You are on page 1of 14

A comparative study of buried structure in soil subjected

to blast load using 2D and 3D numerical simulations


Yong Lu
a,
*
, Zhongqi Wang
a
, Karen Chong
b
a
School of Civil and Environmental Engineering, Nanyang Technological University, Singapore, Singapore 639798
b
Defense Science and Technology Agency, Ministry of Defense, 1 Depot Road, Singapore, Singapore 109679
Accepted 12 February 2005
Abstract
The response of underground structures subjected to subsurface blast is an important topic in protective engineering. Due to various
constraints, pertinent experimental data are extremely scarce. Adequately detailed numerical simulation thus becomes a desirable alternative.
However, the physical processes involved in the explosion and blast wave propagation are very complex, hence a realistic and detailed
reproduction of the phenomena would require sophisticated numerical models for the loading and material responses. In this paper, a fully
coupled numerical model is used to simulate the response of a buried concrete structure under subsurface blast, with emphasis on the
comparative performance of 2D and 3D modeling schemes. The explosive charge, soil medium and the RC structure are all incorporated in a
single model system. The SPH (smooth particle hydrodynamics) technique is employed to model the explosive charge and the close-in zones
where large deformation takes place, while the normal FEM is used to model the remaining soil region and the buried structure. Results show
that the 2D model can provide reasonably accurate results concerning the crater size, blast loading on the structure, and the critical response
in the front wall. The response in the remaining part of the structure shows noticeable differences between the 2D and 3D models. Based on
the simulation results, the characteristics of the in-structure shock environment are also discussed in terms of the shock response spectra.
q 2005 Elsevier Ltd. All rights reserved.
Keywords: Underground structure; Subsurface blast; Soil medium; Blast wave propagation; Numerical simulation; Coupled model
1. Introduction
The response of underground structures subjected to
blast loading is an important topic in protective engineering.
Usually such structures are box shaped concrete structures,
partially or fully buried in soil medium. The loading and
response of the underground structure involve different
mechanisms as compared to the above-ground structures.
Concerning the general response, the above-ground struc-
tures are often modeled using single-degree-of-freedom
(SDOF) system. This formulation offers an efcient method
of analysis for preliminary designs, optimization studies,
and concept evaluations. In the case of underground
structures, the modelling is complicated by the presence
of the surrounding soil. If the underground structure is to be
modeled using SDOF, the dominant mode of the response
should be identied; however, in reality the response of an
underground structure involves the structuresoil inter-
action (SSI). Consequently, the choice of an appropriate
SDOF model and loading function are complicated. Some
modications on the SDOF methods were put forward to
consider the SSI effect [1,2]. These methods focused on the
modication of the free-eld stress function or the
parametric values of the model to t the experimental
data. But the usefulness of these modications is limited
because they could give unreliable results in some ranges
[3]. Moreover, the modications on the loading function and
the model parameters are difcult to determine when the
initial conditions of the problem are unclear. Besides, a
SDOF does not provide detail response information within
the structure.
To overcome the abovementioned difculties, the nite
element method (FEM) may be adopted, so that the structure
and soil can be modeled in a more realistic manner while
the structuresoil interaction can also be incorporated.
Soil Dynamics and Earthquake Engineering 25 (2005) 275288
www.elsevier.com/locate/soildyn
0267-7261/$ - see front matter q 2005 Elsevier Ltd. All rights reserved.
doi:10.1016/j.soildyn.2005.02.007
* Corresponding author. Address: School of Civil and Environmental
Engineering, Nanyang Technological University, 50 Nanyang Avenue,
Singapore, 639798. Tel.: C65 6790 5272; fax: C65 6791 0676.
E-mail address: cylu@ntu.edu.sg (Y. Lu).
Some coupled methods have been proposed for the analyses
of responses of underground structures under blast loading.
The main techniques include FEM, FDM (nite difference
method) and the so-called hybrid method, and some
successful applications have been reported [46]. But in
these methods the explosion process is not included, and
usually the blast loading has to be simplied as a pressure or
velocity function applied on the boundary of the compu-
tational domain. This simplied treatment may be accep-
table in certain cases such as the structure being far away
from the explosion charge, but in more complicated
situations such as a short standoff distance explosion or a
shallowly buried structure case, the above simplication
may be prone to larger errors because the loading functions
are difcult to dene accurately.
In order to model the whole physical process in a more
realistic manner, it is desirable that a fully-coupled method
be devised so that all the processes can be integrated in a
single numerical model. The main challenge to a fully-
coupled method is the needs to model large deformation
zone surrounding the charge while at the same time to be
able to model appropriately the complex geometry of the
buried structure. In the present model, the above problem is
solved by coupling the SPH (smooth particle hydrodyn-
amics) technique, which is superior in modeling large
deformation zone with relatively simple geometry, with the
Lagrange FE elements that are effective in modeling
complex structures.
In the coupled computation, the three-dimensional (3D)
analysis is the ideal approach as it resembles the actual
situation in a more straightforward manner. However, the
computational cost is high with a 3D model, and this can
become a big barrier especially when a large number of runs
are required. As an alternative, in many cases a simplied
2D axisymmetric model may be considered. In this respect,
it is meaningful to provide a general assessment on the
comparative performance of the 2D and 3D models for such
complex problems as predicting the responses of an
underground structure subjected to blast loading. For this
purpose, in this paper the 2D and 3D models are used to
analyze a typical case in which a side-burst is set on a buried
reinforced concrete structure. Through the comparison
between the 2D and 3D models, the response features and
the accuracy of using 2D model for the prediction are
examined. The general characteristics of the damage and the
in-structure shock environment are discussed.
2. Overview of the fully-coupled method
The subsurface blast effect on underground structures
involves many complicated physical processes, namely the
explosion of the charge, formation of the crater, propagation
of the shock wave and stress wave in soil, soilstructure
interaction, and the response of the structure. To handle this
complex problem, the traditional nite element methods
(FEM) will meet many difculties, in particular the large
deformation of soil near the charge. The large deformation
will result in severe distortion of the mesh and thus interrupt
the computation [7]. Some researchers have tried to use the
