Sie sind auf Seite 1von 14

Computers and Structures 87 (2009) 507520

Contents lists available at ScienceDirect

Computers and Structures


journal homepage: www.elsevier.com/locate/compstruc

Response of structures to planar blast loads A nite element engineering approach


T. Brvik a,b,*, A.G. Hanssen a,c, M. Langseth a, L. Olovsson a,d
a

Structural Impact Laboratory (SIMLab), Centre for Research-based Innovation (CRI) and Department of Structural Engineering, Norwegian University of Science and Technology,
Rich. Birkelands vei 1A, NO-7491 Trondheim, Norway
Norwegian Defence Estates Agency, Research and Development Department, PB 405, Sentrum, NO-0103 Oslo, Norway
c
IMPETUS Afea AS, Strandgaten 32, NO-4400 Flekkefjord, Norway
d
IMPETUS Afea AB, Frrdsvgen 18, SE-141 46 Huddinge, Sweden
b

a r t i c l e

i n f o

Article history:
Received 14 March 2008
Accepted 20 February 2009
Available online 29 March 2009
Keywords:
Blast load
Experimental results
Structural response
Numerical simulations
LS-DYNA
Uncoupled and coupled EulerianLagrangian

a b s t r a c t
Design and validation of structures against blast loads are important for modern society in order to protect and secure its citizen. Since it is a challenge to validate and optimise protective structures against
blast loads using full-scale experimental tests, we have to turn our attention towards advanced numerical
tools like the nite element method. Several different nite element techniques can be used to describe
the response of structures due to blast loads. Some of these are: (1) a pure Lagrangian formulation, (2) an
initial Eulerian simulation (to determine the load) followed by a Lagrangian simulation (for the structural
response) and (3) a hybrid technique that combines the advantages of Eulerian and Lagrangian methods
to have a full coupling between the blast waves and the deformation of the structure. Ideally, all blast
simulations should be carried out using the fully coupled EulerianLagrangian approach, but this may
not be practical as the computational time increases considerably when going from a pure Lagrangian
to a fully coupled EulerianLagrangian simulation. A major goal in this study is to investigate if a pure
Lagrangian formulation can be applied to determine the structural response in a specied blast load problem or if more advanced approaches such as the fully coupled EulerianLagrangian approach is required
for reliable results. This is done by conducting numerical simulations of an unprotected 20 ft ISO container exposed to a blast load of 4000 kg TNT at 120 m standoff distance using the three different
approaches presented above. To validate and discuss the results, the simulated response of the container
is compared to available data from a full-scale blast test under such conditions.
2009 Elsevier Ltd. All rights reserved.

1. Introduction
Design and validation of various structures against ballistic impacts and blast loads are important for modern society in order to
protect and secure its citizen. While most ballistic impact scenarios
by KE-projectiles are regarded as complex material problems,
where full-scale component tests can rather easily be accomplished, blast loads are regarded as structural problems. Since it
is difcult, expensive and often impossible to validate and optimise
protective structures or vital infrastructure against blast loads
using full-scale experimental tests, we have to turn our attention
towards advanced numerical tools like the nite element method
(FEM).
Historically, blast load and structural response have been treated separately. Calculations of the blast load propagation and diffraction around structures have normally been carried out using

* Corresponding author. Address: Structural Impact Laboratory (SIMLab), Centre


for Research-based Innovation (CRI) and Department of Structural Engineering,
Norwegian University of Science and Technology, Rich. Birkelands vei 1A, NO-7491
Trondheim, Norway. Tel.: +47 73 59 46 47; fax: +47 73 59 47 01.
E-mail address: tore.borvik@ntnu.no (T. Brvik).
0045-7949/$ - see front matter 2009 Elsevier Ltd. All rights reserved.
doi:10.1016/j.compstruc.2009.02.005

an Eulerian method, e.g. [13], where it is assumed that the blast


loaded structure behaves as a rigid body. The obtained loading
has then often been used together with some simplied analytical
models [4,5] or design manuals [6,7] to calculate the structural response. On the contrary, the structural response due to blast loads
may be simulated using a Lagrangian nite element formulation,
e.g. [810], often by use of a simplied blast load description
(based on ConWep [11] or similar). However, during blast loading
motions of structures and uids are directly coupled due to an
interaction between the deformation of the solid and the motion
of the uid, and they should in fact be considered simultaneously.
Fluid mechanics problems cannot be treated effectively with
Lagrangian meshes due to severe mesh distortions, and hybrid
techniques combining the advantages of Euler and Lagrangian formulation have been developed [12]. Such methods, using an arbitrary LagrangianEulerian (ALE) formulation and a uidstructure
interaction (FSI) contact interface, have during the last decades
been implemented in commercial software like LS-DYNA [13],
ABAQUS [14], AUTODYN [15] and EUROPLEXUS [16]. So far these
methods have not been fully adopted for industrial application
and further development and in particular validation are needed
(see also [1722]). More recently, meshless methods (to reduce

508

T. Brvik et al. / Computers and Structures 87 (2009) 507520

Nomenclature
V0
V

vr
m
c
p0
p
T
T0
Tr

initial volume
volume
V/V0
particle velocity
speed of sound
initial pressure
pressure
temperature
initial temperature
T/T0

mesh distortions and numerical advection errors) based on


Lagrangian formulations have been adopted to describe the propagation of the blast load [23,24].
Most blast load problems can roughly be divided into four major phases, i.e. (1) the detonation phase where the exploding
source produces gases with high pressures and temperatures that
propagate outwards by generating pressure waves which interact
with the surrounding medium, (2) the propagation phase where
high intensity blast waves propagate a certain distance in the surrounding medium, from the source towards the target, (3) the
interaction phase between the propagating waves and the target,
and (4) the response of the target due to the blast waves. Even
though the four phases have been separated above there are direct
couplings between them, and they are all of importance to fully describe the response of structures exposed to blast loads. In the
codes presented above, it is possible to simulate all processes of
the blast load as long as a TNT equivalent of the detonating source
is known (see [1921]). It is still an open question if it is necessary
to model all blast phases in all types of problems, or if some problem can be simplied by modelling only a few of them. In this paper, the detonation phase itself will not be modelled since the blast
load is caused by a plane wave generated from compressed air.
Several different methods can be used to describe blast loads
against structures by FEM. In LS-DYNA [13], at least three different
approaches are possible. These are: (1) A pure Lagrangian formulation. Here the blast load is approximated as an idealized pressure
time curve applied directly to the segments on a given surface
(often modelled with shell elements). During a simulation, the
pressure vector always stays normal to the surface of the shell,
independent of the structural deformation. If the deformations
are moderate the pure Lagrangian approach may give reasonable
results. (2) The second alternative involves running an Eulerian
simulation prior to a Lagrangian simulation. This method can be
used to nd the pressuretime load on all faces surrounding a
structure without taking the full coupling effects into account.
Thus, the method will only give reliable results if the interaction
effects between the load and deforming structure are small (i.e.
essentially no or very small deections of the structure during
blast loading). The advantageous is that the computational time
is considerable reduced compared to a fully coupled simulation.
The procedure is in principle carried out in three simulation steps:
Based on a given blast wave approaching a structure (based on, e.g.
ConWep [11] calculations or FE simulations of the blast process)
the inow properties required in a LS-DYNA simulation are determined. Then a full 3D Eulerian ow model is established and run,
where the structure to be modelled is assumed rigid. From this
simulation, pressuretime data from a number of points on the
surfaces of the structure is registered and mapped. The mapped
data is nally used as pressuretime curves in a pure Lagrangian
simulation to determine the response of the deforming structure.
By doing this, the initial loading onto the whole structure will be

