sim-

ple method is presented to accomplish this task. From each interfacial point, such as point

P in Figure 3.1(b), a probe is inserted in the normal direction to the interface. The coordi-

nates of the point of intersection of the probe (from point P) and the interface, i.e. point IP

on the interface in Figure 3.1(b), can be determined as follows:

~ IP = X
X ~ P + | φP |N
~P . (3.1)

The left and the right states required for assembling the Riemann problem are obtained by

advancing a distance of 1.5 ∆x on either side of the interface from IP. The length of the

probe (1.5 ∆x) is chosen so that the interfacial point P, for which the Riemann problem

is being constructed, bears no or minimal weight in the interpolation procedure involved

in defining the Riemann problem. This enables the Riemann problem to be constructed
 88

with points which are not infused with the errors generated at the interface. In order to

maintain consistency, the point which lies inside the levelset is always denoted as the left

state and the point which lies outside the levelset is always denoted as the right state. Thus

the coordinates of the points corresponding to the left and right states can be computed as

follows:

~L = X
X ~ IP − 1.5∆xN
~P , (3.2)

~R = X
X ~ IP + 1.5∆xN
~P . (3.3)

~ L|R ) corresponding to the left (X~L ) and the right (X~R ) states can
The flow properties (W

then be obtained using a simple bilinear interpolation procedure as shown in Figure 3.1(b).

3.1.3.1 Generalized Riemann Solver for Gas - Water Interface

For the sake of completeness, the analytical solution for the Riemann problem is

briefly described here. Once the Riemann problem is constructed, the solution for the

initial value problem can be obtained using a suitable Riemann solver. In this case the Rie-

mann problem was solved exactly. A good discussion on standard procedures to solve the

Riemann problem can be found in the book by Torro [189]. Haller et al [78] and Cocchil et

al [43] have outlined the procedure to solve the Riemann problem for a gas-water interface

(based on stiffened equation of state). The solution for the Riemann problem consists of

four states separated by three waves as shown in Figure 3.1(c). The nonlinear character-

istic waves (with wavespeeds u + a and u - a) can be either a shock wave or a rarefaction

wave. The linear characteristic wave (which travels at the particle velocity u) represents

the interface separating the interacting materials. The solution for the Riemann problem
 89

determines the intermediate (“⋆”) states sandwiching the interface (contact discontinuity),

across which pressure and normal velocity are continuous but the density is discontinuous.

Leaving the details pertaining to the steps for constructing the functions required to solve

the Riemann problem to Toro et al [189], the functions and the algebraic equation are listed

below:

~ L, W
f (p, W ~ R ) = fL (p, W
~ L ) + fR (p, W
~ R ) + ∆un = 0 where , (3.4)

∆un = unL + unR , where unL|R are the normal velocity components,


W~ L|R = ρL|R , unL|R , pL|R T , and

  ⋆ L|R  γ2γ L|R −1
 raref action 2aL|R p +p∞ ⋆ L|R
} if p⋆ +p∞
 L|R


 fL|R = γL|R −1 {1 − ⋆ L|R L|R < 1.0
 p +p∞ p +p∞



 r  
 AL|R L|R
 shock ⋆ L|R p⋆ +p∞
fL|R
 = “
BL|R + p⋆ +p∞
L|R
” p − p ∞ if L|R ≥ 1.0
p⋆ +p∞
~
fL|R (p, WL|R ) =


 2

 AL|R = γ +1


 ( L|R )ρL|R



   
 γ −1 L|R
BL|R
 = γL|R L|R +1
p L|R + p ∞

The nonlinear algebraic equation (3.4) can be solved via Newton iterations to determine

~ ⋆ . The intermediate states (W

the intermediate states W ~ ⋆ ) obtained from solving the
L|R L|R

Riemann problem are then used to correct the flow properties of the real fluid at the inter-

~ ⋆ obtained from
facial points. For instance, as shown in Figure 3.2(b), the Riemann state W L

solving the Riemann problem constructed at the interfacial point P, is used to correct the

flow properties of the real fluid at point P.

This will ensure that a constant entropy field is maintained throughout the interact-

ing materials (except across shocks). This in turn prevents the diffusion of the entropy field

across the interface and hence prohibits entropy exchange between the interacting (real) flu-
 90

ids. It was pointed out earlier in section 3.1.1, that a simple extension of the entropy field

to correct the real fluid properties at the interfacial points does not comply with the nature

of the characteristic waves arriving/leaving the interface. On the contrary, the correction

procedure enforced in the current approach account for the wave systems interacting with

the interface.

3.1.3.2 Populating the Ghost Points

Once the flow properties at the interfacial points are corrected, the flow properties

are extended along the normal direction to the interface to populate the respective inte-

rior ghost points. To carry out the multi-dimensional extrapolation procedure, the partial

differential equation given in equation (3.5) is solved to steady state.

∂Ψ ~ =0
+ H(φl )~n•∇Ψ (3.5)
∂τ

where Ψ is the variable extended across the interface, τ is the pseudo-time and H(φl ) is the

unit Heaviside function defined as




0 if φl ≤ 0.0,

H(φl ) =



1 if φl > 0.0.

In the current work, the variable Ψ extended across the interface correspond to the flow

~ ⋆ ) obtained from solving the Riemann problem.

properties at the intermediate state (W L|R

As given in equation (3.5) a constant extrapolation method is used for populating the ghost

points, although higher-order extrapolation procedures are possible [15]. Figure 3.2(b)

shows the multi-dimensional correction and extension procedure carried out along the nor-

mal direction to the interface. The point P in Figure 3.2(b) is inside the object i.e. φl ≤ 0
 91

and corresponds to the left state of the Riemann problem. Hence, as mentioned before,

~ ⋆ obtained from solving the Riemann problem is

the Riemann state corresponding to W L

~ ⋆ is
used to correct the flow properties at point P. Then the flow field corresponding to W L

extended along the normal direction to populate the interior ghost points (Figure 3.2(b)).

The one-dimensional version of the correction and the extension procedure is shown in

Figure 3.2(a).

It is important to note that the velocity obtained from the Riemann solver corre-

sponds to the normal velocity component computed in the local curvilinear coordinate. In

order to reconstruct the velocity vector in the global Cartesian coordinates, the slip condi-

tion at the interface is enforced. Thus, the tangential velocities computed for the Riemann

problem (the left and the right states) are extrapolated across the interface. Accordingly,

the velocity vector can be reconstructed as

~u⋆L = un⋆ n̂ + ~utL (3.6)

~u⋆R = un⋆ n̂ + ~utR (3.7)

where n̂ is the normal vector defined in Eq (1.54), un⋆ is the normal velocity obtained from

the Riemann solver and ~utL|R are the tangential velocities (corresponding to the left and

right states of the Riemann problem) computed as follows:

~utL = ~uL − (~uL •n̂)n̂ (3.8)

~utR = ~uR − (~uR •n̂)n̂ (3.9)

The velocity vector can then be decomposed into its corresponding components (u, v, w)

in the Cartesian coordinates.

 92

3.1.3.3 Freshly Cleared Cells

As the interface sweeps through the computational domain, a computational point

which was previously inside the object (i.e. belonging to say phase 1) may now lie out-

side the object (i.e. belonging to phase 2) and vice versa. Identification of these points is

straightforward once the levelset field is updated. Explicitly, points for which φni φn+1
i <0

are tagged as “freshly cleared cells”. Although the values with which the freshly cleared

cells are populated are temporary (as they are overwritten in the correction procedure), de-

pending on the location of these points, they may or may not participate in the interpolation

procedure involved in constructing the Riemann problem. Hence it is required to update

the flow and material properties belonging to these points at the beginning of each time

step. There are several ways to update the properties of the freshly cleared cells. The easi-

est and the most direct method in the GFM framework adopted in this work is to copy the

properties of ghost fluid variables at that point onto the real fluid properties.

3.2 Numerical Examples

3.2.1 One-Dimensional Examples

In this section a series of one-dimensional shock tube (of unit length) problems

are presented. The initial condition corresponds to a singularity in flow variables that re-

solves into a transmitted and a reflected shock or expansion wave or both (depending on the

impedance mismatch between the interacting fluids), and a contact discontinuity. The re-

sulting contact discontinuity is tracked and represented as a sharp interface using the GFM

approach.
 93

For the one-dimensional test cases considered in this section, a computational do-

main corresponding to a shock tube of unit length and unit height was chosen. The grid

spacing for one-dimensional simulations was set at ∆x = 1/200 unless stated to the con-

trary. For LMR calculations, a base mesh of size ∆xg = 0.02 was used for the simulations

and the solutions were evolved with three different (3, 4 and 5 levels of mesh refinement

1 1 1
corresponding to fine mesh size ∆xf = 200
, ∆xf = 400
& ∆xf = 800
respectively) levels

of mesh refinement.

3.2.1.1 Example 1 - Single Fluid Test Case

In this example, a set of one-dimensional test cases in a single fluid medium are

considered.

CASE(A) : Sod’s Shock Tube Problem

The initial conditions for this problem are given below:




(1.0, 10.0, 0.0, 1.4) for x < 0.6,

(ρ, P, u, γ) =



(0.125, 1.0, 0.0, 1.4) for x ≥ 0.6.

P2
A fairly low pressure ratio, P1
= 10 was simulated. The typical solutions after 100 time

steps are shown in Figure 3.3. The figure shows the comparison between Riemann GFM

(henceforth R-GFM) and original GFM with isobaric fix. The solution obtained from tra-

ditional shock-capturing schemes (without explicit interface treatment) for the contact dis-

continuity is also shown in the figure. As is evident from the plots, the methods seem to

produce results that conform to the exact solution. From the entropy plots both original

GFM and R-GFM were found to suffer from slight over-heating errors at the interface. The
 94

R-GFM yields a lower level of over-heating than the original GFM approach.

Next, the pressure ratio ( PP21 ) was increased to 1000. Figure 3.4 shows that the

original GFM severely over-predicts the density field near the contact discontinuity. The

solution obtained from the R-GFM approach shows slight over-heating errors but over-

all the shock, rarefaction and the contact align with the exact solution. As was pointed

out in section 3.1.1, the initial large discontinuity in flow variables (density and pressure

field in this case) results in a strong shock wave and a contact discontinuity traveling to

the right, and an expansion wave traveling to the left. The initial velocity of the inter-

face/contact discontinuity is very high, so that the interface stays in close proximity to the

shock. Hence, during these few instances, the pressure and velocity field across the inter-

face register jumps, which when copied directly onto the ghost points results in inaccurate

representation of the Riemann states generated at the interface. On the other hand, the

R-GFM approach resolves the wave patterns accurately and hence predicts shock and in-

terface locations which agree well with the exact solution, as indicated by the enlarged view

near the interface in Figure 3.4. Figure 3.4 also shows that the traditional shock-capturing

approach results in smearing of the interface over a few grid cells, as opposed to the GFM

approach which retains the interface as a sharp entity.

Comparing the over-heating errors incurred by the R-GFM and the original GFM

approach, the R-GFM approach is found to incur lower over-heating errors as seen from

Figures 3.3(b) & 3.4(b). In particular, Figure 3.4(a) shows that the over-heating errors

incurred by the original GFM approach result in catastrophic oscillations in the density

field and inaccurate predictions of the shock and interface locations. Thus it is clear that
 95

even for a single fluid problem with strong discontinuity considered here, the original GFM

approach fails to maintain a non-oscillatory flow field.

CASE(B) : Woodward Colella “Bang Bang” Problem

This problem was formulated by Woodward et al [200]. Unlike the previous case, this

example involves strong shock reflections and collisions. The initial conditions for the

problem are





 (1.0, 1000.0, 0.0, 1.4) if x < 0.1,




(ρ, P, u, γ) = (1.0, 0.01, 0.0, 1.4) if 0.1 ≤ x ≥ 0.9,








(1.0, 100.0, 0.0, 1.4) if 0.9 < x ≤ 1.0.

Reflective boundary conditions were applied at the two ends of the domain to enforce

the presence of enclosing surfaces. The initial conditions set off strong shock waves that

interact with each other. The solution obtained from the current simulation at T = 0.038

is compared with the reference solution. The reference solution corresponds to a fine grid

calculation obtained without employing GFM. Only the solution obtained using R-GFM

method is presented in Figure 3.5. The plots show that with increase in mesh resolution the

R-GFM solution approaches the reference solution. The peak pressure and the maximum

velocity matches well with the reference solution, especially with finer mesh resolution.

The slight over-/under-heating errors found in the entropy plots displayed in Figure 3.5(d)

are mitigated with improved mesh resolution.

The conservation errors (plotted in Figure 3.6) were minimal and localized to the

instant when the shock impinged on the interface. As pointed out by Liu et. al [118] it
 96

takes very few timesteps for the normal motion of the interface to resume after the impact

of the shock.

3.2.1.2 Example 2

An example of a multi-material shock tube problem is the one-dimensional air-

helium shock tube with the initial conditions given below:




(1.0, 1.0, 0.0, 1.4) for x < 0.5,

(ρ, P, u, γ) =



(0.125, 0.1, 0.0, 1.667) for x ≥ 0.5.

The plots of density, pressure, velocity and entropy after 100 time steps are displayed in

Figure 3.7. As in the previous case, both versions of GFM under-predict the density near

the interface. The insert shows the zoomed-in view of the density profile close to the

interface. The interface position predicted by original GFM is shifted by two/three grid

points relative to the exact solution. In contrast, the interface location predicted by the R-

GFM approach is in line with the exact solution. The pressure and the velocity plots also

show good agreement with the exact solution. The entropy plots show modest over-heating

errors at the interface.

A Note on Conservation Error

A conservation error analysis was carried out for this problem. The total mass, mo-

mentum and energy conservation errors computed using the approach outlined by Wang et

al [197], are presented in Figure 3.8. It was found that the conservation errors were no-

ticeable only during the initial 4 - 5 time steps, when the shock wave was in the vicinity

of the interface. This is consistent with the trend observed by Liu et al [118] and Hu et
 97

al [87]. The conservation errors observed in this case are mainly due to the errors which

are inherent in the GFM approach. These errors arise from the numerical flux computed

based on the ghost field. The numerical flux computed at the interface does not ensure

strict conservation. As shown in Figure 3.9, for all the bulk points explicit numerical flux

conservation can be achieved easily. On the other hand, due to the ghost fluid treatment

from each side of the interface and the construction of fluxes using the ghost field, it is not

possible to compute a unique numerical flux crossing the cell boundary for the interfacial

points. The errors generated due to the non-uniqueness in the numerical flux computed at

the interface becomes significant when strong shocks or detonation waves impinge on the

interface.

It was pointed out by Abgrall et al [6] that a non-oscillatory pressure field can be

produced by methods that are not strictly conservative at the interface. Due to the intrinsic

nature of the ghost fluid treatment, where the ghost field is populated via extension from

the real fluid, it is not straightforward to devise a scheme in the ghost fluid framework that

would completely annihilate these conservation errors. However, Glimm et al [66] have at-

tempted to construct conservative fluxes by locally rearranging the cells cut by the interface

and explicitly enforcing appropriate jump conditions to guarantee equality of the numeri-

cal fluxes at the cell interface. This procedure of dynamically removing the conservation

errors at the interface is appealing and feasible for one-dimensional problems, but becomes

extremely complicated when extended to multi-dimensional problems. A relatively simpler

approach was proposed by Nguyen et. al [47]. In this approach, the conservation errors

were alleviated by redistributing these errors to the numerical fluxes computed at the in-
 98

terface, at the end of each RK sub-step. As pointed out by Hu et al [87], it is not readily

apparent that such a post-processing measure to correct the conservation errors is effective,

particularly when conservation errors have already been incurred in the previous RK time

step. In this work, no such additional measures were enforced to conserve the numerical

fluxes at the interface. Furthermore, the conservation error analysis carried out in this work

points out that these errors are confined to very few time steps when the shock resides

on/near the interface and were found to be benign from the viewpoint of stable computa-

tion over the overall time of calculation. As shown in the insert in the density plots, both

the interface and the shock locations are captured to a good degree of accuracy, indicating

that the conservation errors involved are spatially and temporally localized (close to the

interface and to the duration of shock-interface coincidence), and do not pollute the bulk of

the solution or the long term evolution of the flow field.

3.2.1.3 Example 3

The next example considered in this section is a gas-water shock tube problem. The

initial conditions shown above is analogous to the one-dimensional version of an underwa-

ter explosion. Unlike the previous examples, the initial conditions are normalized with

respect to the properties of water at ambient conditions and unit (1m) length scale. This

example is taken from Tang et al [184]. The problem was also solved by Hu et al [87].



(0.01, 1000.0, 0.0, 1.4) for x < 0.5,

(ρ, P, u, γ) =



(1.0, 1.0, 0.0, 4.4) for x ≥ 0.5
 99

The configuration corresponds to a fast-slow interface as the shock wave is incident in a

material with higher acoustic impedance. Due to high pressure ratio, a relatively strong

shock wave is transmitted into the water medium and a weaker reflected shock wave in

the air medium. The comparison between the numerical simulation obtained after 200

time steps with the corresponding exact solution are shown in Figure 3.10. The numerical

solution closely follows the exact solution.

3.2.1.4 Example 4

The second in the series of the gas-water shock tube problem considered in this

work is taken from Hu et al [87,88]. In comparison with the previous example, the pressure

ratio for this problem is further increased by a factor of 10. The initial conditions are

modified as 


(0.5, 20000.0, 100.0, 2.5) for x < 0.5,

(ρ, P, u, γ) =



(1.0, 1.0, 0.0, 4.4) for x ≥ 0.5.

In this case, a much stronger shock is both reflected and transmitted. Figure 3.11 shows

the plots obtained after 200 time steps. As in the previous case, both the reflected and

transmitted shocks are captured precisely. The position of the contact discontinuity matches

very well with that of the exact solution. These features indicate that the conservation errors

generated, even in the case of stiff fluids, are negligible. Because of very large entropy

disparities between the interacting media, the entropy plots do not reveal any significant

over-/under-prediction of the flow properties.

 100

3.2.1.5 Example 5

The next example considered is a water-air shock tube problem. The water medium

is on the high pressure side and the air is maintained at ambient condition. The initial

conditions normalized with respect to the ambient conditions of water are given below:



(1.3, 10000.0, 0.0, 4.4) for x < 0.5,

(ρ, P, u, γ) =



(0.001, 1.0, 0.0, 1.4) for x ≥ 0.5.

In this example the shock wave travels from the water side into the air side and hence can

be categorized as a fast-slow interface [81]. Due to high impedance mismatch, a strong

rarefaction wave is reflected in the water medium and a weak shock wave is transmitted in

the air medium. The transmitted shock is extremely weak (the pressure ratio across the re-

flected rarefaction wave is much larger than the pressure ratio across the transmitted shock

wave) and hence not readily visible from the density or pressure plots (Figures 3.12(a)

& 3.12(b)). However, the entropy plots shown in Figure 3.12(d) show the jump across the

transmitted shock in air. Also revealed by the entropy plots are the over-heating errors.

Although the over-heating errors observed in this case are relatively large in magnitude,

they are not significant enough to afflict the global solution. The computed position of

the transmitted shock wave and the reflected expansion wave match well with the exact

solution.

It is clear from the examples shown above that the R-GFM approach developed in

this work is able to accurately resolve strong shock interactions with both gas-gas as well as

gas-water interfaces. On the other hand, the original GFM approach generated unphysical

oscillations in the flow field, and in some cases, resulted in complete break-down of the so-
 101

lution (gas-water interfaces). In order for the original GFM approach to apply to gas-water

interfaces, Fedkiw et al [33, 57] modified the GFM approach by extrapolating the veloc-

ity at the interface in the water medium and the pressure from the air medium to define the

ghost states. However, such modifications would render the original GFM approach unsuit-

able for resolving gas-gas interfaces. In contrast, the methodology developed in this work

is found to be robust and accurate in resolving multi-material interfaces without requiring

special treatments at the interface. Furthermore, the plots shown in Figure 3.8 clearly in-

dicate the the conservation errors and the resulting under-/over-heating errors incurred are

localized close to the interface.

In the next few examples, shock tube problems involving multi-fluids are solved

using dynamic mesh adaptation techqniue described in chapter 2.

3.2.1.6 Example 6

The next one-dimensional example considered in this work is taken from Hu et

al [87]. The problem involves interaction between a relatively stiff fluid and helium. The

pressure and density ratio for this problem are also higher than considered before. The

initial conditions for this problem are given below :




(0.384, 100.0, 27.077, 1.667) for x < 0.6,

(ρ, P, u, γ) =



(100.0, 1.0, 0.0, 3.0) for x ≥ 0.6.

The simulation was carried out up to a time T = 0.03 units. The values selected for the

refinement criteria are δ1 = 5.0 and δ2 = 100000.0 (arbitrary large value). The variation of

typical flow variables are compared with the exact solution in Figure 3.13. The agreement
 102

between the two solutions is good. The contact discontinuity is resolved sharply within a

mesh cell without oscillations in the pressure field. The width of the shock wave is confined

to 2-3 mesh cells (for 5 levels of refinement). In addition, the locations of the shock, contact

discontinuity and the rarefaction waves match very well with the exact solution. The L2

errors for the flow variables are displayed in Table 3.1. As can be verified from the table, the

L2 errors and the difference in L2 errors (∆L2 , computed with respect to the corresponding

fine mesh calculations) decrease with increase in the levels of mesh refinement.

3.2.1.7 Example 7

A shock-tube problem involving air and water is solved. The problem is taken from

Wang et al [196]. The initial conditions for the problem are given below:



(1.27, 8000.0, 0.0, 1.4) for x < 0.5,

(ρ, P, u, γ) =



(1.0, 1.0, 0.0, 5.5) otherwise.

The solution is evolved to time T = 0.0015 units. The values selected for the refinement

criteria are δ1 = 100.0 and δ2 = 100000.0 (arbitrarily large value). The results from the

calculations are displayed in Figure 3.14. As can be seen from the figure, the transmitted

shock wave and contact discontinuity along with the reflected rarefaction wave are captured

accurately. The entropy plot (Figure 3.14(d)) reveals modest over-heating errors generated

at the interface. As expected, the over-heating errors decrease with increase in the levels of

mesh refinement. In addition, the over-heating errors are concentrated close to the interface

and are not visible in the density plots. The L2 errors displayed in Table 3.1 show that the

errors incurred in density field are 3 orders of magnitude lower than the errors incurred in
 103

the entropy field. Furthermore, it can be verified that the errors decrease with increase in

mesh refinement levels. The over-/under-heating errors generated in the current approach

are consistent and problem independent, and have been well documented in previous ex-

amples.

3.2.1.8 Example 8

The gas-water shock tube problem considered in this section is taken from Hu et

al [87, 88]. The initial conditions are modified as




(0.5, 20000.0, 100.0, 2.5) for x < 0.5,

(ρ, P, u, γ) =



(1.0, 1.0, 0.0, 4.4) for x ≥ 0.5.

The initial configuration resolves itself into a strong shock wave transmitted into air and

a strong shock wave reflected into water. Figure 3.15 shows the plots obtained after time

T = 0.001 units. The values selected for the refinement criteria are δ1 = 100.0 and δ2 =

100000.0 (arbitrary large value). Both the reflected and transmitted shocks are captured

accurately. The entropy and the density plots do not reveal any significant over-/under-

heating errors. The position of the contact discontinuity and the location of the shock wave

matches very well with that of the exact solution. This indicates that the conservation errors

incurred, even for interfaces separating fluids with high impedance mismatch, is negligible.

As pointed out in section 3.2.1.2, the conservation errors incurred are intrinsic to the GFM

approach [47, 66, 87, 88, 197], and hence cannot be entirely annulled.
 104

3.2.2 Two-Dimensional Examples

3.2.2.1 Spherical Riemann problem

First in the multi-dimensional test cases presented in this work is the single phase

spherical Riemann problem. The initial conditions for this problem are given below:

 q
(1.0, 5.0, 0.0, 0.0, 1.4) for x2 + (y − 0.5)2 ≤ 0.2,


(ρ, P, u, v, γ) =



(1.0, 1.0, 0.0, 0.0, 1.4) otherwise.

The one-dimensional problem in spherical coordinates is reconstructed in cylindrical co-

ordinates (axisymmetric form) as shown in Figure 3.16. The center of the spherical dis-

continuity is located at (0.0,0.6) and a pair of reflecting walls (also modeled as embedded

objects) are located at y = 0.1 and y = 1.1. Reflective Boundary Condition (RBC) aug-

mented with a Riemann solver was employed to model the reflecting walls and the R-GFM

was used to model the initially spherical contact discontinuity. Euler equations in axisym-

metric form were solved in a 1.5 × 1.2 domain with 600 × 450 points. The simulations

were carried out both with (R-GFM) and without explicit interface treatment (i.e. using

ENO shock-capturing) for the contact discontinuity. The simulations were carried out till

T = 0.8. The initial condition corresponds to a jump in pressure in the radial direction.

This results in an inward moving expansion wave, an outward moving shock wave and a

contact discontinuity. Figure 3.17 shows that the R-GFM is able to resolve the contact

discontinuity accurately without modifying the overall wave structure and their relative po-

sitions. The solution obtained from the R-GFM follows closely the trend observed by the

shock-capturing schemes with no explicit interface treatment. Moreover, the enforcement

of the sharp interface treatment accentuates weak features that are lost in the single-field
 105

simulation. As pointed out by Langseth et al [111], the non-physical zone developed due

to wave focusing is visible at the center in both solutions. For a quantitative comparison,

the variation of density and pressure along the centerline (at y = 0.6) at T = 0.5 are shown

in Figure 3.18. The plots are in close agreement with each other.

3.2.2.2 Shock Interacting with Cylindrical Helium Bubble in Air

CASE A : Mach 1.22 Shock interacting with Helium Bubble in Air

The interaction of shock waves with gas bubbles have applications ranging from astro-

physics [37, 140] to cavitation damage of human tissue (shock wave lithotripsy) [92]. The

example considered in this section corresponds to the interaction of a Mach 1.22 shock

with a cylindrical Helium bubble in air. The configuration of this multi-component system

provides a weak shock interacting with fluids of varied stiffness. Haas et al [75] considered

shock waves interacting with R-22 refrigerant and Helium bubbles. They provided a series

of experimental observations which were later verified numerically by Quirk et al [147]

and several others [16, 58, 87, 98, 125]. The initial conditions for this problem are such that

a Helium bubble of 50 mm radius resides in thermal equilibrium with surrounding air at

ambient conditions. The initial conditions normalized based on the property of air at 1 atm

and length scale 1 mm are






 (1.3764, 1.5698, −0.394, 0.0, 1.4) for post-shocked air i.e.x ≥ 225,




(ρ, P, u, v, γ) = (1.0, 1.0, 0.0, 0.0, 1.4) for pre-shocked air,





 q

(0.138, 1.0, 0.0, 0.0, 1.667) for (x − 150)2 + y 2 ≤ 25

 106

The computational domain (325 × 89) is similar that used by Fedkiw et al [58]. The top and

bottom of the domain are prescribed with reflective boundary conditions. The right end of

the domain is prescribed with a Dirichlet inflow condition corresponding to post-shocked

air and the left end with Neumann outflow condition. The grid spacing ∆x = 0.25.

As the speed of sound in Helium is faster than in the surrounding air, the system can

be classified as a slow-fast interface. [3]. Hence, the impinging shock results in a refracted

shock wave and a reflected rarefaction wave. As mentioned in Johnsen et al [98], the re-

fracted shock, upon reaching the other end of the bubble, further degenerates into a weaker

reflected shock wave and a transmitted shock in air. This process of reflection continues

until the resulting wave system degenerates into a weaker wave system [98]. The numerical

Schlieren image and density contours generated from the current simulation are shown in

Figures 3.20 & 3.21 respectively. Figures 3.20(a) & 3.21(a) show the transmitted shock

wave and the reflected expansion. It is clear from the figure that the refracted shock inside

the Helium bubble travels faster than the incident shock. Figures 3.20(b) & 3.21(b) show

the transmitted wave from the Helium bubble and the incident shock in air. The multiply

reflected weak wave system inside the Helium bubble is also clearly visible. A closer exam-

ination of the figure depicts the initial stage of jet formation at the center of the bubble. The

jet becomes more clearly visible in Figures 3.20(c) & 3.21(c). The bubble has now taken

the familiar kidney bean shape. Not so readily visible features in Figures 3.20(c) & 3.21(c)

are the Kelvin-Helmholtz instability [125] reported on the surface of the bubble. These

features may become evident with increase in mesh resolution. Finally, in Figures 3.20(d)

& 3.21(d), the bubble is shattered into fragments due to the impact of the jet. It is worth
 107

noting that the giant vortical structures reported by Marquina et al [125] are not readily

visible in the current simulation. The reason is that Marquina et al have used an extremely

fine mesh (8000 X 600) and a front capturing technique which adds sufficient amount of

numerical diffusion at the interface to capture these vortical structures and the associated

Kelvin-Helmholtz instability. The plots of density and pressure along the horizontal line of

symmetry, along with the pressure history registered at a distance 3 mm downstream of the

initial bubble location are displayed in Figure 3.22.

CASE B : Mach 6 Shock interacting with Helium Bubble in Air

In this example, the Mach number of the impinging shock wave is increased to M = 6.

The configuration of the system, including the domain boundary conditions and the grid

resolution, are the same as in the previous case. The initial conditions are modified as :





 (5.268, 41.83, −5.752, 0.0, 1.4) for post-shocked air i.e.x ≥ 225,




(ρ, P, u, v, γ) = (1.0, 1.0, 0.0, 0.0, 1.4) for pre-shocked air,





 q

(0.138, 1.0, 0.0, 0.0, 1.667) for (x − 150)2 + y 2 ≤ 25


Bagabir et al [16] have analyzed the response of a gas bubble to shocks of different strengths.

In their work, the computations were carried out for a single component flow (with γ = 1.4)

system. Hu et al [88] carried out the corresponding multi-component flow simulation with

a helium bubble and air.

The snapshots of the numerical Schlieren image from the current simulations are

displayed in Figure 3.24. Since the refracted shock wave travels faster inside the bubble, it

reaches the rear end of the bubble before the incident shock wave. The shock wave is then
 108

transmitted to the air medium which is at rest. In the previous case, this transmitted shock

wave coalesces with the incident shock wave to form a single planar shock wave that travels

in the air medium. In this case, because of the time lag developed due to greater celerity of

shock wave inside the bubble, the incident shock wave cannot coalesce with the transmitted

shock wave, leaving behind a curved shock front from the transmitted shock attached to the

planar incident shock wave. The strength of the incident shock is high enough to impart

sufficient momentum to the helium bubble to drag the bubble along with it. As a result

of this, the location of the shock and the interface are always in close proximity to each

other. Moreover, the resultant changes on the topology of the interface are so intricate

that the bubble shatters into tiny fragments due to the impact of the jet. The topology of

the interface at different instants in time are displayed in Figure 3.25. At time T = 15.3

µs, the transmitted shock deforms the front portion of the bubble leaving the leeward side

unaffected. At T = 38.25 µs, the shock has traversed most part of the bubble and at time t =

61.2 µs the bubble starts to deform. At T = 78 µs, the bubble has deformed completely and

starts to form fragments. Once such fragments are formed, it is not possible to continue

the simulation without employing special mesh enrichment techniques. Hu et al [88] have

carried out their simulation by explicitly deleting small pockets of mass. No such additional

measures were taken in this work; given the limitations of a uniform non-adaptive mesh,

the results obtained are in close agreement with that of Hu et al [88].