hybrid method, which integrates the advantage of the nite
element method and nite difference method to overcome
the mesh distortion. But it is not easy to model the interface
condition with this method and also not easy to apply it in
3D analysis. The primary difculty with these kinds of
traditional FEM or hybrid methods comes from the mesh
they rely on. Once a mesh is produced, the elements or
grids that represent the physical region cannot be changed
easily.
To get rid of the difculties arising from the mesh
method, some meshless methods have been put forward.
One of the most important meshless methods is the SPH
(smooth particles hydrodynamics) method [8]. The SPH
method avoids the large distortion of the elements, and can
track the material boundary easily. One of the primary
limitations with the current SPH method is the inefciency
when modeling thin wall structure. This is because the
thickness of a thin wall is very small comparing to other
dimensions, hence small particles would be required and the
time step would thus become very small. This difculty can
be avoided through the incorporation of the SPH method
with traditional FEM approach [9]. The coupled SPH-FEM
approach benets from the advantages of both methods, and
can be efcient in dealing with the aforementioned
complexities. The following gives an overview of the
computational framework of the fully-coupled model.
2.1. Conservation equation
The conservation equations of mass, momentum and
energy can be expressed as [10]
Mass : r Z
r
0
V
0
V
Z
m
V
(1)
Momentum : r_ u
i
Zs
ij;j
Crf
i
(2)
Energy : r_ e Zs
ij
_ 3
ij
Crf
i
u
i
(3)
where r is density, V is the volume, the subscript 0
indicates the initial value, m is the mass, s
ij
are the stresses,
3
ij
are the strains, u is the spatial velocity, e is the energy, f
is the body force, the mark $ is the rst derivative of time,
i and j range from 1 to 3.
The boundary conditions are either specied displace-
ments or traction
x
i
X; t Zg
i
X; t on G
x
; s
ij
n
j
Zt
i
on G
t
(4)
where x is the current coordinate of a point, X is the
reference coordinate, t is time, n is the exterior normal, g
is the specic displacement function, G
x
or G
t
denotes
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 276
the surface where the displacement or traction boundary
condition are applied.
2.2. The smooth particle hydrodynamic (SPH) method
The main advantage of SPH, as a meshless technique, is
to bypass the need of a numerical grid to calculate spatial
derivatives. This avoids the problems associated with mesh
tangling and distortion, which usually occur in FEM
analyses for large deformation impact and explosive loading
events. Although the name includes the term hydrodyn-
hydrodynamic, in fact the material strength can be
incorporated [9].
In SPH methodology, the material is represented by xed
mass particles to follow its motion. Unlike the grid based
methods, which assume a connectivity between nodes to
construct spatial derivatives, SPH uses a kernel approxi-
mation that is based on randomly distributed interpolation
points. The particles carry material quantities such as mass
m, velocity vector v, position vector x, etc. and form the
computational frame for the conservation equations. In this
method, each particle I interacts with all other particles J
that are within a given distance from it (see Fig. 1a). The
interaction is weighted by the so-called smoothing (or
kernel) function. Using this principal, the value of a
continuous function, or its derivative, can be estimated at
any particle I based on known values at the surrounding
particles J using the kernel estimates [8]. Another
important point is that the SPH nodes can use the same
constitutive models as used for the FEM element.
More details about the SPH technique can be found in
related publications [9].
2.3. Coupling of SPH with FEM
Accurate SPH simulations require large number of
particles throughout the SPH region. Hence if high accuracy
is sought or some special geometry is required, such as thin
walls, etc. large run time can become a problem. The
combination of SPH and Lagrange FEM is a good solution
to this problem. The materials in the low deformation
regions can be modeled using the FEM element. The size of
the particles in the SPH region can also be graded, thus
reducing the overall computational demand. Fig. 1b
schematically illustrates how the SPH particles can be
embedded into a traditional Lagrangian FEM mesh.
There are two different ways that the SPH particles can
be coupled with the FEM elements, one is to join the SPH
particles and the FEM elements, the other is not to join them
but to allow the SPH particles to slide along the surface of
the FEM element; in this case, a special sliding interface
algorithm must be used. In the present study, the SPH
particles are joined with the FEM elements because here the
SPH particles are used for the near-eld soil medium (see
Fig. 5); the interface between the SPH mesh and the FEM
mesh is not a material interface.
3. Constitutive models
The material systems of the problem include soil mass,
concrete/reinforcing steel in the structure, and the high
energy charge. In the present fully-coupled analysis,
sophisticated material models are applied. An overview of
these material models is given in this section. More details
can be found from the respective previous publications
[11,12].
3.1. Three-phases soil model
Soil is a multi-phase mixture composed of solid mineral
particles, water and air, and the deformation mechanism
varies with the stress condition. In the process of explosion
in soils and the subsequent blast wave propagation, the
spatial variation and time variation of the stress in soil are
very large. To cater for this drastic change of stress
conditions requires a robust soil model. Recently the authors
have formulated a numerical three-phase soil model which
is capable of simulating explosion and blast wave
propagation in soils [11,12]. The idea stems from a
conceptual model described in [13]. As illustrated in
Fig. 2, in this model the soil is considered as an assemblage
of solid particles that form a skeleton, while the voids are
lled with water and air. In the gure, the element A, B and
C, respectively, represent the deformation of the solid
particles, water and air, and elements D and E describe
2h
J
I
x-x
Neighboring particles of a kernel estimate
Coupled mesh of SPH - FEM
Fig. 1. SPH and coupling with FEM elements.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 277
the friction and resistance of the bond connection between
the solid particles. The model formulation can be roughly
divided into two main parts; the equation of state and the
strength model. In the equation of state, the contribution of
each phase is considered. The damage of soil skeleton is
also included, taking into account the strain rate effect [14].
Satisfying the continuity requirements, in the three-phase
soil system there should be
DV
V
0
Z
DV
w
V
0
C
DV
g
V
0
C
DV
s
V
0
(5)
where V is the volume of a soil element, V
0
is the initial
volume of the element, V
w
is the volume of water, V
g
and V
s
are volumes of air and soil particles, respectively.
Denote the volume of voids as V
p
, V
p
ZV
g
CV
w
, and
hence VZV
s
CV
p
.
The pressure load causes deformation in each phase, as
well as friction between the solid particles and deformation
of the bond between the solid particles. The friction force
and the force due to the bond are all exerted on the solid
phase. Satisfying the equilibrium it follows that
dpK dVK
vV
s
vp
dp

