High explosive simulation using multi-material formulations

A. Alia
*
, M. Souli
Laboratoire de Me canique de Lille, UMR CNRS 8107, Universite des Sciences et Technologies de Lille, 59655 Villeneuve dAscq Cedex, France
Received 29 April 2005; accepted 13 October 2005
Available online 29 November 2005
Abstract
The accurate numerical simulation of air-blast is an industrial concern. Because the experimental investigation of these fast tran-
sient events is expensive and time-consuming, several numerical approaches of air explosion are carried out. In this paper, an air-
blast simulation is described using Eulerian multi-material formulation. In order to validate the numerical approach and prove its
ability for high pressure wave propagation, comparison of two examples with experimental results is performed. Both simulations
lead to the same prediction for the pressure time history. Good agreement between the test results and the predicted pressure
response is achieved.
2005 Elsevier Ltd. All rights reserved.
Keywords: ALE multi-material formulation; High explosive; Shock wave propagation
1. Introduction
Theoretical and experimental studies of blast wave
have been considered by several researchers over the
past decades. When a high explosive is detonated an
inward wave is generated in the explosive material, at
the same time, a shock wave moves through the air med-
ium, which is at lower pressure and a contact disconti-
nuity appears between the rarefaction wave and the
shock wave. Experiments have shown (see [1]) that the
resulting ow is quite complex, involving several physi-
cal phenomena as burning eects and heat transfer. The
detonation of high explosive material converts the
explosive charge into gas at high pressure and tempera-
ture what leads to damage structures. Numerical simula-
tion of high explosive detonation and expansion are very
dicult for classical numerical methods, see [2]. During
the process in the explosion, a very thin reaction zone
divides the domain into inhomogeneous parts and pro-
duces large deformations. Numerical simulations help
to minimise the number of tests required that are very
costly, and also help to interpret test results. Once sim-
ulations are validated by test results, it can be used as
design tool for the improvement of the system structure
involved. Classical Lagrangian nite element methods
cannot resolve large deformations very accurately.
Recent development as ALE
1
or Eulerian multi-material
formulations can be used as an alternative for the simu-
lation of high explosive phenomena. The ALE formula-
tions have been developed to overcome the diculties
due to large mesh distortion.
In this paper the Eulerian multi-material formulation
for the simulation of air-blast problem is presented by
describing the ALE formulation, whose the Eulerian
approach is a particular case. The paper is concentrated
on the validation of the methodology that has been
implemented in an explicit nite element code for struc-
tural dynamic to be able to simulate uid structure inter-
action problems, where the uid can be dened by more
than one material. Auid element can contain more than
1359-4311/$ - see front matter 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.applthermaleng.2005.10.018
*
Corresponding author. Tel.: +33 3 20 33 71 76; fax: +33 3 20 33 71
53.
E-mail address: ahlem.alia@ed.univ-lillel.fr (A. Alia).
1
Arbitrary Lagrange Euler.
www.elsevier.com/locate/apthermeng
Applied Thermal Engineering 26 (2006) 10321042
one uid material. For explosion problem, an element
may contain one or two dierent material, air and gas
produced from explosive detonation. At each time step,
state variables are computed and stored for each mate-
rial. An interface-tracking algorithm based on Youngs
method is used to capture the interface between the
two materials inside the element. This method was used
successfully to model many industrial and academic
applications as the sloshing tank problem [3,4].
In Section 2, the governing equations and the multi-
material algorithms are described. In this section, we
discuss the advection algorithms used to solve mass,
momentum and energy conservation in the multi-mate-
rial formulation. In Section 3, the detonation and dis-
persion processes of high explosive are presented. The
last section is devoted to validate the capability of the
methodology by comparing predicted pressure time his-
tory with experimental data. The most interesting
parameters from these simulations are peak value and
duration of pressure pulse at specic locations, these
pressures can be used for structural response.
2. ALE and Eulerian multi-material formulations
The ALE Multi-Material formulation is a method
allowing the nite element mesh to move independently
from the material ow and where each element in the
mesh can contain a mixture of two or more dierent
materials [5]. The Eulerian Multi-Material formulation
is a method where the material ows through a xed
mesh and where two or more dierent materials can be
mixed within the same element. The material boundaries
are dened and each material is assigned a number.
Material numbers therefore serve a function beyond dis-
tinguishing material with dierent properties. In ALE
approach, after moving the mesh, a remapping of state
variables to the new mesh locations is performed. Advec-
tion methods are essential for accurate remapping. Since
the Eulerian formulation is a particular case of the ALE
nite element formulation, a general ALE point of view
is described for compressible NavierStokes equations.
The ALE approach is based on the arbitrary movement
of a reference domain, which will correspond to the nite
element mesh. As detailed in [6], this domain is intro-
duced as a third domain additionally to the common
material and spatial domains. These domains correspond
to the Lagrangian and Eulerian domains, respectively. In
ALE formulation, the air-blast problem is formulated in
the coordinate of the reference domain. The arbitrary
movement of this reference frame, accompanied by a
mesh-moving algorithm, enables us to rather conve-
niently deal with moving boundaries, free surfaces and
large deformations.
The total time derivative of a variable f with respect
to a reference coordinate can be described as Eq. (1):
df
~
X; t
dt

