0% found this document useful (0 votes)
13 views12 pages

Meta Model Blast

This document presents a metamodel-based shape optimization methodology aimed at enhancing the design of shallow-buried blast-loaded underbody structures for military vehicles. The study demonstrates that reshaping the underbody can significantly reduce impulse response during landmine detonations, with a unique v-shaped design emerging as the optimal solution. The methodology utilizes an empirical model to simulate blast loads and optimize the geometry of the underbody to improve vehicle protection against explosive threats.

Uploaded by

sengozk
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
0% found this document useful (0 votes)
13 views12 pages

Meta Model Blast

This document presents a metamodel-based shape optimization methodology aimed at enhancing the design of shallow-buried blast-loaded underbody structures for military vehicles. The study demonstrates that reshaping the underbody can significantly reduce impulse response during landmine detonations, with a unique v-shaped design emerging as the optimal solution. The methodology utilizes an empirical model to simulate blast loads and optimize the geometry of the underbody to improve vehicle protection against explosive threats.

Uploaded by

sengozk
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 12

International Journal of Impact Engineering 75 (2015) 229e240

Contents lists available at ScienceDirect

International Journal of Impact Engineering


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

A metamodel-based shape optimization approach for shallow-buried


blast-loaded flexible underbody targets
Terrence E. Johnson a, *, Anirban Basudhar b
a
SURVICE Engineering Company, 4695 Millennium Drive, Belcamp, MD 21017, USA
b
Livermore Software Technology Corporation, 7374 Las Positas Road, Livermore, CA 94551, USA

a r t i c l e i n f o a b s t r a c t

Article history: Landmine detonation continues to be a serious threat to Army military vehicles in theater. One strategy
Received 20 February 2014 to mitigate its effects is to use engineering design techniques to reshape the underbody of the hull.
Received in revised form Previous works have shown that reshaping the hull's underbody can change the vehicle's response to
18 August 2014
blast loading. In the presented work, a metamodel-based shape optimization methodology is proposed
Accepted 22 August 2014
Available online 3 September 2014
for optimizing shallow-buried blast-loaded underbody structures. The shallow-buried blast load is
simulated using an empirical model. Additional studies are conducted on target positions with respect to
the blast load and load positioning and the effects of optimization parameters on the optimal solution. It
Keywords:
Shallow-buried
is shown that target vertical position affects impulse response comparison, and for center-buried loads,
Blast the v-shaped underbody target produces a lower impulse than that of other shaped targets. Optimal
Metamodel solutions obtained using an adaptive domain reduction strategy produce more accurate solutions than a
Optimization single iteration strategy. Based on the proposed optimization method, a unique v-like underbody shape is
presented as the optimal underbody solution.
© 2014 Elsevier Ltd. All rights reserved.

1. Introduction protect the drive train [3]. Experimental and numerical research
has shown that a v-shaped target can reduce applied impulse from
Anti-vehicle landmines continue to be a serious threat to Army a charge by deflecting high-pressure gas and soil ejecta into the
military operations in theater. In fact, in previous conflicts on in- ambient. In addition, a portion of the applied pressure has a hori-
ternational soil, loss due to landmine detonation was as much as zontal component that does not contribute to the vertical impulse
70%. A major injury and kill mechanism that results from mine transferred to the vehicle. This improves better protection because
detonation is large gross vehicle movement. This type of movement less load is applied to the vehicle [4].
can cause injury to an occupant's lower leg, spine, and head if the In addition to v-shape targets, other shaped targets have been
occupant is not restrained [1]. Strategies developed to mitigate investigated [5e8]. Of particular interest is a work by Fox et al. [8],
these effects include increasing vehicle standoff (the distance from in which experimental and computational studies were conducted
the bottom of the vehicle to the ground), which reduces impulse on the effect of shallow-buried blast location on shape target
and local loading; adding retrofit kits or a sacrificial mass to the response. The results showed that downwardly convex and concave
underbody of hulls; incorporating v-shaped and monocoque hulls; targets produce lower impulse than a flat plate for a given load.
and sandbagging the floor of trucks to prevent occupant exposure Reasons given for impulse reduction were increased distance be-
to lethal fragments and increase the vehicle's weight [2]. tween the explosive and target and a geometric shaping effect.
The v-shaped hull design is commonly used today to protect Quarter loads generally produced larger target impulse than center
armored military vehicles from improvised explosive devices loads.
(IEDs). Current examples of vehicles that have v-shaped crew Additional works [4,9e12] consider experimental and/or nu-
compartments are the International MaxxPro, Casspir APC, and merical analysis of various shaped targets. Chung Kim Yuen et al.
Cougar H. The MaxxPro also has an additional underbody plate to [4] presented experimental and numerical results for v-shaped
steel plates with different included angles subjected to localized air
blast. Standoff was defined as the distance between the tip of the v
and the closest face of the explosive. It was shown that for a fixed
* Corresponding author. Tel.: þ1 410 273 7722.
E-mail addresses: [email protected], [email protected] standoff and load size, if the v included angle is increased, then
(T.E. Johnson), [email protected] (A. Basudhar). impulse and mid-point deflection are decreased. In addition, for a