vV
g
vp
b
C
vV
w
vp
b

K1
C
vp
a
vV
p
C
vp
c
vV
p

Z0
(6)
where p is the total hydrostatic pressure, p
s
is the pressure
exerted on the solid phase, p
a
is the pressure borne by the
friction between the solid particles, p
b
is the pressure borne
by the water and gas, or the pore pressure, p
c
is the
pressure borne by the bond between the solid particles, and
p
e
is the pressure carried by the soil skeleton which is equal
to the sum of p
a
and p
c
.
Eq. (6) describes the volumetric deformation under the
hydrostatic pressure, in which vV
s
/vp, vV
g
/vp
b
, vV
w
/vp
b
,
vp
a
/vV
p
, vp
c
/vV
p
can be obtained from their independent
equations of state or stressstrain relationship, respectively.
The continuum damage model is applied to describe the
damage of the soil skeleton. The bonds between the solid
particles can be represented by a series of elastic brittle
laments. The resisting stress in each lament obeys the
Hookes law until the lament breaks. Introducing a damage
variable D, it has
p
c
ZE
0
1 KDDV
p
=V
p
(7)
and
D Z1 Kexp K
1
h
b3
eff

h

(8)
where B, h are constants related to the properties of the soil,
b is a constant, 3
eff
is the effective strain
3
eff
Z

2
p
3
3
1
K3
2

2
C3
2
K3
3

2
C3
3
K3
1

1=2
:
On the other hand, the friction between the solid
particles, p
a
, is dependent on the normal stress between
the particles and hence can be assumed to be proportional to
the deformation of the soil skeleton
p
a
ZfK
p
DV
p
(9)
where f is the friction coefcient of the solid particles, K
p
is
the coefcient of proportionality, DV
p
is the incremental
volume of voids.
In the soil model, the viscosity of the water and air is
neglected, so the total shear stress is borne by the soil
skeleton formed by the solid particles. To include the effect
of hydrostatic stress on the shearing resistance of the soil,
the modied Drucker-Prager yield criterion [15] is adopted,
as follows
f Z

J
2

KaI
1
Kk Z0 (10)
in which a and k are material constants related to the
frictional and cohesive strengths of the material, respect-
ively; and I
1
, J
2
are the rst and deviatoric stress invariant,
respectively. Taking into account the strain rate enhance-
ment, the yield function becomes
f Z

J
2

KaI
1
Kk 1 Cb ln
_ 3
eff
_ 3
0

Z0 (11)
where _ 3
0
is the reference effective strain rate, b is the slope
of the strength against the logarithm of strain rate curve, _ 3
eff
is the effective strain rate dened as
_ 3
eff
Z

2
3
d_ 3
ij
d_ 3
ij

:
3.2. Concrete model
The response of the concrete under shock loading is a
complex nonlinear and rate-dependent process. A variety of
constitutive models for the dynamic and static response of
concrete have been proposed in the past. This study adopts
the RHT model developed by Riedel, Hiermaier and Thoma
[16]. This model contains many features known to inuence
the behaviour of brittle materials, namely pressure
hardening, strain hardening, strain rate hardening, third
invariant dependence for compressive and tensile
Fig. 2. Concept of three-phase soil model for shock loading.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 278
meridians, and cumulative damage (strain softening). The
RHT model for concrete has been evaluated successfully in
the modeling of concrete perforation under shock loading,
and systematic parameters have been obtained for several
kinds of concrete.
The RHT model can be generally divided into two parts,
the EOS model and the strength model. The strength model
uses three strength surfaces (Fig. 3); an elastic limit surface,
a failure surface and the remaining strength surface for the
crushed material. Usually there is a cap on the elastic
strength surface.
Following the hardening phase, additional plastic strain-
ing of the material leads to damage and strength reduction.
Damage is accumulated via
D Z

D3
pl
3
failure
p
(12)
3
failure
p
ZD
1
p

Kp

spall

D
2
R3
min
f
(13)
where D
1
and D
2
are damage constants, 3
min
f
is the minimum
strain to reach failure, 3
pl
denotes the plastic strain, p
*
is the
pressure normalized by f
c
, and p

spall
Zp

f
t
=f
c
, where f
t
and
f
c
are tensile and compressive strength, respectively.
The equation of state in the RHT concrete model is P-a
type. The basic Herrmanns P-a model [17] is a phenom-
enological approach, which emphasizes on a correct
behavior at high stresses, but at the same time it also
attempts to provide a reasonable description of the
compaction process at low stress levels. The principal
assumption is that the specic internal energy for a porous
material is the same as that of the same material at solid
density under the same pressure and temperature. Dene the
porosity as aZv/v
s
, where v is the specic volume of the
porous material and v
s
is the specic volume of the material
in the solid state with the same pressure and temperature. If
the equation of state of solid material is given by: pZf(v
s
,e),
then the equation of state of the porous material simply
takes the form: pZf(v/a,e). This equation has been modied
by Carroll and Holt [18] to yield
p Z
1
a
f
v
a
; e