of ~x; t
ot
~v ~w grad
!
f ~x; t 1
where
~
X is the Lagrangian coordinate, ~x is the ALE
coordinate, ~v is the particle velocity and ~w is the velocity
of the reference coordinate, which will represent the grid
velocity for the numerical simulation, and the system of
reference will be later the ALE grid. Thus substituting
the relationship between the total time derivative and
the reference conguration time derivative derives the
ALE equations.
Let X
f
2 R
3
represents the domain occupied by the
uid particles, and let oX
f
denotes its boundary. The
equations of mass, momentum and energy conservation
for a Newtonian uid in ALE formulation in the refer-
ence domain, are given by
oq
ot
qdiv~v ~v ~wgradq 0 2
q
o~v
ot
q~v ~w grad~v div
!
r
~
f 3
q
oe
ot
q~v ~w grad
!
e r : grad~v 4
where q is the density and r is the total Cauchy stress
given by
r p Id lgrad~v grad~v
T
5
where p is the pressure and l is the dynamic viscosity.
For compressible ow, Eqs. (2)(4) are completed by
an equation of state that relies pressure to density and
internal energy.
In the energy Eq. (4), e is the internal energy by unit
volume, the temperature is computed using an average
heat capacity C
v
:
qe C
v
T 6
Eqs. (2)(4) are completed with appropriate boundary
conditions.
One of the major diculties in time integration of the
ALE NavierStokes equations (2)(4) is due to the non-
linear term related to the relative velocity ~v ~w. For
some ALE formulations, the mesh velocity can be solved
using a remeshing and smoothing process. In the Eule-
rian formulation, the problem is formulated in the spa-
tial coordinate, for which the reference frame is xed:
~w
~
0. This assumption eliminates the remeshing and
smoothing process, but does not simplify Eqs. (2)(4).
To solve ALE formulation of NavierStokes Eqs.
(2)(4), there are two ways, and they correspond to
the two approaches taken in implementing the Eulerian
viewpoint in uid mechanics. The rst way solves the
fully coupled equations for computational uid mechan-
ics; this approach used by dierent authors can handle
only a single material in an element. The alternative
approach is referred to as an operator split in the
A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042 1033
literature [7,2] where the calculation, for each time step
is divided into two phases.
First, a Lagrangian phase is performed, using an
explicit nite element method, in which the mesh moves
with the uid particle. In the CFD community, this
phase is referred to as a linear Stokes problem. In this
phase, the changes in velocity, pressure and internal
energy due to external and internal forces are computed.
The equilibrium equations for the Lagrangian phase are
q
d~v
dt
div
!
r
~
f 7
q
de
dt
r : grad~v
~
f ~v 8
From a discretization point of view Eq. (7) is computed
by using one integration point for eciency and in order
to eliminate locking [5]. The zero energy modes are con-
trolled with an hourglass viscosity [8]. The resolution is
advanced in time with the central dierence method,
which provides a second-order accuracy in time using
an explicit method. The mass conservation equation is
used in its integrated form Eq. (9) rather than as a par-
tial dierential equation [3]. Although the continuity
equation can be used to obtain the density in a Lagrang-
ian formulation, it is simpler and more accurate to use
the integrated form Eq. (9) in order to compute the cur-
rent density q:
qJ q
0
9
where q
0
is the initial density and J is the volumetric
strain given by the Jacobian:
J det
ox
i
oX
j