 109

3.2.2.3 Underwater Explosion Near a Free Surface

Numerical simulation of an under-water detonation of gaseous products near a free

surface is considered in this section. Analysis for this application have been carried out

in the past by several authors [38, 73, 117, 136]. The initial configuration of the system

as outlined in Liu et al [117] is shown in Figure 3.26. A cylindrical gaseous explosive

core of unit radius is located 3 units below the free surface, at the center of a 12 X 12

domain. The initial conditions are normalized with properties of water at 1 atm and with

length scale 0.1m (the initial radius of the core). The time sequence of the configuration

of the system is shown in Figure 3.27. The explosion sets off a cylindrical shock front

advancing away from the core. Due to large impedance mismatch between the air - water

interface at the free surface, a relatively weak shock wave is transmitted into the air medium

and a “relief” wave in the form of strong rarefaction wave propagates towards core in the

water medium [38, 117]. The flow features resulting from this explosion are symmetrical

until T = 0.00892. It is at this instant that the reflected rarefaction wave interacts with the

explosive core. At T = 0.024, the free surface is deformed due to the impact of the incident

shock wave. The expansion wave generated in the water has interacted with the bubble and

produces a reflected compression wave (which eventually turns into a shock wave) and an

expansion wave inside the bubble [38]. Due to the interaction of the relief wave with the

bubble, the top portion of the bubble accelerates towards the free surface distorting it from

its initial shape to an elongated ’egg-like’ shape. Also visible at this instant is reflected

shock wave at the bottom of the domain due to the impact of the initial shock wave from

the explosion. At T = 0.045, this reflected wave interacts with the bottom of the bubble
 110

further distorting the shape of the bubble.

3.2.2.4 Shock Loading on a Submerged Structure due to an Underwater Explosion

The next problem considered in this section is the evolution of an underwater ex-

plosion and its impact on a submerged structure. A square structure submerged in water

is impacted by a circular shock wave generated by an explosion. The shock wave im-

parts severe loading on the structure which could potentially damage the structure. This is

a multi-material flow problem consisting of multiple interfaces: air-water, gas-water and

solid-water interface. For the fluid-fluid interfaces, the methodology discussed in the pre-

vious sections was adopted to define the ghost states. For the solid-fluid interface, the

Reflective Boundary Condition (RBC) augmented with a Riemann solver was used to con-

struct the ghost field. This problem is discussed at length in Liu et al [119]. The explosive

core of 0.1 m radius is located at (1.5,2.0) in a 6m × 6m domain. The free surface separat-

ing water from air is located at Y = 5.0. The square structure with length = 1.0m is located

at the center of the domain. The flow conditions, normalized with properties of water at

ambient conditions, are


 q



 (1.27, 9000.74, 0.0, 0.0, 1.4) for the core, (x − 1.5)2 + (y − 2.0)2 ≤ 0.1,




(ρ, P, u, v, γ) = (1.0, 1.0, 0.0, 0.0, 5.5) for quiescent water,y < 5.0,








(0.001, 1.0, 0.0, 0.0, 1.4) for ambient air,y ≥ 5.0.

The bottom of the domain is prescribed with reflective boundary condition and the top,

right and the left boundaries are prescribed with Neumann conditions. The explosion sets

off a circular shock wave that interacts with the structure and later with the free surface.
 111

The incident shock impacts first on the bottom and the left face of the square structure

resulting in a complex shock diffraction pattern as shown in Figure 3.29. Figures 3.29(a)

& 3.29(b) shows the shock diffracting of the structure along with the vortices produced at

the top left and the bottom right corner. At this stage, the shock has not impacted the free

surface. The inward traveling expansion wave and the contact discontinuity separating the

explosive core from water are also clearly visible. The shock wave reflected off the bottom

wall of the domain can be seen advancing towards the explosive core. Figures 3.29(c)

& 3.29(d) show the plots at a later instant in time. The shock wave has impacted the free

surface and the reflected expansion wave traveling in the water medium and the transmitted

shock wave in air are clearly visible. The reflected shock wave from the bottom wall of the

domain has now reached the explosive core, deforming its shape.

The net horizontal and vertical force components were computed by integrating the

pressure force acting on the structure. The force components exerted on the square structure

are plotted are shown in Figure 3.30. The horizontal and the vertical force components have

the same form and they overlap with each other until T = 0.017. This is expected because

the wave structure resulting from the initial shock impact (on the bottom and the left wall)

is symmetric with respect to the structure. Later in time, the vertical force is considerably

modified due to the strong reflected rarefaction wave from the free surface and hence begins

to depart from the horizontal component. The moment exerted on the structure due to these

forces was also computed and plotted in Figure 3.30. For a quantitative comparison, the

force components computed from the current simulation are compared with those obtained

by Liu et al [119]. Liu et al [119] have used MGFM to resolve the fluid-fluid interface and
 112

symmetry based RBC (developed by Forrer et. al [62]) for the solid-fluid interface. As

shown in Figure 3.31, the results from the present simulation compare very well with those

of Liu et al [119].

3.2.2.5 Richtmyer-Meshkov Instability (RMI)

The next example presented in this work is the evolution of the single mode Richtmyer-

Meshkov instability (RMI). The problem consists of an embedded fluid-fluid interface im-

pacted by a shock. The R-GFM approach outlined in chapter 3 is used to capture the

interface conditions. The problem set up is the same as that of Nourgaliev et al [136]. A

computational domain of length 4 units and height 1 unit is chosen. The top and the bot-

tom portions of the domain are prescribed with reflective boundary conditions, depicting

the presence of walls of the shocktube. The right and the left sides are imposed with Neu-

mann conditions. The unperturbed interface separating SF6 and air is initially located at

xI = 2.9. A perturbation with maximum magnitude ǫ = 0.1 and wavelength λ = 1 is

imposed over the interface as given below:

y 1
φ(x) = xI − x − ǫSIN (2π( + ))
λ 4

A planar shock corresponding to Mach 1.24 is initially located at xs = 3.2 with the condi-

tions as follows:





 (1.628, 1.411, −0.39, 0.0, 1.4) x ≥ 3.2,




(ρ, P, u, v, γ) = (1.0, 1.0, 0.0, 0.0, 1.4) φ(xI ) ≥ x < 3.2








(5.04, 1.0, 0.0, 0.0, 1.093) x < φ(xI )
 113

The problem was simulated with a base mesh of size ∆xg = 0.02 with 3 levels of mesh

refinement. The threshold limit for the refinement criteria (δ1 & δ2 ) were set to 10.0 and

10000.0 (arbitrary large value) respectively. The fine mesh size corresponds to 200 points

per unit wavelength of the perturbation. In Figure 3.32(a), the density contours obtained

from the current LMR calculations are juxtaposed over the corresponding uniform fine

mesh calculation. Except for minor discrepancies observed in the coarse mesh regions,

the global flow structure along with the shock and the interface locations are captured to a

good degree of accuracy. The L2 (ρ) error computed at the instant shown in the figure (Fig-

ure 3.32(a)) for this simulation is 3.923 × 10−4 . The evolution of the interface at different

instants in time is displayed in Figure 3.32. As reported in [136], the barotropic generation

of vorticity and the eventual rollups are evident. The contours of density and the mesh

evolution are displayed in Figure 3.33. As can be seen from the figure, the mesh adaptation

follows the interface and the shock maintaining high precision around discontinuities, with

coarse base mesh elsewhere. The OR for this calculation is only 5%, resulting in savings

in computational memory.

3.2.2.6 Mach 1.22 Shock Interacting with Cylindrical (R22 ) Gas Bubble

High resolution computation of a weak shock (Mach 1.22) interacting with a cylin-

drical R22 bubble in air is considered. The initial planar shock is located at x = 155 mm in

a computational domain of size 200 mm × 44.5 mm. The cylindrical bubble of diameter

50 mm is initially located at x = 125 mm. Due to the symmetry of the problem only one

half of the bubble was modeled. The initial conditions for the shocked and un-shocked air
 114

along with the conditions inside the bubble are given below:





 (1.3764, 1.5698, −0.394, 0.0, 1.4) x ≥ 155.0,




(ρ, P, u, v, γ) = (1.0, 1.0, 0.0, 0.0, 1.4) x < 155.0





 q

(3.15385, 1.0, 0.0, 0.0, 1.249) (x − 125)2 + (y)2 ≤ 50


The right end of the domain is prescribed with the Dirichlet condition corresponding to the

post-shocked air. The left end of the domain is prescribed with Neumann conditions, the

top and bottom portions of the domain with reflective conditions. The simulations were

evolved with a base mesh ∆xg = 1.0, with 4 levels of mesh refinement (corresponding

to 800 points per diameter). Due to the weak strength of the shock, the threshold for the

refinement criteria were set at δ1 = 0.01 & δ2 = 0.01.

Since the impedance offered by the R22 bubble is comparable to that of post-

shocked air ((ρc)R22 − (ρc)air ≈ 0.09), the shock-interface interaction resolves into a

transmitted shock wave and a reflected shock wave. As the speed of sound is greater in

the air medium, the incident shock wave travels faster than the weak refracted wave system

(Figures 3.34(a) & 3.34(b)). Furthermore, the velocity of air behind the incident shock

wave is greater than the velocity behind the transmitted shock wave inside the bubble. As

pointed out in [2], this velocity difference introduces a counter-clockwise torque on the in-

terface of the bubble which later develops into Kelvin-Helmholtz (KH) instability at the in-

terface. In addition, the interface is also subjected to Richtmyer-Meshkov instability (RM)

due to the interaction of the shock system with the curved front of the bubble. The unsta-

ble interface eventually rolls up to form vortices and fragments of vortices on the interface

(Figures 3.34(c) & 3.34(d)). To accurately resolve these vortical structures generated on
 115

the interface, extremely fine grid resolution is required as the sharp interface calculations

are devoid of numerical diffusion phenomena that are common with approaches such as

the diffuse interface approaches [38, 125]. As shown in Figure 3.34(d) & 3.35(b), with the

current LMR scheme the vortical patterns and the shock systems are captured. Also seen

in the figure is the R22 jet formed in the trailing edge of the bubble. As pointed out in [2],

due to the formation of the compression wave inside the bubble and the focussing effect

of the incident shock wave, there is a steep rise in the pressure at the trailing edge of the

bubble. This steep rise in pressure causes the onset of a jet at the trailing edge of the bubble.

In Figure 3.35, the topology of the interface at different instants in time are plotted. The

figure clearly shows the instabilities and the R22 jet formed on the interface. The dynamic

adaptation of the mesh is apparent from Figure 3.34. The occupancy ratio computed for

this simulation is 17.3 %.

3.2.2.7 Shock Interacting with Spherical Water Droplet

Shock-droplet interaction is a typical multi-material problem with wide range of

applications such as corrosion effects on turbine blades, fuel-coolant interaction in nuclear

reactors, ablation of space vehicles during re-entry etc. [39]. Effect of planar shock waves

interacting with cylindrical water column(s) have been investigated and well documented

in the past [38, 39, 89, 136]. Here the interaction of planar shock waves with a spherical

water droplet suspended in ambient air is investigated. Three different shock strengths,

Mach 1.47, 3.0 and 6.0 are considered. The spherical liquid mass of radius 0.105 mm is

located at (0.0,0.5) in a 0.5 mm × 1.0 mm domain. Euler equations in the axisymmetric

 116

form are solved for this problem. The simulations are carried out on a base mesh of size

1
∆xg = 200
with 3, 4 and 5 levels of mesh refinement (corresponding to 5.12 million grid

points for the finest level).

The shock systems observed for the interaction of planar shock wave with spherical

and cylindrical droplets are identical, except for the strengths of the reflected and refracted

shock systems. The configuration forms a curved gas-liquid slow-fast interface [136]. As

pointed out in [39], the shock wave upon impacting the water column transmits only a small

fraction of the energy into the water medium due to large acoustic impedance. As a result

of this impedance mismatch, a reflected and a refracted shock systems are generated. The

refracted weak shock wave on the water side travels faster than the incident shock wave

and is reflected back as a “relief wave” by the curved interface. As pointed out by Chang

et al [38], due to the curvature of the interface these rarefaction waves later coalesce to

form a shock wave. Thus the refracted shock wave gets trapped inside the water column

by undergoing multiple reflections off the curved interface and hence does not affect the

surrounding air. In what follows, the shock diffraction patterns generated for different

shock strengths are briefly examined.

Effect of Mach 1.47 Shock Interacting with Spherical Water Droplet:

The threshold for the refinement criteria were set at δ1 = 1.0 and δ2 = 75.0. The

results from the current calculations with 5 levels of mesh refinement are displayed in Fig-

ure 3.36. The reflected, refracted and the incident shock systems along with the “peeling

off” effect on the bubble (reported in [136]) are clearly visible. In Figure 3.36(d), the pres-
 117

sure distribution along the axis of symmetry are plotted for both the cylindrical and the

spherical bubble. In comparison with the cylindrical water column, the strengths (pressure

ratio) of the reflected and the transmitted shock waves are significantly lower for the spheri-

cal droplet (three-dimensional effect). Figure 3.36(d) also confirms the mesh independence

of the solution. The OR for this calculation is 10.0 %.

To validate the current approach, a comparison of the unsteady drag coefcient com-

puted on the surface of a cylindrical droplet is presented in Figure 3.37. The drag coefficient

is compared with that of Igra et al [89], Chen [39] and Terashima et al [187]. The drag co-

efficient distribution predicted with the current method is in good agreement with available

solution.

Effect of Mach 3 Shock Interacting with Spherical Water Droplet:

For the Mach 3 shock wave, the threshold for the refinement criteria were increased

to δ1 = 10.0 and δ2 = 150.0. The results from the present calculations with 4 levels of mesh

refinement are shown in Figure 3.38. The reflected shock wave in the air side forms a “bow

shock”, similar to that of the shock negotiating a solid obstacle, which travels upstream to

transform to a Mach reflection. Figure 3.38 verifies the features of this peculiar type of

anomalous refraction [136]. During the initial stages of shock interaction, a striking simi-

larity was observed with the wave patterns generated by the shock wave diffracting a solid

cylinder (results for shock negotiating cylindrical obstacle are presented in section 5.3.2.3).

The initial Regular Refraction transforms to Mach Reflection forming a Mach stem (Fig-

ure 3.38(a)). In Figure 3.38(d), the shock systems for the cylindrical and the spherical
 118

droplet are compared. The enlarged view shown in the insert confirms the formation of the

Mach stem and the roll up of the slipline emanating from the Mach stem. The OR for this

calculation is 11.3 %.

Effect of Mach 6 Shock Interacting with Spherical Water Droplet:

The threshold for the refinement criteria were set at δ1 = 10.0 and δ2 = 150.0.

The shock diffraction patterns for this case are displayed in Figure 3.39. A double Mach

reflection system, as opposed to a single Mach system, is formed. The double Mach stem

is not visible in Figure 3.39(a), but is visible in Figure 3.39(c) with additional mesh resolu-

1
tion (with ∆xf = 3200
). The pressure distribution along the line of symmetry displayed in

Figure 3.39(d) confirms convergence of solution with mesh refinement. The increase in re-

finement level only accentuates fine structures that are otherwise not visible. The strengths

of the reflected and refracted shock systems are comparable. The OR for this calculation is

13.3 %.

3.2.2.8 Collapse of Cylindrical Air Cavity in Water

In this example, numerical computation of the experimental work carried out by

Bourne et al [28] is presented. Numerical analysis for this problem has been performed

and reported in the past [18, 73, 80, 87]. The initial configuration of the system is given
 119

below:





 (1.31, 19000, 67.32, 0.0, 5.5) for post-shocked water i.e.x ≤ 2.6,




(ρ, P, u, v, γ) = (1.0, 1.0, 0.0, 0.0, 5.5) for pre shocked water i.e.x > 2.6,





 q

(0.0012, 1.0, 0.0, 0.0, 1.4) for (x − 6)2 + (y − 6)2 ≤ 3


A cylindrical air cavity of 6 mm diameter, which is located at the center of the computa-

tional domain, is impacted by a strong shock corresponding to a pressure ratio of 19000.

The shock wave impacting the cylindrical cavity resolves to form a strong reflected “relief”

wave and refracted shock wave. In Figure 3.40, the numerical Schlieren image at the time

of break-up (t = 0.0316) is displayed. The reflected rarefaction wave generates a high ve-

locity jet which impacts the air cavity, which in turn generates an intense blast wave that

splits the air cavity. The numerical Schlieren image reveals oscillations in the density field

close to the interface. These oscillations are present only close to the interface which in-

dicates that these are numerical artifacts arising from the interface treatment technique and

are not physical phenomena. In order to investigate the underlying cause for these oscil-

lations, the one-dimensional version of the problem is analyzed. The initial conditions for

the one-dimensional case are as follows:




(1.31, 19000, 67.32, 0.0, 5.5) for post-shocked water i.e. x ≤ 0.5,

(ρ, P, u, v, γ) =



(0.0012, 1.0, 0.0, 0.0, 1.4) for un-shocked air i.e x > 0.5

A strong shock corresponding to a pressure ratio of 19000 resides on the interface sepa-

rating water and air. The initial conditions correspond to a water-air shock tube problem

which resolves into a weak shock wave traveling in air and a strong rarefaction wave into

the water medium, similar to the two-dimensional version considered above.

 120

Figure 3.41 shows the density and the pressure variation for this case. As can be

seen from the insert in the figure, both density and pressure fields are under-predicted. It

was shown in section 3.2.1, that the approach adopted in this work results in (like other

GFM approaches) consistent over-/under-heating errors that were found not to be problem

dependent. However, in this case, due to the sensitivity of the stiffened equation of state

employed for water, a small perturbation in the density field (and in the internal energy

field) may result in a negative pressure field. As the material properties of water are such

that it can sustain tensile stresses provided the pressure field remains above a certain crit-

ical limit, the computed sound speed is well within the realistic regime and hence does

not cause the break-down of the simulation. Moreover, the negative pressure is observed

close to the interface indicating that these undershoots are truly due to numerical errors

introduced by the interface treatment technique. Furthermore, these under-predictions in

the flow variables were found to decrease with grid refinement demonstrating a strong de-

pendence on the mesh resolution. Thus it is believed that the oscillations revealed by the

two-dimensional simulation and the corresponding under-shoots in the pressure and den-

sity fields in the one-dimensional version, are due to the numerical errors generated at the

interface by the interface treatment technique employed.

The problem is now repeated here with dynamic mesh adaptation. In this simula-

tions, a base mesh of size ∆xg = 0.16 was chosen with 4 levels of mesh refinement (with

effective grid resolution of about 1.44 million points). The threshold for the refinement

criteria were set at δ1 = 1000.0 and δ2 = 1000000.0 (arbitrarily large value). The reflected

relief or expansion wave in the water medium, and the transmitted shock wave in the cavity
 121

are visible in Figure 3.42(a). The pressure ratio across the transmitted shock wave is 33.28

which is 570 times smaller than the pressure ratio across the incident shock wave. Since a

strong expansion wave is reflected off the cavity surface, in addition to capturing the wave

patterns near the air cavity, attention is also paid in capturing the large gradients across

the reflected expansion waves. Figures 3.42(a) & 3.42(d) show that the mesh evolution

follows the head and tail of the rarefaction wave in addition to the transmitted and the in-

cident shock waves. The formation of water jet at the center of the bubble is clearly visible

in Figure 3.42(b). The high speed jet of water results in splitting the air cavity which in

turn sends out a blast wave (Figure 3.42(c)). The enlarged view of the numerical Schlieren

image and the Mach contours at the instant at which the splitting occurs are displayed in

Figures 3.43(a) & 3.43(b) respectively. Figure 3.43(d) shows the velocity of the tail (rear

end) of the cavity as a function of time. The velocity of the water jet at the instant of

collapse is about 2.82 km/s which agrees with the values 2.8 km/s and 2.85 km/s reported

by Hu et al [87] and Nourgaliev et al [136] respectively. The topology of the interface at

different instants in time are plotted in Figure 3.43(d). The plots are in good agreement

with that of Hu et al [87]. Since emphasis was made in capturing weak - strong gradients

occurring near the head and tail of the strong reflected wave, the OR computed for this

simulation is 46.6 %.

3.2.2.9 High Speed Impact of a Liquid Droplet

In this example, high speed impact of a liquid droplet over a flat substrate is con-

sidered. Impact of high speed droplets and the resultant wave mechanisms are important
 122

in thermal spray and coating technologies [40, 78]. A 200µm diameter liquid droplet is

injected with a high velocity of 500 m/s, towards a rigid flat substrate located at the bot-

tom. The initial conditions, normalized with respect to the properties of surrounding air are

given below:



(1000.0, 1.0, 0.0, −1.571, 5.5) for the liquid droplet,

(ρ, P, u, v, γ) =



(1.0, 1.0, 0.0, −1.571, 1.4) for the surrounding air.

Figure 3.44 shows the geometrical setup for this problem. The left boundary is prescribed

with a symmetry condition and the bottom boundary with reflective wall boundary condi-

tion. The top and the right boundaries are prescribed with a Neumann condition. Euler

equations in the axisymmetric form are solved. Haller et al [76–79] have provided a de-

tailed survey of this problem. They have provided theoretical predictions for the jetting

time that matched well with their corresponding numerical model. However, it is not clear

from their analysis how the surrounding air was modeled. In this example the flow inside

the droplet as well as in the surrounding air is considered; i.e. both the droplet and the

surrounding air are injected with the same velocity towards the wall. The snapshots of

the numerical Schlieren image from the current simulation are shown in Figure 3.45. The

fluid droplet, after impacing the rigid substrate, moves off the surface at right angles to its

approach path. This tangential movement of the droplet is due to the momentum transfer

from the droplet to the impacted surface [212]. The jetting of the droplet continues until

all of its kinetic energy is lost in the process. This is in direct contrast to the impact of

an elastic object, such as an elastic ball, which rebounds after transferring its momentum

to the impacted surface [212]. The figures indicate that most of the key features such as
 123

lateral jetting of the liquid droplet along the contact edge, the shock wave traveling in the

droplet and in the surrounding air, the reflected rarefaction wave and focusing of the rar-

efaction wave at the top end of the droplet are captured well. It is worth mentioning that the

fine mesh calculation shown in this example were carried out with roughly million points

(935000 grid points) which is half the number of mesh points used by Haller et al (2 mil-

lion grid points) who used a front tracking solution procedure (FronTier) due to Glimm et

al [64, 65].

The above calculations are repeated with mesh adaptation. The solution is evolved

with 4 levels of refinement on a base mesh of size ∆x = 1.75µm (an effective resolution of

1.13 million grid points). The snapshots of the numerical Schlieren image from the current

simulation are shown in Figure 3.46. The mesh adaptation to the shock diffraction fea-

tures is evident in the figure. The solution also agrees well with the non-LMR calculation

presented in Figure 3.46

3.3 Conclusions

The interaction of strong shock waves with embedded fluid-fluid interfaces in com-

pressible flows was studied. The interface is retained as a sharp entity by virtue of the Ghost

Fluid Method (GFM). The failure of the original GFM, particularly with strong shock in-

teractions was demonstrated. It was shown earlier that solving a local Riemann problem

at the interface alleviated the pitfalls encountered by the original GFM approach. Hence

in this work, a simple procedure to incorporate the Riemann problem at the interface was

developed. The method was found to be robust in handling shocks of varied magnitude
 124

interacting with both gas-gas and gas-liquid interfaces. The (mass, momentum and en-

ergy) conservation error analysis carried out for this method reveal that these errors are

significantly attenuated and localized close to the interface. Under-/Over-heating errors are

mitigated but not entirely eliminated by the present approach. This is due to the intrin-

sic non-conservative nature of the ghost fluid approach in treating sharp interfaces. The

proposed method was shown to generate satisfactory solutions for several complex con-

figurations and shocks interacting with single and multiple interface(s); shocks interacting

with droplets, bubbles and free surface have been computed.

 1 1 1
∆xf ine = 200
(3 Levels) ∆xf ine = 400
(4 Levels) ∆xf ine = 800
(5 Levels)

L2 LMR U-Mesh ∆L2 LMR U-Mesh ∆L2 LMR U-Mesh ∆L2

Example 6

L2 (ρ) 0.3655 0.3654 0.0001 0.2427 0.2427 - 0.1217 0.1217 -

L2 (P ) 2.236 2.233 0.003 1.576 1.573 0.003 0.8256 0.8239 0.0017 -

L2 (u) 0.1146 0.1144 0.0002 0.074 0.074 - 0.0392 0.039 -

L2 (S) 0.732 0.732 - 0.425 0.425 - 0.2203 0.22 0.0003

Example 7

L2 (ρ) E-2 8.291 8.296 147.6 3.476 3.473 32.9 1.943 1.940 28.544

L2 (P ) 23.07 23.076 0.0014 10.52 10.542 0.0265 7.384 7.384 -

L2 (u) E-2 0.1162 0.1162 - 0.0522 0.0522 0.0265 3.457 3.457 -

L2 (S) 2.254 2.257 0.0004 0.979 0.9814 0.001 0.4774 0.4778 0.0004

Table 3.1: L2 errors for the Gas-Gas Shocktube Problem and Air-Water Shocktube Problem I.
125
 126

(a) MGFM approach

(b) Procedure to construct the Riemann problem

(c) Characteristic waves

Figure 3.1: The Riemann problem: (a) Approximate Riemann solver employed in MGFM;
(b) Procedure to construct the Riemann problem at the interface; (c) Typical characteristic
wave structure for Riemann problem.
 127

(a) One-dimensional correction procedure

(b) Multi-dimensional correction procedure

Figure 3.2: Extension procedures: (a) One-dimensional version of the correction and ex-
tension procedure (b) Multi-dimensional correction and extension procedure.
 128

(a) Density

(b) Entropy

Figure 3.3: Example 1 - CASE(A): Plots of (a) density and (b) entropy for single fluid
Sod’s shock tube problem with pressure ratio PP21 = 10.
 129

Figure 3.4: Example 1 - CASE(A): Plots of (a) density and (b) entropy for single fluid
Sod’s shock tube problem with pressure ratio PP12 = 1000.
 130

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.5: Example 1 - CASE(B): Plots of (a) density, (b) pressure, (c) velocity and (d)
entropy for Woodward Colella Bang Bang problem.
 131

4
3.5
3
400
2.5
2

Energy Conservation Error

Mass Conservation Error

1.5
200
1
0.5
0 0
-0.5
-1
-1.5 -200

-2
-2.5
-400
-3
-3.5
-4 500 1000 1500 2000 2500
500 1000 1500 2000 2500
Time Step Time Step

(a) Total Mass Conservation Error (b) Total Energy Conservation Error

(c) Total Momentum Conservation Error

Figure 3.6: Example 1 - CASE(B): Plots of total (a) mass, (b)energy and (c) momentum
conservation errors for Woodward Colella Bang Bang problem.
 132

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.7: Example 2 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy
for air-helium shock tube problem. The inserts correspond to the zoomed in view of the
variation close to the interface.
 133

(a) Total Mass Conservation Error (b) Total Energy conservation Error

(c) Total Momentum Conservation Error

Figure 3.8: Example 2 - Plots of total (a) mass, (b) energy and (c) momentum conservation
errors for air-helium shock tube problem.
 134

Figure 3.9: Non-uniqueness in the numerical flux computed at the interface; Figure re-
produced from “Abgrall, R. and Karni, S., Computations of Compressible Multi-Fluids,
Journal of Computational Physics, 169, 594-623(30), 20 May 2001”.
 135

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.10: Example 3 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-water shock tube problem.
 136

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.11: Example 4 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-water shock tube problem.
 137

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.12: Example 5 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
water-air shock tube problem.
 138

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.13: Example 6 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-gas shock tube problem at time T = 0.03 units.
 139

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.14: Example 7 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
air-water shock tube problem (I) at time T = 0.0015 units.
 140

(a) Density (b) Pressure

(c) Velocity (d) Entropy

Figure 3.15: Example 8 - Plots of (a) density, (b) pressure, (c) velocity and (d) entropy for
gas-water shock tube problem (II) at time T = 0.001 units.
 141

Figure 3.16: Initial configuration for single phase spherical Riemann problem.
 142

(a) t = 0.5, R-GFM

(b) t = 0.5, Shock-Capturing Scheme (without GFM)

(c) t = 0.8, R-GFM

(d) t = 0.8, Shock-Capturing Scheme (without GFM)

Figure 3.17: Snapshots of numerical Schlieren image obtained with and without GFM
treatment for the initially spherical contact discontinuity.
 143

1.1

0.9
Shock Capturing
Riemann GFM
0.8
DENSITY

0.7

0.6

0.5

0.4

0.3

0 0.2 0.4 0.6 0.8 1 1.2 1.4

X

1.2

1.1
PRESSURE

Shock Capturing
0.9 Riemann GFM

0 0.2 0.4 0.6 0.8 1 1.2 1.4

X

Figure 3.18: Spherical Riemann problem: Comparison of (a) density and (b) pressure vari-
ation at y = 0.6 at T = 0.5.
 144

Figure 3.19: Initial configuration for a Mach 1.22 shock impinging on a cylindrical helium
bubble.
 145

(a) Numerical Schlieren Image at T = 85 µs

(b) Numerical Schlieren Image at T = 188 µs

(c) Numerical Schlieren Image at T = 427 µs

(d) Numerical Schlieren Image at T = 803 µs

Figure 3.20: Snapshots of numerical Schlieren image at different instants in time for a
Mach 1.22 shock impinging on a cylindrical He bubble: (a) T = 85 µs, (b) T = 188 µs, (c)
T = 427 µs, and (d) T = 803 µs.
 146

(a) Density Contours at T = 85 µs

(b) Density Contours at T = 188 µs

(c) Density Contours at T = 427 µs

(d) Density Contours at T = 803 µs

Figure 3.21: Snapshots of density contours at different instants in time for a Mach 1.22
shock impinging on a cylindrical He bubble: (a) T = 85 µs, (b) T = 188 µs, (c) T = 427 µs,
and (d) T = 803 µs.
 147

2
85 µ s
188 µ s 2 85 µ s
1.8
256 µ s 188 µ s
341 µ s 256 µ s
427 µ s 341 µ s
1.6
803 µ s 1.8 427 µ s
803 µ s
1.4

1.2 1.6
DENSITY

P
1.4
0.8

0.6 1.2

0.4
1
0.2

0 0.8
0 50 100 150 200 250 300 0 50 100 150 200 250 300
X X

(a) Density Variation along Horizontal Sym- (b) Pressure Variation along Horizontal Sym-
metry Line metry Line
2

1.8

1.6
Pressure

1.4

1.2

1
0 100 200 300 400
Time (µs)

(c) Pressure History at 3mm Downstream of

the Initial Bubble Location

Figure 3.22: Plots of density (Figure 3.22(a)) and pressure (Figure 3.22(b)) along the hor-
izontal line of symmetry, and pressure as a function of time at 3mm downstream of the
initial bubble location (Figure 3.22(c)). The insert displayed in Figure 3.22(c) corresponds
to the plot obtained by Marquina et al[2003].
 148

Figure 3.23: Initial configuration for a Mach 6 shock interacting with a cylindrical helium
bubble.
 149

(a) Numerical Schlieren Image at T = 15.3 µs

(b) Numerical Schlieren Image at T = 38.25 µs

(c) Numerical Schlieren Image at T = 61.2 µs

(d) Numerical Schlieren Image at T = 78 µs

Figure 3.24: Snapshots of numerical Schlieren image for a Mach 6 shock interaction with
a cylindrical He bubble at different instants in time.
 150

Figure 3.25: Topology of the interface (zero level set field) at different instants in time.
 151

Figure 3.26: Initial configuration for underwater explosion near a free surface.
 152

(a) T = 0.00892

(b) T = 0.024

(c) T = 0.045

Figure 3.27: Snapshots of numerical Schlieren image and the corresponding density con-
tours at different instants in time.
 153

Figure 3.28: Initial configuration for underwater explosion near a free surface with im-
mersed structure.
 154

(a) Numerical Schlieren Image at T = 0.01 (b) Pressure Contours at T = 0.01

(c) Numerical Schlieren Image at T = (d) Pressure Contours at T = 0.0197

0.0197

Figure 3.29: Snapshots of numerical Schlieren image (left) and corresponding pressure
contours (right) at different instants in time.
 155

4000
Horizontal Force Component
Vertical Force Component 200

3000

2000
Moment
Force

1000

-200

-1000

0 0.005 0.01 0.015 0.02 0.025 0.03 0 0.01 0.02 0.03

TIME TIME

(a) Horizontal and Vertical Force Compo- (b) Moment on the submerged square
nents

Figure 3.30: Plots of non-dimensionalized force components (Figure 3.30(a)) and moment
(Figure 3.30(b)) exerted on on the submerged square structure.
 156

Liu et. al. Liu et. al.