http://dx.doi.org/10.1016/j.ijimpeng.2014.08.010
0734-743X/© 2014 Elsevier Ltd. All rights reserved.
230 T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240

fixed-plate v angle and load size, if standoff is decreased, then maximize the strain energy absorbed by a coupled aluminum
impulse and mid-point deflection are increased. Barker et al. [12] substructure. The optimized plate was composed of material where
conducted buried blast experiments and numerical analysis, us- the blast load was highest but was removed along the boundary
ing ConWep, on steel, hollow box, v-shaped targets of 10 , 20 , and sides to increase energy absorption. In a second work, Goetz et al.
30 , with a top floor plate using centered and off-centered blast [20] further showed that by modifying the HCA algorithm,
loads. Throughout the experiments, each target had the same convergence could be reached in both one and two material opti-
standoff, which was defined as the distance from the spine of the v mization (multi-material approach). By extension, Hofstetter et al.
to the surface of the soil. Off-center shots were located midway [21] used the Livermore Software Technology Corporation (LSTC)
between the spine of the v and the outermost edge of the target. implementation of HCA to develop optimized one-material sub-
The results showed that as target angle increased, target impulse structures for different objective functions.
decreased. It was suggested that this result was due to a geometric The previously reviewed optimization based approaches to
effect. It was also shown that for a charge size of 800 g, off-center structural design implement air blast loading only. Shape and to-
shots produced a higher impulse. pology optimization of structures for buried blast loading have not
In the works previously cited, target response was primarily been considered. One possible reason is that, until now, buried blast
compared by keeping the standoff constant. Again, standoff is simulations have required ALE-based fluid structure interaction
defined as the distance from the ground, or charge, to the lowest simulations, which require greater computational resources than
point on the target. It was reported that v targets produced a lower ConWep for air blast loading. In this work, buried blast simulations
impulse because of a standoff effect (i.e., underbody structure is will be conducted using a recently developed empirical model. This
farther away from the blast) or a geometric effect (i.e., the under- approach greatly reduces runtime and computational resources as
body vents blast ejecta). One inquiry of interest in this work is the it only requires Lagrangian elements. Validation of the developed
performance of shaped underbodies with alternative methods of empirical model will be conducted using published experimental
comparison. Would a v-shaped object outperform a flat-shaped data. Also, a geometric boundary shape optimization technique will
object if standoff is defined from the ground to the top of the v be implemented in this work. The geometry that will be optimized
instead of the spine? Experimental results from Anderson et al. [13] is a curve. Once developed, the curve is extruded to form a surface
begin to answer this question. Anderson et al. conducted experi- that represents the underbody of target. The surface is meshed and
ments on v-shaped targets where the centers of gravity of the connected to a top structure. The model is completed by applying
targets were held constant. From Anderson's data, impulse com- boundary and loading conditions to the structure. This technique
parisons could also be made between structures with the same can be extended to surfaces that can be extruded into solids.
distance from the ground to the structure's crown. Anderson's work Extruded shapes are ideal for vehicle surrogate structures due to