q
q0
CV
CP

c
E
W
e
L

density of air
initial density of air (atmospheric pressure 1 bar)
specic heat capacity for constant volume
specic heat capacity for constant pressure
CP/CV = 1.4 (air)
total internal energy of air
work
specic internal energy, E/V
length of Euler domain in blast wave propagation direction

well described, but as the structure deforms the accuracy of the


simulation is reduced. (3) The Eulerian algorithms may be applied
to have a full coupling between the blast waves and the deformation of the structure. The multi-material Eulerian or ALE capabilities in LS-DYNA make it possible to follow large ows of various
materials without entering into numerical distortion problems
often experienced in Lagrangian formulations. The drawback is
numerical advective (averaging) errors as material uxes between
elements, which means that Eulerian techniques in general require
a ner mesh discretisation than Lagrangian simulations. LS-DYNA
also has an FSI interface, which is the contact formulation between
Eulerian and Lagrangian parts.
Ideally, all blast simulations should be carried out using the
coupled formulation, but this is not practical as the computational
time increases considerably when going from a pure Lagrangian to
a fully coupled EulerianLagrangian simulation. Therefore, a goal
in this study is to investigate if a pure Lagrangian formulation
can be applied in blast loading problems of exible structures or
if more advanced approaches like the coupled EulerianLagrangian
formulation is required for reliable results. This is done by conducting numerical simulations of an unprotected 20 ft ISO container exposed to a blast load of 4000 kg TNT at 120 m standoff
distance using the three approaches presented above. To validate
and discuss the results, the simulated response of the container
is compared to data from a full-scale blast test under such conditions recently presented in Brvik et al. [25,26].

2. Experimental results
Brvik et al. [25,26] presented a new, cost-effective and lightweight protection concept for a 20 ft ISO container to be used in
international operations. The basic idea was to use extruded aluminium panels, which gives a rather cheap, low-weight solution.
These panels may easily and quickly be xed to a container via a
beam system. To increase the ballistic and blast resistance, a local
mass was lled in the empty cavities of the panels on spot. When
required, the mass is just emptied through a hatch beneath the
panels, and the system retains its low mass. Based on a large number of ballistic impact tests, numerical simulations (using a pure
Lagrangian formulation) of the blast load response and full-scale
validation tests, it was concluded that the proposed add-on protection was able to protect a standard 20 ft ISO container against a
number of specied threats.
Two full-scale blast tests involving the container were also carried out in the study. In the rst test, an unprotected container was
tested as a reference. The purpose of this test was to reveal the
amount of damage in an unprotected container exposed to the design blast load. Then a protected container was tested using the
same set-up and loading as in the rst test to demonstrate the effect of the added protection. Here, only results from the test on the

509

T. Brvik et al. / Computers and Structures 87 (2009) 507520

unprotected container will be applied in the validation of the simulated response using various FEM-based methods. In the following, some main results will be repeated for completeness. It is
referred to Brvik et al. [26] for a full description of all the experimental details.
The blast tests were performed in the Large Blast Simulator
(LBS) at WTD 52 in Oberjettenberg, Germany, which is the biggest
shock tube in Europe. The driver section of the LBS consists of 100
steel bottles, each with a volume of 384 l, which can be lled with
air to a maximum pressure of 200 bars. The air inside the bottles
can be released simultaneously by blasting steel diaphragms at
the front end. The air then expands and forms a plane blast wave
which hits the test object in the test section situated 40100 m
in front of the driver section. As already stated, the design blast
load corresponds to 4000 kg of TNT detonating at a standoff distance of 120 m. ConWep [11] was used to estimate the pressure
time proles shown in Fig. 1 (both for a wall with dimensions as
for the container and a hemispherical surface burst) and to generate the blast load data given in Tables 1 and 2. Based on these data
the LBS was calibrated by conducting fully instrumented tests
against a rigid concrete wall of 3  3 m2 to give the predicted pressuretime curve. The unprotected container was then aligned in
position 95 m from the blast source with its long side facing the
load, xed to the concrete oor (Fig. 2a) and nally instrumented.
The instrumentation consisted of gauges used to measure reected, side-on and internal pressures at various locations
(Fig. 2b), one deection laser gauge and three different high-speed
cameras used for visualization of the blast load response.
After having lled the steel bottles in the LBS with the calibrated design pressure, the air inside the bottles was released.
The air ow expanded and formed a plane blast wave which propagated towards the container. Due to failure in some gauges during
the test, no data is available for the reected pressure versus time
on the front wall (Pfront__) and for the side-on pressure versus
time on the rear side wall (Prear__). However, it is anticipated
that the incipient blast wave in this test was similar to the corresponding test of the protected container [26]. Fig. 3a gives pressuretime curves in kPa for the three operating pressure gauges,
while Fig. 3b gives the deectiontime curve from the laser deection gauge (S__). The laser gauge saturated about 380 ms after
ring, indicating that the deection was above its range of
400 mm. However, the total deection of the wall was estimated
to 485 mm (and for sure within the range 400500 mm) based

Table 1
Reference data from blast calculations against a rigid wall by ConWep [11].
ConWep calculations: container wall size 6.0  2.5 m2
INPUT:
Charge weight: 4000 kg TNT
Horizontal range: 120.00 m

Table 2
Incident data from hemispherical surface burst calculations by ConWep [11] and used
for calibration of inow properties in Eulerian analysis.
ConWep calculations: hemispherical surface burst
INPUT:
Charge weight: 4000 kg TNT
Horizontal range: 120.00 m

Idealised for finite rigid wall (ConWep, Table 1)


40
Exponential decaying for finite
rigid wall (ConWep, Table 1)

Pressure (kPa)

Pressure (kPa)

50
ConWep for hemispherical surface burst (Table 2)
40

30

3D Eulerian simulation for finite


rigid wall (Section 3.3.2)

20
10

20
10
0

-10

-10
0.02

0.04
Time (s)

0.06

0.08

Reflected pressure, one-dimensional model


(Section 3.3.1)