Current Simulation Current Simulation
4000 4000
Horizontal Force Component - Fx

Vertical Force Component - Fy

2000 2000

0 0

-2000 -2000

0 0.005 0.01 0.015 0 0.005 0.01 0.015

Time Time

(a) Horizontal Force Component (b) Vertical Force component

Figure 3.31: Comparison of non-dimensionalized horizontal (Figure 3.31(a)) and vertical

(Figure 3.31(b)) force components exerted on the submerged square structure.
 157

(a) Density contours for LMR and uniform mesh solutions at time T = 2.25

(b) Topology of the interface at different instants in time

Figure 3.32: Richtmyer-Meshkov Instability (RMI): (a) Comparison of density contours

for LMR calculations, with 3 levels of mesh refinement, with the corresponding uniform
fine mesh calculation (b) Topology of the interface at different instants in time.
 158

(a) T = 2.5

(b) T = 4.5

(c) T = 6.5

(d) T = 9.2

Figure 3.33: Snapshots of mesh evolution and density contours for the RMI at different
instants in time.
 159

(a) T = 28µs

(b) T = 106µs

(c) T = 250µs

(d) T =402µs

Figure 3.34: Snapshots of mesh evolution and numerical Schlieren image for a planar shock
(Mach 1.22) interacting with R22 cylindrical bubble at different instants in time.
 160

(a) Interface Evolution

(b) RM/RT Instabilities

Figure 3.35: Mach 1.22 shock interacting with R22 cylindrical bubble: Enlarged view of
the (a) interface evolution at different instants in time and (b) instabilities occurring at the
interface.
 161

(a) Mesh and numerical Schlieren image at (b) Mesh and numerical Schlieren image at
T = 0.376µs T = 0.75µs

(c) Comparison of numerical Schlieren im- (d) Pressure distribution along the line of
age at T = 1.128µs symmetry at T = 0.015µs

Figure 3.36: Mach 1.47 shock interacting with spherical water droplet: (a), (b) and (c)
Snapshots of mesh evolution and numerical Schlieren image computed on a base mesh
1
∆xg = 200 with 5 levels of mesh refinement (d) Pressure distribution along the line of
symmetry at T = 0.015µs.
 162

Figure 3.37: Comparison of unsteady drag coefficient computed on the surface of a cylin-
drical droplet
 163

(a) Mesh and numerical Schlieren image at T = (b) Mesh and numerical Schlieren image at T =
0.02µs 0.037µs

(c) Mesh and numerical Schlieren image at T = (d) Comparison of numerical Schlieren image at
0.06µs T = 0.015µs

Figure 3.38: Mach 3 shock interacting with spherical water droplet: (a), (b) and (c)
Snapshots of mesh evolution and numerical Schlieren image computed on a base mesh
1
∆xg = 200 with 4 levels of mesh refinement (d) Comparison with cylindrical water column
1
computed on a base mesh ∆xg = 200 with 4 levels of mesh refinement.
 164

(a) Mesh and numerical Schlieren image (b) Mesh and numerical Schlieren image
at T = 0.01µs at T = 0.026µs

(c) Comparison of numerical Schlieren (d) Pressure distribution along the line of
image at T = 1.128µs symmetry at T = 0.0031µs

Figure 3.39: Mach 6 shock interacting with spherical water droplet: (a) and (b) Snapshots
1
of mesh evolution and numerical Schlieren image computed on a base mesh ∆xg = 200
with 4 levels of refinement (c) Comparison with cylindrical bubble computed on a base
1
mesh ∆xg = 200 with 4 levels of refinement; (d) Pressure distribution along the line of
symmetry at T = 0.0031µs.
 165

Figure 3.40: Numerical Schlieren image at T = 0.0316, for the collapse of cylindrical air
cavity under the impact of a strong shock wave in water
 166

(a) Density

(b) Pressure

Figure 3.41: One dimensional water-air shock tube problem for the collapse of cylindrical
air cavity in water. Inserts shown in the figures correspond to the variation close to the
interface.
 167

(a) Numerical Schlieren image at T = 0.005 (b) Numerical Schlieren image at T = 0.016

(c) Numerical Schlieren image at T = 0.03 (d) Mesh topology at T = 0.03

Figure 3.42: Numerical Schlieren image for the solution with 4 levels of mesh refinement
on a base of size ∆xg = 0.16.
 168

(a) Numerical Schlieren image at T = 0.03 (b) Mach contours at T = 0.03

(c) Velocity of the tail of the bubble (d) Evolution of the interface

Figure 3.43: Collapse of cylindrical air cavity:(a) Close in view of the numerical Schlieren
image at time T = 0.03 units; (b) Enlarged view of the Mach contours at time T = 0.03
units; (c) Velocity of the tail (rear end) of the bubble as a function of time; (d) Evolution of
the interface (0-levelset).
 169

Figure 3.44: Initial configuration for high speed droplet impacting a flat substrate.
 170

(a) Numerical Schlieren Image at 21.7 ns

(b) Numerical Schlieren Image at 37.54 ns

(c) Numerical Schlieren Image at 84.32 ns

(d) Numerical Schlieren Image at 118.8 ns

Figure 3.45: Snapshots of numerical Schlieren image illustrating the wave patterns gener-
ated from the impact of a high velocity droplet on a flat substrate at different instants in
time.
 171

(a) Numerical Schlieren Image at 31.4 ns

(b) Numerical Schlieren Image at 62.8 ns

(c) Numerical Schlieren Image at 94.2 ns

(d) Numerical Schlieren Image at 125.6 ns

Figure 3.46: Snapshots of numerical Schlieren image illustrating the wave patterns gener-
ated from the impact of a high velocity droplet on a flat substrate at different instants in
time.
 172

CHAPTER 4
SHARP INTERFACE TREATMENT FOR ELASTO-PLASTIC SOLIDS

4.1 Introduction

The phenomenology of high speed impact and penetration mechanics are of inter-

est for plethora of applications including munition-target interactions [17, 212], geological

impact dynamics [8, 150], shock-processing of powders [21, 22], vehicular collision crash-

worthiness [23,214], formation of shaped charges [195,199], etc.. In such applications, the

propagation of elasto-plastic stress waves in the material dictates the deformation of the

colliding bodies. The wave propagation in the interacting media is highly nonlinear, and

often lead to localized phenomena such as shear bands, crack propagation, and complete

failure of the material interface. In addition, the stress and strain fields are related through

nonlinear elasto-plastic yield surfaces, the models for which must be included in the gov-

erning equations [192]. Capturing the response of the material interface subject to such

intense loading conditions is the subject of this chapter.

In this chapter, the sharp interface GFM approach described in section 1.6 is ex-

tended to resolve the response of elasto-plastic solids due to high velocity impact and high

intensity shock loading conditions. As pointed out in chapter 1, the fundamental chal-

lenges to a simulation capability designed to solve problems in the physical phenomena

listed above arise from the large deformations occurring at high strain conditions suffered

by the interacting materials [190]. Traditionally, the methods that have been used to solve

such large deformation, finite strain transient problems have been termed hydrocodes [24].
 173

Hydrocodes may be either based on a Lagrangian formulation, such as in EPIC and DYNA,

where a moving unstructured mesh is used to follow the deformation, or an Eulerian for-

mulation, such as in CTH, where a fixed mesh is used and the boundaries are tracked

through the mesh [190]. The broad range of available hydrocodes has been reviewed by

Anderson [9] and Benson [24]. In this work, an Eulerian fixed grid approach is proposed

to resolve the multi-material high speed impact, collision, penetration and void collapse

problems.

In contrast to the previous Cartesian grid approaches [190, 192], the present effort

is attractive for the following reasons:

i. The interfaces are tracked and represented via traditional level set approach as op-

posed to the hybrid particle level set technique employed in [190] or the marker

particles approach employed in [192].

ii. The GFM approach is applied to treat and represent the embedded interface. In

contrast to [190], where the discretization scheme was modified to incorporate the

boundary conditions at the interface, the present method decouples the discretization

scheme from the interface capturing technique.

iii. A novel approach to infuse the boundary conditions to the ghost cells is proposed.

The interaction of the embedded boundaries with each other and the evolution of free

boundaries is treated by application of appropriate boundary conditions at the result-

ing material-material and material-void boundaries. The proposed method carefully

takes into account of the subcell position and topology of the interface.
 174

iv. A simple and a robust algorithm for tracking and detecting collision is developed.

The algorithm is shown to be effective in resolving collisions at very high speeds of

interaction.

v. Local mesh adaptation is employed for efficient resolution of disparate length scales

and strong gradients.

vi. Plethora of numerical examples encompassing a broad spectrum of speeds of inter-

action (up to 7 km/s) are presented. Although the numerical examples illustrated in

this chapter are limited to planar two-dimensional or axisymmetric problems, the for-

mulation of the approach is carried out in three-dimensions. These examples demon-

strate the robustness and accuracy of the present approach.

4.2 Governing Equations and Constitutive Models

The governing equations and constitutive relations along with the modeling as-

sumptions are discussed at length in section 1.3. Because of the very high speeds involved

in the interaction process, the governing equation constitutes a set of hyperbolic conserva-

tion laws. When cast in Cartesian coordinates, the equations take the form given in Eq 1.13.

The governing equations are solved in its entirety, including the models for the strength of

the materials. These equations are solved in an Eulerian setting on a fixed Cartesian mesh.

Eigenvalue analysis performed in [194] ensure hyperbolicity of the governing equations for

the impact velocity regime of interest. The equations are solved using a third-order TVD-

based Runge-Kutta scheme for time integration and third-order convex ENO scheme [120]

for spatial discretization.

 175

4.3 Capturing the Response of the Material Interface

One of the exemplifying feature of the problems solved in this work is that the

embedded material interface is subject to very severe deformations at high strain-rate con-

ditions. These deformations occur almost instantaneously (typically O(µs)) and hence re-

quire meticulous treatment and resolution of the interface. In this work, the response of the

material interface subject to high velocity impact and shock loading conditions is captured

using the GFM approach outlined in section 1.6. As mentioned before, the GFM approach

requires a definition of band of ghost points corresponding to each phase of the interacting

material. When this band of ghost points are supplied with flow properties the ghost field

together with the real field constitute a single flow field wherein a single equation of state

can be used. Thus in the GFM approach it boils down to populating the ghost cells with

suitable flow conditions.

4.3.1 Populating the Ghost Points

To populate the ghost cells appropriate boundary conditions need to be applied at

the interface location. The conditions that must be enforced at the interface to populate

the ghost cells depend on the material that the interface separates. For instance, the inter-

face could separate a material from another material or void. Depending on the nature of

the interface and the material that the interface separates, the conditions are modified and

the ghost cells are populated accordingly. Section 1.6.2.2 enlists the conditions that must

enforced for interfaces separating different materials. The set of conditions that must be
 176

enforced at the interface can be concatenated in the following general set of conditions.

Dirichlet condition: ΨI = lI (4.1)

∂ΨI
Neumann condition: = mI (4.2)
∂n
Continuity of flow variables: [ΨI ] = nI (4.3)

Discontinuity of flow variables: [ΨI ] 6= 0 (4.4)

where ΨI correspond the flow variables for which the boundary conditions are applied.

The ghost points are defined with flow conditions such that the real field along with the

corresponding ghost field satisfy the boundary conditions (Eqs 4.1 - 4.4) accurately at the

interface. In what follows, the numerical procedure to apply the boundary conditions is

briefly reviewed.

4.3.1.1 Applying the Boundary Conditions at the Material Interface:

To apply the boundary conditions (Eqs 4.1 - 4.4) at the interface, a reflective bound-

ary condition-based approach is adopted. For instance, to define the ghost states at point P

(Figure 4.1), a probe is inserted to identify the reflected point IP1 and the point IP on the

interface. Points IP and IP1 can be identified by using the level set distance function:

~ IP = X
X ~ P + | φP |N
~P (4.5)

~ IP1 = X
X ~ P + 2| φP |N
~P (4.6)

~ is the position vector, φP the level set value at point P and N

where X ~ P is the normal vector

at point P. Once points IP and IP1 are identified, a Vandermonde matrix is constructed on

the surrounding interpolation nodes to determine the flow properties at the ghost point
 177

P. For instance, the surrounding interpolating points 1,2,3 and IP are determined for the

reflected point IP1 (Figure 4.1). At the point IP on the interface, either the value of the

flow variables (Dirichlet conditions) or the flow gradient (Neumann type conditions) is

available. Thus it is necessary to embed the appropriate boundary conditions to complete

the interpolation procedure. For bilinear interpolation we have

Ψ = a1 + a2 x + a3 y + a4 xy (4.7)

where (x,y) are the coordinates of the surrounding interpolation nodes. For Dirichlet con-

dition on IP, the Vandermonde matrix takes the following form:

    
 1 x 1 y1 x1 y1   a1   Ψ1 
    
    
 1 x y2 x2 y2  a   Ψ 
 2  2   2 
  =  (4.8)
    
 1 x y3 x3 y3  a   Ψ 
 3  3   3 
    
    
1 xIP yIP xIP yIP aIP lI

For Neumann condition, the matrix is modified as follows

    
 1 x1 y1 x1 y1   a1   Ψ1 
    
    
 1 x y2 x2 y2  a   Ψ 
 2  2   2 
  =  (4.9)
    
 1 x y3 x3 y3  a   Ψ 
 3  3   3 
    
    
0 nx ny nx yIP + ny xIP aIP mI

The last row of the coefficient matrix in Eq (4.9) is obtained by differentiating Eq (4.7),

noting that
∂ψ ∂ψ ∂ψ
= nx + ny (4.10)
∂n ∂x ∂n

where nx and ny are the normal vector components and mI corresponds to the value of the

normal gradient at the point IP. Once the coefficients are determined, the flow properties at
 178

the reflected point can be deduced using Eq (4.7). For flow variables that are continuous

across the interface, the ghost states at point P are obtained by directly injecting the corre-

sponding flow properties from the real fluid present at point P. The discontinuous variables

are extended to the ghost points via a constant extrapolation approach [15]. Alternatively,

since a constant extrapolation approach ensures zero gradient condition at the interface, the

ghost states corresponding to the discontinuous flow variables can be determined by en-

forcing Neumann condition with mI = 0 in Eq 4.9. Once the flow conditions at the ghost

points adjacent to the interface are determined, the values corresponding to the interior

ghost cells are obtained by solving the extension equation.

4.3.2 Collision Detection Algorithm

In the present work, material interfaces are expected to collide with other inter-

faces or collapse and fragment. Such events need to be tracked and appropriate interface

conditions applied on the interacting parts of the interface. In the current framework, mul-

tiple objects with different material properties are described by different level set functions.

Only one material can possess a given computational grid point at any time, whether it be

void or solid material.

At the beginning of the calculation, the material enclosed inside and outside the

interface are identified as solids, liquids, voids or elsasto-plastic solids. Then one of these

material is classified as bulk material indicating that the objects are immersed in the sur-

rounding bulk material as shown in Figure 4.2(a). Thus unless a collision is detected,

the embedded object is said to interact with the surrounding bulk material and the cor-
 179

responding interface conditions are enforced to populate the ghost cells. For instance, in

Figure 4.2(a) an elasto-plastic solid object is immersed in surrounding bulk material, which

in this case corresponds to a free-surface. Thus for the interface separating the elasto-plastic

solid and the bulk free surface, the conditions corresponding to the material-void interface,

described in section 4.3, are enforced to populate the ghost points.

Various situations may arise when two different objects move toward each other

as shown in Figure 4.2(b). Collisions between multiple objects are inevitable when the

approaching objects are in close proximity. In such cases the interface conditions that must

be applied to populate the ghost points must be different from the material-bulk material

conditions. Thus it is necessary that the colliding objects are detected and the interface is

resolved accordingly. For this purpose the following algorithm is adopted.

To begin with the points straddling the material interface (called the interfacial

points) are tagged as “bulk nodes”, indicating that the interfacial points are nodes that

always interact with the surrounding bulk material. To detect collision, the distance be-

tween the interface of the current object and other interfaces are computed using the level

set functions associated with each interface. For the sake of illustration consider the case

of two levels sets approaching each other as shown in Figure 4.2(b). At any interfacial grid

point, the values of each level set at that point represents the distance to their respective

interfaces. Therefore, the difference in these values represents the distance between the

two interfaces. If the distance between two approaching level sets is less than the tolerance

limit, then the point is marked as a “colliding node”. For instance, with reference to Fig-

ure 4.2(b), all the interfacial points are tagged as bulk nodes. Then the collision algorithm
 180

is invoked to determine the colliding nodes. In Figure 4.2(b), the potential collision point

P is identified by computing the approaching distance between the two colliding objects.

The approaching distance in this case can be computed from the value of the level set field

at point P as indicated in the figure. If this approaching distance is less then the tolerance

limit then point P is tagged as colliding node. Once a point is marked as colliding node,

the conditions corresponding to the material-bulk material are no longer enforced at that

point. Instead the condition corresponding to the material-material interface are enforced

to populate the ghost points. This process is repeated for each level set after the level set is

advected. The tolerance limit to detect collisions are set at κ∆x where κ corresponds to the

CFL number with which the interface is advected. This prevents inter-penetration of level

sets as any grid point can be contained within one level set.

For two level sets, each defining a moving boundary under impending collision,

the normal velocity at the interface is continuous through the impact surface, whereas the

tangential component may be discontinuous as sliding is permitted. No surface friction is

accounted for in this work but can easily be included in the present framework. Thus unlike

in [15], the level set field and level set velocities are unaltered and the inter-penetration of

the level set field is prevented.

 181

4.4 Numerical Examples

4.4.1 Impact of a Copper Rod over a Rigid Substrate - Axisymmetric Taylor Bar

Experiment

4.4.1.1 Impact at 227 m/s - Validation

Taylor test [185] on a copper rod is considered. The Taylor test analysis have been

used as a canonical test problem to verify and validate numerical and experimental obser-

vations. Taylor [185], after an extensive analysis on the impact of cylindrical specimen

over a rigid flat substrate, depicted the deformation process as a sequence of elastic and

plastic wave propagation into the cylinder. In the two-dimensional setting, a cylindrical

rod made of copper with an initial radius of 3.2 mm and a length of 32.4 mm impacts a

rigid flat substrate at 227 m/s (Figure 4.3). A computational domain of radius 8 mm and

length 34.0 mm is chosen for this simulation. The top and right end of the computational

domain are prescribed with Neumann conditions. The presence of rigid wall on the bottom

end of the domain is modeled by enforcing a reflective condition. The left end of the do-

main is prescribed with symmetry condition (with Sxy = 0). The deformation of the rod is
kg
presumed to be axisymmetric. The rod has an initial density of 8930 m3 , Young’s modulus

E = 117GP a, Poisson’s ratio ν = 0.35, and yield stress σY = 400M P a. The material

is assumed to harden linearly with a plastic modulus of 100M P a. The calculations are

carried out up to a time of 80µs (at which point nearly all the initial kinetic energy has been

dissipated as plastic work) on a base mesh of size ∆xG = 0.5mm with 3, 4 and 5 levels of

mesh refinement. The CFL number was set to 0.4 for this computation.

The impact of the rod with the bottom rigid surface results in a precursor, compres-
 182

sive, elastic wave traveling in the bar followed by a slower nonlinear plastic wave front.

The elastic wave travels the entire length and the width of the rod, and is reflected off the

free surface as relief wave. The deformation of the rod ends with the reflected elastic wave

interacting with the plastic wave, since the stress is reduced to zero [131]. Similar to the

impact of the liquid droplet, the initial deformation of the rod is along the line of contact

with the rigid substrate (Figure 4.4(a)). However, in contrast the droplet impact, the jet-

ting of the rod continues along the line of contact up to 40 µs at which point the material

begins to harden (Figure 4.4(b)). With the hardening of the material near the foot of the

rod, the plastic wave (deformation) moves up the rod resulting in the bulging of the base as

shown in Figure 4.4(c). At around 80 µs, the rod reaches a state of rest with the familiar

mushroom like shape (Figure 4.4(d)).

In Figure 4.7(a), the evolution and the topology of the interface at different instants

in time are plotted. The interface topology displayed in the figure corresponds to the solu-

tion obtained using 5 levels of mesh refinement. The interface evolution matches well with

that reported in [180, 190]. To validate the present approach, the results obtained from the

current calculations are compared with previous numerical simulations [180,190,192,210].

The parameters, such as the final radius of the mushroom foot, the final length and the max-

imum effective plastic strain, characterizing the impact of the rod computed in the present

study agree well with the previously reported values (Table 4.1). The methodology de-

veloped in this work is stable and accurate in capturing the extreme stress and velocity

gradients developed in the rod at the moment of impact.

 183

4.4.1.2 Impact Velocity of 400 m/s and 600 m/s

The Taylor test is repeated with impact velocities of 400 m/s and 600 m/s. The

calculations are conducted with 5 levels of mesh refinement. The plots from the present

calculations are displayed in Figures 4.5 & 4.6. As expected, with the increase in the

impact velocity the deformation in the bar is much severe. For the sake of comparison, the

evolution of the interface at different instants in time are plotted in Figure 4.7. Also shown

in the figure is the final configuration of the bar at 80µs. As can be seen from the figures,

for the impact velocity of 600 m/s the bar has completely deformed before coming to rest.

The final radius of the mushroom foot, the final length and the maximum effective plastic

strain, corresponding to different impact velocities are tabulated in Table 4.2.

4.4.2 2D Axisymmetric Penetration of Steel Target by WHA Long Rod

The validation of the present method for two deformable objects with different ma-

terial properties is carried out using a slender Tungsten heavy alloy (WHA) rod projectile

penetrating an initially planar target made of a steel plate. Plates of 29.0 and 49.5 mm

thick were tested at incident velocities of 1250 m/s and 1700 m/s. The thickness represent

the previously determined ballistic limits for the impact velocities [180]. The cylindrical

projectile of 2 mm in radius and 50 mm length impacted a cylindrical steel plate of 40 mm

in radius. A schematic defining the problem is shown in Figure 4.8. The response of the

materials were modeled by using a Johnson-Cook material model and the corresponding

parameters for both materials are shown in Table 1.2. The friction between the two impact-

ing surfaces is neglected in these calculations. The simulations are carried out on a base
 184

mesh of size ∆xG = 0.0005 with 4 levels of mesh adaptation.

In the previous Eulerian calculation reported in [190], the solutions were obtained

on a truncated domain (0.0125 m X 0.02 m) with a grid density of 100x688 mesh points.

With the adaptive mesh refinement facility the current computations were performed on the

actual size as employed in the previous experimental [10] and numerical calculations [180].

In Figures 4.9, 4.10 & 4.11, the contours of effective plastic strain (ǫP ) and velocity, the

evolution of mesh and the interface topology are plotted for the impact velocity of 1250

m/s. As can be seen from the figures, the response of the projectile and the target agrees

well with the calculations reported in [180]. The ejecta, which were not captured by the

previous particle level set approach [190], are predicted very well and matches with the

results presented in [180]. The maximum equivalent plastic strain is found to be around

4.5, occurring mostly near the impact surfaces. The values of equivalent plastic strain are

higher in the WHA material compared to those in the steel material. The plastic strains

obtained in [180] using Lagrangian finite element method with an adaptive mesh agree

very well with the present results, both in terms of the magnitude and distributions of the

plastic strains. In particular, a trough in the plastic strain distribution is noticed in both

results and occurs near the bottom surface in the steel plate at the symmetry axis, as seen in

Figure 4.9(d). The ejection length of the WHA material also agrees well with [180]. The

maximum positive v-component velocity is observed around 40µs, occurring in the ejecting

mass of the WHA material. Around 80µs, the rod comes to rest and only small residual

velocities remain. The maximum temperatures occur around the impacted surfaces as this is

the region of maximum rate of conversion of plastic work to heat. The recorded maximum
 185

temperature is around 1575 K in the WHA material, below the melting temperature of

1777 K for WHA and 1723 K for steel. The largest temperature occurs at around 40µs, and

decreases as the rod goes to rest state. This shows that the largest plastic work done occurs

before this time of 40µs.

Figures 4.12(a) & 4.12(b) show the projectile nose and tail trajectories and veloci-

ties as a function of time, and is compared with the superposed results from experiment [10]

and from numerical calculations [180]. Also plotted are the original rear and impact sur-

faces. The results show excellent agreement with those of experiment and numerical ob-

servations. The predicted penetration depths is also in good agreement with experiments.

In Figures 4.13, 4.14 & 4.15, the contours of effective plastic strain (ǫP ) and ve-

locity, the evolution of mesh and the interface topology are plotted for the impact velocity

of 1700 m/s. Similar to the impact velocity of 1250 m/s, the maximum equivalent plastic

strain recorded is found to be 5.0. The plastic strains obtained in [180] using Lagrangian

finite element method with an adaptive mesh agree very well with the present results, both

in terms of the magnitude and distributions of the plastic strains. In contrast to the lower

impact velocity, projectile is completely absorbed by the target forming a slightly larger

crater and longer penetration length. Similar to the previous case, the rod comes to rest

with only small residual velocities around 80µs. The results are in excellent agreement

with that of [180]. Figures 4.16 & 4.16(b) show the projectile nose and tail trajectories

and velocities as a function of time, and is compared with the superposed results from

experiment [10] and from numerical calculations [180, 190]. The plots are in excellent

agreement.
 186

4.4.3 Shock Wave Interaction with Hemispherical Groove

Shock wave interacting with a hemispherical groove in a Copper matrix is consid-

ered. A planar shock wave generated by contact explosion interacts with a hemispherical

groove of radius 15 mm. The generated shock wave corresponds to a particle velocity of

540 m/s and a pressure ratio of 230 Kbar [46]. The center of the groove is located at 29 mm

from the bottom surface of the plate. The shock is initiated by quadratically accelerating

the velocity of the bottom domain from 0 to 540 m/s. A computational domain of 250 mm

X 30 mm is chosen. A base mesh of size ∆xG = 0.003 with 4 levels of mesh refinement

is selected. The simulation is run till 100 µs. Johnson cook material model is employed to

simulate the response of the Copper matrix.

The interaction of the shock wave with the hemispherical groove results in a re-

flected expansion wave and a formation of jet. The formation of the jet was confirmed in

the experimental work reported in [123] and subsequently in the numerical work reported

in [46]. The model can also be viewed as a prototype for hemispherical explosively formed

projectile (EFP) where the formation of the jet resembles that of the EFP. The jet reaches a

maximum velocity of 2750 m/s at about 12 µs and continues to jet until it reaches the target.

For the sake of validation, the maximum jet velocity and the jet diameter obtained from the

present calculations are compared with the previous computational [46] and experimental

observation [123]. The values are reported in Table 4.3. The agreement is excellent. The

mesh topology and the velocity contours are shown in Figure 4.17. In Figure 4.18, the

interface topology at different instants in time are plotted. The evolution of the interface

closely follows the trend reported in [46].

 187

4.4.4 Void Collapse in a Copper Matrix

In this test case, a cylindrical void with a radius of 1 µm within a Copper matrix

undergoes deformation due to a propagating shock created by imposing a particle velocity

of 550 m/s at the bottom domain boundary. The schematic of the problem is given in [190].

The response of the Copper material is modeled using a Johnson-Cook model. The wave

moves through the copper matrix and is transmitted out through the upper boundary. Both

right and left boundaries are subjected to symmetric conditions. As pointed out in [46], void

collapse is defined as the point at which the lower and upper boundary surfaces come into

contact. The void collapse phenomenon has implications for the initiation process in en-

ergetic materials. Void collapse can occur in different modes depending on the strength of

the impinging shock. As the shock strength increases the void collapse goes from a nearly

cylindrical (visco-plastic) mode to a jet (hydrodynamic) mode where the lower surface of

the void forms a jet which impacts on the upper surface at high velocity. The criterion for

the transition from the visco-plastic to hydrodynamic mode is provided by the analysis of

Khasainov [106] in terms of the ratio of shock passage time to the void deformation time

scales. If the shock passage time scale is larger than the void collapse time scale the mode

of collapse is visco-plastic, and when the shock passage time scale is comparable with the

void collapse time scale the mode of collapse is hydrodynamic [190]. This latter mode is

characterized by the formation of a jet of material which issues from the lower side of the

void and impacts the upper side, leading to large temperatures on the impact location due

to dissipation of kinetic energy of the jet [190].

The results from the current calculations are displayed in Figure 4.19. Formation
 188

of a distinct jet can be observed, with the highest velocity of the jetting material being

around 3360 m/s, much higher than the imposed particle velocity. Similar to the hydro-

dynamic case, the formation of the blast wave at the instant of collapse is also clearly

visible 4.19(d). One can see that the void would be completely collapsed before the shock

wave reaches the top boundary of the domain, i.e. this case represents the hydrodynamic

mode of collapse in agreement with the criterion mentioned above. The evolution of the

interface is displayed in Figure 4.20. As expected, the evolution closely follows the trend

observed in the hydrodynamic case.

4.4.4.1 High Velocity Impact of a Metallic Sphere on a Metal Target

In this example, a high velocity impact of a metallic sphere over a flat metal sub-

strate is studied. Two sets of material compositions are considered in this problem:

• Configuration 1: Impact of a heavy alloy Tungsten sphere on an infinitely thick Steel

target

• Configuration 2: Impact of a Copper sphere on an infinitely thick Copper target

The impact velocity of the sphere is fixed at 2000 m/s for both configurations. Con-

figuration 1 is a simple modification of the WHA rod penetration example considered

in [180,190] whereas configuration 2 corresponds to the problem investigated in [192]. The

dimension of the impinging sphere and the semi-infinite target are displayed in Figure 4.21.

The response of the metallic sphere and the target are modeled using Johnson-Cook mate-

rial model. The calculations are carried out in a domain of size 0.125 mm X 0.175 mm on

a base mesh of size ∆xg = 2µm with 4 levels of mesh adaptation. The results from the
 189

present calculations are presented in Figure 4.22.

Figure 4.22 shows the response of the initially spherical impactor and the target.

The initial response for the two configurations are very similar (Figures 4.22(a) & 4.22(d)).

The impact sets off a compressive wave traveling in both the impactor and the target. Due

to the difference in speed of sound in the media, the subsequent response of the material is

vastly different. As can be seen in Figures 4.22(c) & 4.22(d), the size and shape of the ejecta

for configuration 2 starts to differ from that of configuration 1. The depth of penetration

(Figures 4.22(e) & 4.22(f)) and the width of the resultant crater (Figures 4.22(g) & 4.22(h))

are different for both the configurations. The topology and the evolution of the interface

for the two configurations are displayed in Figure 4.23.