is further discussed in Section 3, as it is used to validate the pre- enhanced manufacturability with respect to highly complex
sented numerical model. shapes, as suggested by Argod [14] and Jain [16].
As already shown, targets of various shapes have been consid- The objective of this work is to present a metamodel-based
ered. These shapes were most likely determined by ad hoc tech- shape optimization methodology for the design of underbody
niques. While this approach may be useful in a down-selection structures for targets subjected to shallow-buried blast loads. The
process between a limited number of designs, it may not result in a objective of optimization is to minimize crew compartment im-
superior structural shape for an objective of interest. Structural pulse calculated using the finite element method. In the sections
optimization techniques provide a mathematically based systemic that follow, the optimization problem, target geometry and blast
approach to satisfy objectives of interest. In 2008, Gurumurthy [3] parameters are defined, and the presented empirical model is
conducted 2D and 3D impulse analysis on air-blast-wave-loaded validated using experimental results. The proposed shape optimi-
vehicle underbody shapes. Optimization was also conducted on a zation methodology is also defined. Numerical results and discus-
2D elliptical underbody shape to reduce maximum impulse. sion are presented for targets with various standoffs and for
Recently, a series of works that address shape optimization of underbody shape optimization.
rectangular panels for air blast load has been presented. Argod et al.
[14,15] developed a nodal coordinate shape optimization method-
2. Model definition
ology for solid panels that is defined by so-called velocity fields. The
basis shape functions for the panels were the product of sine
In this section, the optimization problem, target geometry, and
functions. The optimization technique of choice was the Differential
model components are defined and model validation is presented.
Evolutionary algorithm, which is non-gradient-based. It was shown
The basic optimization problem that is solved is to determine the
that a panel with a double bulge is ideal for minimizing local out-
underbody shape that minimizes crew compartment vertical im-
of-plane displacement and impulse. Jain et al. [16] presented an
pulse. Crew compartment impulse is a local component of the
extension to Argod's work by considering additional nodal coor-
target's global impulse.
dinate shape optimization methodologies and aluminum honey-
The optimization problem is to minimize a variable-dependent
comb sandwich panels under air blast load. Nayak et al. [17,18]
cost function that is subject to constraints. In this work, the vari-
focused primarily on the shape optimization of honeycomb sand-
ables of the cost function are the coefficients of the Fourier series
wich structures and found that shape-optimized solid panels are
expansion that defines the underbody shape. This is further dis-
better at reducing backface deformation, but shape-optimized
cussed in Section 4.1. The standard form of the optimization is as
honeycomb panels are more effective at reducing transmitted
follows:
acceleration.
Find a, b.
Although few examples of topology [19e21] or topography [22]
minimize F(a, b)
optimization of blast-resistant structures exist in the literature,
subject to
works of note were presented by Goetz et al. [19,20] of the Uni-
versity of Notre Dame. Goetz et al. used the Hybrid Cellar Automata Sðx; a; bÞ ¼ 0
(HCA) method to design a, CONWEP loaded, single material sub-
aL < a < aU (1)
structure to reduce the total energy that is transferred from the
L U
blast to the vehicle's interior. Optimization was conducted to b <b<b ;
T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240 231