(14)
where the factor 1/a was included on the basis of an
argument that the pressure in the porous material is nearly
1/a times the average pressure in the matrix material. In the
present study, a polynomial form is adopted for the
functions f, as
f ZA
1
mCA
2
m
2
CA
3
m
3
with m Z
v
0
v
s
K1R0
f ZT
1
mCT
2
m
2
with m Z
v
0
v
s
K1!0
(15)
and
a Z1 Ca
initial
K1
p
s
Kp
p
s
Kp
e

n
(16)
where A
1
, A
2
, A
3
, T
1
, T
2
, are constants, v
0
is initial specic
volume of solid material, a
initial
is initial porous rate, p
e
is
the pressure when the skeleton of the porous material starts
to collapse, p
s
is the pressure when the porous material is
fully compacted, n is a constant.
3.3. Elasticstrain hardening plastic model for steel
Under blast loading, the reinforcing steel may be subject
to strain hardening, strain rate hardening and heat softening
effects. In this study, the John-Cook model [10] is adopted
to model the response of the steel bars in the concrete. The
John-Cook model is a rate-dependent, elasticplastic model.
The model denes the yield stress Y as
Y ZY
0
CB3
n
p
1 CC log 3

p
1 KT
m
H
(17)
Uniaxial Compression
Failure Surface
Elastic Limit Surface
Residual Surface
Uniaxial Tension
P
Tensile
Elastic
Strength
Compressive
Elastic Strength
f
c
f
t
Y
Fig. 3. Three strength surfaces for concrete.
Air
Soil
Blast
D
R
Target point
(a) Free field without buried structure
Air
Soil
Underground structure
Blast
D
R
(b) Coupled field with buried structure
Fig. 4. Schematic description of model congurations.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 279
where Y
0
is the initial yield strength, 3
p
is the effective plastic
strain, 3

p
is the normalized effective plastic strain rate, B, C, n,
m are material constants. T
H
is homologous temperature,
T
H
ZTKT
room
=T
melt
KT
room
, with T
melt
being the melting
temperature and T
room
the ambient temperature.
3.4. JWL equation of state for explosive charge
The Jones-Wilkens-Lee (JWL) equation of state [19]
models the pressure generated by the expansion of the
detonation product of the chemical explosive, and it has
been widely used in engineering calculations. It can be
written in the form
P ZC
1
1 K
u
R
1
v

expKr
1
v
CC
2
1 K
u
R
2
v

expKr
2
v C
ue
v
(18)
Fig. 5. Numerical models for 2D and 3D free-eld analyses.
Table 1
Parameters used in the three-phase soil model for numerical calculations
Soil Air phase
Solid particles a
1
Z0.58 Initial density
r
g0
Z1.2 kg m
3
Water a
2
Z0.38 Initial sound speed
c
g0
Z340 m/s
Air a
3
Z0.04 Constant k
g
Z1.4
Initial density r
0
Z1.92!10
3
kg m
3
Soil skeleton
Solid particles phase Shear modulus GZ55 MPa
Initial density r
s0
Z2.65!10
3
kg m
3
Bulk modulus K
p
Z165 MPa
Initial sound speed c
s0
Z4500 m/s fZ0.56
Constant k
s
Z3 aZ0.25
Water phase kZ0.2
Initial density r
w0
Z1.0!10
3
kg m
3
_ 3Z1%=min
Initial sound speed c
w0
Z1500 m/s bZ0.1
Constant k
w
Z7 hZ1.0
E
0
Z20 MPa
bZ5.0
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 280
where v is the specic volume, e is specic energy. The
values of constants C
1
, R
1
, C
2
, R
2
, u for many common
explosives have been determined from dynamic
experiments.
3.5. Soilconcrete interface
According to the experimental results from Huck et al.
[20], in general the soilstructure (concrete) interface
strengths may be described by Coulomb failure laws. On a
smooth soilconcrete interface failure is initiated when
the shear stress parallel to the surface exceeds the failure
law. On a rough soilconcrete interface, failure is initiated
when the maximum soil shear stress exceeds the failure law.
The experimental results from Mueller [21] indicate that the
strength properties of the interface are close to the strength
properties of the soil. For this reason, in the present study
the interface between the (rough) concrete and soil is
modeled using regular FEM elements with completely
joined surface.
4. Numerical model
The response of an underground structure depends on
the input load. The input load is transmitted from the soil
medium and may be measured in terms of ground shock
or stress wave. Therefore when different analytical
approaches are considered, such as 2D or 3D model, it
is important to rst examine the prediction of the stress
wave in the free-eld soil before going into the soil
structure coupled analysis. Fig. 4(a) and (b) show
schematically the scenarios without or with a buried
structure. Considering the example structure size (to be
described later) and the observations from some trial
analyses, the computational domain is chosen to be on
the order of 50 m wide and 30 m deep. The explosive
charge is chosen to be 50 kg (TNT equivalent),
embedded at 4.8 m deep.
The calculations are carried out using the programme
Autodyn [22]. Specic material models, such as the three-
phase soil model, are implemented by incorporating user
subroutines.
4.1. Free eld analysis
In order to minimize the effect of mesh sizes, three
different meshes were tried for both the 2D and 3D models.
Fig. 5 shows the nal meshes that produce stable results
with a tolerable computation time. A SPH zone is arranged
for the area surrounding the charge (including the charge
itself). The element size for the FEM region is about 0.5 m.
For the 3D model, only half of the eld is modeled
considering the symmetry about the yZ0 plane. The
transmission boundary condition is applied at all the
articial boundaries to minimize the stress wave reection
at these computational boundaries. The parameters used in
the models for the soil and the charge are listed in Tables 1
and 2 based on existing literature [13,19].
Several target points are arranged along the radial
direction from the charge at embedded depth of 4.8 m to
record the propagation of the stress wave in the soil.
Theoretically speaking, the situation can be very well
simulated as an axis-symmetrical problem so the 2D and 3D
models are expected to produce practically the same results,
given appropriate model settings.
Fig. 6(a) and (b) show the computed shape of the crater
formed in the soil at about 100 ms after the detonation of the
charge. The radius of the crater is about 2.2 m. The shape
Table 2
JWL parameters used for modeling TNT in the present study
C
1
(GPa) C
2
(GPa) R
1
R
2
u e
0
(MJ m
K3
) VOD (m s
K1
) r
0
(kg m
K3
)
3.738!10
2
3.747 4.15 0.9 0.35 6.0!10
3
6.93!10
3
1.63!10
3
e
0
, the initial C-J energy per volume; VOD, the C-J detonation velocity.
Fig. 6. Computed craters in soil.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 281
and the size of the crater from the 2D and 3D models are
comparable. The estimated range of the radius of the crater
according to Henrych [13] is about 1.53.5 m for different
shapes of charge and properties of the soil, which are
consistent with the current simulation results.
Fig. 7 shows the computed pressure time histories at a
target of 10 m away from the charge. The results from the
2D and 3D analysis are in good agreement.
The attenuations of the peak pressure, peak particle
velocity (PPV) and peak particle acceleration (PPA) in
soil with increasing distance from the charge are
compared in Fig. 8. The respective empirical equations
recommended by TM5 [2] are also shown in the graphs
(straight lines) for a comparison, namely
Function 1 peak pressure : PP 6!
R