30

OUTPUT:
Peak incident overpressure: 22.13 kPa
Normally reected overpressure: 48.25 kPa
Time of arrival: 237.9 ms
Positive phase duration: 69.41 ms
Incident impulse: 643.2 kPa ms
Reected impulse: 1270 kPa ms
Shock front velocity: 370.5 m/s
Peak dynamic pressure: 1.668 kPa
Peak particle velocity: 48.63 m/s
Shock density: 1.41 kg/m3
Specic heat ratio: 1.4

on post-test measurements/assumptions and simplied calculations, while the permanent deection was measured to be about
400 mm [26]. These curves were used to generate the data collected in Table 3 about duration, peak pressure and impulse of both
the positive and negative phase of the blast load at different locations. Note that data from the test on a protected container is also
given in parenthesis in Table 3 for comparison.
Fig. 4 shows some pictures of the interaction between the blast
wave and the unprotected container from the high-speed camera
located inside the tunnel. From the pictures it is seen that the front
wall rst deforms globally by the planar blast wave. Then the
deformation changes into localised buckling, causing a plastic

50

OUTPUT:
Peak pressure: 47.36 kPa at X = 0 m and Y = 0 m
Minimum pressure: 47.25 kPa at X = 6 m and
Y = 2.5 m
Total impulse: 11660 N s
Average impulse: 777.4 kPa ms
Equivalent uniform peak pressure: 47.34 kPa
Equivalent uniform impulse: 791.5 kPa ms
Decay coefcient of equivalent uniform pressure
a: 0.05981

0.02

0.04

0.06

0.08

Time (s)

Fig. 1. (a) Idealized and exponential decaying reected pressuretime prole for a blast of 4000 kg TNT at 120 m standoff distance based on ConWep for a nite rigid wall
(Table 1) compared to the average pressure to the front face wall from a 3D Eulerian simulation and (b) comparison of reected pressuretime curves from ConWep for a
hemispherical surface burst (Table 2) and the one-dimensional uid-ow Eulerian simulation.

510

T. Brvik et al. / Computers and Structures 87 (2009) 507520

Fig. 2. (a) Unprotected container mounted inside the LBS before the blast and (b) position of the various pressure and laser gauges used in the test [26].

40
Proof__
Pside__
Pinside
Pfront__
Prear__

Pressure [kPa]

30
20
10
0
-10

0
S__

-50

Deflection [mm]

-20

In this area the laser gauge was outside its working area of 400 mm.
Based on post-test analysis/measurements the maximum deflection
was estimated to 485 mm.

-100
-150
-200
-250
-300
-350

-30

-400
350

400

450

500

550

600

650

350

400

450

Time [ms]

500

550

600

650

Time [ms]

Fig. 3. (a) Pressuretime curves from various pressure gauges and (b) measured deectiontime curve from the laser gauge obtained during the blast test on an unprotected
container [26].

Table 3
Data obtained from the pressure gauges for unprotected container (data for protected container in parenthesis for comparison) [26].
Gauge

Duration pos. phase


(ms)

Max peak pressure


(kPa)

Impulse pos. phase


(kPa ms)

Duration neg. phase


(ms)

Min peak pressure


(kPa)

Impulse neg. phase


(kPa ms)

Pfront__
Pfront_1=4 _
Pside__
Proof__
Prear__
Pinside

(85.0)
(86.0)
64 (65)
65 (67)
(68)
81 (86)

(37.4)
(36.7)
23.4 (23.8)
37.7 (24.0)
(23.8)
22.7 (15.9)

(1204)
(1159)
691 (681)
1156 (713)
(522)
710 (536)

()
()
236 (235)
>235 (164)
(113)
>214

(6.2)
(5.9)
20.4 (19.5)
29.4 (19.2)
(8.6)
8.8 (10.3)

(93)
(129)
1733 (2010)
>2118 (1193)
(531)
486 (>1126)

collapse of the front wall. After that the roof goes down and the
doors go in and out, before the roof rebounds and collapses. The
maximum permanent deformation of the front wall was located
about 1.5 m above the ground and in the centre of the wall as illustrated in Fig. 5a. The upper horizontal beam on the front side
received a plastic deformation downwards of about 110 mm and
inwards of 60 mm. The upper beam on the rear side received a
plastic, horizontal deformation of 40 mm inwards. Fig. 5b shows
a picture of the inside of the container after the blast. The roof
had a permanent plastic deformation upwards of about 300 mm.
Fracture was not observed in any parts of the container after the
test, even though some parts were considerably plastically deformed. In any case, one main conclusion from this test was that
an unprotected container will be severely damaged by the prescribed blast load, and will not operate satisfactory afterwards. It

should nally be mentioned that when the container was protected with the proposed aluminium panels, no such damage
was observed in the front wall after the blast [26].

3. Numerical modelling approaches


Three numerical models with increasing level of complexity and
expected accuracy concerning the blast load description are investigated. The aim is to generate a reference basis which can be used
for engineering design of blast loaded structures. The Lagrangian
model of the 20 ft ISO container, that forms the basis for all simulations, is described in Section 3.1. All simulations are carried out
using the explicit solver of the non-linear nite element code LSDYNA [13].

T. Brvik et al. / Computers and Structures 87 (2009) 507520

511

Fig. 4. Pictures from the high-speed camera located inside the tunnel showing the interaction between the blast wave and the unprotected container [26].

Fig. 5. (a) Front side wall and (b) inside of unprotected container after the blast [26].

The rst model for blast loading description is a pure Lagrangian approach, Section 3.2, where we apply a predened pressure
time load from ConWep [11] calculations to all nodes of the frontal
(blast loaded) face of the container. In the second approach we use
an uncoupled EulerianLagrangian approach. This is described in
Section 3.3. Here, a 3D computational uid dynamics (CFD) model
provides pressuretime curves for all faces of a rigid container.
These curves are then mapped and used to load the container on
all faces in a subsequent Lagrangian simulation, in the same way
as in Section 3.2 for only the front face. In Section 3.4 a fully coupled EulerianLagrangian simulation is carried out, i.e. the blast
loading is transferred from the air ow to the structure by a
uidstructure interface (FSI).
To describe the loading in this study, ConWep was used to calculate the equivalent uniform peak pressure and impulse from a
charge weight of 4000 kg TNT detonating at a standoff distance
of 120 m towards a nite rigid (or close to rigid) wall with dimensions equal to the front face of the container (Table 1). In a similar
way as in Brvik et al. [25], we idealize the loadtime curve from
ConWep calculations to a linear relationship for the pure Lagrangian analyses in Section 3.2, see Fig. 1a. This approach is reasonable
if the loading is completed before the maximum deformation of
the container is reached, an assumption often used in design. In

Fig. 1a the idealized loadtime curve is compared to the exponential decreasing curve using the decay coefcient stated in Table 1.
ConWep was also used to provide incident and reected impulses,
peak pressure, particle velocity and duration of loading as listed in
Table 2 for a hemispherical surface burst against an innite rigid
surface. The reected loadtime curve from this calculation is plotted in Fig. 1b. Such information is required to calibrate the inow
properties in an EulerianLagrangian simulation. It should at this
point be mentioned that the estimated impulse by ConWep against
the container is underestimated, while the peak pressure is somewhat overpredicted compared to the experimental data. This will
be further discussed in Section 4.
For the simulations in Sections 3.2 and 3.3 we also study the effect of internal air-pressure in the container. The inner pressure
was described by a control volume approach in the Lagrangian
and uncoupled EulerianLagrangian models. The pressure buildup was then estimated as for an ideal gas with c = 1.4. For the fully
coupled simulation in Section 3.4, the inner air-pressure build-up
is automatically present in the simulation due to the FSI. However,
in this model we studied the effect of inner pressure by introducing
venting from the container. Table 4 shows the simulation programme with cross-reference to the section where the models
are presented.