10
Before the second phase, which is the advection phase,
an interface-tracking algorithm is performed in order
to compute accurately the material interfaces in the
ALE elements containing several uid materials. For
these elements, the volume fraction of one uid satises:
V
f
6 1 11
The total stress by r is weighed by volume fraction to
get the uid stress elds:
r
f
r V
f
12
A way of tracking interfaces is the use of the volume
fractions of the elements, or the Young method [9].
The Young or volume of uid (VOF) method is devel-
oped to track an interface in elements containing two
materials for two-dimensional problems. This method
is adapted in this paper for the three-dimensional prob-
lems. In this method, the material layout is described so-
lely by the volume fraction repartition of the uid
material in the ALE elements. Specically, a straight line
using the simple linear interface calculation (SLIC) tech-
nique of Woodward and Collela [10] approximates the
interface in the cell. Interfaces are initially drawn paral-
lel to the element faces. Then nodal volume fraction is
computed to each node based on the fraction volumes
of elements that share the same node. This nodal volume
fraction repartition determines the slope of the material
interface inside the element. The normal vector to the
interface inside the element is dened by
~n
grad
!
f
j grad
!
f j
13
where f is the nodal volume fraction. The position of the
interface is then adjusted so that it divides the element
into two volumes, which correctly matches the element
volume fraction. The interface position is used to calcu-
late the volume of the uid owing across cell sides. As
the X-advection, Y-advection and Z-advection are cal-
culated in separate steps, it is sucient to consider the
ow across one side only.
In the second phase, called advection or transport
phase, the transportation of mass, momentum and
energy across element boundaries are computed. This
may be thought of as remapping the displaced mesh at
the Lagrangian phase back to its initial position. The
transport equations for the advection phase are
o/
ot
~c grad
!
/ 0
/~x; 0 /
0
x
14
where~c ~v ~w is the dierence between the uid veloc-
ity ~v, and the velocity of the computational domain ~w,
which will represent the mesh velocity in the nite ele-
ment formulation. In some papers ~c is referred as the
convective velocity.
Eq. (14) is solved successively for the conservative
variables: mass, momentum and energy with initial con-
dition /
0
(x), which is the solution from the Lagrangian
calculation of Eqs. (7) and (8) at the current time. In Eq.
(14), the time t is a ctitious time: in this paper, time step
is not updated when solving for the transport equation.
There are dierent ways of splitting the NavierStokes
problems. In some split methods, each of the Stokes
problem and transport equations are solved successively
for half-time step.
The hyperbolic system Eq. (14) is solved for mass
momentum and energy by using a nite volume method.
Either a rst-order upwind method or second-order Van
Leer advection algorithm [11] can be used to solve Eq.
(14). The second method is employed in the numerical
applications.
3. Material models for air and explosive
Air is modelled with 8th node nite elements using
the hydrodynamic material model. The model requires
1034 A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042
an equation of state, density, pressure cut-o and viscos-
ity coecient to be dened. The viscosity and pressure
cut-o are set to zero, because pressure cannot be nega-
tive and the viscosity forces are negligible. The ideal gas
law (gamma law) is used as the equation of state for air.
This polytropic equation of state is given by considering
the general linear polynomial equation of state [12]:
p C
0
C
1
l C
2
l
2
C
3
l
3
EC
4
C
5
l C
6
l
2