where a, b 2 ℝn . Term ℝn is the n-dimensional real number space.


The design variables are a and b. F(a, b) represents vertical mo-
mentum transfer; and S(x, a, b) represents a constraint that defines
the Fourier series equation. The constraints aL and bL and aU and bU
are the lower and upper bounds for the design variables, respec-
tively. As will be discussed in Section 4, a and b represent the co-
efficients of the Fourier series equation. These constraints also
implicitly regulate the mass of the resulting structures.
The base numerical model configuration is given in Fig. 1. The
target, a simplified representation of a vehicle, is composed of four
Fig. 2. The finite element representation of the target. The target is discretized into
parts (as shown in Figs. 1 and 2). Part 1 represents the crew five separate flexible and rigid parts.
compartment. The impulse of Part 1 will be minimized in optimi-
zation. Part 2 presents the vehicle floor. Part 3 represents an elastic
plastic energy storing system. Part 4 represents an external sacri- The range of applicability of the model as derived by Westine is a
ficial mass. The surface area of Part 1 is 100 cm  100 cm and is function of explosive energy release, cross-sectional area of the
modeled as a rigid material. The thickness of Part 1 is 42 cm. Part 2 charge, soil density, standoff distance measured form the target
is modeled as a 3.8 cm-thick sheet of A36 steel. Parts 3 and 4 are from to center of the charge, depth of burial measured from the soil
modeled using aluminum 6061. The thickness of Part 3 is 2.6 cm. surface to the center of the charge, and lateral distance to the point
The thickness of Part 4 is 3.2 cm. The total average weight of the of interest beneath the target.
target is approximately 2069 kg. The distance from the v spine to
the ground is 27.5 cm. The distance from the v crown to ground is 3. Model validation
39.3 cm. The charge is Composition (Comp) B and weighs 625 g. The
charge's shape is flat and cylindrical. Its height is 3.7 cm, and its The shallow-buried blast empirical model *IIM is validated by
diameter is 11.3 cm. The top of the charge is buried 5.08 cm into the comparing experimental results from Anderson [13] to simulation
soil and centered beneath the target. The soil density is 1370 kg/m3 results. Anderson conducted well-characterized shallow-buried
and is identical to that described in Anderson et al. [13]. blast experiments on flat and v-shaped plates. The experiments
The structure is loaded using *INITIAL_IMPULSE_MINE (*IIM), consisted of a round cardboard barrel that contained the soil,
which is an empirical model, implemented in LS-DYNA, for buried charge, flat (Fig. 3) and v-shaped (Fig. 4) steel plates, and instru-
mine blast loading. *IIM was developed in 2012 and is unique to LS- mentation for measurement. The v-shaped plates were constructed
DYNA. While not as accurate as the Multi-Material Arbitrary of flat plates that were cut in half and butt-welded together. The
Lagrange Eulerian (MM-ALE) approach, which considers fluid plates contained small bracing plates that added stiffness and
structure interaction, *IIM requires less computation time, memory, served as a mount for instrumentation. The cardboard barrel had a
and storage and can produce a first-order approximation of struc- height of 85 cm, a diameter of 63 cm, and a wall thickness of 3 mm.
tural response to blast loading. As is shown later, *IIM produces The volume of the barrel was 255 L. Anderson showed through
reasonable relative impulses for various underbody shapes. *IIM is pretest simulations that a semi-infinite test bed could be approxi-
based on Westine [23] and Tremblay's [24] impulse data equations, mated using a confined test bed, which is defined by the cardboard
which were derived from shallow-buried blast experiments on flat barrel. The charge was 625 g of Comp B and buried so that its top
and oblique targets and calculates specific impulse for each finite was 5 cm below the sand. The charge was cylindrical with a height-
element segment where load is applied. The calculated value is to-diameter ratio of approximately 1e3 (a height of 3.7 cm and a
multiplied by the area of the segment to determine the impulse for diameter of 11.3 cm). The soil used in the experiments was common
that segment. The total impulse is calculated by summing the im- silica sand with a grain size of up to 1 mm. Different moisture
pulses of the segments. Impulse is applied to the nodes of the finite contents of the sand were used during experimentation: 7%, 14%,
element target as an initial velocity. Because *IMM is an engineer- and 22%. An abbreviated version of Anderson's experimental test
ing model and derived from maximum impulse experimental data, matrix is given in Table 1. The impulse of the plates was determined
it loads the target with maximum predicted impulse at time t ¼ 0. by multiplying the plate's mass by its initial velocity resulting from
blast. This velocity was calculated from maximum plate height, H,
using equation.
pffiffiffiffiffiffiffiffiffi
V¼ 2gH ; (2)

Fig. 1. The configuration of the model that is used in FEA simulations and Fig. 3. The picture on the left shows Anderson's [13] base experimental setup. The
optimization. picture on the right shows the base setup with a plate.
232 T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240

Fig. 4. A depiction of Anderson's [13] setup for v plate targets.

where g is the gravitational constant (9.81 m/s2).