512

T. Brvik et al. / Computers and Structures 87 (2009) 507520

Table 4
Simulation program.
Run

Description

Sections

1a
1b
2
3
4
5
6

Lagrangian, no inner pressure, linear decay


Lagrangian, no inner pressure, exponential decay
Lagrangian, with inner pressure, linear decay
Uncoupled EulerianLagrangian, no inner pressure
Uncoupled EulerianLagrangian, with inner pressure
Coupled EulerianLagrangian
Coupled EulerianLagrangian with venting

3.2
3.2
3.2
3.3
3.3
3.4
3.4

3.1. Container model


The geometrical model of the standard 20 ft ISO container (with
external dimensions of about L  B  H = 6  2.5  2.5 m3) has earlier been described in some detail by Brvik et al. [25], and it follows in large part the geometrical specications for a real
container. The model consists of the main faces (4 walls + roof)
made of corrugated steel panels. The corrugated panels are connected to a top rectangular frame made of hollow steel sections.
The oor of the container is made up of two main bottom rails with
18 transverse beams and a thin steel plate on top. The edges of the
container are given by four columns which are connected to the
corrugated panels and the oor.
All parts are modelled by the Belytschko-Tsay shell element
type in LS-DYNA [13] with two integration points through the
thickness of the shell. A tied-contact interface was used to connect
the different parts. This means that there is no nodal sharing between the different part numbers. According to the technical specications of the container, all parts are made from anti-corrosive
steels with a minimum yield stress of 355 MPa and a tensile
strength of 500 MPa (i.e. modest strain hardening). The material
behaviour during blast loading was modelled as elastic-plastic

with isotropic strain hardening (Material Type 3 in LS-DYNA) with


density q 7850 kg=m3 , Youngs modulus E = 210,000 MPa, Poissons ratio m = 0.3, yield stress r0:2 355 MPa and a tangential
hardening Et 1000 MPa [25]. The possible effect of strain rate
hardening of the material was neglected, since simulations showed
that the strain rate during blast loading was in general small and
below 100 s1 . This is also a conservative estimate, since no additional target strength giving a reduced container deection is
added due to the increased strain rate.
For all numerical studies we apply mass damping at the end of
the simulation to obtain the nal, deformed shape of the container.
The element mesh density used in the container model was higher
than that used by Brvik et al. [25], see also Section 3.5.1. They
used an element side length of approximately 25 mm, while the
current study is based on an element side length of 10 mm. This
was done to better describe localised buckling of the corrugated
plates. Using this mesh size, the container model consisted of
approximately 620,000 shell elements. Plots of the numerical model of the container to illustrate the mesh density are given in Fig. 6.
3.2. Lagrangian approach
All nodes in the numerical model of the front face of the container, see Fig. 7a, are given the idealized ConWep pressuretime
curve specied by Fig. 1. Both linearly (Run 1a, Table 4) and exponentially decaying loads (Run1b, Table 4) are investigated for the
case of no inner pressure. The displacement-time curves of Runs
12 (without and with inner container pressure see Table 4)
are given in Fig. 15. Note that a 2025% decrease in deection is
obtained if the air inside the container is modelled. Also, note that
there is a signicant difference between Run 1a and Run 1b even if
the total impulse is the same, i.e. the shape of the pressuretime
curve plays a signicant role for the displacementtime curve in

Fig. 6. (a) Numerical model of container and (b) corner detail after blast simulation (Run 3) to show the mesh density and localised buckling of the corrugated plates.

Fig. 7. Pure Lagrangian approach; (a) nodal loads on front face only and (b) nal shape of the container after loading.

513

T. Brvik et al. / Computers and Structures 87 (2009) 507520

these simulations. This is due to that the loading must be considered as dynamic and not impulsive. If the simulated nal shape
of the front wall of the container facing the blast (Fig. 7b) is compared to a picture taken after the corresponding test (Fig. 5a), the
qualitative agreement using this approach (linearly decaying load)
is found to be good. However, since only the Lagrangian solver was
used, the buckling of the roof was not obtained (as also discussed
in [26]). This simulation (up to 0.1 s) typically took 5 CPU-hours
running on a single AMD dual-core 2.2 GHz processor on a Linux
cluster.

where c is the speed of sound. For the full Eulerian reference frame
used in Step 2 we must take into account reections from the sides
and rear side, hence the dimensions as given by Fig. 8. L = 13 m was
used for the current simulations.
For the inow side we know (from ConWep) the pressuretime
history given by p(t). For LS-DYNA we need to specify the internal
variables (relative volume vr and relative temperature Tr) of the inow which corresponds to this pressuretime history. Assuming
adiabatic conditions this can be done by using the following formulas (see Appendix A for details)

3.3. Uncoupled EulerianLagrangian approach

v r t

The uncoupled EulerianLagrangian approach is based on three


different simulation steps. The required input is incident blast
loading characteristic from ConWep [11] or similar, see Table 2.
Note that the following description of the general nite element
aspects of the uid-ow model is the same as for the fully coupled
model that will be presented in Section 3.4.

Here V0 is initial volume, V is current volume, p0 is initial pressure, p


is current pressure, T0 is initial temperature, T is current temperature, and c is the ratio of specic heat capacities.
The nodal velocitytime curve v(t) of the inow part must now
be prescribed. We have done this by trial and error (by running
simulations iteratively) with the incident particle velocity of Table
2 as a known value. The correct velocitytime relationship is the
one that gives a pressure pulse that does not distort as it propagates towards the container, see sketch of Fig. 9. We base the initial, peak particle velocity on the ConWep estimate in Table 2.
Fig. 10 shows normalised volume and temperature as function
of time as well as the prescribed nodal velocity of the calibrated
model. This ensures a proper non-distorted pressuretime prole,
see data from LS-DYNA simulation in Fig. 11 (note the arrival of the
reected wave in Element 140 this is after the relevant part of
loading of 0.1 s). The air was then represented by the material
*
MAT_NULL in LS-DYNA using an initial density of 1.2 kg/m3 at
1 bar = 100 kPa initial pressure. The ideal gas law was represented
by *EOS_IDEAL_GAS using CV = 719 J/kg K, CP = 1007 J/kg K and an
initial temperature of 293 K. The reected pressuretime curve
from the reaction end based on this simulation is plotted in
Fig. 1b, where it is compared to the pressuretime prole from

3.3.1. Step 1 one-dimensional uid-ow analysis


We need to calibrate the numerical model of the air ow so that
the incoming pressuretime history matches that of the ConWep
data in Table 2. In principle, this means that the boundary conditions on the inow side of our Eulerian domain must be chosen
to match the desired incident pressuretime history. The timedependent parameters in LS-DYNA for the inow boundary are
two internal variables controlling the gas behaviour, which in
our case are the relative volume vr and the relative temperature
Tr, as well as the particle velocity v.
The inow properties were calibrated by considering a onedimensional representation of the gas (one string of brick elements), see Figs. 8 and 9, which is divided into one inow part
and one part representing the main volume. The length L is approximated by the time of duration of the blast loading Dt, i.e. L c Dt


1=c
Vt
p0

;
V0
pt

T r t

Tt

T0

V0
Vt

Full Eulerian reference frame


Gas inflow with
relevant boundary
conditions

Container
One-dimensional Eulerian reference frame

L/2

Fig. 8. Calibration of inow properties to match ConWep data.