As can be seen from the figure, the depth of penetration for configuration 1 is much

greater than that of configuration 2. On the other hand, the width of the crater for configu-

ration 1 is significantly smaller than configuration 2. The heavier Tungsten sphere, due to

its larger initial kinetic energy penetrates much deeper and comes to rest at about 250 µs.

The lighter Copper sphere comes to rest at 200 µs. It is worth pointing out that the previous

calculation reported in [192] failed to capture the formation of crater and ejecta. With the

present framework, these features that are typical of hyper-velocity impact phenomena are

depicted and predicted accurately.

4.4.5 Very High Speed Impact of Aluminum Sphere over Aluminum Target

In this example, a very high speed impact of Aluminum sphere into an effectively

semi-infinite Aluminum target is investigated. The Aluminum sphere impacts the target
 190

at 7000 m/s. An Aluminum sphere of 6.35 mm in diameter impacts a target also made

of Aluminum. The length and the radius of the semi-infinite cylindrical target are 10.0

mm and 2.9 mm respectively. The simulations are carried out with a base mesh of size

∆xG = 0.0005 with 4 levels of mesh adaptation. A Johnson-Cook material model is

employed to compute the response of the Aluminum target and the impactor.

The problem was previously investigated in [213] using a two-dimensional explicit

Lagrangian FEM hydrocode ZeuS. The zoomed in view of the results obtained from the

current calculations are reported in Figures 4.24 & 4.25. As expected, due to the very

high velocity involved in the impact process, a shock wave is formed that travels both in

the impactor and the target. These shock waves are visible in Figure 4.25(a) in which the

mesh adaptation and the contours effective plastic strain (ǫP ) are displayed. The mesh

adapting to the shock wave is also clearly visible in the figure. Subsequent evolution

of the mesh and the deformation of the impactor and the target are displayed in Fig-

ures 4.25(b), 4.25(c), & 4.25(d). As pointed in [213], the projectile is almost consumed

by 3 µs after impact (Figure 4.24(b) ). Due to very large incident kinetic energy, the width

of the crater and the depth of the ejecta are very large compared to the corresponding low

velocity impact processes. Aluminum impactor eventually forms a very thin layer that set-

tles on the crater formed on the target (Figure 4.24(d)). Since the target is sufficiently large,

there are no reflected relief waves from the target to relieve the compressive strain in the

impactor. As a result, the impactor is always in a state of compression and continues to

penetrate the target. At the instant shown in Figure 4.24(d), the impactor has almost come

to a complete stop. The velocity of the head of the impactor at this instant is about 500
 191

m/s which is about 14 times smaller than the initial impact velocity. The evolution and

the topology of the interface at different instants in time are plotted in Figure 4.26. The

example demonstrates the robustness of the present approach in capturing the high velocity

impact phenomena.

4.4.6 Perforation of Aluminum Plates by Conical Nose Projectile

Numerical simulations of conical nosed projectile perforating cylindrical target

plates are performed in this study. The conical nosed projectile is made of Tungsten and the

cylindrical target plate is made of 5083-H131 Aluminum. The geometry and the problem

set up can be found in the experimental studies conducted in [63]. Prior numerical calcu-

lations for this problem have been reported in [180]. Two plates of thickness 12.7 mm and

50.8 mm impacted at velocities 1195 m/s and 1176 m/s respectively are considered. The

initial problem configuration is displayed in Figure 4.27. Johnson cook material model

was employed to simulate the response of the projectile and the target. In consistent with

the parameters selected in [180], the Taylor-Quinney coefficient β is set to zero. Both the

1
simulations were carried out on a base mesh of size ∆x = 1000
with four levels of mesh

refinement. The histories of effective plastic strain (ǫP ) along with the mesh evolution are

displayed in Figures 4.28 & 4.29.

The conical nosed projectile upon impacting the target forms a hole with cavity

diameter equal to the shank diameter of the projectile [63]. As reported in [63, 180], the

impacting projectile is practically undeformed for both incident velocities. This can be

readily seen from the snapshots displayed in the Figures 4.28 & 4.29. The sharp conical
 192

nose of the projectile is retained in tact even when the interface is moved using the tra-

ditional level set advection procedure (with no Lagrangian particles for correction). This

clearly indicates that the diffusion errors that are intrinsic in the level set technique have

very little influence in altering the shape and topology of the interface. Furthermore the

adaptive mesh refinement facility employed in this work is effective in mitigating these dif-

fusion errors. The mesh adaptation and evolution displayed in the figures clearly show that

the regions with fine mesh are concentrated to those region with significant plastic strain.

The maximum effective plastic strain computed for the 12.7 mm thick Aluminum plate is

1.52 which is close to the value (1.50) reported in [180]. However, the value registered for

the 50.8 mm thick Aluminum plate is 2.05 which is greater than the value reported in [180].

Nevertheless, the response of the plate closely follows the trend reported in [63, 180].

4.4.6.1 Explosion Inside a Copper Tube

In this example, an explosion inside a Copper tube and the response of the Copper

metal walls is simulated. The initial configuration of the problem is shown in Figure 4.30.

A high pressure explosive core of gas is located at the center of a Copper tube containing

water at ambient conditions. The initial conditions are given below:




(4000.0kg/m3 , 500000.0atm.0, 0.0, 0.0, 7.15) for

(ρ, P, u, v, γ) =



(1000.0kg/m3 , 1.0atm, 0.0, 0.0, 1.4) for water at ambient conditions.

The size of the computational domain is 6m X 6m and the size of the base mesh employed

is ∆xg = 0.06. The computations are carried out with four levels of mesh refinement. The

explosive core is modeled using an ideal gas law and the surrounding bulk fluid (water) is
 193

modeled using a stiffened equation of state (Eq 1.60). The Copper tube is idealized as a

perfectly flowing plastic with no hardening model. Both the explosive core and the walls

of the Copper tube are tracked and represented by level sets. The evolution of the explo-

sive core is captured by solving the Riemann problem at the interface (Section 1.6.2.1)

and the response of the metal tube is computed by enforcing the conditions listed in Sec-

tion 1.6.2.2. A similar problem was investigated in [115] using the MGFM approach.

In [115], the deviatoric response of the material was not computed explicitly but included

in the hydro-elasto-plastic (HEP) equation of state used to model the dilatational response

of the solid phase. Their model was based on the assumption that under such explosive

loading conditions, the response of the material is predominantly compressive in nature.

Hence the dilatational response is orders of magnitude larger than the deviatoric response

of the structure. With such HEP equation of state, the MGFM approach developed in [118]

was extended to model the structural response. In subsequent work [202, 203], the MGFM

approach was used to couple the Eulerian fluid calculation to a Lagrangian solid calcula-

tion for exclusively modeling fluid-structure interaction problems. In direct contrast to the

past efforts, the calculations reported in this section are performed with the generic sharp

interface Eulerian framework. The results are displayed in Figure 4.31

The explosion sets off a shock wave that travels in the water medium and an ex-

pansion wave traveling in the core (Figure 4.31(a)). The shock wave in the water medium

upon impacting the Copper tube walls, results in a transmitted and a reflected shock wave

(Figure 4.31(b)). As seen in Figure 4.31(b), the impinging shock wave is strong enough to

cause the Copper walls to deform. The reflected shock wave subsequently interacts with
 194

the expanding explosive core and results in a secondary system of waves; a reflected relief

wave in the water medium and a transmitted shock wave in the explosive core. The multiple

reflection causes the fluid near the tube walls to cavitate and hence the calculation could not

be carried out without adequate cavitation model. Development of such a model is beyond

the scope of the present work and hence the computations were deliberately stopped before

the cavitation phenomenon could arise. Figures 4.31(c) & 4.31(d) display the deformation

of the tube walls at later instants in time. The deformation in this case is governed by the

multiply reflected wave systems. Also seen in the figures is the expanding gas core with a

significantly distorted shape.

4.4.6.2 Shock Wave Interacting with Deformable Metal Particle(s)

Shock waves interacting with metal particles and related momentum transfer are

important in applications related to detonation and deflagaration in condensed media. Tra-

ditional models for detonation phenomena were based on frozen shock assumption wherein

the momentum transfer and subsequent motion of the particles were ignored. The impor-

tance of such an assumption for heavy and light particles were analyzed in [209]. In this

example, similar calculations are repeated in the current sharp interface framework. In

particular, the planar shock wave interacting with spherical Aluminum and Tungsten par-

ticle of radius 5 µm are considered. An initially planar shock wave in water with initial

conditions,



(1639kg/m3 , 1.013Kbar, 1987m/s, 0.0, 5.5) for y ≤ 2.6,

(ρ, P, u, v, γ) =



(1000kg/m3 , 1.0bar, 0.0, 0.0, 5.5) for y > 2.6
 195

is positioned at Y = 20 µm. The spherical particle is positioned at (0 µm, 25 µm) in a

computational domain of size 25 µm X 50 µm. A base mesh of size ∆xg = 0.5µm with

four levels of mesh refinement is used. The materials are assumed to behave perfectly

plastic with no hardening model. The initial configuration of the problem is displayed in

Figure 4.32.

The results from the current calculations are displayed in Figure 4.33. The initial

interaction of the shock wave with the particle results in a reflected and a transmitted shock

wave (Figure 4.33(a)). The transmitted shock wave upon arriving at the other end of the

particle results in a reflected relieve wave and a transmitted shock wave (Figure 4.33(b)).

The negative pressure values (inside the Aluminum particle) indicated in Figure 4.33(b)

correspond to the tensile wave traveling in the particle. Such negative pressure contours

may not be visible in [209] as a HOM equation of state was employed with no explicit

correction for relief waves. As evident from Figures 4.33(c) & 4.33(d), the response of

Aluminum particle is vastly different from that of Tungsten. This is because the shock

wave imparts significantly higher pressure than the yield stress of the material and hence

the response of the particle varies with materials [209]. The Aluminum particle not only

undergoes considerable lateral expansion and distortion but also moves significantly with

the incident shock wave. The evolution and the topology of the interface for Aluminum

particle is shown in Figure 4.34(a). The location of the leading and trailing edge as a

function of time is plotted in Figure 4.34(b). The transmission coefficient (the ratio of the

velocity of the center of mass of the particle with the velocity of the incident shock wave)

computed for Aluminum particle is 0.63 and the value reported in [209] is 0.6. Thus the
 196

present calculations agree well with previous efforts [209].

Similar to the calculations performed in [209], shock wave interacting with a cluster

of Aluminum particles are considered next. The strength of the shock wave is the same as

the single particle. The cylindrical Aluminum particles of 5 µm radius are seeded in an

orderly fashion in a computational domain of size 70 µm X 125 µm. A base mesh of size

∆xg = 0.25µm with four levels of mesh refinement is chosen for this calculation . The left

and the right end of the domain are imposed with symmetry conditions to depict a periodic

arrangement of particles. The solution obtained from the current calculations are displayed

in Figure 4.35.

As expected the shape and topology of the particles is much different from the

single particle case. This is due to the complex nature of reflected wave patterns generated

from the neighboring particles. Although the present configuration is slightly different

from [209], the overall interface evolution patterns are in good qualitative agreement.

4.4.7 High Velocity Impact Induced Explosion

When a high energy explosive material is impacted at sufficiently high velocity,

the resulting energy deposition can cause the initiation of explosion. The facility to com-

pute such high velocity impact driven explosion of high explosives is demonstrated in this

example. The problem consist of a cylindrical WHA Tungsten projectile impacting on a

cylindrical Steel casing containing HMX explosive. Three level sets, each for the Tungsten

projectile, the steel casing and the HMX explosive, are employed to track and represent the

interfaces. The problem demonstrates the robustness of the current approach in resolving
 197

multiple interface collisions. The dimensions and initial configuration of the problem are

shown in Figure 4.36. Johnson-cook material model are used for modeling the response

of Steel casing and the Tungsten projectile. For the HMX explosive, perfect plastic flow

model is assumed (neglecting the effects of viscosity). A base mesh of size ∆xG = 0.0005

along with 4 levels of mesh adaptation is used.

The impact of the projectile on the steel casing results in a momentary compressive

wave, that later coalesce to form a shock wave, traveling in the projectile and the steel cas-

ing (Figure 4.37(a)). The shock wave generated as a result of this impact process (in the

front end of the steel casing) is transmitted to the HMX explosive at the instant shown in

Figure 4.37(a). This transmitted shock propagates through the HMX media and gets re-

flected and transmitted as a shock wave upon arriving at the back end of Steel casing. The

secondary shock reflection process is visible Figure 4.37(b). This multiple shock reflection

initiates a detonation phenomena in the HMX explosive and results in the deformation of

back end of the steel casing (the part of the casing not in contact with the impactor). This

deformation process is visible in Figures 4.37(c) & 4.37(d). For the sake of clarity, the evo-

lution and topology of the interface are plotted in Figure 4.39. This deformation of the Steel

enclosure and the projectile may not continue endlessly as depicted in Figure 4.37(d). With

sufficient damage model, the shattering and collapse of the materials can be adequately

modeled. Incorporation of damage models in the present framework is straightforward and

the efforts in this direction will be reported in the future. In Figure 4.38, the mesh evolu-

tion and effective plastic strain (ǫP ) contours at different instants in time are displayed. The

mesh adaptation process is apparent in the figures.

 198

4.4.8 Axisymmetric Dynamic-Tensile Large-Strain Impact-Extrusion of Copper

The experimental study on the influence of grain size on the response of Copper was

conducted in [91]. In this section, the corresponding numerical simulations are presented.

The example problem considered here consists of a Copper sphere of 7.6 mm in diameter

undergoing a tensile extrusion process. The extrusion process is carried out by impact the

Copper sphere fired at 400 m/s towards extrusion die. The Extrusion die made of hardened

Steel is designed with an entrance diameter of 7.62 mm and an exit diameter of 2.28 mm;

a reduction of 70 % as shown in the Figure 4.40. A base mesh of size ∆xG = 0.0005 is

chosen with 4 levels of mesh adaptation. A Johnson Cook material model is employed to

capture the response of the sphere and the extrusion die.

The evolution of the effective plastic strain (ǫP ) and velocity contours at different

instants in time are displayed in Figure 4.41. The initial impact of the sphere (at about

10 µ s corresponding to the instant shown in Figure 4.41(a)) on the extrusion die results

in the acceleration of the leading edge of the sphere as it exits the die [91]. At about 20

µ s, corresponding to the instant shown in Figure 4.41(b), the conical-shaped portion of

the sphere comes to rest in the extrusion die. This can be easily verified from the velocity

contours registered in Figure 4.41(b) and in the subsequent Figures 4.41(c) & 4.41(d).

The portion of the sphere continues to remain at rest while the leading edge of the sphere

stretches to form a shape-charge jet. Figure 4.41(c) shows the onset of the initial necking

process which subsequently forms three-major particle (as verified by the experimental

observation conducted in [91]). These particles are clearly visible in the Figure 4.41(d).

The jet continues to stretch and results in further splitting up of particles that can only be
 199

captured with explicit damage models. Despite the lack of adequate damage model, the

present calculations are able to predict the overall behavior of the extrusion process that

matches well with the experimental predictions reported in [91]. The maximum equivalent

plastic strain (ǫP ) was observed during the jetting phase and corresponds to a value of 9.3.

The numerical computations conducted in [91] reports a value of 9.0, which is in close

agreement with the current predictions.

Figure 4.42 shows the evolution of the mesh. The enlarged view of the mesh adapt-

ing to the shape-charge jet is shown in Figure 4.42(c). The mesh adaptation is excellent

and is apparent from the figures. In Figure 4.43, the evolution of the interface at different

instants in time are displayed. It is clear that throughout the extrusion process, the shape of

the extrusion die is unaltered.

 200

Case Final Length (mm) Final Base Radius (mm) Maximum ǫp

Current (3 Levels) 21.35 6.75 2.84

Current (4 Levels) 21.50 7.01 3.087

Current (5 Levels) 21.53 7.05 3.169

Tran et al [190] 21.15 7.15 2.86

Udaykumar et al [192] - 6.8 6.97-7.24

Camacho et al [180] 21.42-21.44 7.21-7.24 2.97-3.25

Zhu et al [210] 21.26-21.40 6.97-7.18 2.75-3.03

Table 4.1: Comparison of results for the axisymmetric impact of Copper rod at 227 m/s.
 201

Case Final Length (mm) Final Base Radius (mm) Maximum ǫp

227 m/s (5 levels) 21.53 7.05 3.169

400 m/s (5 levels) 10.56 12.81 5.01

600 m/s (5 levels) - 24.56 5.3

Table 4.2: Comparison of parameters for the axisymmetric impact of Copper rod for dif-
ferent impact velocities
 202

Velocity (cm/s) Diameter (δc , cm)

Present 0.275 0.58

Computation [46] 0.264 0.16

Experiment [123] 0.27 0.6

Table 4.3: Comparison with experimental and computational results for the jet velocity and
diameter.
 203

Figure 4.1: Embedding the boundary conditions with the interpolation procedure.
 204

(a) Classification of materials

(b) Collision detection algorithm

Figure 4.2: Collision detection algorithm: (a) An elasto-plastic solid object immersed in
a free surface. The surrounding free surface is flagged as bulk material (b) Algorithm to
detect collision between any two level sets; φl indicate the value of the level set function
corresponding to the lth material interface and δ is the distance between the approaching
level sets at point P.
 205

Figure 4.3: Initial configuration for two-dimensional axisymmetric Taylor test on a Copper
rod.
 206

(a) (b)

(c) (d)

Figure 4.4: Snapshots of pressure and effective plastic strain (ǫP ) contours at different
instants in time for the axisymmetric impact of Copper rod at 227 m/s.
 207

(a) (b)

(c) (d)

Figure 4.5: Snapshots of pressure and effective plastic strain (ǫP ) contours at different
instants in time for the axisymmetric impact of Copper rod at 400 m/s
 208

(a)

(b)

(c)

(d)

Figure 4.6: Snapshots of pressure and effective plastic strain (ǫP ) contours at different
instants in time for the axisymmetric impact of Copper rod at 600 m/s
 209

(b) 400 m/s

(a) 227 m/s

(c) 600 m/s (d) Final Interface Configuration

Figure 4.7: Axisymmetric Taylor bar experiment: Figures (a), (b), and (c) show the evolu-
tion and the topology of the interface at different instants in time for three different impact
velocities (d) Final configuration of the interface at 80µs.
 210

Figure 4.8: Initial configuration for the two-dimensional axisymmetric penetration of Steel
target by WHA long rod
 211

(a) 20 µs (b) 40 µs

(c) 60 µs (d) 80 µs

Figure 4.9: Contours of equivalent plastic strain (ǫp ) and velocity of a Tungsten rod pene-
trating a steel plate at 1250 m/s.
 212

(a) 20 µs (b) 40 µs

(c) 60 µs (d) 80 µs

Figure 4.10: Contours of equivalent plastic strain (ǫp ) and mesh evolution of a Tungsten
rod penetrating a steel plate at 1250 m/s.
 213

(a) 20 µs (b) 40 µs

(c) 60 µs (d) 80 µs

Figure 4.11: Snapshots of the interface topology of a Tungsten rod penetrating a steel plate
at 1250 m/s.
 214

(a) Projectile nose and tail trajectories (b) Projectile nose and tail velocity histories

Figure 4.12: Tungsten rod penetrating steel plate at 1250 m/s: (a) Trajectories and (b)
velocities of the nose and tail of the projectile.
 215

(a) 20 µs (b) 40 µs

(c) 60 µs (d) 80 µs

Figure 4.13: Contours of equivalent plastic strain (ǫp ) and velocity of a Tungsten rod pene-
trating a steel plate at 1700 m/s.
 216

(a) 20 µs (b) 40 µs

(c) 60 µs (d) 80 µs

Figure 4.14: Contours of equivalent plastic strain (ǫp ) and mesh evolution of a Tungsten
rod penetrating a steel plate at 1700 m/s.
 217

(a) 20 µs (b) 40 µs

(c) 60 µs (d) 80 µs

Figure 4.15: Snapshots of the interface topology of a Tungsten rod penetrating a steel plate
at 1700 m/s.
 218

(a) Projectile nose and tail trajectories (b) Projectile nose and tail velocity histories

Figure 4.16: Tungsten rod penetrating steel plate at 1700 m/s: (a) Trajectories and (b)
velocities of the nose and tail of the projectile.
 219

(a) (b)

(c) (d)

Figure 4.17: Snapshots of velocity contours and mesh evolution at different instants in time
for the response of a hemispherical groove to a shock wave
 220

Figure 4.18: Topology and evolution of the interface at different instants in time for the
response of a hemispherical groove to a shock wave
 221

(a) 0.35 ns (b) 0.5 ns

(c) 0.75 ns (d) 1.0 ns

Figure 4.19: Snapshots of velocity and equivalent plastic strain (ǫp ) contours at different
instants in time for the collapse of a cylindrical void in a Copper matrix
 222

Figure 4.20: Topology and evolution of the interface for the collapse of a cylindrical void
in a Copper matrix
 223

Figure 4.21: Initial configuration for the impact of a metallic sphere on a semi-infinite
metallic target
 224

(a) Configuration 1 (b) Configuration 2

(c) Configuration 1 (d) Configuration 2

(e) Configuration 1 (f) Configuration 2

(g) Configuration 1 (h) Configuration 2

Figure 4.22: High velocity impact of metal sphere on a metal Surface: (a), (c), (e), and (g)
correspond to configuration 1 - impact of a heavy alloy Tungsten sphere on an infinitely
thick Steel target and (b), (d), (f), and (h) correspond to configuration 2 - impact of a Copper
sphere on an infinitely thick Copper target
 225

(a) Configuration 1

(b) Configuration 2

Figure 4.23: Toplogy and location of the interface at different instants in time: (a) configu-
ration 1 - impact of a heavy alloy Tungsten sphere on an infinitely thick Steel target and (b)
corresponds to configuration 2 - impact of a Copper sphere on an infinitely thick Copper
target
 226

(a) 0.4 µ s (b) 2 µ s

(c) 4 µ s (d) 6 µ s

Figure 4.24: Snapshots of effective plastic strain (ǫP ) and velocity contours at different
instants in time for the dynamic tensile extrusion of Copper
 227

(a) 0.4 µ s (b) 2 µ s

(c) 4 µ s (d) 6 µ s

Figure 4.25: Snapshots of effective plastic strain (ǫP ) and mesh topology at different in-
stants in time for the 7000 m/s impact of Aluminum sphere on Aluminum target.
 228

Figure 4.26: Topology and evolution of the interface at different instants in time for the
7000 m/s impact of Aluminum sphere on Aluminum target.
 229

Figure 4.27: Initial configuration for the perforation of Aluminum plate by a conical nose
projectile.
 230

(a) (b)

(c) (d)

Figure 4.28: Snapshots of effective plastic strain (ǫP ) contours and mesh evolution at dif-
ferent instants in time for the perforation of 12.7 mm thick Aluminum plate at an incident
velocity of 1195 m/s.
 231

(a) (b)

(c) (d)

Figure 4.29: Snapshots of effective plastic strain (ǫP ) contours and mesh evolution at dif-
ferent instants in time for the perforation of 50.8 mm thick Aluminum plate at an incident
velocity of 1176 m/s.
 232

Figure 4.30: Initial configuration for the explosion inside a Copper tube.
 233

(a) T = 112.5 µs (b) T = 300.0 µs

(c) T = 600.0 µs (d) T = 900.0 µs

Figure 4.31: Snapshots of numerical Schlieren image, at different instants in time, for the
response of a Copper tube to an explosion in water.
 234

Figure 4.32: Initial configuration for a shock wave interacting with metal particle.
 235

(a) T = 1 ns (b) T = 2 ns

(c) T = 3 ns (d) T = 5 ns

Figure 4.33: Snapshots of pressure contours for a Tungsten (left) and an Aluminum (right)
particle subject to a shock wave in water.
 236

(a) Toplogy and location of the interface (b) Locus of leading and trailing edge

Figure 4.34: Aluminum particle subjected to a planar shock wave: (a)Topology and loca-
tion of the interface at different instants in time; (b) locus of leading and trailing edge as
function of time.
 237

(a) (b)

(c) (d)

Figure 4.35: Snapshots of pressure contours for a cluster of Aluminum particles subject to
a planar shock wave in water.
 238

Figure 4.36: Initial configuration for the explosion initiated due to high velocity impact.
 239

(a) 0.75 µ s

(b) 1.5 µ s

(c) 4.5 µ s

(d) 10 µ s

Figure 4.37: Snapshots of effective plastic strain (ǫP ) and velocity contours at different
instants in time for the impact driven explosion problem.
 240

(a) 0.75 µ s

(b) 1.5 µ s

(c) 4.5 µ s

(d) 10 µ s

Figure 4.38: Snapshots of effective plastic strain (ǫP ) contours and mesh evolution at dif-
ferent instants in time for the impact driven explosion problem.
 241

Figure 4.39: Topology and evolution of the interface at different instants in time for the
impact driven explosion problem.
 242

Figure 4.40: Initial configuration for the axisymmetric dynamic-tensile impact-extrusion

of Copper.
 243

(a) 10 µ s (b) 20 µ s

(c) 30 µ s (d) 40 µ s

Figure 4.41: Snapshots of effective plastic strain (ǫP ) and velocity contours at different
instants in time for the dynamic tensile extrusion of Copper.
 244

(a) 10 µ s (b) 20 µ s

(c) 30 µ s (d) 40 µ s

Figure 4.42: Snapshots of velocity contours and mesh evolution at different instants in time
for the dynamic tensile extrusion of Copper.
 245

Figure 4.43: Topology and evolution of the interface at different instants in time for the
dynamic tensile extrusion of Copper.
 246

CHAPTER 5
SHARP INTERFACE TREATMENT FOR SOLID OBJECTS

5.1 Introduction

Recently, the use of embedded boundary methods for the representation of solid

boundaries in incompressible [86, 124, 148, 177] and compressible [57, 58, 118, 133, 135,

190] flows has become popular. The Ghost Fluid Method (GFM) [58] (explained in the

previous chapter) provides a simple framework to implicitly transmit the presence of an

embedded interface to the flow field. However, when applied to shock interactions with

stationary or moving solid objects embedded in compressible flows, the GFM was found

to suffer from acute over-/under-heating numerical errors [60, 118]. In section 3.2.1.1, it

was demonstrated that the nav̈e approach of entropy extrapolation (isobaric fix proposed

by Fedkiw and coworkers [58]) failed to maintain non-oscillatory pressure field, particu-

larly for strong shocks interacting with fluid-fluid interfaces. Furthermore, it was found

that the failure of GFM to accurately resolve the interface can be alleviated by populat-

ing the ghost points by solving a local Riemann problem normal to the interface [118].

This was later adopted by several researchers both in the GFM and non-GFM frame-

work [13, 14, 42, 49, 87, 118]. However, these methods were mostly developed to treat

fluid-fluid interfaces and hence it is not clear how one could extend such methods to rep-

resent embedded solid objects. A variant of the GFM, the Characteristic Based Match-

ing(CBM) method proposed by Nourgaliev et al [133] was found to be effective in re-

ducing the shock-interface interaction errors significantly. In the present work, a unified
 247

formulation to treat both solid-fluid and fluid-fluid interfaces is presented. Briefly, a simple

Reflective Boundary Condition (RBC) method is used to populate the ghost points and the

resulting heating errors are corrected by augmenting the solutions by solving a Riemann

problem between the real field and the reflected ghost field. The Riemann problem con-

structed normal to the interface is solved using an exact Riemann solver and the resulting

Riemann states are used to populate the ghost points. This approach of solving a Riemann

problem is simpler to implement than the CBM approach and is also shown later to be com-

putationally less expensive. As demonstrated in the results from the current simulations,

the proposed method is found to yield satisfactory solutions for several complex configu-

rations and shock diffraction phenomena. Shocks interacting with multiple particles and

complex shapes have been computed and the method is currently being applied to study the

dynamics of dense particulate compressible flows.

5.2 Boundary Conditions

The tracked interface has to be coupled with the flow solver such that the jump in the

mass, momentum and energy fluxes along with the material properties across the interface

are depicted accurately. In the GFM approach, this translates to suitably populating the

ghost points. However, there is a significant difference between the treatment of fluid-fluid

and (rigid) solid-fluid interfaces in the GFM framework. In the case of fluid-fluid interfaces

the flow variables, such as velocity and pressure, that are continuous across the interface

are injected into the ghost field directly. In the latter case, due to the absence of flow fields

on one side of the interface (the solid side), it is not immediately clear how to populate the
 248

ghost points with the virtual flow properties.

At the interface of a solid body immersed in compressible flows, the boundary

conditions that hold true are enlisted in section 1.6.2.3. The set of boundary conditions

given in Eqs (1.92) - (1.96) govern the behavior of the flow near the embedded solid body

and must be enforced on the real fluid by suitably populating the ghost points. There are

several approaches for applying these boundary conditions within the spirit of the Ghost

Fluid Method. However, as will be demonstrated in the numerical examples (section 5.3),

not all of them are successful in computing shock-interface interactions in a robust fashion.

5.2.1 Boundary Condition Type I: Reflective Boundary Condition (RBC)

The Reflective Boundary Condition (RBC) approach is widely adopted to depict

the presence of a rigid body in the fluid [11, 48, 62, 109, 204]. RBC, as the name indicates,

reflects the real flow fields across the interface. In the GFM framework, the fields from the

real fluid are reflected onto the ghost points in the solid phase. The reflection operation

is carried out such that the boundary conditions (Eqs (1.92) - (1.96)) are enforced exactly

at the interface. Since the boundary conditions involve the flow properties in the normal

and tangential directions, the reflection process must be carried out in the local body-fitted

curvilinear coordinates (vn , vt1 , vt2 ), instead of the global Cartesian coordinates (u,v and

w).

In order to carry out the reflection process accurately, the reflected point on the

real fluid side for each interfacial ghost point on the solid phase must be determined. The

reflected point IP1 on the real fluid side, corresponding to ghost point P, can be obtained
 249

by inserting a probe through point P as shown in Figure 5.4. IP is the point of intersection

of the probe with the interface. The absolute value of the levelset (| φP |) gives the normal

distance of point P from the interface. This value φP can then be used to march into the

real fluid to determine the coordinates of the reflected point IP1,

X~IP1 = X~P + 2| φP |N~P . (5.1)

The flow properties at the reflected point can be obtained using a regular bilinear inter-

polation from the surrounding computational points as shown in Figure 5.4. It must be

noted that one or more of these neighboring computational points may lie inside the solid,

in which case the corresponding ghost fluid properties are used in the interpolation proce-

dure. Once the flow properties at the reflected point are determined, it is straightforward to

reflect the flow properties on to point P.