15
For ideal gas, this equation can be reduced using appro-
priate coecients (C
0
= C
1
= C
2
= C
3
= C
6
= 0, C
4
=
C
5
= c 1):
p c 1
q
q
0
E 16
where l
q
q
0
1, q
0
and q are the initial and current
densities of air, and E is the specic internal energy
(units of pressure) and c is the polytropic ratio of specic
heats. For the diatomic molecules comprising air, this
adiabatic expansion coecient is c = 1.4. The numerical
values of Table 1 are used for air. Note that Eq. (16)
gives at time t = 0, an initial pressure P
0
= 1 bar, for
c = 1.4 and E
0
= 2.5 bar.
The polytropic form Eq. (16) is related to the ideal
gas law via
Pv RT 17
where R is the gas constant and T is absolute tempera-
ture. This later depends on the specic internal energy
e with the relation Eq. (6). To be thermodynamically
consistent, the air material must be initialized with a
non-zero internal energy so that its initial pressure is
non-zero. The air can leak out of the mesh if appropri-
ate boundary conditions are not imposed at the external
boundary, to avoid initial air leakage, a 1 bar pressure
boundary condition is assumed.
As clearly described in [13] for underwater explosion
phenomena, the high explosive process is divided in two
processes, the detonation process, and the interaction
process between the produced gas and the surrounding
air. The detonation process consists of burning eects
and propagation of reactive wave with constant velocity
inside the explosive, this process is very fast. It only
takes few microseconds for the explosive to be burned.
Two dierent detonation can be used, programmed
burn based on time, or compression burned where deto-
nation is based on material compression. In this process,
the heat capacity ratio c is also taken as constant,
although it varies slightly in the temperature range asso-
ciated with explosions.
The explosive material model requires density, deto-
nation velocity V
d
, ChapmanJouguet pressure p
CJ
,
and an equation of state for pressure. Throughout this
section, various subscripts will be used. These are s for
an isentrope, h for a Hugoniot, CJ for the CJ state,
and 0 for the initial state. ChapmanJouguet pressure
generally refers to the detonation pressure, which is
somewhat lower than the initial shock front pressure
[14]. The assumption of Chapman and Jouguet [15]
states (see Fig. 1) that for a plane detonation wave to
be propagated steadily, the Rayleigh line, which is
derived from the mass and momentum conservation,
must be tangent at the CJ point to the Hugoniot curve
of the gaseous products that is derived from the energy
conservation. In a steady state process, these later are
assumed be produced with an innite reaction rate in
order to attain the chemical equilibrium. As shown in
Fig. 1, the shock front, which advances through the
explosive with a detonation velocity V
d
compresses the
explosive particles from an status point (P
0
, v
0
) to
another one (P
1
, v
1
), dened by the intersection of the
Rayleigh line and the Hugoniot curve for the explosive.
After the completion of the detonation process, the
interaction process takes place. A produced gas with
high pressure and temperature expands outward by gen-
erating a pressure wave. Due to the high pressure, the
gaseous products can be assumed to be inviscid, and
thus viscous forces are ignored. Unlike underwater
explosion, where the pressure wave velocity is higher
than the gas bubble expansion velocity, in air explosion,
the pressure wave moves with the gasair interface. The
uid layer at the pressure wave is a mixture of gas and
air, which makes it more complex from a simulation
point of view.
Table 1
Parameters of air
c E
0
(Mbar) q
0
(g/cm
3
)
1.4 2.5 10
6
1.293 10
3
Fig. 1. Hugoniot curves and Rayleigh line.
A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042 1035
An essential part of the numerical models, used to
simulate high explosives and their detonating products,
is the equation of state (EOS) relating energy, pressure
and density. Both theoretical and empirical approaches
have been employed to describe explosive equations of
state [16]. Various types of EOS describe the state of det-
onation products. JonesWilkinsLee (JWL) EOS is
widely used because of its simplicity [17] and due to
the fact that most high explosives are well modelled by
this equation of state.
The derivation of the JWL equation of state is out-
lined here, with the following discussion largely plagiar-
ised from [18].
The denition of the JWL equation of state can start
from its isentropic form, namely:
p
s
AexpR
1
V BexpR
2
V CV
x1
18
where p is the pressure and the subscript s denotes ref-
erence to isentropic compression or expansion. A, B, R
1
,
R
2
and x are constants which the performance values
are presented in Table 2. These performance properties
are based on the cylinder expansion test in controlled
conditions [14,17]. They are slightly dierent from those
calculated by thermodynamic equilibrium codes such as
TIGER, IdeX [19] or CHEETAH [2022]. For reacted
products, R
1
is chosen to be about four or so times lar-
ger than R
2
, so that the rst term dominates at high
pressures, the second term is signicant at intermediate
pressures, and the third term prevails at low pressures.
By integrating the thermodynamic identity Eq. (19),
Eq. (18) leads to the energy on the isentrope [23] given
by Eq. (20).
oe
ov