The experimental results from Anderson are used as the stan-
dard for comparison against the developed finite element analysis
(FEA) simulations that implement *IIM. The goal is to validate *IIM
loading model for shallow-buried blast loading problems. The nu- Fig. 5. The target plate geometries are shown clockwise from top left: (a) 13 convex
merical models that were developed for comparison were only down; (b) 13 concave down; (c) 21 concave down; (d) flat; (e) 21 convex down.[8]
composed of Lagrangian targets. The flat targets had a mass of
300 kg and a size of 80 cm  80 cm  6 cm. Each half section of the
v target was 40 cm  80 cm  6 cm. The center of mass of the plates blast experiments and numerical studies on flat, downwardly
with respect to the charge varied in Anderson's experiments. It concave, and downwardly convex aluminum alloy 6061 targets
must be noted that plate distance to surface is measured from the (Fig. 5). The charge had a mass of 636 mg, was cylindrical shaped,
plate's center of mass to the soil surface. The moisture content of and was buried in wet sand. It was noted that the experimental
the sand is implicitly modeled in *IIM through variable RHOS-the components could be scaled to full-size by Hopkinson-Cranz
density of overburden soil. Results from Anderson's experiment scaling. Using scaling principles, the full-size charge was deter-
and from the developed numerical model are given in Table 1. The mined to be 4.54 kg, depth of burial was 593 mm, and standoff was
averages of Anderson's experimental results are compared to 416 mm. Results from Fox are given in Table 2. The results, labeled
simulation results. In general, the results show good agreement Experimental, have been normalized by the impulse for target
with Anderson's experiment. The largest percent difference occurs bottom angle 0 and charge location center (6.64 N-s). For valida-
for v plate of 90 . In addition, it is also seen that *IIM produces the tion, shaped targets, as shown in Fig. 5, were developed using the
same impulse trends as the experimental data. This demonstrates finite element method and simulated. As previously mentioned, the
that *IIM provides a load to the target that produces reasonably buried charge was simulated using *IIM. The results from the nu-
accurate results for target impulse. merical simulation are also presented in Table 2. As with the
A second set of experimental data, from Fox et al. [8], was used experimental results, the numerical results are normalized by the
to further validate *IIM. Fox et al. conducted scaled shallow-buried impulse of a plate angle of 0 . Because the target size, charge

Table 1
The experimental test matrix from Anderson [13] and results and simulation comparison.

Test series Moisture Mean density Plate type Plate distance to Impulse experiment Impulse simulation Percent
content (%) (kg/m3) surface (cm) (kg-m/s) (kg-m/s) difference (%)

1 7 1370 Flat 20 1963


1 2025
1 1950
Average 1979 2317 17
2 7 1370 Flat 30 1729
2 1627
2 1553
Average 1636 1767 8
3 7 1370 V-90 25 830
3 849
3 758
Average 812 485 40
4 7 1370 V-120 25 1285
4 1130
4 1130
Average 1182 984 17
5 14 1490 Flat 20 2170
5 2190
5 2102
Average 2154 2416 12
6 22 1670 Flat 20 2275
6 2540
6 2719
Average 2511 2557 2
T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240 233