W
3
p

K2:4
MPa
Function2peakparticlevelocity : PPVZ7
R

W
3
p

K2:4
m=s
Function 3 peak particle acceleration :
PPA Z600
R

W
3
p

K3:1
g
in which R is the distance away from the charge, W is
the weight of the charge.
The 2D and 3D results both agree well with the above
empirical equations.
It is also interesting to compare the computing time
needed in running the 2D and 3D analysis for problems of
this scale. With a PC of P4-CPU 2.66 GHz, RAM 1G and
Hard disk 88 GB, the computing times needed for an
Time /ms
0 20 40 60 80 100 120
0.0
0.2
0.4
0.6
P
r
e
s
s
u
r
e

/
M
P
a
Target at 10m
2D model
3D model
Fig. 7. Typical pressure time history in soil at target of 10 m from the
charge.
Scaled distance (R/W
1/3
) /mkg
-1/3
0.4 0.6 0.8 1 2 4 6
0.1
1
10
100
Function 1
P
e
a
k

P
r
e
s
s
u
r
e

/
M
P
a
3D mesh
2D mesh
0.4 0.6 0.8 1 2 4 6
0.1
1
10
100
Function 2
P
P
V

/
m
/
s
Scaled Distance (R/W
1/3
)/mkg
-1/3
2D model
3D model
0.4 0.6 0.8 1 2 4 6
1
10
100
1000
10000
Function 3
P
P
A

/
g

Scaled Distance (R/W
1/3
) /mkg
-1/3
2D model
3D model
Fig. 8. Attenuation of the peak pressure, peak particle velocity (PPV) and peak particle acceleration (PPA) in soil (straight lines based on TM5).
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 282
analysis of duration 130 ms are approximately 45 min for
the 2D model and about 13 h for the 3D model. With the
inclusion of the buried structure as will be described in
Section 4.2, the computing time is about 1 h 30 min for the
2D analysis and about 20 h for the 3D analysis.
4.2. Coupled analysis with buried structure
A representative buried structure is considered for the
analysis. The structure has a box shape of overall dimensions
of 23 m(length)!20 m(width)!5 m(depth), andis buriedin
the soil so that the roof is on the ground surface level. Fig. 9
shows the details of the structure. The charge prole is the
same as in the earlier free-eld analysis (50 kg at embedded
depth of 4.8 m), and the charge standoff distance to the front
wall is 10 m (scaled distance y2.7 m/kg
1/3
)
As the structure is relatively wide in the third direction,
the problem may be simplied as a 2D axis-symmetric
problem. Fig. 10 shows the coupled models for the 2D and
3D analysis, respectively. In the coupled models the charge
and soil are modeled with the same mesh as in the free-eld
analysis, while the structure is modeled using ne FE
elements. It is noted that in the RC structure the reinforcing
steel is modeled by equivalent shell elements with the same
steel volume content. Complete binding condition is
imposed at the interface between the soil and structure.
For concrete with a cylinder compressive strength of
35 MPa and reinforcing steel of yield strength 350 MPa,
the parameters used in modeling the concrete and reinfor-
cing steel materials are summarized in Tables 3 and 4 based
on [16,10]. Several target points are arranged throughout the
2D and 3D structure to record the response of the structure
for analysis and comparison, as indicated in Fig. 9.
(a) 2D view of structure and target points
(b) 3D view of structure (halved) and target points
23m
10m
0.5m
5
.
0
m