s
p or
oE
oV

s
p with E q
0
e 19
where e is the internal energy per unit mass (in
MJ kg
1
= GPa mm
3
mg
1
) and E is the internal per
unit volume (in GPa) used in the equation of state.
E
s

A
R
1
expR
1
V
B
R
2
expR
2
V
C
x
V
x
20
A full equation of state is then constructed by expanding
the thermodynamic identity Eq. (21):
oe
op

v

v
C
or
oE
op

V

V
C
21
where c is the Gruneisen gamma, in a Taylor expansion
to give the following equation:
p p
s
V
C
V
E E
s
V 22
This has the form identied as a M eGru neisen equa-
tion of state. It is required that the Gru neisen gamma
is held constant independent of the specic volume,
and is identied as C = x. By substituting Eqs. (18)
and (20) into Eq. (22), the desired p(E, V) form is nally
achieved, namely:
p
JWL
V ; E
AexpR
1
V BexpR
2
V CV
1x

x
V
A
R
1
expR
1
V
B
R
2
expR
2
V
C
x
V
x

x
V
E
p
JWL
V ; E
A 1
x
R
1
V

expR
1
V B 1
x
R
2
V

expR
2
V
x
V
E
23
The rst term of JWL equation, known as high pressure
term, dominates rst for V close to one (see Fig. 2). The
second term is inuential in the JWL pressure for V
close to two. Observe that in the expanded state
(V ! 1), the JWL equation of state reduces only to
the third term:
p
x
V
E 24
The last term is the polytropic equation of state Eq. (16)
for air. With x = c 1, the JWL pressure matches
asymptotically the ideal gas pressure for large volumes.
Table 2
Parameters of C-4
A (Mbar) 5.98155
B (Mbar) 0.13750
R
1
4.5
R
2
1.5
E (Mbar) 0.087
x 0.32
V
d
(cm/ls) 0.804
q (g/cm
3
) 1.601
p
CJ
(Mbar) 0.281
0 1 2 3 4 5
0.00
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
Relative volume (V)
P
r
e
s
s
u
r
e