Element 140

Reaction end

Inflow end: Element 190

5m

Direction of propagation

p (t ), v(t )

p+(t-t*) = p+(t)

p+(t)

Incoming impulse:

t
t*

Fig. 9. Undistorted incoming pressure pulse. The element reference is used in Fig. 11.

c1
1

514

T. Brvik et al. / Computers and Structures 87 (2009) 507520

1.1

50

Tr

40

20

0.9

vr

v
(m/s)

T r , vr

30

10

0.8

0
0

0.02

0.04

0.06

0.08

Fig. 12. 3D Eulerian model for uid-ow analysis.

0.1

Time (s)
Fig. 10. Inow properties: normalised volume and temperature and plot of
absolute nodal velocity.

130

E190
E140

Pressure (kPa)

120

110

100
Fig. 13. Sampling of pressure on container surface during 3D Euler simulation.

90

80
0

0.04

0.08

0.12

0.16

Time (s)
Fig. 11. Pressuretime at inow Element 190 compared to Element 140 (5 m
downstream).

ConWep for a hemispherical surface burst. The agreement is as


seen excellent and indicates that our use of the inow data has
been reasonable.
3.3.2. Step 2 full 3D Eulerian phase
In this step we generate a 3D Eulerian model of the air around a
rigid container volume, see Fig. 12. The model is divided into two
parts, one small inow part and one part comprising the main volume, as was done for the one-dimensional model described in Section 3.3.1. The 1D ow model was a string of 200 solid elements in
the direction of wave propagation, while the 3D model contains
nearly 2 million solid elements. On this model we apply exactly
the same inow conditions as for the 1D ow model (Step 1). On
each face of the surface of the container volume we sample pressuretime data for 8  8 evenly spaced points, see Fig. 13. The container part of the model in Fig. 12 is simply a restriction on the
uid-ow domain where we apply boundary conditions to prevent
uid ow in the normal direction of the container. The computational time for Step 2 was about 0:15 CPU-hours on a single
AMD dual-core 2.2 GHz processor.
3.3.3. Step 3 Lagrangian analysis
For the nal Lagrangian analysis in Step 3 it is pragmatic to use
pressure loading by parts, i.e. we average all the sampled pressure

time curves so that we get one pressuretime history for each of


the main container faces, namely the front, rear, top and end sides
of the container (Fig. 14). The gure also shows the resulting average curves for use with the structural analysis based on the outcome of the 3D Eulerian simulation in Step 2. The average
pressure curve on the front face of the nite rigid wall is compared
to similar curves from ConWep calculations in Fig. 1a. It is again
seen that the agreement between curves from ConWep and the
curve from the full 3D Eulerian simulation is good. Obtained
displacement-time curves of Runs 34 (both without and with
inner container pressure see Table 4) are given in Fig. 15. In a
similar way as for the Lagrange simulation in Section 3.2, this
simulation (up to 0.1 s) typically took 5 CPU-hours running on a
single AMD dual-core 2.2 GHz processor on a Linux cluster.
3.4. Coupled EulerianLagrangian approach
The full 3D Eulerian reference frame was divided into nearly 2
million elements. Inow properties as indicated in Fig. 8 are as described in Step 1, Section 3.3.1. The penalty based uidstructure
interaction algorithm *CONSTRAINED_LAGRANGE_IN_SOLID was
used to model contact between the blast load and the deforming
container.
The obtained displacement-time curve of Run 5 (see Table 4) is
given in Fig. 15. In Run 6 we introduced venting from the container.
The results from Run 5 and Run 6 were practically equal up to the
initial peak deformation, and Run 6 is therefore not included in
Fig. 15. The venting was modelled by removing the FSI from the
rear corrugated panel of the container. These simulations (up to
0.1 s) took about 70 CPU-hours on 8 AMD 2.2 GHz processors on
a Linux cluster. Compared with the pure Lagrangian simulation,
the computational time for the fully coupled simulations was increased by a factor of more than 100.

515

T. Brvik et al. / Computers and Structures 87 (2009) 507520

60
ConWep

Pressure (kPa)

40

Front
Rear
Side

20

Top

-20
0

0.01

0.02

0.03

0.04

0.05

Time (s)
Fig. 14. Average pressure curves applied to main faces of container compared to triangular pressure pulse from ConWep.

Uncoupled Euler Lagrange, no inner pressure

0.6

Fully coupled Euler Lagrange

Uncoupled Euler Lagrange, with inner pressure

-0.2

Fully coupled Euler Lagrange (double impulse)


Measured data

Estimated data (dotted curve)


Lagrangian, no inner pressure (exponential decay)

-0.4

Lagrangian, with inner pressure (linear decay)

-0.6
Lagrangian, no inner pressure (linear decay)

-0.8
0

0.1

0.2
Time (s)

0.3

0.4

Fig. 15. Displacement of the centre of the front wall of the container as function of
time. Note that the dotted part of the measured data curve is estimated since the
laser gauge saturated in this area during the test.

Max displacement front wall center (m)

Max displacement front wall center (m)

0.2

Only front face refined

0.4

Complete model refined (10mm elements)

0.2

0
0

10000

20000

30000

40000

50000

Mesh density (Elements/m2)


Fig. 16. Mesh density convergence study.