Steel rebar
1
5
2
3
4
6
0.6m
1.3m
23m
1.3m
1
5
2
3
4
5
m
0.6m
0.5m
Fig. 9. Buried structure conguration in 2D and 3D models.
Fig. 10. Numerical model meshes for 2D and 3D analyses.
Table 3
Parameters used in the RHT model for concrete
Initial density
r
0
(kg m
K3
)
2.314!10
3
Specic heat C
v
(J/kg K)
6.54!10
2
Reference den-
sity r
s
(kg m
K3
)
2.75!10
3
The RHT strength model
f
c
(Mpa) 35 3
pl(elasticplastic)
1.93!10
K3
A 1.6 B 1.6
N 0.61 M 0.61
f
t
(MPa) 3.5 D
1
0.04
n1 0.036 D
2
1.0
n2 0.032 3
min
f
0.01
Q
2,0
0.6085 P
e
(MPa) 2.33!10
1
G
initial
(MPa) 1.67!10
4
P
s
(MPa) 6.0!10
3
G
residual
(MPa) 2.17!10
3
P-a EOS
A
1
(MPa) 3.527!10
4
T
1
(MPa) 3.527!10
4
A
2
(MPa) 3.958!10
4
T
2
(MPa) 0.0
A
3
(MPa) 9.04!10
3
n 3.0
Table 4
Parameters used for modeling reinforcement steel bar
Reference density r
0
(kg m
K3
)
7.896!10
3
B (MPa) 2.75!10
2
Bulk modulus K (MPa) 2.0!10
5
C 0.022
Specic heat C
v
(J/kg K) 4.52!10
2
n 0.36
Shear modulus G (MPa) 8.18!10
4
m 1.0
Y
0
(MPa) 3.5!10
2
T
room
(K) 3.0!10
2
T
melt
(K) 1.811!10
3
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 283
4.2.1. Damage of the structure
Fig. 11 shows the damage in the front part of the structure
facing the detonation. It can be seen that the damage pattern
of the 2D model is similar to that of the 3D model at the
middle section; however, the maximum value of damage
from the 2D analysis is notably less than that of the 3D
results, especially at the top and bottom connection regions.
The above difference may be attributable to the fact that,
by the 2D axis-symmetric model, the structure is treated
effectively as a circular ring. This introduces some articial
constraint in the circumferential direction as compared to
the actual box-shaped structure, leading to certain under-
estimation of the critical damage to the front wall of the
structure. In this respect, the use of the 3D model certainly
produces more realistic damage distribution, including that
in the transverse direction. The maximum damage in the 3D
model for this case is about 0.20, which corresponds to a
state with some minor cracking in the concrete.
Fig. 11. Computed damages to buried structure (front part facing the charge).
0 20 40 60 80 100 120 140
-1.0
-0.5
0.0
0.5
1.0
1.5
2.0
P
r
e
s
s
u
r
e

/
M
P
a
Time /ms
Center of Front Wall
2D model
3D model
0 20 40 60 80 100 120 140
-0.5
0.0
0.5
1.0
1.5
2.0
P
r
e
s
s
u
r
e

/
M
P
a
Time /ms
Corner of the section
2D model
3D model
0 20 40 60 80 100 120 140
-0.20
-0.15
-0.10
-0.05
0.00
0.05
0.10
0.15
P
r
e
s
s
u
r
e

/
M
P
a
Time /ms
Center of ground plate
2D model
3D model
Fig. 12. Computed blast pressure on the front wall/bottom plate centre.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 284
4.2.2. Computed blast loading on structure
Fig. 12 shows the computed reected pressure on the
front wall and the bottom plate. As can be seen, the results
for the front wall, whether at the centre or near the lower
corner, are very similar between the 2D and 3D analyses.
Marked difference is observed in the pressure on the bottom
plate, where the 3D results exhibit much larger peak
pressure than from the 2D analysis.
In fact, the stress wave impinging the structure will
experience complex reection processes which depend on
the prole of the structure, particularly around the edge and
corner areas. Since in the 2D axis-symmetric model the
structure is treated as a ring in which only the cross-section
geometry is preserved, the blast wave path is somewhat
altered, and consequently the loading conditions on other
parts of the structure (except the front wall) are affected.
4.2.3. In-structure shock
The in-structure shock environment will affect the safety
and functionality of the equipment in the structure and
hence is an important aspect to be evaluated from the
analysis. The time histories of the acceleration and velocity
obtained from the 2D and 3D models at representative target
points are plotted in Figs. 13 and 14. From these curves the
following may be observed: (a) The 2D and 3D predictions
of the shock response at the front wall centre are
comparable. The peak values of the acceleration and
velocity in the horizontal (x) direction agree reasonably
well between the two models. The difference in the
z-direction (vertical) is larger. (b) The difference of the
responses at the bottom plate centre between the 2D and 3D
models is more obvious, particularly in the vertical
direction. This echoes the observation of markedly larger
pressure at the bottom plate from the 3D model as compared
to the 2D model. Besides the difference in the blast pressure,
the support condition of the bottom plate in the idealized
ring shape in the 2D model, as well as the distribution of
the load along y-direction that is not represented in the 2D
model, all affect the accuracy of the computed bottom plate
response in the vertical direction.
Based on the results from the 3D model, some general
observations regarding the in-structure shock environment
can be deduced:
(i) The maximum acceleration and velocity response
occur around the center of the front wall. For
Target 5, x-dir
-60
-40
-20
0
20
40
60
0 20 40 60 80 100 120 140 160
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
2D model
3D model
2D model
3D model
2D model
3D model
2D model
3D model
2D model
3D model
Target 5, z-dir
-30
-20
-10
0
10
20
30
0 20 40 60 80 100 120 140 160
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
2D model
3D model
(a) Front wall centre
Target 2, x-dir
-30
-20
-10
0
10
20
30
0 20 40 60 80 100 120 140 160
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
Target 2, z-dir
-30
-20
-10
0
10
20
30
0 20 40 60 80 100 120 140 160
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
(b) Front wall corner
Target 3, x-dir
-15
-10
-5
0
5
10
15
0 20 40 60 80 100 120 140 160
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
Target 3, z-dir
-15
-10
-5
0
5
10
15
0 20 40 60 80 100 120 140 160
Time (ms)
A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
(c) Bottom plate centre
Fig. 13. Computed acceleration time histories at selected target points.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 285
Target 5, x-dir
-2
-1
0
1
2
0 20 40 60 80 100 120 140 160
Time (ms)
V
e
l
o
c
i
t
y

(
m
/
s
)
2D model
3D model
Target 5, z-dir
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120 140 160
Time (ms)
V
e
l
o
c
i
t
y