(
M
b
a
r
)
P
JWL
JWL first term
JWL second term
JWL third term
Fig. 2. Variation of JWL pressure with respect to the relative volume.
1036 A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042
4. Computational results
In the explicit time integration, the time step length is
determined by the smallest element size in the model.
Therefore, the mesh density should be as uniform as
possible. Unnecessarily small elements should be
avoided, because they make the time step size small, thus
increasing the computing time. Extremely large elements
should be avoided as they decrease accuracy. Mixing of
small and large elements in the same model should be
also avoided as much as possible because such models
tend to reduce simulation accuracy. Fluids are especially
dicult to model because they undergo large deforma-
tions, and element shapes and sizes can change consider-
ably during the Lagrangian cycle of the ALE time step.
The Van Leer advection algorithm is applied to remap
the conservative variables. An explosion is a complex
phenomenon, which requires good modelling tech-
niques. In this section, two examples are treated. The
rst one is dealing with the propagation of the air-blast
wave in free space. The second concerns the reection of
the shock wave by a rigid wall. In both cases, a spherical
high explosive C-4 is surrounded by air and the ignition
point is placed at the centre of the explosive. The com-
putations are performed with the Eulerian multi-mate-
rial formulation.
4.1. Detonation in free space
This rst problem is the propagation of the air-blast
wave in an innite domain. In order to alleviate calcula-
tions, the air and explosive are modelled by l/8th of
sphere with three symmetric planes (see Fig. 3), which
consisted of 56,916 elements. Another coarse mesh is
composed of 27,972 elements.
The 8-node nite elements have no shear strength but
can represent correctly the detonation pressure. The
high explosive C-4 is assumed to be spherical in shape
(see Fig. 3). The ignition point is placed at the centre
of the sphere. The radius of this later is 4.07 cm and
its weight is 1 lb (454 g). Since the burn velocity is
0.8 cm/ls, it takes about 5 ls for the burn front to con-
sume all the explosive material. Sixteen elements are
needed to span the radius of the sphere to adequately
build up the detonation pressure during the explosive
Fig. 3. The explosiveair model mesh.
Fig. 4. Pressure wave propagation at t = 0.3 ms.
A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042 1037
burn. The spherical charge is surrounded with the air
mesh such that there is one-to-one node match at the
boundary between the explosive model and the air mod-
els (see Fig. 3). The small elements in the gure are
explosive material and the long radially extending ele-
ments are air.
During the explosion simulation, a gas bubble forms
and expands. Consequently, air adjacent elements are
violently pushed in front of the bubble. A high velocity
shock front departs from the explosive source to the sur-
rounding air. A shock is a narrow discontinuity in the
pressure wave, and therefore would require ne mesh
resolution in order to capture a reasonably accurate
shock peak pressure. The spherical air mesh is ne
enough to match accurately the shock pressure that
originates from the explosive.
The pressure wave propagation simulated by the
Eulerian multi-material formulation is shown in Figs.
46 at dierent times.
Fig. 7 shows two predictions of the overpressure at a
point situated at 5 ft (152.4 cm) from the charge. The
Eulerian multi-material pressure increase of 2.9 bars
occurs at t = 1.5 ms for the ne mesh. For the coarse
one, the birth time is 1.44 ms and the predicted overpres-
sure is 2.7 bars. In Fig. 8, the numerical overpressure of
the ne mesh is compared with the experimental result.
The experimental curve presents a pressure peak of
2.96 bars at t = 1.5 ms. The numerical results converge
to the experimental ones. The best relative errors in arri-
val time and peak magnitude are less or equal to 2%
which represents a good approximation for this
problem.
Fig. 5. Pressure wave propagation at t = 0.6 ms.
Fig. 6. Pressure wave propagation at t = 0.9 ms.
1038 A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042
4.2. Detonation in presence of a wall
In this application, the explosive charge is 4 ft from a
wall. As previously, the spherical high explosive C-4