3.5. Sensitivity studies


3.5.1. Mesh sensitivity
The simulation results reported in Sections 3.23.4 were carried
out using an element side length of 10 mm for all corrugated parts
of the container. Fig. 16 shows a mesh sensitivity study on Run 3
(uncoupled EulerLagrange, no inner pressure), using element side
lengths of 25 mm, 13 mm, 10 mm and 5 mm. Only the mesh density of the front panel was varied in these simulations, while all
other parts had a constant element side length of 25 mm. A result
using the FE model with an element side length equal to 10 mm for
all panel parts is also shown for comparison. In practice, the results
seem to have converged for the 10 mm element side length. A plot
of the container model both before and after blast loading revealing the mesh density is given in Fig. 6. The deformation pattern
at the centre of the front panel is as seen local buckling. This means
that some small variations in the maximum displacement as function of mesh density must be expected due to the mesh dependency of the exact location of the buckling lobes.
3.5.2. Blast wall response one-dimensional study
It is evident from Fig. 15 that the fully coupled Eulerian
Lagrangian approach gives considerable less structural response

compared to the uncoupled EulerianLagrangian simulation and


in particular to the pure Lagrangian solution. For the uncoupled approach we have obtained pressuretime curves from Step 2, Section 3.3.2, using a completely rigid container volume. This means
that the reected pressure relates to a blast wave reecting from
a stationary, non-moving container surface. In the fully coupled
simulation, the reected pressure will be inuenced by the movement of the deformable container surface. We have done a study
using the one-dimensional model presented in Section 3.3.1 to
illustrate the inuence of exible blast loaded surfaces on the magnitude of the reected impulse. On this simple model we varied the
velocity conditions on the reaction end, which is opposite of the inow end (see Fig. 9). We conducted 4 simulations using reaction
end velocities of 0, 1, 5 and 10 m/s in the blast loading inow direction. As seen from Fig. 17, the reaction end velocity has a major
inuence on the reected impulse. From the fully coupled simulation we have an average front wall velocity of at least 10 m/s until
maximum response is gained. In the one-dimensional model this
reduces the reected impulse by 23%. This is a clear indication
for why the fully coupled EulerianLagrangian simulation gives
much less maximum response compared to the other modelling
approaches.

516

T. Brvik et al. / Computers and Structures 87 (2009) 507520

1.6

Pressure impulse (kPas)

0 m/s
1.2

1 m/s
5 m/s
10 m/s

0.8

0.4

0
0

0.04

0.08

0.12

0.16

Time (s)

3.6. Discussion of numerical models

Fig. 17. Effect on reaction wall velocity on reected impulse (one-dimensional


study).

Table 5
Maximum response from stiffer container.

Model
Lagrangian, linear decay
Uncoupled EulerianLagrangian
Coupled EulerianLagrangian

3.5.5. Effect of model complexity on the CPU-time


It has been stated that the computational time increases by a
factor of more than 100 when going from a pure Lagrangian approach to a fully coupled EulerianLagrangian approach. For the
uncoupled EulerianLagrangian approach the CPU-time is similar
to the Lagrangian approach, but several modelling and simulation
steps are now required. It should nally be noticed that the method used to determine the inow properties to LS-DYNA as described in Section 3.3.1 also are required for the fully coupled
approach. Thus, it is quite clear that both the uncoupled and in particular the fully coupled approaches are considerably more time
consuming than the pure Lagrangian approach. Another important
recognition is that it is hardly possible to run the fully coupled simulation on a standard computer (using a single processor), and that
massive parallel processing (MPP) using a cluster or a supercomputer is almost always required.

Original
container

Increased container
wall thickness

Normalised
max. displ.
1
0.65
0.22

Normalised
max. displ.
1
0.60
0.57

Normalised
impulse
1
0.65

3.5.3. Blast wall response stiffer container


Considering the results of the one-dimensional model in Section
3.5.2, it is natural to assume that a stiffer container will lead to less
difference between the three computational models since the
velocity of a stiff container wall during blast loading is reduced.
We have redone the simulations by doubling the wall thickness
of the main faces consisting of corrugated panels (from 2 mm to
4 mm). The new, normalised peak displacements from all models
are compared in Table 5. As seen, the coupled and the uncoupled
EulerianLagrangian analyses now give almost the same peak displacement. However, the pure Lagrangian approach (Run 1a) is still
giving larger deformations compared to the other methods. This is
mainly due to the differences in impulse loading depicted in
Fig. 14. The pure Lagrangian simulation reaches the maximum displacement at approximately time t = 0.020 s in Fig. 14, whereas the
uncoupled EulerianLagrangian model has the maximum for time
t = 0.017 s (i.e. before the loading is over so it must be considered
as dynamic and not impulsive). The corresponding ratio between
pressure impulses on the container wall at the time of maximum
displacement is 0.65. The ratio of impulses (0.65) closely matches
the ratio of maximum displacement of the front wall centre (0.60),
which is plausible since the material behaviour of the stiffer container walls is almost elastic.

We have shown that the uncoupled EulerianLagrangian approach and the fully coupled EulerianLagrangian model give the
same response for a container with increased stiffness, Section
3.5.3. Also, for a more exible container, the fully coupled EulerianLagrangian approach was found to give less maximum deformation compared to the uncoupled approach. The exercise on
the stiffened container therefore indicates that (1) the FSI functions
seems to work satisfactorily and (2) the structural exibility reduces the pressure impulse and thus reduces the nal magnitude
of the structural response in a fully coupled EulerianLagrangian
simulation. In other words, for exible structures an uncoupled
EulerianLagrangian approach (where the load is applied directly
to the nodes of the blast loaded surfaces) will give too large pressure loads. The exercises in Section 3.5.3 showed why the pure
Lagrangian approach gives larger structural response than the
other approaches. It also demonstrated that the Lagrangian and
uncoupled EulerianLagrangian elastic responses were similar
when comparing normalised maximum structural response with
corresponding normalised impulse loads.
This demonstrates that the difference in response between the
models is fully explainable and that it is the load specication for
rigid versus exible structures itself that mainly creates this difference (i.e. based directly on ConWep [11] or an initial 3D CFD analysis). By comparing Fig. 1 with Fig. 14, we see that ConWeps
triangular pressure loading used for the Lagrangian simulations
with linearly decaying load in fact is bound between an upper
curve, which is the reected pressure based on the one-dimensional CFD analysis (Fig. 1) and a corresponding lower pressure
curve based on the three dimensional CFD analysis (Fig. 14 front).
This observation appears to be reasonable and shows that it is not
correct to use this pressure load directly in a pure Lagrangian approach for exible structures, nor is it correct to apply pressure
loads found from a 3D uncoupled EulerianLagrangian simulation
of the uid ow against a rigid body to a subsequent structural
analysis on a deforming body.

4. Comparison between numerical and experimental results


3.5.4. Effect of multiple surface loading
A nal study was done by using the uncoupled Eulerian
Lagrangian model (i.e. Run 3 in Table 4), but removing the pressure
loading on the sides, rear and roof. That is, we only applied loading
to the front wall of the container (as for the pure Lagrangian simulation). The study was done using the original wall thickness of
the container. The results showed that by loading only the frontal
face reduced the peak displacements by 10%, i.e. from originally
505 mm to 456 mm.