(
m
/
s
)
2D model
3D model
2D model
3D model
2D model
3D model
2D model
3D model
2D model
3D model
(a) Front wall centre
Target 2, x-dir
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120 140 160
Time (ms)
V
e
l
o
c
i
t
y

(
m
/
s
)
Target 2, z-dir
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120 140 160
Time (ms)
V
e
l
o
c
i
t
y

(
m
/
s
)
(b) Front wall lower corner
Target 3, x-dir
-0.4
-0.2
0
0.2
0.4
0 20 40 60 80 100 120 140 160
Time (ms)
V
e
l
o
c
i
t
y

(
m
/
s
)
Target 3, z-dir
-1
-0.5
0
0.5
1
0 20 40 60 80 100 120 140 160
Time (ms)
V
e
l
o
c
i
t
y

(
m
/
s
)
(c) Bottom plate centre
Fig. 14. Computed velocity time histories at selected target points.
bottom plate, x-dir
-20
-10
0
10
20
Time (ms) A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
target 2
target 3
target 4
bottom plate, z-dir
-20
-10
0
10
20
0 20 40 60 80 100 120
Time (ms) A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
target 2
target 3
target 4
(a) Bottom plate
Front wall, x-dir
-60
-40
-20
0
20
40
0 20 40 60 80 100 120
0 20 40 60 80 100 120
0 20 40 60 80 100 120
Time (ms) A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
target 2
target 3
target 4
Front wall, z-dir
-20
-10
0
10
20
Time (ms) A
c
c
e
l
e
r
a
t
i
o
n

(
g
)
target 2
target 3
target 4
(b) Front wall
Fig. 15. Spatial variation of acceleration within the structure.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 286
the present case where the scaled standoff distance
(from the charge to the front wall) is about 2.7 m/kg
1/3
,
the maximum acceleration reaches 58 g (horizontal)
with a primary frequency around 100 Hz, which
reects the fundamental natural frequency of the
front wall. The maximum velocity reaches 1.2 m/s
(horizontal direction).
(ii) The maximum x-direction acceleration and velocity of
the bottom plate are 10.2 g and 0.3 m/s, respectively.
The x-direction response of the bottom plate can be
considered as representing the motion of the whole
structure in the horizontal direction.
(iii) There exists noticeable spatial variation of the shock
environment throughout the structure, as also shown in
Fig. 15. This is primarily attributable to the propa-
gation of the shock wave in the structure (within the
plane of each side) and the dynamic response of the
structural components.
For a rough verication of the computed magnitude of
the in-structure shock, the simplied approach in code
TM-5 [2] is used to provide an empirical estimation of the
in-structure acceleration and velocity. This estimation is
based on free-eld ground shock. By integrating the
empirical acceleration-range function over the span of the
structure, an estimate of the average acceleration is
obtained. The estimate of the velocity can be found in a
similar manner. For the kind of soil considered in the current
example, the empirical formulas for the free-eld velocity
and the acceleration, according to TM-5, are in the range of
V
avg1
Z7:38
R

W
3
p

Kn
; V
avg2
Z4:62
R

W
3
p

Kn
;
a
avg1
Z651
R

W
3
p

KnK1
; a
avg2
Z406
R

W
3
p

KnK1
where V
avg1
, a
avg1
correspond to the attenuation coefcient
nZ2, V
avg2
, a
avg2
correspond to the attenuation coefcient
nZ2.5. The average velocity and acceleration of the
structure, estimated from the integration of these functions
over the span of the structure, are found to be V
avg1
Z
0.31 m/s, V
avg2
Z0.09 m/s (velocity), and a
avg1
Z6.45 g,
a
avg2
Z6.45 g (acceleration).
Compared with the computer simulation results, the
above empirical evaluation appears to signicantly under-
estimate the in-structure shock amplitudes, especially for
the front wall (comparing to 58 g and 1.3 m/s from
numerical calculations). This is not surprising as the
evaluation according to TM-5 is only an indication of the
nominal free-eld ground shock over the space occupied by
the embedded structure and it does not reect the dynamic
response of the structure. In fact, as can be seen from the
response histories at different locations on the structure
shown in Figs. 1315, under the excitation of the blast
loading the structure exhibits considerable vibration
response all over the structure, particularly on the front
wall and the bottom plate in the case of a side burst. As a
result, the vibration amplitudes on these components are
signicantly higher. Nevertheless, at locations where the
vibration response is less signicant such as at the corners of
the structure (e.g. target point 2 in Figs. 1315), the
acceleration and velocity amplitudes tend to show reason-
able agreement with the empirical evaluation results.
To provide another perspective of the shock environment
concerning their effects on equipment in the structure,
Fig. 16 shows the shock response spectra for the accelera-
tions computed from the 3D model at different locations of
the structure. It can be seen that the shock response
spectrum at the centre of the front wall is critical in the
entire frequency range of interest, although in the vertical
direction the trend is not that clear. Furthermore, the spectral
acceleration is sharply reduced at the lower frequency
range. Therefore, for acceleration sensitive equipment
devices, the possible shock damage may be much alleviated
if appropriate isolation measures are taken so that the
natural frequency of the individual equipment installations
are reduced to below a certain level, for example below
30 Hz for the case under consideration.
5. Conclusions
In this paper, the response of underground structure
subjected to explosion in soil is analyzed using 2D and 3D
0
20
40
60
80
100
120
0 100 200 300 400 500 600
Frequency (Hz)
S
p
e
c
t
r
a
l

a
c
c
e
l
e
r
a
t
i
o
n

(
g
) Front wall centre
Front wall central corner
bottom plate centre
Rear wall corner
Roof centre
Front wall edge corner
(a) Horizontal (radial) direction
0
10
20
30
40
50
0 100 200 300 400 500 600
Frequency (Hz)
S
p
e
c
t
r
a
l