Table 2 performed by first considering reasonable underbody shapes


A comparison of experimental results from Fox et al. and numerical results [8]. shown in Fig. 3: flat, v, u, inverse u, and w. Each shape satisfies
Case Target Charge Experimentally Normalized Normalized boundary conditions, and points above and below the horizontal
bottom location determined experimental numerical axis cannot exceed 0.13 m. The flat shape represents the case where
angle [ ] impulse imparted no sacrificial mass is added to the underbody. All other shapes
to target [N-s]
represent the addition of a sacrificial mass to the underbody. The v
1 21 Center Average 6.02 0.91 0.81 shape is a popular shape for underbody target geometry because it
2 13 Center Average 6.68 1.01 0.87
has been shown to deflect the blast wave, thus reducing target
3 0 Center Average 6.64 1 1
4 13 Center 5.26 0.79 0.73 impulse. The inverse u shape's midsection naturally provides a
5 21 Center 4.84 0.73 0.58 greater standoff than its side sections, which may provide advan-
6 0 Quarter Average 7.90 1.19 0.92 tages for mobility and a possible decrease in target impulse.
7 13 Quarter 6.94 1.05 0.74
However, because of its inverse u shape, the concave underbody
8 21 Quarter 5.74 0.86 0.63
may trap the blast wave and offset the effects of increased standoff.
The maximum number of Fourier series terms was determined
by overlaying shape data from Fig. 6 with several Fourier function
properties, and standoff in the numerical simulations were plots. An example of a Fourier series function plot for shape w is
different from what Fox used in experimentation, a direct com- given in Fig. 7. The Fourier function plot labels in Fig. 7 represent the
parison of the results cannot be made. However, a comparison of individual number of an and bn terms. One term denotes a total of
the impulse trends is sufficient to further validate the numerical two Fourier terms, one for a and another for b; two terms denotes a
model. The experimental results show that impulse for target total of four Fourier term; and so on. As expected, the accuracy of
angles 13 and 0 are nearly identical, and impulse is decreased as the Fourier function improves as the individual number of an and bn
the concaved angle is increased to 21. It was also shown that the terms is increased. Yet, as the number of terms is increased, the
downwardly convex targets have a lower impulse than the flat design space of the optimization problem is increased, which re-
target. It was suggested by Fox et al. that the observed reduction in sults in increased runtime and problem complexity. As with many
loading when compared to the flat plate is most likely the result of engineering optimization problems, this problem possesses a
increased distance between the explosive and target. It was further tradeoff between computational efficiency and accuracy. As a
shown that the quarter-located loads produced a larger impulse result, the number of individual an and bn chosen for the problem is
than the center loads. For center shots, the numerical results show five.
the same trends except when comparing targets 13 and 0 . In The range of an and bn terms was determined by computing
comparing the numerical results of 0 plate angle for the center and these values for the five preselected basic shapes, as shown in
quarter-located load, it was shown that the center load impulse is Table 3. It must be noted that a1 and a2 have the largest values for
greater than the quarter load. This difference is not seen experi- the select shapes, which suggests that they have the greatest in-
mentally. Overall, a comparison of the experimental and numerical fluence. The quantities b1eb5 are zero, which demonstrates that
results shows that the developed numerical model can reasonably Fourier sine terms are not necessary to produce the given shapes.
predict impulse trends for downwardly concave and convex targets. However, a range is given to b1eb5 so as to not further limit the
design space to previously generated shapes. The limits of the an
4. Shape optimization methodology and bn are given in column 7 of Table 3.

In this section, the shape function is defined and the shape 4.2. Optimization methodology
optimization methodology is given. The presented methodology is
partitioned into eight steps and is design for finite element struc- Metamodel-based optimization was conducted to determine
tures. The developed methodology is also implemented using the optimal shape of the target's underbody. Metamodels present a
LSTC's optimization code LS-OPT. simple and inexpensive surrogate (approximated mathematical
expression) of the actual design and, once created, can be used to
4.1. Shape function find the optimum [26]. In this work, the software LS-OPT is used for
optimization. A summary of the entire metamodel-based
The shape optimization methodology implemented in this work
is a type of geometric boundary method [25]. The geometry that is
0.15
optimized is a curve. The curve is designed by the Fourier series
expansion which is.
0.1

a X
k
0.05
yðxÞ ¼ 0 þ ðan cos nux þ bn sin nuxÞ (3)
Length (m)

2 n¼1
0
0 0.2 0.4 0.6 0.8 1
where u ¼ 2p/x is the fundamental frequency and an and bn are real
-0.05
valued coefficients. The quantity x represents the length of the field.
The boundary conditions for the target are that y ¼ 0, x ¼ 0 and -0.1
y ¼ 0, x ¼ x. The quantity x is defined as x ¼ 1. The conditions are
P
satisfied by setting a0 ¼ 2ð an Þ. This also removes an extra -0.15
variable from the analysis. Length (m)
The choice of Fourier series expansion as the shape function
v inverse u flat u w
allows simple and complex shapes to be included in the design
domain. However, the number of Fourier terms (k) and the bounds Fig. 6. The curves represent basic shapes that can be derived from Fourier series
of an and bn must be determined. This determination was expansion.
234 T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240