All simulations so far are based on blast load data estimated


from ConWep calculations (see Fig. 1, Tables 1 and 2). In the following, results from these simulations will be compared to the experimental observations presented in Section 2 from the full-scale blast
test on an unprotected container. It should at this point be mentioned that instead of using inow properties from ConWep, the
explosion process itself could have been modelled (as done in, e.g.
[1820]). However, this has not been done in the present study

T. Brvik et al. / Computers and Structures 87 (2009) 507520

since ConWep also gave inow properties for the experimental


tests presented in [26].
Fig. 15 shows maximum displacementtime curves of the front
wall as predicted by the various FE approaches. As already seen,
considerable variations in predicted results between the applied
methods are present. The measured displacementtime curve
(Fig. 3b) is also included in the gure for comparison (where also
the estimate maximum displacement is shown as the dotted part
of the curve). It is found that the uncoupled EulerianLagrangian
approach without inner pressure in the container gives the best
t to the estimated maximum response, while the permanent displacement is far off (indicating outwards instead of inwards permanent deection). The pure Lagrangian approaches with inner
pressure (linearly decaying pressure) in the container and without
inner pressure (exponentially decaying pressure) both seem to describe the permanent displacement of the front wall quite well, but
they overestimates the maximum displacement. This approach
without inner pressure (linear decaying pressure) overestimates
the response the most. The latter methodology was used in [25]

517

with reasonable accuracy, but with a coarser element mesh (see


also the mesh sensitivity study in Fig. 16). Note further that the
Lagrangian simulations seem to give conservative results, while
non-conservative results are in general obtained using the other
approaches. Especially the fully coupled model seems to underestimate the structural response signicantly. This is further illustrated in Fig. 18, showing plots of the deformed shapes of the
container from simulations applying the various FE methods. If
these shapes are compared to the pictures taken by the high-speed
camera during the test (Fig. 4), quantitative agreement is only
found for the Lagrangian solution. For the EulerianLagrangian approaches, where also the roof and the sidewalls were loaded, the
deections are in general too small. None of the applied methods
are able to describe the outwards buckling of the roof seen
experimentally.
That a simple Lagrangian model combined with ConWep calculations quantitatively seems to give better results than the fully
coupled approach is in line with the conclusions of, e.g. Mullin
and OToole [17]. However, one reason for the large variation in

Fig. 18. Plots of deformed shape of container (with inner pressure) versus time during blast load from simulations applying various FEM methods.

T. Brvik et al. / Computers and Structures 87 (2009) 507520

60

Head-on pressure
(front side)
Head-on pressure
(rear side)
Curve from ConWep
(infinite rigid wall)
Curve from ConWep
(idealised - finite rigid wall)

Pressure (kPa)

40

20

60

Head-on pressure Pfront__


(Calbration test)
Head-on pressure Pfront__
(Protected container test)

40

Pressure (kPa)

518

20

Curve from ConWep


(infinite rigid wall)
Curve from ConWep
(idealised - finite rigid wall)

-20

-20

0.2

0.4

0.6

0.8

Time (ms)

0.02

0.04

0.06

0.08

Time (ms)

Fig. 19. Reected pressuretime proles from ConWep compared with measured pressuretime curves from the calibration test, from the protected container test [26] and
from the one-dimensional Eulerian simulation.

Fig. 20. Plots of deformed shape of container versus time during blast load from simulations with a fully coupled EulerianLagrangian approach and increased impulse.

predicted results between the applied methods has been found due
to the load specication for rigid versus exible structures, and in
previous sections it was concluded that it is not correct to use a
pressure load estimate for a rigid wall in response calculations of
exible structures. Other explanations for the deviation between
the approach with the highest expected accuracy and the experimental results may be the simplied material description used in
the simulations or an imprecise description of the blast load by
ConWep compared to the actual one measured in the test. To check
these possibilities, some new simulations were carried out using
the fully coupled EulerianLagrangian approach.
First, a simulation was carried out to examine the sensitivity of
the structural response due to variations in applied material properties. This was done by reducing the tangential hardening to
Et 100 MPa, i.e. a rather distinct reduction in strain hardening
of the steel material in the container. With such a reduction in
strain hardening the maximum response in the front wall increased to 185 mm, which is less than a 10% increase compared
to the simulation using Et 1000 MPa. Thus, this cannot explain
the large deviation obtained between the fully coupled Eulerian
Lagrangian simulation and the blast test results.
Secondly, Fig. 19a shows reected pressuretime curves from
ConWep for a blast load of 4000 kg TNT at 120 m standoff distance
(both for a nite and an innite rigid wall). Unfortunately, the gauge
supposed to measure the reected pressure versus time on the front
wall in the actual test failed during the test so no data is available.
Therefore, the curves from ConWep are compared with measured
pressuretime curves from the calibration test (using a rigid concrete wall) and from the protected container test (using a wall of extruded aluminium panels to protect the front wall of the container)
in Fig. 19a. These tests are described in more detail in [26]. At rst
sight, the agreement seems to be reasonable (especially for the

innite rigid wall). However, a closer look at these curves (see


Fig. 19b) reveals major differences between them. The difference
in reected impulse is more than doubled between the calibration
test and the idealized curve from ConWep used in the Lagrange
calculations. To check this effect, a nal simulation on the unprotected container was run with the fully coupled model where the
incoming pressuretime curve was doubled, i.e. doubling the incident impulse. Note that in this simulation, using the unprotected
container, the front wall is more exible than in the tests plotted
in Fig. 19.
The obtained maximum displacementtime curve of the front
wall from this simulation is plotted in Fig. 15 where it is compared to the experimental results and the other approaches. As
seen, by doubling the impulse the maximum displacement increases to about 500 mm and the response becomes similar to
the measured data. The deformed shape of the container versus
time from this simulation is plotted in Fig. 20. Now the upwards
buckling of the roof and the movement of the sidewalls become
well described (see also Fig. 4). Thus, both the quantitative and
qualitative response of the container is in good agreement with
the experimental data when the impulse is doubled. Note nally
that if this loading is applied in the pure Lagrange simulation,
the blast load response of the container would become far too
large.
The above exercise revealed that it is difcult to base the design
of blast loaded structures on ConWep calculations alone, because
the load description may be too imprecise for complex problems.
Thus, even though more costly it seems necessary to run fully coupled EulerianLagrangian simulations. The study also shows that
full-scale blast tests on exible structures are difcult to control
and instrument, and that more full-scale test results are required
in order to validate FEM-based blast simulations.