5.2.1.1 Embedding Boundary Conditions with the Interpolation Procedure

Sometimes, an interfacial ghost point, like point P in Figure 4.1, may lie very close

to the interface. In such cases, using a direct bilinear interpolation procedure will involve

the (as yet unknown) ghost point. Hence in order to avoid interpolating from the ghost

point in consideration, the methodology outlined in section 4.3.1.1 is employed to embed

the boundary conditions in the interpolation procedure. Accordingly, the value of the flow

variables at the reflected point IP1 can be obtained using a bilinear interpolation of the form

ψ = a1 + a2 x + a3 y + a4 xy (5.2)

where (x, y) are the coordinates of the reflected point IP1 and ψ corresponds to the flow

variables. In order to determine the coefficients a1 − a4 , Eq (4.7) is applied to the sur-

 250

rounding points chosen for the interpolation procedure. Hence, for Dirichlet type boundary

conditions Eq (5.2) reduces to,

    
 1 x 1 y1 x1 y1   a1   ψ1 
    
    
 1 x y2 x2 y2  a   ψ 
 2  2   2 
  =  (5.3)
    
 1 x y3 x3 y3  a   ψ 
 3  3   3 
    
    
1 xIP yIP xIP yIP a4 ψ4

and for Neumann type conditions, we have

    
 1 x1 y1 x1 y1   a1   ψ1 
    
    
 1 x y2 x2 y2  a   ψ 
 2  2   2 
  =  (5.4)
    
 1 x y3 x3 y3  a   ψ 
 3  3   3 
    
    
0 nx ny nx yIP + ny xIP a4 S4

The last row of the coefficient matrix in Eq (5.4) is obtained by differentiating Eq (4.7),

noting that
∂ψ ∂ψ ∂ψ
= nx + ny (5.5)
∂n ∂x ∂y

where nx and ny are the components of the normal vector at point P and S4 corresponds

to the value of the normal gradient at the interface. After determining the coefficients, the

flow properties at the reflected point can be determined using Eq (4.7).

5.2.1.2 Determining the Flow Properties at the Interfacial Ghost Points

Once the flow properties at the reflected point IP1 are determined, the next step is

to incorporate the boundary conditions (Eqs (1.92) - (1.96)) at the interface. In the RBC

framework, the boundary conditions take the following form: The no-penetration condition
 251

is satisfied by reflecting the normal velocity, as given by,

vnP = 2UnIP − vnIP1 , (5.6)

where UnIP is the velocity of the center of mass of the solid body. For the tangential veloci-

ties, Eq (1.93) must be satisfied at the interface. The Neumann condition for the tangential

velocities in the RBC framework then becomes,

vt1P = 2Ut1IP − vt1IP1 , (5.7)

vt2P = 2Ut2IP − vt2IP1 , (5.8)

where, for slip,

Ut1IP = vt1IP1 , (5.9)

Ut2IP = vt2IP1 , (5.10)

and for no slip,

Ut1IP = 0, (5.11)

Ut2IP = 0, (5.12)

where Ut1IP and Ut2IP are the tangential velocities at IP. Along the same lines, the pressure

field can be obtained from Eq (1.96),

ρp vt21
PP = PIP1 − ( P
− ρp anP )2 | φP |, (5.13)
RP

where anP is the acceleration of the solid. The density ρp used in Eq (5.13), is obtained by

reflecting the value at IP1, i.e.,

ρP = ρIP 1 . (5.14)
 252

Once the flow properties corresponding to the interfacial ghost points in the solid phase

are determined, these properties are then extended to populate the interior ghost points

using a suitable multi-dimensional extrapolation procedure. Linear and other higher-order

extrapolation procedures have been discussed in Aslam et al [15]. However, the numerical

experiments carried out in this work indicate that the order of the extrapolation procedure

has little or no significant influence over the accuracy of the solution, and hence a simple

constant extrapolation procedure was found to be sufficient for the problems attempted

in this paper. Thus in the current formulation the subcell postion of the interface and

the topology of the interface are carefully embedded in the interpolation procedure. With

the reflected ghost field, the problem is reduced effectively to a single fluid (real fluid -

ghost fluid) problem which can be solved with known (fluid - fluid) interface treatment

techniques.

5.2.1.3 Correction for Over-/Under-Heating Problems

The reflection of strong shocks from moving or stationary walls can cause numer-

ical artifacts such as excessive over-/under-heating errors [127, 132]. The over-/under-

heating errors are the result of lack of dissipation in the numerical model. In reality, the

viscous effects are not negligible in the regions occupied by the shock and when a shock

reflects of the interface, the excessive heat generated are dissipated due the viscous ef-

fects (conducted away from the wall). Since the numerical schemes, do not have such

strong dissipative mechanism, the heat generated accrue over time and result in over-/under-

prediction of field variables. Donat et al [52] developed a flux-splitting scheme which had a
 253

built-in mechanism to dissipate the excessive heat generated at the wall. Fedkiw et al [60]

proposed an isobaric fix, which is a simple extrapolation of entropy from the real fluid

side onto the corresponding ghost field (Figure 5.2(a)), to suppress the over-/under-heating

errors.

Several interface treatment techniques developed in the past [58, 60, 133, 135] were

prone to these heating errors and RBC is no exception. As mentioned above, with RBC,

the solid-fluid interface is converted to a relatively simple (real) fluid - (ghost) fluid inter-

face problem. The interaction between the real and the corresponding ghost fields result

in strong shock or rarefaction waves or both, which travel into the real and ghost fields.

The problem becomes accentuated for a strong shock impacting the interface due to the

following three reasons.

i. Due to the implementation of reflective conditions, which uses bilinear interpolation

to obtain the flow properties at the reflected point: Using any form of interpolation

techniques across strong discontinuities will result in interpolation errors. These er-

rors, although generated locally near the interface, accrue over time and afflict the so-

lution way from the interface. It is important to point out that the errors arising from

the interpolation procedure can be significantly reduced by employing lower/higher-

order MUSCL [193]/ENO [163] based adaptive polynomial interpolation procedure.

However, such schemes require wider stencils and gradient evaluation, which be-

comes complicated to compute and computationally expensive particularly when the

mesh becomes highly unstructured; i.e. when additional mesh enrichment techniques

like Adaptive Mesh Refinement (AMR) [26] or tree-based Local Mesh Refinement
 254

(LMR) [108] is employed.

ii. Due to the over-/under-heating errors arising from shock reflection phenomena.

iii. Due to over-/under-specification of boundary conditions: The characteristic analy-

sis (discussed in section 5.2.4) of the governing hyperbolic equations dictates that

the variables corresponding to incoming characteristic waves must be supplied with

flow conditions independent of the flow field, and the variables corresponding to the

outgoing characteristic waves must be evolved in time based on the flow properties

within the computational domain [133, 135]. Since the RBC approach does not con-

form to the restriction imposed by the characteristic waves at the interface (except for

supersonic inflow condition), it results in over-/under-specification of the boundary

conditions, which in turn causes spurious waves to enter the computational domain.

The isobaric fix [60] is one approach to annihilate the growth of over-/under-heating shock

interaction errors. However, it was shown in the previous chapter that the isobaric fix does

not perform adequately for strong shocks interacting with a fluid-fluid interface. Further-

more, it was shown that the most effective way of dynamically correcting such errors was

to accurately resolve the wave interactions occurring at the interface. To this effect, a Rie-

mann problem was solved locally, normal to the interface, to decompose the singularities

in wave patterns and material properties. In section 3.1.2, the methodology to construct the

Riemann problem at the interface of a fluid object immersed in the flow was reviewed. The

same methodology can be extended to correct the heating errors arising due to shocks in-

teracting with the solid-fluid interfaces. Moreover, the Riemann solver constructed locally
 255

at the interface essentially solves the mass, momentum and energy balances at the interface

(Rankine-Hugoniot jump conditions), and implicitly adheres to the nature of the waves ar-

riving/leaving the interface. Hence the Riemann states obtained from the Riemann solver

are devoid of shock interaction errors and could be used to mitigate the growth of such er-

rors. The procedure described in section 3.1.2 is adopted here. Once the flow properties at

the interfacial ghost points are determined, the flow field corresponding to the interior ghost

points (located inside the solid) can be obtained by extending the flow properties using a

suitable multi-dimensional extrapolation procedure [15].

5.2.2 Boundary Condition Type II: Simple Boundary Condition (SBC)

SBC approach is the most direct way to enforce the boundary conditions (Eqs (1.92)

- (1.96)) accurately at the interface. This is a simple extension of the approach discussed in

Marella et al [124], where a methodology was developed to treat the presence of embedded

objects in an incompressible flow. A version of this approach was also discussed by Arienti

et al [11]. As shown in Figure 5.2(b), a probe is inserted through each interfacial ghost

point P in the solid phase. Coordinates of IP, the point of intersection of the probe with the

interface, are determined from Eq (3.1). Along the same lines, the coordinates of points

IP1 and IP2 are determined as follows:

~ IP 1 = X
X ~ IP + ∆xN
~P , (5.15)

~ IP 2 = X
X ~ IP + 2∆xN
~P . (5.16)

Flow properties at points IP1 and IP2 are obtained using bilinear interpolation from the

surrounding computational points. The normal derivatives required in Eqs (1.92) - (1.96)
 256

can now be easily computed as

∂ψ ψIP − 4ψIP 1 + ψIP 2

= , (5.17)
∂n 3∆x

where ψ corresponds to temperature T, or tangential velocities vt1 & vt2 or the pressure P.

The density at the interface can be obtained from the equation of state. The velocity of

the center of mass is specified as the normal velocity at the point IP. After determining the

flow properties at the interface, the values are extended to define the ghost field in the solid

phase.

It is important to note that this method suffers from severe over-/under-heating er-

rors. In particular, these errors are amplified during the transient shock-interface interac-

tion process. Evaluation of derivatives with the use of probes, especially during the initial

stages of the shock interaction with the interface, result in severe interpolation errors. If

these start-up errors are left uncorrected, they are fed back into the real fluid through the

ghost field. It was found that these errors cannot be corrected by augmenting this approach

with a local Riemann solver. This is because, populating the ghost points with SBC results

in an abrupt change in flow properties. Employing a Riemann solver would yield interme-

diate states which together with the real fluid would not satisfy the boundary conditions

(Eqs (1.92) - (1.96)) at the interface. Therefore, for strong shock interactions with the

embedded solid boundaries, the SBC approach is ill-advised.

 257

5.2.3 Boundary Condition Type III: Simple Boundary Condition using Coordinate

Transformation (SBC-CT)

Another method for treating the presence of embedded solid bodies is to use the

coordinate transformation/rotation technique [133, 135]. In this method, the discretization

of Eqs (1.92) - (1.96) are carried out in global Cartesian coordinates for each interfacial

ghost point in the solid phase, and then transformed to the local (body-fitted) curvilinear

coordinates by means of a suitable transformation matrix. Thus the normal and tangential

derivatives can be computed as follows;

   
∂ψ ∂ψ
 ∂n   ∂x 
   
   
 ∂ψ  = J ∂ψ , (5.18)
 ∂t1   ∂y 
   
   
∂ψ ∂ψ
∂t2 ∂z
 
 nx ny nz 
 
 
J=
 t1x t1y t1z  .
 (5.19)
 
 
t2x t2y t2z

where J is the Jacobian matrix and the unit vectors n̂, tˆ1 and tˆ2 are the normal and tan-

gential vectors computed from the level-set field [101, 133, 135]. This method minimizes

interpolation across the discontinuity and hence localizes the over-/under-heating errors to

few points close to the interface. Although this method appears to be very simple and

robust, there is a slight smearing of the interface due to the following reason: Eqs (1.92)

- (1.96) are valid at the interface and hence the normal and tangential derivatives must be

evaluated at the interface itself. Evaluating the derivatives at the interfacial ghost points

results in diffusing the interface over few grid points and hence using Eq (5.18) would not
 258

ensure the application of boundary conditions strictly at the interface.

5.2.4 Boundary Condition Type IV: Characteristic Based Matching (CBM) [133, 135]

Another promising approach to treat the presence of embedded objects in com-

pressible flows is to use the characteristic wave analysis at the interface [133, 135, 188].

As the governing equations are a set of hyperbolic conservation laws, the equations can be

recast in characteristic form using a suitable transformation matrix. The sign of the eigen-

values determine whether the waves are incoming or outgoing, and hence the direction of

propagation of information. The incoming waves carry information into the flow field and

hence correspond to the physical boundary conditions that need to be supplied indepen-

dent of the flow field. The outgoing waves carry information from within the flow field

and hence correspond to the numerical boundary conditions (obtained via extrapolation)

that need to computed from within the flow field. The characteristic analysis was carried

out by Thomson [188] for the domain boundaries and was extended by Poinsot et al [141]

for compressible turbulent flows. The method was applied for complex configurations in

curvilinear coordinates by Kim et al [107]. Nourgaliev et al [133,135] extended the method

to treat interfaces embedded in a Cartesian mesh in the GFM framework. The discussion in

this section follows from Thomson [188] and Nourgaliev et al [135] and interested readers

should refer to the original papers for additional information.

5.2.4.1 Coordinate Transformation

In order to obtain a non-oscillatory pressure field and to maintain sharp transition

in the field variables across the interface, it is desirable to work with primitive variables.
 259

Hence it is required to transform the conservative variable formulation of the governing

equations to the corresponding primitive (non-conservative) variable formulation [5, 102].

The associated transformation matrix along with the governing equations in the primitive

variable formulation is

∂U~ ∂ F~ ~
∂G ~
∂H
+ + + ~
= S, (5.20)
∂t ∂x ∂y ∂z
∂U ~
~ ∂ Ũ ~
∂ F~ ∂ Ũ ∂G ~
~ ∂ Ũ ∂H ~
~ ∂ Ũ
+ + + ~
= S. (5.21)
~
∂ Ũ ∂t ~
∂ Ũ ∂x ~
∂ Ũ ∂y ~
∂ Ũ ∂z

where  
 ρ 
 
 
 P 
 
 
~ 
Ũ = 

(5.22)
 u 

 
 
 v 
 
 
 
w
∂U~
Thus there exist an unique transformation matrix [P] = ~ with inverse given by [P ]−1 =
∂ Ũ
~
∂ Ũ
∂U~ However, in order to apply the the boundary conditions (Eqs (1.92) - (1.96)) at the

~ must be transformed to the corresponding primitive

interface, the conservative variable U

variable formulation in the local curvilinear coordinates U~˜⋆ , where U~˜⋆ is defined as
 
 ρ 
 
 
 P 
 
 
~˜⋆  
U = vn , (5.23)
 
 
 vt1 
 
 
 
vt2
 260

where vn , vt1 and vt2 are defined in Eq (1.97). The Jacobian matrix that enables the trans-

formation to the local curvilinear coordinates is defined in Eq (5.19). The transformation

~ to the primitive variable formulation U~˜⋆ in the local body

matrix [P], that transforms U

fitted coordinate, then becomes

 
 1 0 0 0 0 
 
 
 KE 1
 γ−1
ρvn ρvt1 ρvt2 

 
∂U ~  
[P ] = =
 u 0 ρnx ρt1x ,
ρt2x  (5.24)
∂ U~˜⋆ 



 v 0 ρny ρt1y ρt2y 
 
 
 
w 0 ρnz ρt1z ρt2z

where
1 2
1

KE = u + v 2 + w2 = vn 2 + vt1 2 + vt2 2 . (5.25)
2 2

Thus, pre-multiplying Eq (5.20) with the transformation matrix [P] we get,

∂ U~˜⋆ ∂ U~˜⋆ ∂ U~˜⋆ ∂ U~˜⋆

[P ] + [Ã⋆ ] + [B̃ ⋆ ] + [C˜⋆ ] − S~⋆ = 0, (5.26)
∂t ∂n ∂t1 ∂t2

where
 
∂ F U~˜⋆
~
[Ã⋆ ] = , (5.27)
∂U~˜⋆ 
~ U~˜⋆
∂G
[B˜⋆ ] = , (5.28)
∂U~˜⋆ 
~ U~˜⋆
∂H
[C˜⋆ ] = , (5.29)
∂ U~˜⋆
 261

     
and F U~˜⋆ , G U~˜⋆ , H U~˜⋆ are the projections of the Cartesian flux vectors on to the

local curvilinear coordinates [135],

 “   
~
”
∂F U˜⋆ ∂F (U
~)
 ∂n   ∂x 
   
~
“ ”
∂G U˜⋆
   ∂G(U~) 
  = [J]  . (5.30)
 ∂t1   ∂y 
   
~
“ ”
 ∂H U˜⋆   ~) 
∂H (U
∂t2 ∂z

Pre-multiplying Eq (5.26) with [P ]−1 we obtain

∂ U~˜⋆ ∂ U~˜⋆ ∂ U~˜⋆ ∂ U~˜⋆

+ [P ]−1 [Ã⋆ ] + [P ]−1 [B˜⋆ ] + [P ]−1 [C˜⋆ ] − [P ]−1 S~⋆ = 0. (5.31)
∂t ∂n ∂t1 ∂t2

where
 
 1 0 0 0 0 
 
 
 (γ − 1) KE (γ − 1) − (γ − 1) u − (γ − 1) v − (γ − 1) w 
 
∂ U~˜⋆
 
−1
 
[P ] = = − vn
0 nx ny nz .
∂U ~ 
 ρ ρ ρ ρ 

 
 vt1 t1x t1y t1z 
 − ρ 0 ρ ρ ρ 
 
 t2y

vt2 t2x t2z
− ρ 0 ρ ρ ρ
(5.32)

It is important to note that in the derivation of [P] and [P ]−1 , an ideal gas equation of

state has been used. With a more generalized equation of state (such as the Mie-Gruneisen

equation of state), the computation of matrices [P] and [P ]−1 may become exceedingly

complicated. Equation (5.31) is the required primitive variable formulation in the local

curvilinear coordinate.

5.2.4.2 Compatibility Relation

Equation (5.31) is a hyperbolic equation which can be diagonalized to obtain the

compatibility relation. However, it is not always possible to simultaneously diagonalize

 262

the matrices [P −1 ][Ã⋆ ] , [P −1 ][B˜⋆ ] , [P −1 ][C˜⋆ ]. Since we are interested in analyzing the

characteristic waves arriving at the interface, diagonalization of Eq (5.31) can be carried

 
out with respect to the normal direction (i.e., with respect to [P −1 ][Ã⋆ ] [188]. Grouping

all the transverse terms together Eq (5.31) can be written as,

∂ U~˜⋆ ∂ U~˜⋆ ~
+ [χ] + [S˜⋆ ] = 0. (5.33)
∂t |{z} ∂n |{z}
[P ]−1 [A˜⋆ ] ~
˜⋆ ~
˜⋆
[P ]−1 ([B˜⋆ ] ∂∂t
U
+[C˜⋆ ] ∂∂t
U
−S~⋆ )
1 2

Diagonalizing Eq (5.33) with respect to [χ] we get the compatibility relation [84, 188],

~˜⋆
−1 ∂ U ~
[L] + [L] + [L]−1 [S˜⋆ ] = 0. (5.34)
∂t

where  
 ρ 
 
 
 P 
 
 
~˜⋆  
U = vn , (5.35)
 
 
 vt1 
 
 
 
vt2
   
∂p ∂vn


 L1   λ1 ∂n − ρc ∂n 
   
  

∂ρ ∂p
  λ2 c2 ∂n − ∂n
 L2   
 
   
     
[L] =   =  ∂vt1 , (5.36)
 L3   λ3 ∂n 
   
     
 ∂v
 L4 
 
 λ4 ∂nt2 

   
  

∂p
L5 λ5 ∂n + ρc ∂v∂n
n

and

λ1 = vn − c, (5.37)
 263

λ2 = v n , (5.38)

λ3 = v n , (5.39)

λ4 = v n , (5.40)

λ5 = vn + c. (5.41)

are the eigenvalues, c is the speed of sound, [L]−1 is the matrix with left eigenvectors as

its rows and [L] is correction term matrix [133, 135] which is used to infuse the boundary

conditions at the interface. The eigenvalues λ1 and λ5 represent the sound waves moving

in the negative and positive direction normal to the interface, λ2 corresponds to the velocity

of the entropy advection , λ3 and λ4 correspond to the velocities with which vt1 and vt2 are

advected normal to the interface [188]. At any point on the interface, the sign of the eigen-

values determine the direction (positive for inflow and negative for outflow) of propagation

of characteristic waves and the Mach number determines the flow condition at that point.

For example, if the Mach number M ≤ 1, then the flow condition is subsonic. If the normal

velocity vn ≥ 0, then there are four incoming waves corresponding to λ2 , λ3 , λ4 , & λ5 , and

one outgoing wave corresponding to λ1 . This combination of flow with wave direction is

categorized as “subsonic inflow” with respect to the interface. Therefore, for a subsonic

inflow condition, four boundary conditions corresponding to the three velocity components

and temperature must be supplied. Equation (5.34) can be written in a convenient form by

pre-multiplying with [L] to obtain,

∂ U~˜⋆ ~
+ [L][L] + [S˜⋆ ] = 0, (5.42)
∂t
 264

 
1 1


 c2
L2 + (L5 + L1 ) , 
2
 
 
 1 
 2
(L 5 + L 1 ) 
 
 
[L] = [L][L] = 

1
(L 5 − L 1 ) .
 (5.43)
2ρc
 
 
 L3 
 
 
 
L4

The correction terms (L) given in Eq (5.43) are used to evolve variables corresponding to

the outgoing waves in time. The variables corresponding to incoming waves are specified as

boundary conditions. This way of applying the boundary conditions prevents over-/under-

specification of boundary conditions which in turn prevents spurious disturbances from

entering the flow field [188]. To solve Eq (5.42) in Cartesian coordinates at the interface of

the embedded object, requires the evaluation of the tangential and normal derivatives and

fluxes. This calls for a meticulous treatment of cut cells and explicit conservation of fluxes

at the interface which becomes a monumental task especially for a system of equations

like Eq (5.42). However, the difficulties associated with the cut cell treatment and with

the evaluation of derivatives can be alleviated if the characteristic analysis of Eq (5.42) is

carried out for the interfacial ghost points, as done in the CBM approach of Nourgaliev et

al [133, 135].

As the fluxes in Cartesian coordinates are computed for all points in the compu-

tational domain, one can make use of these fluxes to compute the normal and tangential
 265

derivatives occurring in Eq (5.42). Carrying out the diagonalization procedure yields,

∂ U~˜⋆ ~
+ [δL] + [S˜⋆ ] = 0. (5.44)
∂t |{z} “
|{z} ”
~ ~ ~
0 1

[P ]−1 ∂F
+ ∂∂y
G
+ ∂∂z
H ~
−S
B 1 ∂x
1
B 2
B c
δL2 + (δL5 + δL1 )
2
C
C
C
B C
B C
B C
1
B
B
B 2
(δL5 + δL1 ) C
C
C
B C
B C
B C
1
B
B
B 2ρc
(δL5 − δL1 ) C
C
C
B C
B C
B C
δL3
B C
B C
B C
B C
B C
B C
@
δL4 A

The form of Eq (5.44) is similar to that of Eq (5.42) except now the source term in Eq (5.44)

includes fluxes in Cartesian coordinates pre-multiplied with the transformation matrix.

The characteristic analysis can now be performed on the interfacial ghost points using

Eq (5.44) [135]. As mentioned before, at each interfacial ghost point, the Mach number

and the sign of the normal velocity determine the flow condition and the flow direction

respectively. Based on the flow condition and the flow direction, the boundary treatment at

the ghost point fall into one of the four categories;

i. subsonic inflow - four incoming waves and one outgoing wave,

ii. subsonic outflow - four outgoing waves and one incoming wave,

iii. supersonic inflow - All five waves are incoming waves,

iv. supersonic outflow - All five waves are outgoing waves

Thus for the variables corresponding to the incoming waves, the respective boundary con-

ditions from Eqs (1.92) - (1.96) must be used to find the values. Once the variables cor-

responding to the incoming waves are evaluated, the correction terms can be computed by
 266

inverting Eq (5.44). For the outgoing waves, the corresponding variables must be evolved

in time depending on the flow field at that instant. The correction term for the outgoing

wave is set to zero (as the normal derivative terms are taken into account in the modified

source term defined in Eq (5.44), the explicit evaluation of the correction terms will be

annulled by the corresponding terms in the source term). For example, for a subsonic in-

flow case, there are four incoming waves corresponding to λ2 , λ3 , λ4 , & λ5 and hence the

correction terms δL2 , δL3 , δL4 , & δL5 must be evaluated from Eq (5.44), and one outgoing

wave corresponding to λ1 and hence δL1 is set to zero [135].

δL1 = 0, (5.45)
 
∂vt1 ˜⋆
n
δL3 =− −S 4 , (5.46)
∂t
 
∂vt2 ˜⋆
n
δL4 =− −S 5 , (5.47)
∂t
 
∂vn ˜⋆
n
δL5 = δL1 − (2ρc) −S 3 , (5.48)
∂t
   
1 2 ˜⋆
n ∂ρ
δL2 =− (δL5 + δL1 ) − c S 1 + (5.49)
2 ∂t

Using the correction terms δL1 − δL5 , pressure term in Eq (5.44) can be computed as

follows
n 1
P n+1 = P n − ∆t(S˜⋆ 2 + (δL1 + δL5 )). (5.50)
2

Density can then be retrieved by using the equation of state. Iterations are performed until

convergence in the computed values of pressure and density are achieved [135]. The same

procedure is repeated for all interfacial ghost points and all flow variables can be evaluated.

These flow variables are then extended to the interior ghost points in the ghost band.
 267

5.2.4.3 Other Methods

Other methods for applying interface conditions on embedded solid boundaries

have been discussed at length by Arienti et al [11]. The methods include mirroring tech-

nique, reflection method and direct injection method. In the mirroring technique [62], the

flow properties on the real fluid side are mirrored onto the ghost fluid by multiplying with

a suitable rotation/reflection matrix. The reflection method discussed in Arienti et al [11]

simply extends the flow properties at the real fluid interfacial nodes, but with an inverted

sign of the normal velocity to satisfy the no-penetration boundary condition. The injection

method formulated by Arienti et al [11] is a variation of SBC. In this approach, flow vari-

ables at the real fluid interfacial points are extended to the ghost points, but now the normal

velocity is set to zero, preserving the tangential velocities to depict the slip condition at the

interface. These methods are mentioned here only for the sake of completeness and will

not be considered in the numerical examples to be discussed in the next section.

5.3 Numerical Examples

5.3.1 One Dimensional Examples

The one-dimensional examples shown here demonstrate the performance of various

(interface treatment) methods in handling regular shock reflections. It is found that the

choice of method impacts significantly on the accuracy and stability of the computations.

5.3.1.1 Rotated Shock Tube

A variant of the one-dimensional shock tube problem due to Nourgaliev et al [135]

is considered first. As shown in Figure 5.3, the walls of the shock tube are embedded in
 268

the computational domain by defining a suitable level set field. The shock tube is oriented

at an angle to the regular Cartesian mesh so that a two-dimensional problem is solved, but

the problem is one-dimensional along the centerline of the shock tube.

The discontinuity at T = 0 is at the center of the shock tube. The shock tube is

placed at different orientations relative to the grid lines. Rotation angles corresponding to

αrot = 30o , αrot = 45o , and αrot = 60o with respect to the horizontal have been consid-

ered. The size of the computational domain is 2 × 2 units. The length and the height

of the shock tube were chosen to be 1.5 and 1.0 units respectively. The simulations were

carried out to T = 0.164. The results obtained from the current simulations are shown in

Figure 5.4. RBC augmented with local Riemann solver (henceforth RBC-RS), SBC and the

CBM approaches were used to simulate this problem. Figure 5.4 shows good agreement

between these methods and with the exact solution, in capturing the shock and the contact

discontinuity. The simulations were stopped to prevent the shock from reflecting off the

wall, so that the resulting numerical solution could be compared with the exact solution.

5.3.1.2 Woodward Colella (Bang Bang) Problem

Unlike the previous case, this example involves strong shock reflections and shock

collisions. The initial conditions along with the geometrical set up is given in Figure 5.5.

As in the previous example, the shocktube is placed at different orientations (αrot =

30o , and αrot = 60o ) so that the problem is not strictly one-dimensional and the impact

of the interface treatment and mesh orientation can be examined. The simulations were run

to time T = 0.038 units and the plots of density, pressure and entropy along the centerline
 269

of the shocktube are displayed in Figure 5.6. Note that the SBC, unable to suppress the

initial transient shock development errors, failed in the first few times steps of the shock

reflection process.

From the entropy plots displayed in Figures 5.6(e) & 5.6(f), it is evident that RBC

suffers from extensive over-heating errors that appear to be problem dependent (over-

/under-heating errors were not observed in the previous example). Clearly the over-heating

errors associated with the RBC cannot be neglected and hence must be corrected in order

to obtain better solutions in the presence of strong shocks. On the other hand, the RBC-

RS shows no significant over-/under-heating errors. Both CBM and the SBC-CT produce

under-heating errors which are small. This example clearly illustrates the requirement of

an additional correction procedure to suppress the errors associated with shock-interface

interactions.

5.3.1.3 Shock Reflecting off a Stationary Wall

This is strictly one-dimensional test case with Mach 10 and Mach 20 shocks reflect-

ing off a stationary wall. The wall is positioned at x = 0.25 and the incident shock is initially

located at x = 0.35 in a computational domain of unit length. The shock impinges on the

wall and reflects back bringing the oncoming fluid to rest. Again SBC failed exactly at the

instant when the shock impinged on the wall. Clearly, this shows that the SBC suffers from

severe interpolation errors especially in the presence of strong shocks near the interface.

The results obtained from the current simulations can be found in Figure 5.8. The position

of the reflected shock obtained from the RBC with and without the Riemann correction
 270

differs by a fraction of a grid cell from that obtained using CBM. This discrepancy in the

shock location can be attributed to the mass conservation errors occurring during the initial

shock formation process. Once the shock has developed and detaches from the surface, the

interface recovers back to its normal motion and hence the conservation errors are localized

to the few time steps [118] during which the shock impinges on the interface.

Thus, it is clear from the one-dimensional examples presented in this section that

the SBC approach suffered from severe over-/under-heating errors that had catastrophic

consequences in the final outcome of the solution. The RBC approach incurred over-/under-

heating errors that were problem dependent (Figures 5.6(e), 5.6(f), & 5.8). In contrast, the

RBC-RS and CBM method resulted in lower levels of over-/under-heating errors and were

consistent in predicting satisfactory solutions. This clearly favors the use of CBM and RBC

approach with Riemann solver correction over other methods.

5.3.2 Two-Dimensional Examples

In the one-dimensional test cases considered in the preceding section, the shocks

interacted with interfaces to produce regular reflections. Multi-dimensional shock diffrac-

tion patterns can be irregular (consisting of Mach stem, slip lines emanating from Mach

stem and eventual roll up of slip lines due to baroclinic vorticity generation, etc.). In this

section a series of multi-dimensional test cases are presented to illustrate the performance

of the GFM-based approaches in such situations.

 271

5.3.2.1 Shock Diffraction over Wedge

The first two-dimensional test case considered in this work, is the diffraction of a

Mach 1.3 shock wave past a wedge. The details of the initial configuration of the prob-

lem can be found in Sivier et al [166]. This problem has received considerable attention

in the past, and extensive numerical and experimental documentations are available for

comparison [166]. Snapshots of the numerical Schlieren image, at different instants in

time, obtained from using both the RBC-RS and the CBM approach are displayed in Fig-

ures 5.9 & 5.10. Although the contour plots generated using the two approaches look

very similar, there are subtle differences that are worth noting. As mentioned in Sivier et

al [166], the planar shock, upon impacting the wedge, forms a cylindrical reflected “bow”

shock centered at the wedge tip with a Mach stem attached to the wedge surface (Fig-

ures 5.9(a) and 5.9(b)). All typical features of Mach reflections are captured well by both

(RBC-RS & CBM) methods. However, both methods fail to capture the slip line origi-

nating from the wedge surface which is clearly visible in Sivier et al. But in Sivier et al,

an Adaptive Mesh Refinement (AMR) technique was employed which provides sufficient

mesh enrichment to capture such fine details. Figures 5.9(c) and 5.9(d) correspond to the

instant when the incident shock wave has passed over the wedge surface. As observed by

Siver et al, the roll up of slip layers to form vortices that engulf the vertex of the entropy

fan are clearly visible. At this instant, the opposite shock systems from the symmetrical

side of the wedge have crossed and have begun to interact with these vortices. The ex-

pansion fan, which is produced as a result of the diffraction of the Mach stem around the

corner [166] and which subsequently moves down the wedge surface, is also visible in the
 272

figures. Figures 5.9(e) - 5.9(h) show that the vortices have shifted to the right, downstream

of the wedge. The slip layer emanating from these vortices are now visible along with

a new triple point formed at the vortex center. All of these features typical of the shock

diffraction phenomena past a wedge are consistent with other numerical results [166] and

with experimental observations [156].