values. The output of the LS-PrePost step is an LS-DYNA input


deck. A unique input deck is created for each sample. The eight
sub-steps used to develop the models are as follows.
3-1. Create line shape of target underbody
3-2. Extrude line to make a surface
3-3. Mesh the surface
3-4. Create 2D geometry of remainder of target
3-5. Extrude to make surfaces
3-6. Mesh the surfaces
3-7. Assign target underbody surface as load-receiving surface
3-8. Merge coincident nodes.
4. Solver execution and result extraction: In step four, LS-OPT calls a
solver to perform the simulation for each sample and then ex-
tracts the values for prescribed responses. The solver used in
this work is LS-DYNA R7.0.0 finite element software. The input
file for LS-DYNA is the one created in step three.
Fig. 7. The curves demonstrate the effect of number of Fourier coefficients on the 5. Metamodeling: In step five, metamodels for all the responses are
shape.
created over the design space for the purpose of design explo-
ration. In this work, the metamodels where created using Radial
optimization procedure is depicted in Fig. 8. It can be organized into
Basis Function Network [27].
eight basic steps:
6. Optimization: In step six, numerical optimization is conducted
using the metamodel constructed in step five to find the optimal
1. LS-OPT problem setup: In step one, the design problem is defined.
solution that minimizes the objective while satisfying the con-
This step includes definition of the task (e.g., metamodel-based
straints. The numerical optimizers can be gradient or non-
optimization), strategy (e.g., single iteration, sequential or
gradient based.
sequential with domain reduction), objectives, constraints, and
In this work, a genetic algorithm (GA) was selected as the
variables. A point selection scheme and a metamodel type are
optimizer, which is a population-based method that does not
also defined. The optimization problem is:
require derivative information and seeks the global optimal
solution. A detailed description of GAs and their associated
Find a1,a2,a3,a4,a5,b1,b2,b3,b4,b5
options or parameters can be found in the LS-OPT manual [27].
minimize F(a, b)
7. Result storage and report generation: In step seven, the optimized
subject to
solution is saved and reported. In an iterative strategy, steps 2e7
ðP2P4Þ are repeated until a termination criterion is met.
dmin >0 (4) 8. Metamodel accuracy validation using check points: An optional
important step that is not formally a part of the optimization
dðP2Þ < d0 process but that is useful to validate the accuracy of the opti-
mized solution is called check adequacy. In this step, the accu-
where objective function F is the vertical impulse of the crew racy of the metamodel is checked using uniformly sampled
compartment (part 1 of Fig. 1) of a simplified vehicle structure, check points. An error value is calculated for each check point by
ðP2P4Þ comparing the simulated result to the metamodel prediction for
dmin is the minimum distance between parts 2 (target floor) and
4 (target underbody), and d(P2) is the deflection of part 2, which is every response. If the maximum error is below a prescribed
constrained to be less than a constant value d0. By constraining value, then the metamodel is deemed sufficient and the results
ðP2P4Þ from optimization are taken as optimal.
dmin to be strictly positive, it is ensured that the floor and un-
derbody do not come in contact.
In summary, the optimization problem is solved using Space
2. Point selection: Once the LS-OPT setup is complete and it is Filling criterion for point selection, Radial Basis Function Network
executed, in step two, LS-OPT generates a set of points, each of as the metamodel, and GA for optimization over the metamodel.
which represents a distinct design configuration (distinct
Fourier coefficient values). In this work, the optimization 5. Numerical results and discussion
problem is solved using Space Filling criterion for point
selection. In this section, the effects of target standoff and charge location
3. Pre-processing: In step three, LS-OPT calls the pre-processor to on target impulse are examined. Following that, shape optimization
generate the model for each sample. The model for each design is conducted to determine the target's optimal underbody shape
configuration (or sample) is created automatically using LS- that minimizes crew compartment impulse. The simulations are
PrePost command files with parameterized Fourier coefficient conducted using the LS-DYNA finite element solver. The developed
targets are loaded by *IIM, which is used here to simulate shallow-
Table 3 buried blast. Each simulation takes approximately 30 s to run. The
This gives the numerical values for the Fourier constants and boundary conditions. impulse results in Figs. 11, 14 and 17 are normalized by 2934 kg-m/s.
V Inverse u Flat U W Selected range
5.1. Target standoff and charge location
a1 0.0514 0.0507 0 0.0507 0.0257 0.0514 to 0.0514
a2 0 0.0134 0 0.0134 0.0386 0.0386 to 0.0386
a3 0.0057 0.006 0 0.006 0.0029 0.006 to 0.006 An important factor in comparing impulse between various tar-
a4 0 0.0034 0 0.0034 0 0.0034 to 0.0034 gets is their relative position. For example, a comparison of two
a5 0.0021 0.0022 0 0.0022 0.001 0.0022 to 0.0021 targets with uniform center of gravity (c.g.) height may show that
b1eb5 0 0 0 0 0 0.006 to 0.006
target A has a lower impulse than target B. However, it is conceivable
T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240 235

Fig. 8. Summary of the problem setup and execution process using LS-OPT.