(radius 3.23 cm, weight 0.5 lb (227 g)) modelled with
32 hexahedral nite elements (see Fig. 9) is ignited at
its centre and 16 elements are needed to span the radius
of the sphere to adequately build up the detonation pres-
sure during the explosive burn. The burn front con-
sumes all the explosive material with about the same
time as the preceding application. Two kind of meshes
are employed: a coarse mesh composed of 143,202 ele-
ments while another ne one comprised 216,648 ele-
ments. The spherical charge is surrounded with the air
mesh such that there is one-to-one node match at the
boundary between the explosive and the air models
(see Fig. 9). The other boundaries of the air mesh are
plane and orthogonal themselves (see Fig. 9). A plane
of blocked nodes is placed at a distance of x = 4 ft
(121.62 cm) from the explosive charge. Two planes of
symmetry are considered what involves a modelling with
a quarter of uid computational mesh. Non-reecting
boundary conditions are imposed on the other planes.
The pressure wave propagation simulated by the
Eulerian multi-material formulation is shown in Figs.
10 and 11. The pressure wave encounters the wall in less
Fig. 7. Numerical overpressure for two dierent meshes.
Fig. 8. Comparison of the experimental and numerical overpressure.
A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042 1039
than 1.5 ms, reects back into the uid from the bound-
ary and returns to its origin. Figs. 10 and 11 show,
respectively, the propagating shock wave just before
and after reection on the plane. The computation is
stopped at 6.5 ms after the detonation. By that time all
the signicant events take place. In particular, Figs. 12
and 13 show the pressure response at a point located
at 5 ft from the static charge. The experimental over-
pressure of 2.17 bars occurs at t = 2.0 ms whereas the
numerically predicted ones appear at t = 1.85 ms with
a magnitude of 2.13 bars for the ne mesh and with a
magnitude of 1.95 bars for the coarse one. For this rst
peak, the result is similar to the last case. At t = 5.0 ms,
the experimental curve shows a second peak of 1.2 bars
resulting from the reection of the overpressure on the
plane. The numerical approach predicts an amplitude
of 1.26 bars for the ne mesh whereas the coarse one
gives a second numerical overpressure of 1.23 bars.
The birth times of the second overpressure are not so
sharp as the experimental one. As previously for both
pressure peaks, the numerical prediction converges with
the mesh renement to the experimental ones. Note that
Fig. 9. Quarter of high explosive and air mesh.
Fig. 10. Air-blast before the reection (t = 1 ms).
Fig. 11. Reecting air-blast at t = 2 ms.
1040 A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042
the rst overpressure peak compares fairly closely with
the experimental one. For the subsequent peak, the rel-
ative error in term of pressure magnitude is about 8%.
This error is due to the advection cycle. While the Van
Leer advection method is a second-order scheme, the
initial steep shock is smeared out by the inherent diu-
sive error of the advection scheme and the mesh size.
Moreover, an accurate computation of the reected
pressure is really tricky since the magnitude of the shock
vanishes with large distance from the ignition point.
This level of agreement is sucient to valid the Eulerian
multi-material simulation.
5. Conclusion
High explosion process is simulated using an Eulerian
formulation, with a xed mesh. Unlike classical
Lagrangian methods, the mesh is xed and the material
ows though the mesh. For explosion problem in air,
we consider two dierent materials, with interface recon-
struction method between the two materials. The inter-
face is computed inside each element based on volume
fraction of each material within the element. Two dier-
ent cases have been tested, a free propagation pressure
wave, and pressure wave with rigid wall reection. For
the advection algorithm that follows the Lagrangian
phase in the split method, we used the second-order algo-
rithms. The pressure impulses computed by time integra-
tion of pressure time history are very close for both
simulations. Good agreements for the pressure transient
in the explosion with experimental data are obtained and
major explosion physical phenomenon can well simu-
lated using the Eulerian multi-material method. The
development of multi-material formulations has opened