Plots of the primary shock distance from the wedge surface are shown in Fig-

ure 5.11. Shown in the figure, in accordance with Sivier et al, are the horizontal shock

detachment distance “a” of the reflected bow shock from the foot of the wedge (Fig-

ure 5.11(a)), vertical distance “r” from the tip of the wedge to the diffracting shock (Fig-

ure 5.11(b)), and the horizontal distance “x” between the tip of the wedge and the incident

shock (Figure 5.11(c)). The plots show excellent agreement between the two methods in

determining the position of the primary shocks and are consistent with the predictions made

by Sivier et al [166].

5.3.2.2 Shock Diffraction over a Circular Cylinder - Comparison with CBM

A Mach 2.81 planar shock wave negotiating a cylindrical obstacle in a shock tube

is considered. As pointed out by Bazhenova et al [20], the initial interaction of the planar

shock with the cylinder is a regular reflection with the shock reflecting normally along the

axis of symmetry. The reflected shock begins traversing the cylinder before transitioning to

a Mach reflection. The Mach reflection consists of the incident shock, the reflected shock,

Mach stem and slipstream. Ripley et al [149] pointed out that this curved Mach stem

travels down the cylinder and interacts with the shock system from the symmetrical part of
 273

the cylinder to form a wake. Several papers documenting this phenomenon are available

for comparison [20, 31, 149, 158, 176, 204].

Due to the symmetry of the problem, only one-half of the problem was modeled

using a symmetry condition at the bottom wall. Reflective boundary condition for the top

wall and Neumann condition for outlet were imposed. At the inlet, a Dirichlet condition

corresponding to Mach 2.81 shock was enforced. The semi-circular obstacle with radius

0.105 units was positioned at x = 0.5 and y = 0.0. The simulations were performed with

500 × 250 mesh points in a 1 × 0.5 computational domain. The initial discontinuity was

located at x = 0.385 units so that the sharp discontinuity relaxes to a numerical shock profile

before impacting the obstacle.

Results from the current simulations are shown in Figures 5.12, 5.13 and 5.14. Ex-

perimental results obtained by Bazhenova et al [20] are shown in Figures 5.12(e) - 5.13(e).

Figure 5.12 shows the curved Mach stem, with the reflected and incident shock, just before

interacting with the opposite shock system. Both CBM and RBC-RS agree well with each

other and also with the experimental results. Figure 5.13 shows the triple point and wake

formed after the shock interaction. The contact discontinuities, upper and lower triple

points, and the Mach systems are captured well. Although the density contours in Fig-

ure 5.14 shows a somewhat wider wake for the CBM approach, there is no other discern-

able difference between the results. The figures also indicate that the large scale features of

the diffraction pattern observed particularly in the wake region appear more prominent for

the RBC-RS approach in comparison with CBM method. For a quantitative comparison

between the two methods, the bow shock detachment distance over time and the locus of
 274

upper triple point were tracked [20]. Figure 5.15 shows the plots obtained from the cur-

rent calculations. Figure 5.15(a) predicts a parabolic behavior for the shock detachment

distance which is consistent with the numerical results obtained by Ripley et al [149]. The

angle, made by the locus of the upper triple point with the horizontal, obtained using the

RBC-RS is 31.3o and that obtained using CBM is 30.5o . The correlation obtained from

experimental observations predicts an angle of 33o [100, 149] which is in close agreement

with the current numerical calculation.

5.3.2.3 Diffraction of a Planar Shock Wave Negotiating a Cylindrical Obstacle - LMR

Calculations

The shock-cylinder interaction problem is repeated but with dynamic mesh adapta-

tion. The purpose of this example is to investigate the impact of LMR on the final solution,

present guidelines for selecting refinement criteria and to validate the proposed RBC-RS

approach. To validate the current method, the coefficient of drag (CD ) obtained from the

current simulations are plotted against experimental [182] and numerical [53] results (Fig-

ure 5.16) for two different shock strengths (Mach 1.3 and 2.6). The simulations were

1
carried out with a base mesh of size ∆x = 200
with 4 levels of refinement. As is evident

from the Figure 5.16, the drag vs time obtained is in good agreement with experimental

and numerical predictions. The peak values are predicted accurately for both cases. For

the Mach 1.3 shock wave, the trend starts to depart when the shock wave starts to diffract

off the lee side of the cylinder, i.e. in the wake region. This is because, in the current

computations viscous effects are not taken into account, whereas the numerical predictions
 275

reported in [53] were obtained by solving Navier Stokes equations. Effect of viscosity be-

comes predominant in the wake region, particularly for shocks of moderate strength. For

strongly shocked flows, the drag coefficient CD is dominated by form drag. This can be

verified by comparing the trend observed for the Mach 2.6 case. By doubling the Mach

number, the prediction for drag coefficient obtained by solving (inviscid) Euler equations

approaches more closely the trend obtained from experimental observations.

High Fidelity Calculation for Mach 2.81 Shock Wave Negotiating a Cylindrical Ob-

stacle:

Next, the diffraction patterns observed for a Mach 2.81 shock wave negotiating a station-

1
ary cylindrical obstacle is considered. For this purpose, a fine base mesh of size ∆xg = 300

with 4 levels of mesh refinement was selected (an effective resolution of 3 million grid

points). δ1 and δ2 were set at 10.0 and 250.0 respectively. The results from the current

calculations are presented in Figure 5.17. The Mach reflection consisting of the incident

shock, the reflected shock, the Mach stem and the slipline are visible in Figure 5.17(a).

The slipline emanating from the Mach stem, that was not captured in Figure 5.20, is clearly

visible in Figure 5.17(a). In addition, the roll up of slipline is also captured accurately.

The curved Mach stem travels down the cylinder and interacts with the shock system from

the symmetrical part of the cylinder to form the wake region (Figures 5.17(b), 5.17(c)

& 5.17(d)). The enlarged view of the wake regions at two different instants in time are

displayed in Figure 5.18. In Figure 5.18(b), the onset of Kelvin-Helmholtz instability in

the wake region, that were previously not reported in the literature, can be seen. For the
 276

sake of comparison, the enlarged view of the wake region from a uniform mesh calculation

1
(∆x = 800
) is presented (Figure 5.18(a)). As can be seen from Figures 5.18(a) & 5.18(b),

the onset of instability is only apparent with sufficiently high mesh refinement. The sub-

sequent development of the instability at a later instant in time (T = 2.4 µs) is shown in

Figure 5.18(d). Figure 5.18(c) shows the enlarged view of the mesh in the wake region at T

= 1.1 µs. The mesh adaptation is excellent and closely follows the delicate contours in the

wake region. The OR computed for this simulation is only 8.7 %. In Figure 5.19, the locus

of upper and lower triple point are plotted with experimental [182] and numerical [53] pre-

dictions. The trends are in excellent agreement. In Figure 5.19(b), a perfect parabolic fit

is obtained which is consistent with the predictions made in [149]. Furthermore, the angle,

made by the locus of the upper triple point with the horizontal is 33o and which is the exact

value reported in the experimental observations in [100].

Computational Time Study:

As the diffraction patterns are rich with features typical of shock-interface interaction

phenomena, the problem is selected as a prototype for performing timing analysis. The

simulations were evolved to time T = 6.28 µs. The refinement criteria for the LMR calcu-

lations were set to δ1 = 10.0 & δ2 = 25.0. For a representative measure of computational

time, the percentage CPU time ratio were computed with respect to the CPU time taken by

1
the uniform fine mesh calculation (∆x = 800
). The solutions were evolved for 4 different

(3, 4, 5 & 6) levels of mesh refinement while maintaining the size of the fine mesh cells

1
constant, i.e. ∆xf = 800
& ∆xg = ∆xf × 2k−1 where k is the refinement level. The details
 277

from the current calculations are displayed in Table 5.1.

As is evident from the table, although the OR remains almost constant, the % CPU

time increases with increase in the levels of mesh refinement. Particularly for the case

with 6 levels of mesh refinement (on a coarse base mesh of size ∆xg = 0.04), the CPU

time is well above 100%. The reason for this increase in CPU time can be attributed to a

coarse base mesh used in the calculation, and the interpolations performed over different

grid levels at the fine-coarse mesh boundary. Hence, for fast computations with a given OR,

the recommended limit on the mesh refinement level is 5. Furthermore, considering speed

and accuracy of the calculation, 4 levels of mesh refinement can be chosen as an optimal

refinement level.

In Figure 5.20, the LMR-based solution with 6 levels (corresponding to a refine-

ment factor of 32) of mesh refinement (top portion of the figure) are compared with the

1
corresponding fine mesh calculations with ∆x = 800
(bottom reflected portion of the fig-

ure). As can be seen from the figure, the shock locations and the wake regions are resolved

accurately. The density contours (Figures 5.20(a), 5.20(c) & 5.20(e)) reveal striking simi-

larity between the two solutions. The numerical Schlieren image (Figure 5.20(f)), indicate

that the LMR-based calculation was not able to adequately resolve the slipline (visible in

Figure 5.20(d)) in the wake region. Features such as slip lines and other weak structures

in the flow domain can be accentuated by suitably tuning the threshhold limit for the re-

finement criteria. The pressure distribution on the surface of the cylinder are plotted in

Figure 5.21. As the interface is always maintained at the finest level, the pressure distri-

bution for different levels of mesh refinement are in excellent agreement with the uniform
 278

mesh solution.

Sensitivity to Refinement Criteria:

Next, the sensitivity of the solution to the refinement criteria is examined. For this pur-

pose, the calculations were repeated on a base mesh of size ∆xg = 0.01 with 4 levels of

refinement. To examine the impact of δ1 on the solution, simulations were performed for

a range of values of δ1 with δ2 held constant (at arbitrary large value). A similar approach

was adopted to examine the effect of δ2 on the solution. The two sets of data are presented

in Figure 5.22. As is apparent from the figure, with the decrease in the threshold limit,

the L2 (S) error decreases with increase in the occupancy ratio. The trend observed with

respect to the L2 (S) error for δ1 & δ2 are similar. This is because the primary shock sys-

tems are resolved with decrease in threshold limit. In contrast, the trend observed with the

occupancy ratio differ significantly from one another. This is because, with the decrease in

the threshold limit for δ2 , weak features/gradients are accentuated and resolved to the finest

level. Thus to balance accuracy and speed of the calculation, the suggested range for the

1.0
threshold limits, δ1 and δ2 , are δ1 ∆xg ≈ (∆xg − ∆xg
) and δ2 ∆xg ≈ (100∆xg − 2.5∆xg )

respectively, where ∆xg is the size of the global base mesh.

5.3.2.4 Shock Diffraction over Multiple Cylinders

This test case is considered to demonstrate the ability of the interface treatment

methods to handle multiple embedded objects. This example is similar to the previous one

except that, in this case there are three asymmetrically placed cylindrical obstacles in the

path of a planar shock wave. The cylinders of radius 0.1 units were placed at (0.3,0.3),
 279

(0.35,0.7), and (0.5,0.1) in a 1 × 1 computational domain with 500 mesh points in each

direction. A Mach 3 planar shock was initially positioned at x = 0.19. The results obtained

from the current simulations are shown in Figure 5.23. Table 5.2 shows a comparison of

CPU time taken by the two methods for one complete time step. Since the simulations were

carried out on a fine mesh and due to the presence of multiple embedded objects, the CPU

time for this example is a good representation of typical computation time for the problems

attempted in the present work. Note that the transformation matrix computed in obtaining

the compatibility relation is a function of time and when subject to strong transient shocks,

the inherent linearization process assumed in obtaining the characteristic equations does not

hold true [133]. As pointed out by Nourgaliev et al [133], these errors can be minimized

by employing sub-iterations, i.e. by advancing in a series of small time steps for each RK

time iteration. Although the CPU time taken by CBM without sub-iterations and RBC-RS

are comparable, the CBM method with sub-iterations takes larger amount of CPU time to

converge. Therefore the CBM and RBC-RS methods are comparable in terms of compu-

tational time. The choice of RBC-RS is however attractive due to the relative simplicity

of implementation of the approach, since one can operate in the primitive-variable Carte-

sian formulation. Despite the simplicity of formulation and implementation of RBC-RS

approach, the results produced by these two methods are in good agreement.

5.3.2.5 Shock Diffraction over Rectangles and Squares

The next example considered is the interaction of a planar shock with a square and

rectangular cylinder respectively. Previous attempts to simulate the shock interactions with
 280

rectangular cylinders have been reported by Sun et al [176] and Ripley et al [149]. Both

Sun et al and Ripley et al have used unstructured body fitted grids while in the present work

the interfaces are embedded in an underlying Cartesian mesh. The problem consists of a

Mach 2 shock impinging on a rectangular cylinder with a length to breadth ratio of 4. The

cylinders were placed at an angle of 60o (angle made by the minor axis with horizontal).

The center of the cylinder was located at (2.75,1.5) in a 7.5X3.3 domain. A 750 X 303 mesh

was used and the simulations were run till non-dimensional time T = 3.0. The top and the

bottom walls were imposed with reflective boundary condition and the inlet with Dirichlet

condition corresponding to a Mach 2 shock. The density contours and numerical Schlieren

image obtained from the current simulations are shown in Figure 5.24. Both methods

employed are in good qualitative agreement in capturing most of the typical features of

shock diffraction phenomena, such as the multiple shock reflections, diffractions, and the

vorticity generated due to baroclinic torque [176].

5.3.2.6 Dynamics of a Cylinder Subject to Mach 3 Shock

The first two-dimensional moving boundary problem demonstrated in this work

is the dynamics of a cylindrical object subject to a Mach 3 shock. This problem was

considered by Falcovitz et al [55] and later by several others [11,61]. The problem consists

of a Mach 3 shock impacting a cylindrical object which is initially at rest and located at

the bottom of the domain. The subsequent motion of the cylinder results in the cylinder

lifting off the lower surface. The motion of the cylinder is computed by integrating the

fluid forces acting on the cylinder at each instant. The numerical Schlieren image from
 281

the current calculations are displayed in Figures 5.26(a) & 5.26(b). The locus of the center

of mass of the cylinder is shown in Figure 5.26(c) for comparison. Thus, it is clear from

Figure 5.26(c) that the numerical predictions of the postion of the cylinder made by the

two approaches are in excellent agreement with each other. Furthermore, the numerical

Schlieren image displayed in Figures 5.26(a) & 5.26(b) closely follows the trend observed

by Forrer et al [61].

Thus, despite of few subtle differences, it can be inferred from the numerical ex-

amples shown above that there are no significant discrepancies in the numerical results ob-

tained using CBM and RBC-RS methods. Considering the relative ease of implementation

and speed of computation, RBC-RS is preferred over the CBM method. In the follow-

ing examples, the robustness and versatility of the RBC-RS approach are demonstrated for

shocks of different strengths interacting with embedded (stationary and moving) objects of

complex shapes and orientations.

5.3.2.7 Double Mach Reflection of a Strong Shock

This is a popular test case and has been used in the past to test the numerical ac-

curacy of several hydrocodes. Extensive discussion on the performance of different shock

capturing schemes for this problem can be found in Woodward et al [200]. Pember et

al [138] have used this example to test the accuracy of their adaptive refinement method.

The problem consist of a Mach 10 shock impacting a ramp. The Ramp is inclined at 30o

with the horizontal and located at x = 0.3. The shock, which is initially located at the foot

of the ramp, upon impacting the ramp forms a double Mach reflection. Several grid sizes
 282

were considered in the current simulation, and the resulting plot of density variation along

the ramp are shown in Figure 5.28. For all grid sizes, the position of the reflected shock

and the double Mach reflection formed are in close agreement. The contours of density and

the corresponding Schlieren image for this case can be found in Figure 5.27. The results

concur well with those obtained by Woodward et al [200] and Pember et al [138].

5.3.2.8 Mach 3 Wind Tunnel with a Forward Facing Step

This example was introduced by Emery et al [54] and since then has been used ex-

tensively to test the accuracy of several numerical schemes [200]. A Wind tunnel 3 units

long and 1 unit high, has a backward facing step of height 0.2 units, located at 0.6 units

from the entrance of the tunnel. The tunnel is continuously supplied with a gas having

properties, ρ = 1.4, u = 3.0, v = 0.0, P = 1.0 and γ = 1.4. The presence of the step in the

tunnel constricts the flow forming a bow shock, which interacts with the top wall and the

horizontal portion of the step forming multiply reflected shock structure. Initially the reflec-

tions are regular, but as time elapses these reflections become Mach reflections with Mach

stem formed both at the top wall and at the horizontal portion of the step. The slip streams

emanating from these Mach stems are subject to Kelvin-Helmoholtz instability which can

be captured with sufficiently high mesh resolution. The results from the current simulations

are shown in Figure 5.29. The results show the Mach stem along with the slipstreams gen-

erated during the shock reflection phenomenon. Also, the numerical Schlieren image show

the formation of the Kelvin-Helmoholtz instability. Clearly, as can be seen from plots, with

the increase in mesh resolution, the instability becomes more predominant.

 283

5.3.2.9 Shock Diffracting of a Backward Facing Step

A wide range of numerical and experimental data can be found for a shock diffract-

ing around a backward facing step. The resulting shock diffraction pattern consists of

features which are self similar and Mach number dependent. For instance, if the post shock

flow is supersonic then the reflected supersonic rarefaction wave demarcates the shocked

and unshocked region with the reflected wave being on the edge of the Prandtl Meyer ex-

pansion wave otherwise the reflected wave forms a smooth transition between the shocked

and the unshocked region. For a detailed discussion of the flow features in this system,

the reader is directed to Hiver et al [85]. In the current work, we have considered three

examples; Mach 1.3 shock, a Mach 2.4 shock and a Mach 20 shock diffracting of a 90o

corner. The first two cases can be compared with numerical [149] and experimental [157]

results available in literature, while the third case is considered mainly to illustrate the

performance of the method in high Mach number flows. The density contours and the

corresponding Schlieren images generated from the current simulations have accurately

captured the vortex roll up mechanism, the associated Prandtl Meyer expansion waves, the

secondary shock wave formed in the diffraction phenomenon and the curved incident shock

wave.

5.3.2.10 Mach 2.1 Shock Negotiating a Curved Channel

In this example, a Mach 2.1 shock traverses through a curved channel with a diverg-

ing exit. The resulting shock diffraction phenomenon consist of a curved Mach stem, slip

line roll up and several other interesting features. The problem has been discussed in detail
 284

by Nourgaliev et al [135] and hence only the results obtained from the current simulations

(using RBC-RS approach) are presented here. Interested readers may refer to Nourgaliev

et al [135] for additional details. The simulations were carried out with a particle negoti-

ating a curved channel with a prescribed motion [135]. Figure 5.35 shows the numerical

Schlieren image and the density contours obtained from the current simulations. As can be

readily seen, the results obtained using RBC-RS approach compare very well with that of

Nourgaliev et al [135].

5.3.2.11 Mach 3 shock interacting with multiple (120) stationary objects

A Mach 3 shock wave interacting with randomly oriented stationary obstacles is

considered. The problem consists of 120 arbitrarily oriented and randomly positioned ellip-

tical and cylindrical obstacles in a 25 × 6 computational domain. Although the coordinates

of the center of mass of the particles were generated randomly, care was taken in placing

the particles particularly on the bottom boundary so as to depict a symmetrical problem

of 60 particles that are distributed randomly in the x direction. These stationary particles

which are immersed in quiescent air, were subject to a Mach 3 shock. As seen from Fig-

ure 5.36, the shock after negotiating through these (fixed) obstacles develops a complex

diffraction pattern, in addition to the reflected and the transmitted shock wave that emerge

from the diffraction pattern. This example demonstrates the potential of the present method

for the application to state of the art Direct Numerical Simulations (DNS) of a shock wave

traversing through a porous medium or a dusty gas [152, 168].

 285

5.3.2.12 Mach 8 spherical shock interacting with multiple (24) moving objects

The second example considered in this section, is an explosion corresponding to

Mach 8 cylindrical shock in the presence of 24 symmetrically positioned particles that are

free to move under the influence of the shock wave. Again, due to the symmetry of the

problem, only one quadrant of the problem was modeled. As in the example considered

in section 5.3.2.6, the velocity of the center of mass of the particles were obtained from a

simple force balance. The external fluid force acting on the particles were computed by

integrating the pressure force around each particle. The density and size of the particles

were (randomly) chosen so that the collisions between the moving particles and with the

domain boundaries were averted. Hence collision models, such as in Nourgaliev et al [134],

in the presence of shocks were not included in this calculation. Figures 5.37 & 5.38 show a

series of numerical Schlieren image and vorticity contours at different instants in time. The

initial cylindrical shock, upon interacting with the particles disperse the particles from their

initial positions. The cylindrical shock front, after reflecting off the walls of the domain,

interacts again but now in the opposite direction to the motion of the particles, thereby

slowing the particles.

5.3.2.13 Diffraction Patterns Generated by Supersonic Bluff-body Projectile(s)

Previous investigations [95, 97, 155] on ballistics of projectiles were limited to sta-

tionary or moving projectiles with simple square or rectangular shapes that conform with

the computational mesh. The experimental and the numerical simulations documented for

this problem were focused on the initial unsteady shock wave diffraction phenomena (such
 286

as the blast wave, the jet flow, the vortex ring etc.) generated due to the movement of the

projectile at the vicinity of the outlet of the launch tube. In this example, a realistic scenario

of a bluff body projectile fired at transonic Mach number (M = 1.22) moving through the

mesh (as opposed to solving in a moving coordinate system attached to the projectile) is

studied. Attention is paid to the later part of the projectile motion and the resultant diffrac-

tion patterns. Configuration of the system and the computational domain are displayed in

the Figure 5.39. The simulation was carried out with base mesh of size ∆xg = 0.02 with 5

levels of mesh refinement. The threshhold for the refinement criteria were set at δ1 = 1.0

& δ2 = 1.0.

The numerical Schlieren image along with the mesh topology generated from the

current simulations are displayed in Figure 5.40. The initial movement of the projectile

results in a series of compression waves that coalesce to form the “bow shock” that stands

off from the nose of the projectile (Figure 5.40(a)). The diffraction pattern observed at the

rear end of the projectile is similar to that of shock wave negotiating 90o step. Figure 5.40(a)

shows the Prandtl-Meyer expansion waves and slip lines generated at the rear end of the

projectile. As expected, the slip lines produced eventually roll up to form vortices (vortex

rings in this case) that are exemplary of diffraction patterns of shock wave negotiating

a 90o step [175]. These vortices are clearly visible in Figure 5.40(b). Figures 5.40(c)

& 5.40(d) correspond to the instant at which the projectile has moved sufficiently far away

from the launch tube/pad. At the front end of the projectile, the bow shock formed stands

off sufficiently far away from the nose of the projectile. At the rear end of the projectile,

the baroclinic vortices generated are convected downstream relative to the movement of the
 287

projectile (Figure 5.41(a)). In Figure 5.41(b) the front end bow shock stand-off distance

is plotted tracked as a function of time. The shock stand-off distance increases linearly

with time for time T ≤ 0.5 ms and subsequently reaches a steady state value of 0.22 units

(0.55 body diameter) which is typical of projectile fired at transonic Mach numbers [1].

Figures 5.40(a) through 5.40(d) indicate that evolution of the mesh closely follows the

primary features in the flow field. The occupancy ratio computed for this simulation is

23.25 %.

5.3.2.14 Dynamics of Particles in High Speed Flows

The shock wave processes in particle laden flows are significantly different from

the single phase gaseous counterpart, owing to the momentum and energy exchange be-

tween the solid and the gaseous phase which modifies the shock wave diffraction pattern.

Typical applications include shock interaction and resultant ignition/detonation of solid

propellants/explosives [112], explosive dispersion of pollutants and particles [114, 208],

hypervelocity thermal spray and coating technologies [51] etc. Previous simulations re-

ported in the literature were limited to heuristic models that predict forces due to planar,

and in some cases, spherical shock - (point) particle interactions. The corresponding DNS

simulations were restricted to single, stationary or moving particle [53, 90, 153, 174]. In

this example, an attempt is made towards DNS simulations of multiple particles interacting

with planar and curved bow shocks. The example focuses on determining the dynamics

of particles introduced in a (planar) shocked flow, and later impacted by a Reflected Bow

Shock (RBS) wave. The configuration of the problem along with the initial location of the
 288

shock are depicted in Figure 5.42. A computational domain of size 10 × 10 is chosen for

the current simulation. A planar shock wave of Mach 2.81 is located at x = 4.75 units.

Cylindrical particles with densities 10 times the fluid (air in this case) density are intro-

duced in the post-shocked flow as shown in the Figure 5.42. A stationary bluff body with

a semicircular front edge is positioned at the bottom of the domain. The presence of the

stationary body generates the RBS wave that later interacts with the oncoming particles. A

base mesh of size ∆xg = 0.1 with 4 levels of refinement are employed for the simulation.

The threshold for the refinement criteria were set at δ1 = δ2 = 10.0. The simulation was

run to time T = 10.0 ms.

The results from the current simulations are presented in Figure 5.44. The initial

motion of the particles is due to the relative velocity between the fluid and the particles.

Due to this relative motion, a bow shock wave (with pressure ratios 2.67 for objects 4-5

and 1.86 for objects 1-3) develops that is typical of shock wave interacting with cylindri-

cal/spherical particles. As a result of the development of the bow shock wave, a steep rise

in the drag force on the particle is observed. This can be verified from the plots shown in

Figure 5.43(c). The trend observed is similar to a shock interacting with single particle.

These features are clearly visible in Figure 5.44(a). Also visible in Figure 5.44(a) is the

initial stage of the development of the RBS wave and the incident shock wave near the

bluff body. The pressure ratio across the RBS wave is about 4.13 which corresponds to a

Mach number of 2. At about T = 3.5 ms, particle 1 is initiated by the RBS (Figure 5.43(d)).

As can be seen, until the instant at which the RBS impacts particle 1, the CD &CL trends

are symmetrical. The impact of the RBS on particle 1 generates a steep increase in the
 289

lift coefficient. Particle 2, which is always under symmetrical loads does not experience

any lift force until T = 5.9 ms. This is the instant at which particle 2 is impacted by the

RBS. The numerical Schlieren image at this instant is shown in Figure 5.44(b). The figure

shows that the vortices generated in the wake regions of particles 1 & 2, are shifted up-

wards due to the impact of the RBS wave. At the instant shown in the figure, the incident

normal shock has passed through the computational domain and hence is not visible in the

figure. Figure 5.44(c) corresponds to the instant at which all the particles are impacted by

the RBS. The mesh adaptation during this process is apparent from the figures. The OR for

this simulation is 22.3 %.

5.3.2.15 Shock Diffraction Patterns in a Dusty Cloud

In this example, the DNS of a planar shock wave traversing through a dusty gas

layer is performed. In [56], the problem was investigated by solving the Euler equations

using a mixture model. Here the corresponding DNS calculations for two different Mach

numbers (1.67 & 3) are performed. The initial configuration of the system is shown in

Figure 5.45. A layer of (100) particles, with density ρ = 1000, were randomly seeded

(Figure 5.45) in a computational domain of size 60mm×40mm. The radius of the particles

(d = 0.2mm) and the mass fraction (M = 0.005) chosen for the current calculations are

slightly higher than in [56]. The present computations were performed on a base mesh of

size ∆xg = 0.25d with 4 levels of mesh refinement (13 points per diameter) resulting in

an effective resolution of about 2.45 million grid points. The threshold for the refinement

criteria were set at δ = 1.0 & δ = 25.0. Figures 5.46 & 5.47 show the numerical Schlieren
 290

image and vorticity contours for shocks of Mach 1.67 and 3.0 respectively. The inserts

show in Figure 5.46 correspond to the enlarged view of the complex shock diffraction and

the vortex shedding pattern observed in the wake region. The mesh topology shown in

the inserts closely follow the intricate vortices and the shock wave diffraction pattern. The

wake patterns shown in Figures 5.46 & 5.47 were not reported in the previous mixture

model-based calculations [56]. Furthermore, the effect of viscosity on the wake patterns

is an open subject of investigation. It is believed that the current effort is a step towards

DNS of such large scale computations. Presently, efforts are underway for large scale

computations (with large number of particles (O(1000)) with smaller radius (O(1 µm)))

and the results will be reported in a separate work.