a
c
c
e
l
e
r
a
t
i
o
n

(
g
) Front wall centre
Front wall central corner
bottom plate centre
Rear wall corner
Roof centre
Front wall edge corner
(b) Vertical direction
Fig. 16. Shock response (acceleration) spectra at different locations of the
structure.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 287
fully-coupled numerical models. The models incorporate
the SPH technique with FEM to form an efcient
combination, where the SHP is employed for its superior
ability in handling large deformations while the FEM is
suited for modeling the buried structure and the relatively
low deformation soil regions. The general response
characteristics are discussed. The computational efciency
and the accuracy of using 2D model as compared to the 3D
model are examined.
The 2D model is shown to be able to predict satisfactorily
the blast wave propagation in the soil medium. With a buried
structure, the 2D model can predict the blast loading and the
response of the front wall of the structure with reasonable
accuracy. The blast wave reection is complicated by the
actual 3D shape of the structure, especially around the edge
and corner areas. As a result, the loading conditions and the
response of the remaining part of the structure using the 2D
model are less satisfactory as compared to the 3D model.
Therefore, in situations where mainly the critical responses in
the structure are of major concerns, the 2D model can be
acceptable for the analysis of such buried structures subjected
to blast loading.
In general, for a side burst scenario considered in this study,
the maximum acceleration and velocity response are found to
take place around the center of the front wall. The maximum
damage also occurs on the front wall at top and bottomregions
as well as around the centre. For a side burst at a standoff
distance about 2.7 m/kg
1/3
, the maximum acceleration is
found to be about 58 g (horizontal) and the maximumvelocity
is about 1.2 m/s (horizontal). Based on the acceleration time
histories, the in-structure shock environment is also depicted
by the shock response spectra within the structure. It is
observed that the peak spectral acceleration occur in the front
wall when the frequency of the oscillator is around 100 Hz.
The shock response spectra drop drastically when the
frequency of the oscillator is reduced to below a certain
level (e.g. 30 Hz for the case herein). This observation can be
useful in the design of equipment installation for reducing the
in-structure shock hazard to the equipment.
References
[1] Biggs JM. Introduction to structural dynamics. New York: McGraw-
Hill; 1964.
[2] TM5-855-1. Fundamental of protective design for conventional
weapons. US Army Engineer Waterways Experiment Station,
Vicksburg; 1984.
[3] Hinman EE. Single degree of freedom solution of structuremedium
interaction. In: Proceedings of international symposium on the
interaction of conventional weapons with structures, March 913,
vol. 1. Mannheim, West Germany: Federal Minister of Defense; 1987,
p. 4459.
[4] Nelson I. Numerical solution of problems involving explosive. In:
Proceedings of dynamic methods in soil and rock mechanics,
September 516, vol. 2. Rotterdam: A.A. Balkema; 1977, p. 23997.
[5] Stevens DJ, Krauthammer T. Analysis of blast-loaded, buried RC arch
response. Part I. Numerical approach. J Struct Eng ASCE 1991;
117(1):197212.
[6] Stevens DJ, Krauthammer T, Chandra D. Analysis of blast-loaded,
buried arch response. Part II. Application. J Struct Eng ASCE 1991;
117(1):21334.
[7] Benson DJ. Computational methods in Lagrangian and Eulerian
hydrocodes. Comput Methods Appl Mech Eng 1992;99:235394.
[8] Hayhurst CJ, Clegg RA. Cylinderically symmetric SPHsimulations of
hypervelocity impacts on thin plates. Int J Impact Eng 1997;20:
33748.
[9] Johnson GR, Petersen EH, Stryk RA. Incorporation of a SPH option
into the EPIC code for a wide range of high velocity impact
computations. Int J Impact Eng 1993;14:38594.
[10] Meyers MA. Dynamic behavior of materials. New York: Wiley; 1994.
[11] Wang Z, Lu Y. Numerical analysis on dynamic deformation mechanism
of soils under blast loading. Soil Dyn Earthq Eng 2003;23:70524.
[12] Wang Z, Hao H, Lu Y. The three-phase soil model for simulating
stress wave propagation due to blast loading. Int J Numer Anal
Methods Geomech 2004;28:3356.
[13] Henrych J. The dynamics of explosions and its use. Amsterdam:
Elsevier; 1979.
[14] Prapahanran S, Chameau JL, Holtz RD. Effect of strain rate on
undrained strength derived from pressuremeter tests. Geotechnique
1989;39(4):61524.
[15] Drucker DC, Prager W. Soil mechanics and plastic analysis or limit
design. Q Appl Math 1952;10(2):15764.
[16] Riedel W, Thoma K, Hiermaier S. Numerical analysis using a new
macroscopic concrete model for hydrocodes. In: Proceedings of ninth
international symposium on interaction of the effects of munitions
with structures, p. 31522.
[17] Herrmann W. Constitutive equation for the dynamic compaction of
ductile porous materials. J Appl Phys 1969;40(6):24909.
[18] Carrol MM, Holt AC. Static and dynamic pore collapse relations for
ductile porous materials. J Appl Phys 1972;43(4):1626 et seq.
[19] Lee EL, Hornig HC, Kury JW. Adiabatic expansion of high explosive
detonation products, UCRL-50422.: Lawrence Radiation Laboratory,
University of California; 1968.
[20] Huck PJ, Saxena SK. Response of soil-concrete interface at high
pressure. In: Proceedings of the Tenth International Conference on
Soil Mechanics and Foundation Engineering, June 1519, Stockholm:
A.A. Balkema; 1981, 2: 141144.
[21] Mueller CM. Shear friction tests support program; laboratory friction
test results for WES ume sand against steel and grout: Report 3.
USAE WES, Technical Report, SL-86-20; 1986.
[22] AUTODYN Theory Manual, revision 3.0. Century Dynamics, San
Ramon, CA; 1997.
Y. Lu et al. / Soil Dynamics and Earthquake Engineering 25 (2005) 275288 288