a wide range of simulations that were not possible with
the Lagrangian formulation due to the inherent features
such large mesh distortion. These new formulations
allow the simulation of a variety of shock events with dif-
ferent transmitting media. In the next stage of the present
study, we will make precise numerical investigations to
predict a more accurate ground reection and will also
attempt to provide a more realistic model by replacing
the rigid wall by a exible structure.
Acknowledgements
The authors thank Dr. David L. Kennedy for sharing
his expertise in the area of explosive detonation.
References
[1] D.W. Boyer, An experimental study of the explosion generated by
a pressurised sphere, J. Fluid Mech. 9 (1960) 401429.
[2] M. Souli, A. Ouahsine, L. Lewin, ALE and uid-structure
interaction problems, Comput. Meth. Appl. Mech. Eng. 190
(2000) 659675.
[3] T. Belytschko, W.K. Liu, B. Moran, Nonlinear Finite Elements
for Continua and Structures, John Wiley & Sons, New York,
2000.
Fig. 12. Numerical overpressure for dierent meshes.
Fig. 13. Comparison of the experimental and numerical overpressure.
A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042 1041
[4] N. Aquelet, M. Souli, J. Gabrys, L. Olovsson, A new ALE
formulation for sloshing analysis, Struct. Eng. Mech. 16 (4) (2003)
423440.
[5] D.J. Benson, A mixture theory for contact in multi-material
eulerian formulations, Comput. Meth. Appl. Mech. Eng. 140
(1997) 5986.
[6] T.J.R. Hughes, W.K. Liu, T.K. Zimmerman, Lagrangian Eulerian
nite element formulation for viscous ows, J. Comput. Meth.
Appl. Mech. Eng. 21 (1981) 329349.
[7] D.J. Benson, Computational methods in Lagrangian and Eulerian
hydrocodes, Comput. Meth. Appl. Mech. Eng. 99 (2) (1992) 235
394.
[8] D.P. Flanagan, T. Belytschko, A uniform strain hexahedron and
quadrilateral and orthogonal hourglass control, Int. J. Numer.
Meth. Eng. 17 (1981) 679706.
[9] D.L. Young, Time-dependent multi-material ow with large uid
distortion, in: K.W. Morton, M.J. Baines (Eds.), Numerical
Methods for Fluids Dynamics, Academic Press, New York, 1982.
[10] P.R. Woodward, P. Collela, The numerical simulation of two-
dimensional uid ow with strong shocks, Lawrence Livermore
National Laboratory, Report: UCRL-86952, 1982.
[11] B. Van Leer, Towards the ultimate conservative dierence scheme.
IV. A new approach to numerical convection, J. Comput. Phys. 23
(1977) 276299.
[12] B. Langrand, E. Deletombe, J.-L. Charles, J.-F. Sobry, S. Martin,
H. Chazal, Armoured vehicles subject to mine explosionsan
analysis method for operationability and survivability, J. Phys. IV
France 110 (2003) 621626.
[13] G.F. Kinney, K.J. Graham, Explosive Shocks in Air, Springer-
Verlag, New York, 1985.
[14] D.L. Kennedy, Real air equation of state for LS-DYNA, private
communication, 16 February 2000.
[15] M.B. Liu, G.R. Liu, Z. Zong, K.Y. Lam, Computer simulation of
high explosive explosion using smoothed particle hydrodynamics
methodology, Comput. Fluids 32 (2003) 305322.
[16] V.M. Thiel, E.L. Lee, S. Cochran, Eects of overdriven shock
states on the equation of state of 9404 explosive, J. Appl. Phys. 54
(11) (1983) 67606763.
[17] S. Itoh, H. Hamashima, K. Murata, Y. Kato, Determination of
JWL parameters from underwater explosion test, in: 12th Inter-
national Detonation Symposium, San Diego, CA, 2002.
[18] W.C. Davis, in: J.A. Zukas, W.P. Walters (Eds.), Explosive
Eects and Applications, Springer-Verlag, New York, 1998.
[19] F.H. Su, S.S. Cheng, Theoretical estimation for unreacted and
reacted states of composite propellant, Prop. Expl. Pyrotech. 23
(1998) 193199.
[20] P.C. Souers, S. Anderson, R. Avara, L. Fried, J. Janzen, S.
McGuire, B. Wu, Bigplate: an oblique angle explosive eos test,
Lawrence Livermore National Laboratory, Report: UCRL-ID-
131282, 1998.
[21] P.C. Souers, J.W. Forbes, L.E. Fried, W.M. Howard, S. Ander-
son, S. Dawson, P. Vitello, R. Garza, An improved simple method
of deducing JWL parameters from cylinder expansion test, Prop.
Expl. Pyrotech. 26 (2001) 180190.
[22] L. Fried, P.C. Souers, CHEETAH: a next generation thermo-
chemical code, Report: UCRL-ID-117240, Lawrence Livermore
National Laboratory, 1994.
[23] P.C. Souers, L.C. Haselman Jr., Detonation equation of state at
LLNL, Lawrence Livermore National Laboratory, Report
UCRL-ID-116113, 1993.
1042 A. Alia, M. Souli / Applied Thermal Engineering 26 (2006) 10321042