5.4 Conclusions

The issue of treatment of a rigid solid-fluid interface interacting with strong shock

waves was studied. Several approaches for numerical treatment of such interactions were

considered and evaluated for a range of test problems. A Riemann solver based GFM ap-

proach formulated to resolve the interface dynamics for embedded fluid-fluid interfaces

was extended in conjunction with the Reflective Boundary Conditions (RBC) to treat the

presence of solid objects in compressible flows. It was shown in the numerical examples

that this method was robust, accurate and significantly reduced (but did not entirely elimi-

nate) the over-/under-heating errors at the material interface. Both the Characteristic-Based

Matching method (CBM) and the RBC augmented with Riemann correction (RBC-RS) ap-

proaches were found to be consistent in generating satisfactory solution for several com-
 291

plex configurations and shock diffraction phenomena, and significantly reduced the shock

interaction errors. However, the associated complexity involved in carrying out the char-

acteristic decomposition and the wave analysis in the characteristic space at the interface

render the CBM approach less attractive than the RBC-RS approach. Shocks interacting

with multiple particles and complex shapes have also been computed.
 292

Ref. Levels ∆xg CPU Time (in %) OR L2 (ρ)

1 0.0025 100 100 -

3 0.005 36.6 46.72 3.918 × 10−5

4 0.01 39.15 46.1725 4.198 × 10−5

5 0.02 60.98 47.286 4.746 × 10−5

6 0.04 163.8 51.88 5.439 × 10−5

Table 5.1: Timing analysis for a Mach 2.81 shock negotiating a stationary cylindrical ob-
1
stacle, for different levels of refinement with ∆xf = 800 .
 293

Interface Treatment Methods CPU Time ratio

Reflective Boundary Condition With Riemann Solver 1.0

CBM without sub-iterations 1.0483

CBM with five sub-iterations 1.2276

Table 5.2: CPU time taken for one complete (third-order) RK time iteration
 294

Figure 5.1: Illustration of grid points used in RBC

 295

(a) Isobaric fix

(b) SBC procedure

Figure 5.2: Isobaric fix and the SBC procedure: (a) Simple entropy extrapolation to cor-
rect the density field for the interfacial points; (b) Evaluating the normal derivatives at the
interface.
 296

Figure 5.3: Configuration of a rotating shock tube embedded in the computational domain
 297

1 Exact Solution
1
Reflective with Riemann solver
Exact Solution
CBM
Reflective with Riemann Solver SBC
CBM
SBC
0.8 0.8
DENSITY

DENSITY
0.6 0.6

0.4 0.4

0.2 0.2

0.6 0.8 1 1.2 1.4 1.6 1.8 0.8 1 1.2 1.4 1.6 1.8 2
X X

(a) α = 300 (b) α = 450

1
Exact Solution
Reflective with Riemann Solver
CBM
SBC

0.8
DENSITY

0.6

0.4

0.2

1.4 1.6 1.8 2 2.2 2.4 2.6

X

(c) α = 600

Figure 5.4: Plots of density along the centerline of the shock tube placed at different orien-
tations
 298

Figure 5.5: Configuration of a rotating shock tube embedded in the computational domain
 299

CBM CBM
SBC Using CT SBC Using CT
Reflective with Riemann Reflective with Riemann Solver
Reflective Reflective
4 4

3 3

DENSITY

DENSITY
2 2

1
1

0
0.6 0.8 1 1.2 1.4 1.4 1.6 1.8 2 2.2
X X

(a) α = 300 (b) α = 600

CBM CBM
SBC Using CT SBC Using CT
Reflective with Riemann Reflective with Riemann Solver
Reflective Reflective
300 300
PRESSURE

200 PRESSURE
200

100 100

0.6 0.8 1 1.2 1.4 1.4 1.6 1.8 2 2.2

X X

(c) α = 300 (d) α = 600

CBM
SBC Using CT 1400 CBM
Reflective with Riemann SBC Using CT
1500 Reflective Reflective with Riemann Solver
Reflective
1200

1000

1000
ENTROPY

ENTROPY

800

600

500
400

200

0
0.6 0.8 1 1.2 1.4 1.4 1.6 1.8 2 2.2
X X

(e) α = 300 (f) α = 600

Figure 5.6: Woodward Colella “Bang Bang” Problem: Plots of Density (Figures 5.6(a)
and 5.6(b)), Pressure (Figures 5.6(c) and 5.6(d)) and Entropy (Figures 5.6(e) and 5.6(f))
along the centerline for different orientations (α = 300 on the left and α = 600 on the
right) of the shock tube
 300

CBM 0.11 CBM

Norugaliev et. al. Norugaliev et. al.
25 Reflective BC Reflective BC
Reflective With Riemann Reflective With Riemann
0.1
TEMPERATURE

20 0.09
DENSITY

0.08

15
0.07

0.06
10

0.3 0.4 0.5 0.6 0.7 0.8 0.9 1 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X

Figure 5.7: Plots of density and temperature for a Mach 10 shock reflecting off a stationary
wall
 301

CBM CBM
SBC Using Coordinate Transformation 0.45 SBC Using Coordinate Transformation
Reflective BC with Riemann Solver Reflective BC with Riemann Solver
Reflective BC Reflective BC

25 0.4
TEMPERATURE

0.35
DENSITY

20

0.3

15

0.25

10
0.2
0.3 0.4 0.5 0.6 0.7 0.8 0.9 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X X

Figure 5.8: Mach 20 shock reflecting off a stationary wall: Plots of density (Figure 5.8(a))
and temperature (Figure 5.8(b))
 302

(a) RBC-RS; T = 1.9 (b) CBM; T = 1.9

(c) RBC-RS; T = 4.9 (d) CBM; T = 4.9

(e) RBC-RS; T = 6.9 (f) CBM; T = 6.9

(g) RBC-RS; T = 8.9 (h) CBM; T = 8.9

Figure 5.9: Snapshots of numerical Schlieren image of a Mach 1.3 shock diffracting a
wedge: Results from RBC-RS are shown in Figures 5.9(a), 5.9(c), 5.9(e) and 5.9(g) and the
the corresponding plots obtained from CBM are displayed in Figures 5.9(b), 5.9(d), 5.9(f)
and 5.9(h).
 303

(a) RBC-RS T = 1.9 (b) CBM; T = 1.9

(c) RBC-RS; T = 4.9 (d) CBM; T = 4.9

(e) RBC-RS; T = 6.9 (f) CBM; T = 6.9

(g) RBC-RS; T = 8.9 (h) CBM; T = 8.9

Figure 5.10: Snapshots of density contours of a Mach 1.3 shock diffracting a wedge:
Results from RBC-RS are shown in Figures 5.9(a), 5.9(c), 5.9(e) and 5.9(g) and the the
corresponding plots obtained from CBM are displayed in Figures 5.9(b), 5.9(d), 5.9(f)
and 5.9(h).
 304

(a) Horizontal shock detachment distance of (b) Vertical distance from the tip of the
the reflected bow shock from the foot of the wedge to the diffracting shock
wedge

(c) Horizontal distance between the tip of

the wedge and the incident shock

Figure 5.11: Comparison of primary shock detachment distance over time between CBM
and RBC-RS
 305

(a) Density Contours (b) Numerical Schlieren Image

(c) Density Contours (d) Numerical Schlieren Image

(e) Experimental Schlieren Im-

age from Bazhenova et al [20]

Figure 5.12: Contours of density and corresponding numerical Schlieren image at time T =
0.11: Results from RBC-RS are shown in figures 5.12(a) and 5.12(b) and CBM in 5.12(c)
and 5.12(d); Figure 5.12(e) corresponds to the experimental Schlieren image
 306

(a) Density Contours (b) Numerical Schlieren Image

(c) Density Contours (d) Numerical Schlieren Image

scan0001.jpg

(e) Experimental Schlieren Im-

age from Bazhenova et al [20]

Figure 5.13: Contours of density and corresponding Schlieren image at time T = 0.2:
Results from RBC-RS are shown in Figures 5.13(a) and 5.13(b) and CBM in 5.13(c)
and 5.13(d); Figure 5.13(e) corresponds to the experimental Schlieren image.
 307

(a) Density Contours (b) Numerical Schlieren Image

(c) Density Contours (d) Numerical Schlieren Image

Figure 5.14: Contours of density and corresponding Schlieren image at time T = 0.5:
Results from RBC-RS are shown in Figures 5.14(a) and 5.14(b) and CBM in 5.14(c)
and 5.14(d)
 308

0.2
Reflective With Riemann Solver
CBM

Shock Detachment Distance

0.15

0.1

0.05

0
0 0.1 0.2 0.3 0.4 0.5
Time

(a) Shock detachment distance vs time

0.5

0.45
Reflective With Riemann Solver
0.4 CBM
33 degree Correlation
0.35

0.3
Ycoordinate

0.25

0.2

0.15

0.1

0.05

0
0.3 0.4 0.5 0.6 0.7 0.8 0.9 1
X coordinate

(b) locus of upper ripple point

Figure 5.15: Comparison between CBM and the RBC-RS : Shock detachment distance vs
time (a) and locus of upper triple point (b)
 309

Figure 5.16: Comparison of drag coefficient (CD ) with experimental and numerical pre-
dictions, for Mach 1.3 and 2.6 shock waves negotiating stationary cylindrical obstacles.
The experimental image was obtained from “Takayama, K. and Itoh, K.,Unsteady drag
over cylinders and aerofoils in transonic shock tube ows, Proceedings of the 15th Interna-
tional Symposium on Shock Waves and Shock Tubes, Stanford, California, U.S.A., PP 439
485, 1985”. The corresponding numerical image was obtained from ”D. Drikakis and D.
Ofengeim and E. Timofeev and P. Voionovich, Computation Of Non-Stationary Shock-
Wave/Cylinder Ineraction Using Adaptive-Grid Methods, Journal of Fluids and Struc-
tures,11(6),665 - 692,1997“.
 310

(a) Numerical Schlieren image at T = (b) Numerical Schlieren image at T =

0.314 µs 0.612 µs

(c) Numerical Schlieren image at T = (d) Numerical Schlieren image at T =

1.1 µs 2.4 µs

Figure 5.17: Numerical Schlieren image for the solution with 4 levels of mesh refinement
1
on a base of size ∆xg = 300 at different instants in time: (a) T = 0.314 µs; (b) T = 0.612
µs; (c) T = 1.1 µs; (d) T = 2.4 µs;
 311

(a) The wake region at T = 1.1 µs (Uniform (b) The wake region at T = 1.1 µs
Mesh Calculation)

(c) The Mesh in the wake region at T = 1.1 (d) The wake region at T = 2.4 µs
µs

Figure 5.18: Enlarged view of the wake region of the numerical Schlieren image at two
1
different instants in time: (a) Uniform mesh calculation with ∆x = 800 at 1.1 µs; (b) LMR
calculation at T = 1.1 µs; (c) Mesh topology in the wake region at T = 1.1 µs; (d) LMR
calculation at T = 2.4 µs;
 312

(a) Locus of upper and lower triple (b) Polynomial Fit

point with experimental [182] and numeri-
cal [53] predictions

Figure 5.19: Shock-cylinder interaction validation: (a) Comparison of locus of upper and
lower triple point (b) A parabolic fit for the locus of the upper and lower triple point
 313

(a) T = 0.314 µs (b) T = 0.314 µs

(c) T = 0.47 µs (d) T = 0.47 µs

(e) T =0.565 µs (f) T = 0.565 µs

Figure 5.20: Comparison of density contours ((a), (c) and (e)) and numerical Schlieren
image ((b), (d) and (f)) for LMR-based calculation with 6 levels of mesh refinement, with
1
the uniform fine mesh computations (with ∆x = 800 ) at different instants in time: (a) and
(b) at T = 0.314 µs, (c) and (d) at T = 0.47 µs and (e) and (f) at T = 0.565 µs
 314

Figure 5.21: Comparison of pressure distribution on the surface of the cylinder for different
levels of mesh refinement at time T = 5.65 µs
 315

(a) L2 (S) (b) Occupancy Ratio

Figure 5.22: Sensitivity of the solution to the refinement criteria: (a) L2 (S) for entropy (b)
% Mesh occupancy ratio
 316

(a) RBC-RS; T = 0.09 (b) CBM; T = 0.09

(c) RBC-RS; T = 0.125 (d) CBM; T = 0.125

(e) RBC-RS; T = 0.25 (f) CBM; T = 0.25

Figure 5.23: Snapshots of numerical Schlieren image for a Mach 3 shock interacting with
multiple stationary cylinders at different instants in time: Results from RBC-RS are shown
in Figures 5.23(a), 5.23(c) and 5.23(e) and the corresponding plots for the CBM approach
are displayed in Figures 5.23(b), 5.23(d) and 5.23(f).
 317

(a) RBC-RS

(b) CBM

(c) RBC-RS

(d) CBM

Figure 5.24: Mach 2 shock interaction with a square inclined at 60o with the horizontal:
Figures (a) and (c) correspond to numerical Schlieren image and figures (d) and (e) corre-
spond to the density contours
 318

(a) RBC-RS

(b) CBM

(c) RBC-RS

(d) CBM

Figure 5.25: Mach 2 shock interaction with a square inclined at 60o with the horizontal:
Figures (a) and (c) correspond to numerical Schlieren image and figures (d) and (e) corre-
spond to the density contours
 319

(a) RBC-RS

(b) CBM

0.7

RBC-RS dx = 1/1000
0.6 RBC-RS dx = 1/500
CBM dx = 1/1000
0.5

0.4
Xc, Yc

0.3

0.2

0.1

0
0.05 0.1 0.15 0.2 0.25 0.3
TIME

(c) Locus of center of mass as a function of time

Figure 5.26: Cylinder subject to Mach 3 shock: Numerical Schlieren image (RBC-RS in
Figure 5.26(a) and CBM in Figure 5.26(b)) and locus of center of mass (Figure 5.26(c))
 320

2.5

1.5
Y

0.5

0
0 0.5 1 1.5 2 2.5 3
X

(a) Density contours at T = 0.22

2.5

1.5
Y

0.5

0
0 0.5 1 1.5 2 2.5 3
X

(b) Entropy Contours at T = 0.22

Figure 5.27: Mach 10 shock interacting with a 30o ramp at T = 0.22: Contours of (a) density
and (b) entropy.
 321

0.1
22 dx = 1/40 dx = 1/40
dx = 1/80 dx = 1/80
20 dx = 1/120 dx = 1/120
dx = 160 dx = 160
18 0.08

16

TEMPERATUR
14 0.06
DENSITY

12

10
0.04
8

6
0.02
4

0 0
0 0.5 1 1.5 2 2.5 3 0 0.5 1 1.5 2 2.5 3
X X

(a) Density (b) Pressure

12
dx = 1/40
dx = 1/80
dx = 1/120
10 dx = 160

8
ENTROPY

0
0 0.5 1 1.5 2 2.5 3
X

(c) Entropy

Figure 5.28: Grid refinement study - (a) Density, (b) Pressure and (c) Entropy plots along
the ramp at T = 0.22
 322

1 (b) Numerical Schlieren image for ∆x =

(a) Density contours for ∆x = 120 1
120

1 (d) Numerical Schlieren image for ∆x =

(c) Density contours for ∆x = 160 1
160

1 (f) Numerical Schlieren image for ∆x =

(e) Density contours for ∆x = 240 1
240

1 (h) Numerical Schlieren image for ∆x =

(g) Density contours for ∆x = 300 1
300

Figure 5.29: Mach 3 wind tunnel with a backward facing step: Density contours ((a), (c)
and (e)) and numerical Schlieren image ((b), (d) and (f))
 323

(a) Density Contours (b) Numerical Schlieren Image Contours

(c) Experimental Schlieren Image

Figure 5.30: Mach 1.3 shock diffracting a 90o corner: (a) Density contours, (b) numerical
Schlieren image and (c) experimental Schlieren image
 324

(a) Density contours (b) Numerical Schlieren Image

(c) Experimental Schlieren Image

Figure 5.31: Mach 2.4 shock diffracting a 90o corner: (a) Density contours, (b) numerical
Schlieren image and (c) experimental Schlieren image
 325

(a) Density contours (b) Numerical Schlieren Image

Figure 5.32: Mach 20 shock diffracting a 90o corner: (a) Density contours and (b) numeri-
cal Schlieren image
 326

(a) Density contours at T = 0.0205 (b) Numerical Schlieren image at T = 0.0205

(c) Density contours at T = 0.035 (d) Numerical Schlieren image at T = 0.035

Figure 5.33: Mach 2.1 shock negotiating a curved channel without obstacles: Density
contours ((a) and (c)) and numerical Schlieren image ((b) and (d)) at two different instants
in time
 327

(a) Density contours at T = 0.0156 (b) Numerical Schlieren image at T = 0.0156

(c) Density contours at T = 0.035 (d) Numerical Schlieren image at T = 0.035

Figure 5.34: Mach 2.1 shock negotiating a curved channel with stationary obstacles: Den-
sity contours ((a) and (c)) and numerical Schlieren image ((b) and (d)) at two different
instants in time
 328

(a) Density contours at T = 0.014 (b) Numerical Schlieren image at T = 0.014

(c) Density contours at T = 0.025 (d) Numerical Schlieren image at T = 0.025

Figure 5.35: Mach 2.1 shock negotiating a curved channel with a moving obstacle: Density
contours ((a) and (c)) and numerical Schlieren image ((b) and (d)) at different instants in
time
 329

(a) Numerical Schlieren Image at T = 0.36

(b) Numerical Schlieren Image at T = 4.0

(c) Numerical Schlieren Image at T = 9.16

Figure 5.36: Series of numerical Schlieren image for a Mach 3 shock traversing through
120 randomly oriented stationary particles
 330

(a) Numerical Schlieren at T = 0 (b) Vorticity Contours at T = 0.725

(c) Numerical Schlieren at T = 1.62 (d) Vorticity Contours at T = 3.0

Figure 5.37: Series of numerical Schlieren image (Right) and vorticity contours (Left) for
Mach 8 spherical shock interacting with 12 particles; Series correspond to the shock wave
dispersing initially stationary particles
 331

(a) Numerical Schlieren Image at T = (b) Vorticity Contours Image at T =

0 0.725

(c) Numerical Schlieren Image at T = (d) Vorticity Contours Image at T =

1.62 3.0

Figure 5.38: Series of numerical Schlieren image (Right) and vorticity contours (Left) for
Mach 8 spherical shock interacting with 12 particles; Series correspond to the shock wave
reflecting off the domain walls.
 332

Figure 5.39: Configuration of a projectile travelling at Mach 1.22

 333

(a) T = 0.25 ms (b) T = 0.565 ms (c) T = 4.65 ms (d) T = 7.03 ms

Figure 5.40: Numerical Schlieren image and the mesh topology for a supersonic projectile
fired at Mach 1.22 at different instants in time.
 334

(a) Numerical Schlieren image at T = 7.03 ms

(b) Shock stand off distance with respect to time

Figure 5.41: Bluff body projectile fired at Mach 1.22: (a) Numerical Schlieren image at
time T = 7.03 ms; (b) Shock stand off as a function of time
 335

Figure 5.42: Problem configuration for the dynamics of multiple particles introduced in
high speed flow
 336

(a) Horizontal coordinate of the center of (b) Vertical coordinate of the center of
mass mass

(c) Drag Coefficient (d) Lift Coefficient

Figure 5.43: Dynamics of multiple particles introduced in high speed flow: (a) Locus of
horizontal coordinate of the center of mass; (b) Locus of vertical coordinate of the center
of mass; (c) Drag coefficient (d) Lift coefficient
 337

(a) Mesh Topology at T = 1.79 (b) Mesh Topology at T = 5.9 ms

ms

(c) Mesh Topology at T = 10.0

ms

Figure 5.44: Numerical Schlieren image and the mesh topology of cylindrical particles
introduced in high speed flows (a) T = 1.79 ms; (b) T = 5.9 ms; (c) T = 10.0 ms
 338

40

Quiescent Air Layer

30

Shock Wave Dusty Layer (Seeded with particles)

(60,22)
Y

20

10

0
(10,0)
0 10 20 30 40 50 60
X

Figure 5.45: Initial configuration of the system for DNS of shock wave traversing through
a dusty layer of gas
 339

(a) Numerical Schlieren Image

(b) Vorticity Contours

Figure 5.46: Mach 1.67 shock wave traversing through a dusty layer of gas at time T =
76.15 µ s: (a) Numerical Schlieren image and (b) vorticity contours
 340

(a) Numerical Schlieren Image

Figure 5.47: Numerical Schlieren image for a Mach 3 shock wave traversing through a
dusty layer of gas at time T = 41.17 µ s
 341

CHAPTER 6
CONCLUSIONS AND RECOMMENDATIONS FOR FUTURE WORK

6.1 Conclusions

The primary goal of this thesis is to develop a robust, high-fidelity, Eulerian, Carte-

sian grid framework to solve high-speed multi-material interaction problems. The novelty

of the work lies in the fact that an Eulerian approach is adopted for resolving the response

of both solids and fluids. Another unique and distinguishing feature of this work is that the

framework developed is applicable for a wide class of materials and for a broad spectrum

of speeds of interactions. In addition, the robustness, versatility and simplicity of the ap-

proach makes it an attractive choice for solving several complex multi-material interaction

problems. The examples that are solved in this thesis cover a plethora of applications and

the stability of the framework in solving these problems has been well established.

6.2 Significant Contributions of this Thesis

To summarize some of the main contributions of this thesis are listed below:

• A generic sharp interface Cartesian grid method is formulated to capture the response

of the embedded interface subject to intense loading conditions

• A robust Riemann problem approach is devised to resolve shock-droplet/shock-bubble

interactions. The approach is found to be stable and accurate in decoupling the nature

of the waves interacting at the embedded interface

 342

• Physically consistent boundary conditions are formulated to capture the response of

elasto-plastic solids under intense loading conditions

• A simple, robust and general formulation is devoloped to infuse interface conditions

to the ghost nodes

• A conservative approach is devised to enforce flux conservation at mesh interfaces

6.3 Recommendations for Future Work

The framework can be extended in the following aspects:

• Although the implementation of the framework was carried out to include three-

dimensions, the computations reported in this theses were limited to two-dimensional

applications. In order to to perform real time, three-dimensional computations, multi-

core computing facility must be incorporated in the present implementation. The

efforts in this direction are currently underway and the results from such calculations

will be reported in the future.

• The high-velocity impact and penetration studies conducted in this work typically in-

volve fragmentation and failure of the interacting media. To accurately simulate these

phenomena suitable fracture, damage and failure models have to be incorporated in

concomitant with the interface capturing technique. To emphasize on the importance

of these models, a sample computation of an Explosively Formed Projectile (EFP) is

presented.

As shown in Figure 6.1(a), a conventional EFP consists of a metal liner held against
 343

an explosive core. When the explosive core is initiated a detonation wave sets in the

explosive charge. When this detonation wave impinges on the metallic liner, the liner

material undergoes very large deformation resulting in the formation of high speed

jet. The formation of high speed jet is shown in Figure 6.1(b). This hypervelocity

jet of metal is capable of penetrating great depths into steel armor; however, in travel

over some distance the jet breaks up along its length into particles that drift out of

alignment, greatly diminishing its effectiveness at a distance.

In Figure 6.2(a), the initial formation of the jet process computed in this work is pre-

sented. For the sake of comparison, the jet formation process as depicted in [131] is

displayed in Figure 6.2(b). As can be seen from the two figures, the present method-

ology is able to predict the initial jet formation process quite accurately. However,

the present methodology was not able to predict the evolution of the jet at later stages

as this would require the material to break apart from the explosive mold. To model

this process a suitable void growth model has to be incorporated. In addition, the

fragmentation of jet requires sophisticated damage and fracture models. These mod-

els exist and can be directly coupled with the numerical facility developed in this

thesis.

• The computations reported in this thesis were performed for inert configurations,

where-in the energetics of the interacting materials were neglected. In those cases,

the interacting materials where treated as thermo-mechanically non-participating re-

gion. In reality, the resulting temperatures are very high, particularly for high shock

loadings and high velocity impact problems, that the participating materials undergo
 344

phase compression and chemical reactions that significantly alter the energy deposi-

tion mechanism. Thus for performing chemically reactive compressible flow calcula-

tions, the numerical facility developed in this thesis can be coupled with appropriate

detonation models.
 345

(b) Radiograph Images of Jet formation [131]

(a) Components of EFP [131]

Figure 6.1: Explosively Formed Projectile: (a) Components of EFP and (b) Radiograph
images of the jet formation process
 346

(a) Current Calculation of EFP

(b) Shaped Jet Formation as depicted

in [131]

Figure 6.2: The jet formation process: (a) Current calculations (b) Theoretical prediction
 347

REFERENCES

[1] Shock Waves and Man. The University of Toronto Press, 1974.

[2] Analysis and Numerics for Conservation Laws, chapter Wave Processes at Inter-
faces. Springer Berlin Heidelberg, 2005.

[3] A.M. Abd-El-Fattah and L.F. Henderson. Shock waves at a slow-fast gas interface.
Journal of Fluid Mechanics, 89:79–95, 1978.

[4] R. Abgrall. On essentially non-oscillatory schemes on unstructured meshes: analysis

and implementation. J. Comput. Phys., 114(1):45–58, 1994.

[5] R. Abgrall. How to prevent pressure oscillations in multicomponent flow calcula-

tions: A quasi conservative approach. Journal of Computational Physics, 125:150–
160, 1996.

[6] R. Abgrall and S. Karni. Computations of compressible multi-fluids. Journal of

Computational Physics, 169:594–623(30), 20 May 2001.

[7] G. Agresar, J. J. Linderman, G. Tryggvason, and K. G. Powell. An adaptive, carte-

sian, front-tracking method for the motion, deformation and adhesion of circulating
cells. J. Comput. Phys., 143(2):346–380, 1998.

[8] Thomas J. Ahrens and John D. O’Keefe. Impact on the earth, ocean and atmosphere.
International Journal of Impact Engineering, 5:13 – 32, 1987.

[9] Charles E. Anderson. An overview of the theory of hydrocodes. International Jour-

nal of Impact Engineering, 5:33 – 59, 1987.

[10] Charles E. Anderson, Volker Hohler, James D. Walker, and Alois J. Stilp. Time-
resolved penetration of long rods into steel targets. International Journal of Impact
Engineering, 16(1):1 – 18, 1995.

[11] M. Arienti, P. Hung, E. Morano, and Shepherd J.E. A levelest approach to Eulerian-
Lagrangian coupling. Journal Of Computational Physics, 185:213–251, 2003.

[12] M. Arienti, E. Morano, and Shepherd J.E. Shock and detonation modeling with the
mie-grüneisen equation of state. Technical report, Graduate Aeronautical Laborato-
ries Report, 2004.
 348

[13] T.D. Aslam. A level set algorithm for tracking discontinuities in hyperbolic conser-
vation laws I: Scalar equations. Journal Of Computational Physics, 167:413–438,
2001.

[14] T.D. Aslam. A level set algorithm for tracking discontinuities in hyperbolic conser-
vation laws II: Systems of equations. Journal of Scientific Computing, 19(1-3):37–
62, 2003.

[15] T.D. Aslam. A partial differential equation approach to multidimensional extrapola-

tion. Journal Of Computational Physics, 193:349–355, 2004.

[16] A. Bagabir and D. Drikakis. Mach number effects on shock-bubble interaction.

Shock Waves, 11:209–218, 2001.

[17] J. Baker and A. Williams. Hypervelocity penetration of plate targets by rod and rod-
like projectiles. International Journal of Impact Engineering, 5:101 – 110, 1987.

[18] G.J. Ball, B.P. Howell, T.G. Leighton, and M.J. Schofield. Shock induced collapse of
a cylindrical air cavity in water : A free Lagrange simulation. Shock Waves, 10:265
– 276, 2000.

[19] J. W. Banks, D. W. Schwendeman, A. K. Kapila, and W. D. Henshaw. A high-

resolution godunov method for compressible multi-material flow on overlapping
grids. J. Comput. Phys., 223(1):262–297, 2007.

[20] T.V. Bazhenova, L.G. Gvozdeva, and M.A. Nettleton. Unsteady interactions of shock
waves. Progress in Aerospace Sciences, 21:249–331, 1984.

[21] D. J. Benson and W. J. Nellis. Dynamic compaction of copper powder: Computation

and experiment. Applied Physics Letters, 65(4):418–420, 1994.

[22] D. J. Benson, W. Tong, and G. Ravichandran. Particle-level modeling of dynamic

consolidation of ti-sic powders. Modelling and Simulation in Materials Science and
Engineering, 3(6):771–796, 1995.

[23] D. L. Benson, J. O. Hallquist, M. Igarashi, K. Shimomoki, and M. Mizuno. The

application of DYA3D to large scale crashworthiness calculations. In Proceedings
of the ASME Computers in Engineering Conference, New York, July 20 - 24 1986.
ASME.

[24] David J. Benson. Computational methods in lagrangian and eulerian hydrocodes.

Comput. Methods Appl. Mech. Eng., 99(2-3):235–394, 1992.
 349

[25] David J. Benson. Volume of fluid interface reconstruction methods for multi-
material problems. Applied Mechanics Reviews, 55(2):151–165, 2002.

[26] M. J. Berger and P. Colella. Local adaptive mesh refinement for shock hydrodynam-
ics. J. Comput. Phys., 82(1):64–84, 1989.

[27] Marsha J. Berger and Joseph E. Oliger. Adaptive mesh refinement for hyperbolic
partial differential equations. Technical report, Stanford, CA, USA, 1983.

[28] N.K. Bourne and J.E. Field. Shock induced collapse of single cavities in liquids.
Journal of Fluid Mechanics, 244:225 – 240, 1992.

[29] Jean Braun and Malcolm Sambridge. A numerical method for solving partial dif-
ferential equations on highly irregular evolving grids. Nature, 376:655–660, Aug
1995.

[30] Greg L. Bryan. Fluids in the universe: adaptive mesh refinement in cosmology.
Comput. Sci. Eng., 1(2):46–53, 1999.

[31] A.E. Bryson and R.W.F. Gross. Diffraction of strong shocks by cones, cylinders and
spheres. Journal of Fluid Mechanics, 10:1–16, 1960.

[32] R. Eatock Taylor C. W. Smith, J. Zang. Wavelet-based adaptive grids as applied to

hydrodynamics. International Journal for Numerical Methods in Fluids, 2008.

[33] R. Caiden, R.P. Fedkiw, and C. Anderson. A numerical method for two-phase flow
consisting of separate compressible and incompressible regions. Journal of Compu-
tational Physics, 166:1–27, 2001.

[34] Jay Casper and H. L. Atkins. A finite-volume high-order eno scheme for two-
dimensional hyperbolic systems. J. Comput. Phys., 106(1):62–76, 1993.

[35] Thomas C. Cecil, Stanley J. Osher, and Jianliang Qian. Simplex free adaptive tree
fast sweeping and evolution methods for solving level set equations in arbitrary di-
mension. J. Comput. Phys., 213(2):458–473, 2006.

[36] Thomas C. Cecil, Stanley J. Osher, and Jianliang Qian. Essentially Non-Oscillatory
adaptive tree methods. J. Sci. Comput., 35(1):25–41, 2008.

[37] J. W. Chalmersa, S. W. Hodsona, K. H. A. Winklera, P. R. Woodward, and N. J.

Zabusky. Shock-bubble interactions : Generation and evolution of vorticity in two-
dimensional supersonic flows. Fluid Dynamics Research, 3(4):392–394, 1988.
 350

[38] C.H. Chang and M.S. Liou. A robust and accurate approach to computing com-
pressible multiphase flow: Stratified flow model and AUSM+ - up scheme. Journal
of Computational Physics, 225:840 – 873, 2007.

[39] H. Chen and S.M. Liang. Flow visualization of shock/water interactions. Shock
Waves, 17:309 – 321, 2008.

[40] D. Cheng, G. Trapaga, J.W. McKelliget, and E.J. Lavernia. Mathematical mod-
elling of high velocity oxygen fuel thermal spraying of nanocrystalline materials:
An overview. Modelling and Simulation in Materials Science and Engineering, 11:1
– 31, 2003.

[41] Dae-Il Choi, J. David Brown, Breno Imbiriba, Joan Centrella, and Peter MacNe-
ice. Interface conditions for wave propagation through mesh refinement boundaries.
Journal of Computational Physics, 193(2):398–425, 2004.

[42] J.P. Cocchi and R. Saurel. A Riemann problem based method for the resolution of
compressible multi-material flows. Journal of Computational Physics, 137:265–298,
1997.

[43] J.P. Cocchi, R. Saurel, and J.C. Loraud1. Treatment of interface problems with
Godunov-type schemes. Shock Waves, 5(6):347–357, 1996.

[44] William J. Coirier and Kenneth G. Powell. An accuracy assessment of Cartesian-

mesh approaches for the euler equations. J. Comput. Phys., 117(1):121–131, 1995.

[45] William John Coirier. An Adaptively-Refined, Cartesian, Cell Based Scheme for the
Euler and Navier-Stokes Equations. PhD thesis, Ann Arbor, MI, USA, 1994.

[46] S. R. Cooper, D. J. Benson, and V. F. Nesterenko. A numerical exploration of the

role of void geometry on void collapse and hot spot formation in ductile materials.
International Journal of Plasticity, 16(5):525 – 540, 2000.

[47] Nguyen D., F. Gibou, and Fedkiw R.P. A fully conservative Ghost Fluid Method
and stiff detonation waves. In Proceedings of the 12th International Detonation
Symposium, San Diego, CA, August, 11 - 16 2002.

[48] A. Dadone and B. Grossman. Ghost cell method for inviscid three dimensional flows
on Cartesian grids. In Presented at the 43rd AIAA Aerospace Sciences Meeting and
Exhibit, Reno, NV,USA, January 10-13 2005. AIAA 2005-0874.

[49] S. F. Davis. An interface tracking method for hyperbolic systems of conservations

laws. Applied Numerical Mathematics, 10:447 – 472, 1992.
 351

[50] M. D. de Tullio, P. De Palma, G. Iaccarino, G. Pascazio, and M. Napolitano. An

immersed boundary method for compressible flows using local grid refinement. J.
Comput. Phys., 225(2):2098–2117, 2007.