T. Brvik et al. / Computers and Structures 87 (2009) 507520

519

5. Conclusions

dW pdV c  1edV

A methodology has been proposed for the creation of inow


properties in uncoupled and fully coupled EulerianLagrangian
LS-DYNA simulations of blast loaded structures. The inow properties are assumed planar and calibrated based on output data from a
hemispherical surface burst of 4000 kg TNT at 120 m standoff distance by ConWep [11]. The structural responses based on the
uncoupled or coupled EulerianLagrangian approaches were found
to be less compared to a pure Lagrangian simulation with a pressure load provided by ConWep based on corresponding nite rigid
wall reected pressures. It was demonstrated that structural exibility reduces the pressure loads and thus the structural response.
Care must therefore be taken when applying pressure loads originating from rigid or stationary references to exible structures as
the response will be overestimated. However, in the case of exible
containers the application of a too high pressure loading leads to a
conservative design. Pressuretime proles from ConWep calculations both for a nite and an innite rigid wall have been compared
to corresponding curves from FE simulations, and the agreement is
found to be excellent. This indicates that both our approach and
ConWep give a good description of the reected pressuretime
prole for the specied loading against rigid surfaces.
The effect of inner pressure build up in a non-vented container
is signicant for large volumetric deformations and was found to
reduce the structural response of the Lagrangian simulation by almost 25%. This effect is inherent and thus directly accounted for in
the fully coupled EulerianLagrangian model. It was also shown
that the shape of the pressuretime prole plays a signicant role
for the displacementtime curve in these simulations, where the
loading should be considered as dynamic and not impulsive.
When compared to available full-scale experimental test data
on a blast loaded unprotected container, the Lagrangian approach
seemed to give better results than the fully coupled Eulerian
Lagrangian approach when the design blast load was solely based
on ConWep calculations. However, the test data revealed that the
actual blast load was higher than the one assumed in the simulations. If the reected impulse was doubled, both the quantitative
and qualitative response of the container was found to be in good
agreement with the experimental data using the fully coupled approach. This shows that it is difcult to base the design of exible
blast loaded structures on ConWep calculations alone and that
more full-scale experimental tests are required in order to validate
FEM-based blast simulations.

The internal energy of the gas is reduced by an equal amount, i.e.

Acknowledgement
The nancial support of this work from the Structural Impact
Laboratory (SIMLab), Centre for Research-based Innovation (CRI)
at the Norwegian University of Science and Technology (NTNU),
is gratefully acknowledged.
Appendix A. Inow properties for Eulerian analyses
For our system of air we consider a given volume V and a given
total internal energy E.
The specic internal energy e per volume is then

e qC V T E=V

A1

The corresponding pressure is

p qC P  C V T


CP
 1 qC V T c  1e
CV

A2

For an adiabatic gas expansion we get the following work done by


the gas

dE dW pdV

A3

A4

The change of specic internal energy is then

 
 
E
de
E
1 dE
d 1
dE 1
E

)
d
E

V
dV
V
V dV
dV V
dV V V 2

A5

Eqs. (A1), (A3), (A4), and (A5) now give

de dE 1
E
pdV 1 E 1
1
1


  p e  ce
dV dV V V 2
dV V V V
V
V

A6

This differential equation has the following solution

 c
de
dV
e
V0
c
)
e
V
e0
V

A7

Combining Eqs. (A2) and (A7) give

p0

V0
V

c
()

 1=c
V
p0

V0
p

A8

The change of density is controlled by

q q0

V0
V

A9

The relation between pressure and temperature gives

 
V0
C P  C V T
p qC P  C V T q0
V
 c
 c
V0
V0
p p0
q0 C P  C V T 0
V
V

A10a
A10b

Dividing the two expressions in Eq. (A10) eventually gives

T0

V0
V

c1
A11

References
[1] Remennikov AM, Rose TA. Modelling blast loads on buildings in complex city
geometries. Comput Struct 2005;83:2197205.
[2] Luccioni B, Ambrosini D, Danesi R. Blast load assessment using hydrocodes.
Eng Struct 2006;28:173644.
[3] Zhou XQ, Hao H. Prediction of airblast loads on structures behind a protective
barrier. Int J Impact Eng 2008;35:36375.
[4] Biggs JM. Introduction to structural dynamics. New York: McGraw-Hill Inc.;
1964.
[5] Baker WE, Cox PA, Westine PS, Kulesz JJ, Strehlow RA. Explosion hazards and
evaluation. New York: Elsevier Scientic Publishing Company Inc.; 1988.
[6] US Department of Army Technical Manual (TM5-1300). Design of structures to
resist the effects of accidental explosions. Washington, DC; 1990.
[7] US Department of Army Technical Manual (TM5-855-1). Fundamentals of
protective design for conventional weapons. Washington, DC; 1986.
[8] Langdon GS, Schleyer GK. Deformation and failure of proled stainless steel
blast wall panels. Part III: nite element simulations and overall summary. Int J
Impact Eng 2006;32:9881012.
[9] Balden VH, Nurick GN. Numerical simulation of the post-failure motion of steel
plates subjected to blast loading. Int J Impact Eng 2005;32:1434.
[10] Dharmasena KP, Wadley HNG, Xue Z, Hutchinson JW. Mechanical response of
metallic honeycomb sandwich panel structures to high intensity dynamic
loading. Int J Impact Eng 2008;35:106374.
[11] US Army Engineers Waterways Experiment Station. ConWep conventional
weapons effects. USA; 1991.
[12] Belytschko T, Liu WK, Moran B. Nonlinear nite elements for continua and
structures. Chichester, UK: John Wiley & Sons, Ltd.; 2000.
[13] LSTC. LS-DYNA Keyword Users Manual, Version 971. Livermore Software
Technology Corporation, January, 2007.
[14] ABAQUS. <http://www.simulia.com/products/abaqus_fea.html>.
[15] AUTODYN. <http://www.ansys.com/products/autodyn.asp>.
[16] EUROPLEXUS. <http://www.samcef.com/en/pss.php?ID=32&W=products>.
[17] Mullin MJ, OToole BJ. Simulation of energy absorbing materials in blast loaded
structures. In: Proceedings of eighth international LS-DYNA users conference,
May 24, 2004.

520

T. Brvik et al. / Computers and Structures 87 (2009) 507520

[18] Luccioni B, Ambrosini D, Danesi R. Analysis of building collapse under blast


loads. Eng Struct 2004;26:6371.
[19] Wang Z, Lu Y, Hao H, Chong K. A full coupled numerical analysis approach for
buried structures subjected to subsurface blast. Comput Struct 2005;83:
33956.
[20] Lu Y, Wang Z. Characterization of structural effects from above-ground
explosion using coupled numerical simulation. Comput Struct 2006;84:
172942.
[21] Hu W, Chen Z. Model-based simulation of the synergistic effects of blast and
fragmentation on a concrete wall using the MPM. Int J Impact Eng 2006;32:
206696.
[22] Guilke JE, Harman TB, Banerjee B. An EulerianLagrangian approach
for simulating explosions of energetic devices. Comput Struct 2007;85:
66074.

[23] Omang MG, Brve S, Trulsen JK. Numerical simulations of shock wave
reection phenomena in non-stationary ows using regularized smoothed
particle hydrodynamics (RSPH). Shock Waves 2006;16:16777.
[24] Omang MG, Brve S, Trulsen JK. Shock collisions in 3D using an axisymmetric
regularized smoothed particle hydrodynamics code. Shock Waves 2007;16:
46775.
[25] Brvik T, Hanssen AG, Dey S, Langberg H, Langseth M. On the ballistic and blast
load response of a 20 ft ISO container protected with aluminium panels lled
with a local mass Phase I: Design of protective system. Eng Struct
2008;30(6): 160520.
[26] Brvik T, Burbach A, Langberg H, Langseth M. On the ballistic and blast load
response of a 20 ft ISO container protected with aluminium panels lled with a
local mass Phase II: Validation of protective system. Eng Struct 2008;30(6):
162131.