[51] A. Dolatabadi, J.. Mostaghimi, and V. Pershin. Modeling dense suspension of solid
particles in highly compressible flows. International Journal of Computational Fluid
Dynamics, 18:125 – 131, 2004.

[52] R. Donat and A. Marquina. Capturing shock reflections: An improved flux formula.
Journal of Computational Physics, 125:42–58, 1996.

[53] D. Drikakis, D. Ofengeim, E. Timofeev, and P. Voionovich.

[54] A.F. Emery. An evaluation of several differencing methods for inviscid fluid flow
problems. Journal of Computational Physics, 2(3):306–331, 1968.

[55] J. Falcovitz, G. Alfandary, and G. Hanoch. A two dimensional conservation laws

scheme for compressible flows with moving boundaries. Journal of Computational
Physics, 138:83–102, 1997.

[56] A. V. Federov, Yu. V. Kharlamova, and T. A. Khmel. Reflection of a shock wave in

a dust cloud. Combustion, Explosion and Shock Waves, 43(1):104 – 113, 2007.

[57] R. Fedkiw. Coupling an Eulerian fluid calculation to a Lagrangian solid calculation

with the Ghost Fluid Method. Journal of Computational Physics, 175:200 – 224,
2002.

[58] R.P. Fedkiw, T.D. Aslam, B. Merriman, and S. Osher. A non oscillatory Eulerian
approach to interfaces in multi-material flows (the Ghost Fluid Method). Journal of
Computational Physics, 152(2):457 – 492, 1999.

[59] R.P. Fedkiw, T.D. Aslam, and S. Xu. The Ghost Fluid Method for deflagration and
detonation discontinuities. Journal of Computational Physics, 154(2):393 – 427,
1999.

[60] R.P. Fedkiw, A. Marquina, and B. Merriman. An isobaric fix for the overheating
problem in multi-material compressible flows. Journal of Computational Physics,
148:545–578, 1999.

[61] H. Forrer and M. Berger. Flow simulations on Cartesian grids involving com-
plex moving geometries flows. International Series of Numerical Mathematics,
129(315):139–178, 1998. Birkhuser, Basel 1.
 352

[62] H. Forrer and R. Jeltsch. A higher order boundary treatment for Cartesian grid
methods. Journal Of Computational Physics, 140:259–277, 1998.

[63] M.J. Forrestal, V.K. Luk, and N.S. Brar. Perforation of aluminum armor plates with
conical-nose projectiles. Mechanics of Materials, 10(1-2):97–105, 1990.

[64] J. Glimm, J. W. Grove, X. Li, K. M. Shyue, Y. Zeng, and Q. Zhang. Three-

dimensional front tracking. SIAM Journal on Scientific Computing, 19(3):703–727,
1998.

[65] J. Glimm, C. Klingenberg, O. McBryan, B. Plohr, D. Sharp, and S. Yaniv. Front

tracking and two-dimensional Riemann problems. Advances in Applied Mathemat-
ics, 6(3):259–290, 1985.

[66] J. Glimm, X. L. Li, and N. Zhao. Conservative front tracking and levelset algorithms.
Proceedings of the National Academy of Sciences, 98(25):14198 – 14201, 2001.

[67] J. Glimm, D. Marchesin, and O. McBryan. A numerical method for two phase flow
with an unstable interface. Journal of Computational Physics, 39:179 – 200, 1981.

[68] D. M. Greaves and A. G. L. Borthwick. On the use of adaptive hierarchical meshes

for numerical simulation of separated flows. International Journal for Numerical
Methods in Fluids, 26(3):303 – 322, 1998.

[69] D. M. Greaves and A. G. L. Borthwick. Hierarchical tree-based finite element mesh

generation. International Journal for Numerical Methods in Fluids, 45(4):447 –
471, 1999.

[70] Deborah M. Greaves. Hierarchical tree-based finite element mesh generation. Inter-
national Journal for Numerical Methods in Fluids, 44(10):1093 – 1117, 2004.

[71] Deborah M. Greaves. A quadtree adaptive method for simulating fluid flows with
moving interfaces. J. Comput. Phys., 194(1):35–56, 2004.

[72] Boyce E. Griffith, Richard D. Hornung, David M. McQueen, and Charles S. Pe-
skin. An adaptive, formally second order accurate version of the immersed boundary
method. J. Comput. Phys., 223(1):10–49, 2007.

[73] J.W. Grove and R. Menikoff. Anomalous reflection of a shock wave at a fluid inter-
face. Journal of Fluid Mechanics, 219:313–336, 1990.

[74] Shan Guang-Jiang and Wang Chi-Shu. Efficient implementations of Weighted ENO
schemes. Technical report, 1995.
 353

[75] J.F. Haas and B. Sturtevant. Interaction of weak shock waves with cylindrical and
spherical inhomogeneities. Journal of Fluid Mechanics, 181:41–76, 1987.

[76] K. K. Haller. High-Velocity Impact of a Liquid Droplet on a Rigid Surface: The

Effect of Liquid Compressibility. PhD thesis, Swiss Federal Institute of Technology
Zurich, 2002.

[77] K.K Haller, D. Poulikakos, Y. Ventikos, and P. Monkewitz. Shock wave formation in
droplet impact on a rigid surface: Lateral liquid motion and multiple wave structure
in the contact line region. Journal of Fluid Mechanics, 490:1–14, 2003.

[78] K.K Haller, Y. Ventikos, and D. Poulikakos. Computational study of high speed
liquid droplet impact. Journal of Applied Physics, 92(5):2821–2828, 2002.

[79] K.K Haller, Y. Ventikos, and D. Poulikakos. Wave structure in the contact line region
during high speed droplet impact on a surface: Solution of the Riemann problem for
the stiffened gas equation of state. Journal of Applied Physics, 93(5):3090–3097,
2003.

[80] R.K.S. Hankin. The Euler equations for multiphase compressible flow in conserva-
tion form. Journal of Computational Physics, 172:808 – 826, 2001.

[81] L.F. Henderson. On the refraction of shock waves. Journal of Fluid Mechanics,
198:365–386, 1989.

[82] W.D. Henshaw and D. W. Schwendeman. Moving overlapping grids with adaptive
mesh refinement for high-speed reactive and non-reactive flow. J. Comput. Phys.,
216(2):744–779, 2006.

[83] Stefan Hiermaier. Structures Under Crash and Impact. Springer US, 2008.

[84] Charles Hirsch. Numerical Computation of Internal and External Flows - Volume:II,
Computational Methods for Inviscid and Viscous Flows. Wiley, 1988.

[85] R. Hivier. Computation of shock wave diffraction at a ninety degrees convex edge.
Shock Waves, 1:89–98, 1991.

[86] J.M. Hong, T. Shinar, M. Kang, and R. Fedkiw. On boundary condition capturing
for multiphase interfaces. Journal of Scientific Computing, 31:99 – 125, 2007.

[87] X. Y. Hu and B. C. Khoo. An interface interaction method for compressible multi-

fluids. Journal of Computational Physics, 198:35–64, 2004.
 354

[88] X. Y. Hu, B. C. Khoo, N. A. Adams, and F.L. Huang. A conservative interface

method for compressible flows. Journal of Computational Physics, 219(2):553 –
578, May 2006.

[89] D. Igra and K. Takayama. Numerical simulation of shock wave interaction with
water column. Shock Waves, 11:219–228, 2001.

[90] O. Igra and K. Takayama. Shock tube study of the drag coefficient of a sphere
in a non-stationary flow. Proceedings: Mathematical and Physical Sciences,
442(1915):231–247, 1993.

[91] G. T. Gray III, E. Cerreta, C. A. Yablinsky, L. B. Addessio, B. L. Henrie, B. H.

Sencer, M. Burkett, P. J. Maudlin, S. A. Maloy, C. P. Trujillo, and M. F. Lopez. Influ-
ence of shock prestraining and grain size on the dynamic-tensile-extrusion response
of copper: Experiments and simulation. AIP Conference Proceedings, 845(1):725–
728, 2006.

[92] A. R. Jamaluddin, G. J. Ball, and T. G. Leighton. Free-Lagrange simulations of

shock/bubble interaction in shock wave Lithotripsy. In Shock Waves : Proceedings
of the 24th International Symposium on Shock Waves.

[93] Leland Jameson. AMR vs high order schemes. J. Sci. Comput., 18(1):1–24, 2003.

[94] P. Jenny, B. Muller, and H. Thomann. Correction of conservative Euler solvers for
gas mixtures. Journal of Computational Physics, 132:91–107, 1997.

[95] Xiaohai Jian, Zhihua Chen, Baochun Fan, and Hongzhi Li. Numerical simulation
of blast flow fields induced by a high-speed projectile. Shock Waves, 18:205 – 212,
2008.

[96] Z. Jiang, K. Takayama, and B. W. Skews. Numerical study on blast flowfields in-
duced by supersonic projectiles discharged from shock tubes. Physics of Fluids,
10(1):277–288, 1998.

[97] Zonglin Jiang, Yonghui Huang, and Kazuyoshi Takayama. Shocked flows induced
by supersonic projectiles moving in tubes. Computers & Fluids, 33(7):953 – 966,
2004.

[98] E. Johnsen and T. Colonius. Implementation of WENO schemes in compressible

multicomponent flow problems. Journal of Computational Physics, 219:715–732,
2006.

[99] G. R. Johnson and W. H. Cook. Fracture characteristics of three metals subjected

 355

to various strains, strain rates, temperatures and pressures. Engineering Fracture

Mechanics, 21(1):31 – 48, 1985.

[100] J. Kaca. An interferometric investigation of the diffraction of a planar shock wave

over a semicircular cylinder. Technical Report 269, University of Toronto, Institute
for Aerospace Studies (UTIAS), 1988.

[101] M. Kang, R.P. Fedkiw, and X.D. Liu. Boundary condition capturing method for
multiphase incompressible flow. Journal of Scientific Computing, 15(3):323 – 360,
2000.

[102] S. Karni. Hybrid multi-fluid algorithms. SIAM Journal of Scientific Computing,

17:1019–1039, 1996.

[103] Smadar Karni, Alexander Kurganov, and Guergana Petrova. A smoothness indicator
for adaptive algorithms for hyperbolic systems. J. Comput. Phys., 178(2):323–341,
2002.

[104] R. Keppens, Nool. M., G. Tóth, and J.P. Goedbloed. Adaptive mesh refinement for
conservative systems: multi-dimensional efficiency evaluation. Computer Physics
Communications, 153(3):317 – 339, 2003.

[105] Akhtar S. Khan and Sujian Huang. Continuum Theory of Plasticity. Wiley-
Interscience, 1995.

[106] B.A. Khasainov, A.A. Borisov, B.S. Ermolaev, and A.I. Korotkov. Two phase visco-
plastic model of shock initiation of detonation in high density pressed explosives. In
Seventh Symposium (International) on Detonation, Naval Surface Weapons Center.

[107] J.W. Kim and D.J. Lee. Generalized characteristic boundary conditions for compu-
tational aeroacoustics. AIAA Journal, 38:2040–2049, 2000.

[108] S. Krishnan. An Adaptively Refined Cartesian Grid Method for Moving Boundary
Problems Applied to Biomedical Systems. PhD thesis, The University of Iowa, 2006.

[109] L. Krivodonova and M. J. Berger. High-order accurate implementation of solid

wall boundary conditions in curved geometries. Journal Of Computational Physics,
211:492–512, 2006.

[110] Alexander Kurganov and Eitan Tadmor. Solution of two-dimensional Riemann prob-
lems for gas dynamics without Riemann problem solvers. Numerical Methods for
Partial Differential Equations, 18:584 – 608, 2002.
 356

[111] J.O. Langseth and R.J. Leveque. A wave propagation method for three dimensional
hyperbolic conservation laws. Journal of Computational Physics, 165:125 – 166,
2000.

[112] V. S. Lanovets, V. A. Levich, N. K. Rogov, Yu. V. Tunik, and K. N. Shamshev.

Dispersion of the detonation products of a condensed explosive with solid inclusions.
Shock Wave, 29(5):638–641, 1993.

[113] Guillaume Layes, Georges Jourdan, and Lazhar Houas. Distortion of a spherical
gaseous interface accelerated by a plane shockwave. Physical Review Letter, 91,
2003.

[114] Y. Ling, A. Haselbacher, and S. Balachandar. Modeling and simulation of explosive

dispersal of particles in a multiphase explosion.

[115] T. G. Liu, W. F. Xie, and B. C. Khoo. The modified ghost fluid method for coupling
of fluid and structure constituted with hydro-elasto-plastic equation of state. SIAM
J. Sci. Comput., 30(3):1105–1130, 2008.

[116] T.G. Liu, B.C. Khoo, and C.W. Wang. Ghost Fluid Method for compressible gas-
water simulation. Journal of Computational Physics, 204:193–221, 2005.

[117] T.G. Liu, B.C. Khoo, and K.S. Yeo. The simulation of compressible multi-medium
flow II. Application to 2D underwater shock refraction. Computers and Fluids,
30:315 – 337, 2001.

[118] T.G. Liu, B.C. Khoo, and K.S. Yeo. Ghost Fluid Method for strong shock impacting
on material interface. Journal of Computational Physics, 190:651–681, 2003.

[119] T.G. Liu, B.C. Khoo, K.S. Yeo, and C. Wang. Underwater shock-free surface-
structure interaction. International Journal for Numerical Methods in Engineering,
58:609–630, 2003.

[120] X.D. Liu and S. Osher. Convex ENO high order multi-dimensional schemes without
field by filed decomposition or staggered grids. Journal Of Computational Physics,
142:304–330, 1998.

[121] Frank Losasso, Ronald Fedkiw, and Stanley Osher. Spatially adaptive techniques for
level set methods and incompressible flow. Computers and Fluids, 35:2006, 2005.

[122] Frank Losasso, Frédéric Gibou, and Ron Fedkiw. Simulating water and smoke with
an octree data structure. In SIGGRAPH ’04: ACM SIGGRAPH 2004 Papers, pages
457–462, New York, NY, USA, 2004. ACM.
 357

[123] V.L. Mali. Combustion, Explosion, and Shock Waves., 214, 1975.

[124] S. Marella, S. Krishnan, H. Liu, and H.S. UdayKumar. Sharp interface Cartesian grid
method I : An easily implemented technique for 3D moving boundary computations.
Journal Of Computational Physics, 210:1–31, 2005.

[125] A. Marquina and P. Mulet. A flux-split algorithm applied to conservative models for
multicomponent compressible flows. Journal of Computational Physics, 185:120 –
138, 2003.

[126] Daniel F. Martin, Phillip Colella, and Daniel Graves. A cell-centered adaptive pro-
jection method for the incompressible navier-stokes equations in three dimensions.
J. Comput. Phys., 227(3):1863–1886, 2008.

[127] R. Menikoff. Errors when shock waves interact due to numerical shock width. SIAM
Journal on Scientific Computing, 15:1227–1242, 1994.

[128] Chohong Min and Frédéric Gibou. A second order accurate level set method on
non-graded adaptive cartesian grids. J. Comput. Phys., 225(1):300–321, 2007.

[129] J. F. Molinari. Finite element simulation of shaped charges. Finite Elements in

Analysis and Design, 38(10):921 – 936, 2002.

[130] F.J. Mont’ans and R.I. Borja. Implicit j2-bounding surface plasticity using pragers
translation rule. International Journal for Numerical Methods in Engineering,
55(10):1129 – 1166, 2000.

[131] Marc André Myers. Dynamics Behaviour of Materials. John Wiley and Sons, Inc.,
1994.

[132] W.F. Noh. Errors for calculations of strong shocks using an artificial viscosity and
an artificial heat flux. Journal Of Computational Physics, 72:78–120, 1987.

[133] R.R. Nourgaliev, T.N. Dinh, and T.G. Theofanous. The Characteristic Based Match-
ing (CBM) method for compressible flow in complex geometries. In Presented at
the 41st AIAA Aerospace Sciences Meeting and Exhibit, Reno, NV,USA, January
6-9 2003. AIAA 2003-0247.

[134] R.R. Nourgaliev, T.N. Dinh, and T.G. Theofanous. The direct numerical simulation
of disperse multiphase high-speed flows. In Presented at the 42nd AIAA Aerospace
Sciences Meeting and Exhibit, Reno, NV, USA, January 5-8 2004. AIAA 2004-1204.

[135] R.R. Nourgaliev, Dinh T.N., and T.G. Theofanous. The Characteristic Based Match-
 358

ing (CBM) method for compressible flow with moving boundaries and interfaces.
ASME Journal of Fluids Engineering, 126:586–604, 2004.

[136] R.R. Nourgaliev, Dinh T.N., and T.G. Theofanous. Adaptive Characteristics-Based
Matching for compressible multi-fluid dynamics. Journal of Computational Physics,
213:500 – 529, 2006.

[137] S. Osher and J.A. Sethian. Fronts propagating with curvature dependent speed algo-
rithms based on Hamilton - Jacobi. Journal Of Computational Physics, 79:12–49,
1988.

[138] Richard B. Pember, John B. Bell, Phillip Colella, William Y. Crutchfield, and
Michael L. Welcome. An adaptive cartesian grid method for unsteady compress-
ible flow in irregular regions. J. Comput. Phys., 120(2):278–304, 1995.

[139] G. Perigaud and R. Saurel. A compressible flow model with capillary effects. Jour-
nal of Computational Physics, 209:139–178, 2005.

[140] J. M. Pittard, T. W. Hartquist, and J. E. Dyson. Self-similar evolution of wind-blown

bubbles with mass loading by hydrodynamic ablation. Astronomy and Astrophysics,
373:1043–1055, 2001.

[141] T.J. Poinsot and S. K. Lele. Boundary conditions for direct simulations of compress-
ible viscous flows. Journal of Computational Physics, 101:104–129, 1992.

[142] Jean-Philippe Ponthot. An extension of the radial return algorithm to account for
rate-dependent effects in frictional contact and visco-plasticity. Journal of Materials
Processing Technology, 80-81:628 – 634, 1998.

[143] Jean-Philippe Ponthot. Unified stress update algorithms for the numerical simulation
of large deformation elasto-plastic and elasto-viscoplastic processes. International
Journal of Plasticity, 18(1):91 – 126, 2002.

[144] Stéphane Popinet. GERRIS: a tree-based adaptive solver for the incompressible Eu-
ler equations in complex geometries. Journal of Computational Physics, 190(2):572
– 600, 2003.

[145] James J. Quirk. A parallel adaptive grid algorithm for computational shock hydro-
dynamics. Appl. Numer. Math., 20(4):427–453, 1996.

[146] J.J. Quirk. An Adaptive Grid Algorithm For Computational Shock Hydrodynamics.
PhD thesis, Cranfield Institute of Technology, 1991.
 359

[147] J.J. Quirk and S. Karni. On the dynamics of shock bubble interaction. Journal of
Fluid Mechanics, 318:129–163, 1996.

[148] Mittal R. Immersed Boundary Methods. Annual Review of Fluid Mechanics, 37:239
– 261, 2005.

[149] R.C. Ripley, F.S. Lien, and M.M. Yovanovich. Numerical simulation of shock
diffraction on unstructured meshes. Computers and Fluids, 35:1420–1431, 2006.

[150] David J. Roddy, Sheldon H. Schuster, Martin Rosenblatt, Lisa B. Grand, Paul J. Has-
sig, and Kenneth N. Kreyenhagen. Computer simulations of large asteroid impacts
into oceanic and continental sites - preliminary results on atmospheric, cratering and
ejecta dynamics. International Journal of Impact Engineering, 5:525 – 541, 1987.

[151] Alexandre M. Roma, Charles S. Peskin, and Marsha J. Berger. An adaptive version
of the immersed boundary method. J. Comput. Phys., 153(2):509–534, 1999.

[152] T. Saito, M. Mauramoto, and K. Takayama. Numerical investigation of shock waves

through gas-particle mixtures. Shock Waves, 13:299–322, 2003.

[153] T. Saito, M. Saba, M. Sun, and K. Takayama. The effect of an unsteady drag force
on the structure of a non-equilibrium region behind a shock wave in a gas-particle
mixture. Shock Waves, 17:255–262, 2007.

[154] S. Sambasivan and H.S. UdayKumar. Ghost Fluid Method for strong shock in-
teractions part 2: Immersed solid boundaries. AIAA Journal, 47(12):2923–2937,
December 2009.

[155] R. Saurel, P. Cocchi, and P. B. Butler. Numerical study of cavitation in the wake of a
hypervelocity underwater projectile. Journal of Propulsion and Power, 15:513–522,
1999.

[156] H Schardin. Stossrohre [Shock Tubes]. Springer Verlag, New York, 1966.

[157] H Schardin. Optische StromungsmeβTechnik. Braun Verlag, Karlrushe, 1988.

[158] D. Scott Stewart and S.L. Yoo. High resolution numerical simulation of ideal and
non-ideal compressible reacting flows with embedded internal boundaries. Combus-
tion Theory and Modeling, 1:113–142, 1997.

[159] Kurt Sebastian and Chi-Wang Shu. Multidomain weno finite difference method with
interpolation at subdomain interfaces. J. Sci. Comput., 19(1-3):405–438, 2003.
 360

[160] J. A. Sethian. Level Set Methods and Fast Marching Methods: Evolving Interfaces
in Computational Geometry, Fluid Mechanics, Computer Vision, and Materials Sci-
ence. Cambridge University Press, 2 edition edition, 1999.

[161] J.A. Sethian. Evolution, implementation, application of levelsets and fast marching
methods for advancing fronts. Journal Of Computational Physics, 169:503–555,
2001.

[162] J.A. Sethian and P. Smereka. Levelset methods for fluid interfaces. Annual Review
of Fluid Mechanics, 35:341–372, 2003.

[163] C.W. Shu and S. Osher. Efficient implementation of Essentially Non-Oscillatory

shock capturing schemes II. Journal Of Computational Physics, 83:32–78, 1989.

[164] J.C. Simo and T. J. R. Hughes. General return mapping algorithms for rate inde-
pendent plasticity. In C. S. Desai, E. Krempl, P. D. Kiousis, and T. Kundu, editors,
Constitutive Laws for Engineering Materials: Theory and Applications. Elsevier
Science Publishing Co.

[165] J.C. Simo and T.J.R. Hughes. Computational Inelasticity. Springer; Corrected edi-
tion, 2000.

[166] S. Sivier, E. Loth, J. Baum, and R. Lohner. Vorticity produced by shock wave diffrac-
tion. Shock Waves, 2:31–41, 1992.

[167] Vassili Sochinikov and Shlomo Efrima. Level set calculations of the evolution of
boundaries on a dynamically adaptive grid. International Journal for Numerical
Methods in Engineering, 56(13):1913 – 1929, 2003.

[168] M. Sommerfield. Unsteadiness of shock waves propagating through gas-particle

mixtures. Experiments in Fluids, 3:197–206, 1985.

[169] John Strain. Fast tree-based redistancing for level set computations. J. Comput.
Phys., 152(2):664–686, 1999.

[170] John Strain. Semi-Lagrangian methods for level set equations. J. Comput. Phys.,
151(2):498–533, 1999.

[171] John Strain. Tree methods for moving interfaces. J. Comput. Phys., 151(2):616–648,
1999.

[172] John Strain. A fast modular semi-Lagrangian method for moving interfaces. J.
Comput. Phys., 161(2):512–536, 2000.
 361

[173] N. Sukumar, Brian Moran, and Ted B. Belytschko. The natural element method
in solid mechanics. International Journal Numerical Methods in Engineering,
43(5):839, Nov 1998.

[174] M. Sun, T. Saito, K. Takayama, and H. Tanno. Unsteady drag on a sphere by shock
wave loading. Shock Waves, 14:3–9, 2003.

[175] M. Sun and K. Takayama. Vorticity production in shock diffraction. Journal of Fluid
Mechanics, 478(1):237–256, 2003.

[176] Mingyu Sun. Numerical and Experimental Investigation of Shock Wave Interactions
with Bodies. PhD thesis, Tohoku University, 1998.

[177] M. Sussman, K.M. Smith, Ohta M. Hussaini, M.Y., and R. Zhi-Wei. A sharp inter-
face method for incompressible two-phase flows. Journal of Computational Physics,
221(2):469 – 505, 2007.

[178] Mark Sussman and Emad Fatemi. An efficient, interface-preserving level set redis-
tancing algorithm and its application to interfacial incompressible fluid flow. SIAM
Journal on Scientific Computing, 20(4):1165–1191, 1999.

[179] Mark Sussman, Emad Fatemi, Peter Smereka, and Stanley Osher. An improved level
set method for incompressible two-phase flows. Computers & Fluids, 27(5-6):663 –
680, 1998.

[180] Camacho. G. T. and M. Ortiz. Adaptive lagrangian modelling of ballistic penetra-

tion of metallic targets. Computer Methods in Applied Mechanics and Engineering,
142(3-4):269 – 301, 1997.

[181] H. Takahira and S. Yuasa. Numerical simulations of shock bubble interactions us-
ing an improved Ghost Fluid Method. In 2005 ASME Fluids Engineering Division
Summer Meeting and Exhibition, Houston, TX, USA, January 19-23 2005. ASME,
FEDSM2005-77119.

[182] K. Takayama and K. Itoh. Unsteady drag over cylinders and aerofoils in transonic
shock tube ows. In Proceedings of the 15th International Symposium on Shock
Waves and Shock Tubes.

[183] H. S. Tang and F. Sotiropoulos. A second-order godunov method for wave problems
in coupled solid-water-gas systems. Journal of Computational Physics, 151(2):790–
815, 1999.

[184] H.S. Tang and D. Huang. A second order accurate capturing scheme for 1D invis-
 362

cid flows of gas and water with vacuum zones. Journal of Computational Physics,
128(2):301 – 318, 1996.

[185] Geoffrey I. Taylor. The use of flat-ended projectiles for determining dynamic yield
stress. i. theoretical considerations. Proceedings of the Royal Society of London.
Series A. Mathematical and Physical Sciences, 194(1038):289–299, 1948.

[186] Geoffrey I. Taylor and H. Quinney. The latent energy remaining in a metal after cold
working. Proceedings of Royal Society, Series A, 143:307–326, 1934.

[187] Hiroshi Terashima and Grtar Tryggvason. A front-tracking/ghost-gluid method

for fluid interfaces in compressible flows. Journal of Computational Physics,
228(11):4012 – 4037, 2009.

[188] K.W. Thomson. Time dependent boundary conditions for hyperbolic systems II.
Journal of Computational Physics, 89:439–461, 1990.

[189] E.F. Toro. Riemann Solvers and Numerical Method for Fluid Dynamics - A Practical
Introduction. Springer, 2 edition, 1997.

[190] L. B. Tran and H.S. UdayKumar. A particle levelset based sharp interface Cartesian
grid method for impact, penetration, and void collapse. Journal Of Computational
Physics, 193:469–510, 2004.

[191] H.S. UdayKumar, S. Krishnan, and S. Marella. Adaptively refined, parallelised sharp
interface cartesian grid method for three-dimensional moving boundary problems.
International Journal of Computational Fluid Dynamics, 23:1061–8562, 2009.

[192] H.S. UdayKumar, L. Tran, D.M. Belk, and K.L. Vanden. An Eulerian method for
computation of multi-material impact with ENO shock-capturing and sharp inter-
faces. Journal Of Computational Physics, 186:136–177, 2003.

[193] B van Leer. Towards the ultimate conservative difference scheme, V. A second order
sequel to Godunov’s method. Journal of Computational Physics, 32:101136, 1979.

[194] K. J. Vanden. Characteristic analysis of the uniaxial stress and strain governing
equations with thermal-elastic and mie-grüneison equation of state. Technical Mem-
orandum, AFRL, Eglin AFB, FL, 1998.

[195] W. P. Walters and Jonas A. Zukas. Fundamentals of Shaped Charges. Wiley-

Interscience, 1989.

[196] C. Wang, H. Tang, and T .G. Liu. An adaptive ghost fluid finite volume method for
compressible gas-water simulations. J. Comput. Phys., 227(12):6385–6409, 2008.
 363

[197] C.W. Wang, T.G. Liu, and B.C. Khoo. A real Ghost Fluid Method for the simu-
lation of multi-medium compressible flow. SIAM Journal of Scientific Computing,
28(1):278–302, 2006.

[198] Z. J. Wang. A quadtree-based adaptive Cartesian/quad grid flow solver for Navier-
Stokes equations. Computers & Fluids, 27(4):529 – 549, 1998.

[199] Alex B. Wenzel. A review of explosive accelerators for hypervelocity impact. Inter-
national Journal of Impact Engineering, 5:681 – 692, 1987.

[200] P. Woodward and P. Colella. The numerical simulation of two dimensional fluid flow
with strong shocks. Journal of Computational Physics, 54:115–173, 1984.

[201] Han-Chin Wu. Continuum Mechanics and Plasticity. Chapman & Hall, CRC Press,
2004.

[202] W. F. Xie, Y. L. Young, and T. G. Liu. Multiphase modeling of dynamic fluid-

structure interaction during close-in explosion. International Journal for Numerical
Methods in Engineering, 74(6):1019 – 1043, Oct 2006.

[203] Wenfeng Xie, Zhanke Liu, and Yin Lu Young. Application of a coupled eulerian-
lagrangian method to simulate interactions between deformable composite structures
and compressible multiphase flow. International Journal for Numerical Methods in
Engineering, 9999(9999):609–630, July 2009.

[204] S. Xu, T.D. Aslam, and D. Scott Stewart. High resolution numerical simulation of
ideal and non-ideal compressible reacting flows with embedded internal boundaries.
Combustion Theory and Modeling, 1:113–142, 1997.

[205] Darren De Zeeuw and Kenneth G. Powell. An adaptively refined Cartesian mesh
solver for the euler equations. J. Comput. Phys., 104(1):56–68, 1993.

[206] Darren De Zeeuw and Kenneth G. Powell. An adaptively refined Cartesian mesh
solver for the Euler equations. Journal of Computational Physics, 104(1):56–68,
1993.

[207] Darren L. De Zeeuw. A quadtree-based adaptively-refined Cartesian-grid algorithm

for solution of the Euler equations. PhD thesis, Ann Arbor, MI, USA, 1993.

[208] F. Zhang, D.L. Frost, P.A. Thibault, and S.B. Murray. Explosive dispersal of solid
particles. Shock Waves, 10(2):431 – 443, 2001.

[209] Fan Zhang, Paul A. Thibault, and Rick Link. Shock interaction with solid particles
 364

in condensed matter and related momentum transfer. Proceedings: Mathematical,

Physical and Engineering Sciences, 459(2031):705–726, March 8 2003.

[210] Y. Y. Zhu and S. Cescotto. Unified and mixed formulation of the 4-node quadrilateral
elements by assumed strain method: Application to thermomechanical problems.
International Journal for Numerical Methods in Engineering, 38(4):685–716, 1995.

[211] Udo Ziegle. A three-dimensional cartesian adaptive mesh code for compressible
magnetohydrodynamics. Computer Physics Communications, 116(1):65 – 77, 1999.

[212] J. A. Zukas, T. Nicholas, H. F. Swift, L. B. Greszczuk, and D. R. Curran. Impact

Dynamics. Krieger Publishing Company, 1992.

[213] J. A. Zukas and S. B. Segletes. Numerical modeling of hypervelocity impact phe-

nomena with desktop computers. Adv. Eng. Softw., 14(1):77–84, 1992.

[214] Jonas A. Zukas. Introduction to Hydrocodes. Elsevier Science, 2004.