Fig. 9. The targets are positioned so that the center of gravity of the targets underbodies are vertically aligned.

Fig. 10. The targets are positioned so that the center of gravity of the targets are vertically aligned.
236 T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240

1.6 1.6
1.4 1.4

Normalized Impulse
Normalized Impulse

1.2 1.2
1 1
0.8 0.8
0.6 0.6
0.4 0.4
0.2 0.2
0 0
flat inverse u u v w flat inverse u u v w
underbody c.g. total c.g. floor height standoff
Fig. 11. The data for targets with uniform underbody center of gravity and uniform Fig. 14. The data are for targets with uniform floor height and uniform standoff.
target center of gravity.

since its curved shape is closer to the blast and absorbs a higher
that target B could produce a lower impulse than target A if standoff impulse. It is interesting that the w underbody shaped target pro-
was made uniform. This section presents a set of comparisons that vides a slightly higher impulse than the flat. As shown in Fig. 9, the
start to address the question of how to best position targets for the top of the w underbody section is further away from the blast
evaluation of impulse. Four different methods of positioning the (31 cm) than the flat (24 cm), but its double nadirs are closer.
targets for impulse comparison are presented: uniform underbody The second method of comparing target impulse is by setting
c.g. height, uniform global c.g. height, uniform standoff, and uniform the global c.g. height for each target to be uniform. This method is
target floor height. The loading cases considered are centered, offset, slightly different than the previous case where the c.g. of each
and offset with target forward velocity. underbody sacrificial mass was made uniform. The resultant effect
The first two set of comparisons that are considered are uniform of prescribing a uniform global c.g. is that the bottom left corner of
underbody c.g. (Fig. 9) and uniform global c.g. positioned (Fig. 10) the targets are within a few centimeters of each other, as shown in
over a centered blast load. As a base for comparison, the distance Fig. 10. The results in Fig. 11 show that the target impulses are
from the charge to the target is measured in the vertical direction generally higher than in the uniform underbody c.g. method. This
and starts from the bottom left edge of the target and ends at the result is reasonable in light of the observation that the nadirs of
charge surface. As shown in Figs. 9 and 10, positioning by uniform these targets are generally closer to the blast. In comparing inverse
c.g. results in varying distances of closeness to the charge. The u target impulses between both positioning cases, the results show
numerical simulation results for vertical impulse of the targets are that target impulse has decreased since the target is now farther
given in Fig. 11. Fig. 11 shows that, when underbody c.g. is uniform, away from the charge. It must be noted that *IIM does not account
the v target has a lower impulse than the flat, as generally expected, for confinement effects; thus, any increase in impulse due to the
and the inverse u target has the largest impulse of all the targets. It potential confinement of high-pressure gas beneath the inverse u
is expected that the u target produces a higher impulse than the v target is not taken into account. The u target has a higher impulse

Fig. 12. The targets are positioned so that the floor of the targets are vertically aligned.

Fig. 13. The targets are positioned so that the standoff between the targets are the same.
T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240 237

1.2 0.7

Normalized Impulse
Normalized Impulse

1 0.6

0.8 0.5
0.4
0.6
0.3
0.4
0.2
0.2
0.1
0
0
0 200 400 600 800 1000
flat inverse u u v w
Charge Size (g)
static dynamic
center off center
Fig. 17. The data are for targets that are static and dynamic.
Fig. 15. This data is from a work by Barker et al [12].

than the v because its round edges result in a higher impulse. The w v-shaped target (Fig. 15). The charge was located halfway between
target has the highest impulse because it is closest to the blast. the spine of the v and the outermost edge of the target and along
The next positions for comparison are uniform floor height the horizontal line of the center load. The results showed a po-
(Fig. 12) and uniform standoff (Fig. 13) under center blast load. The tential for increased impulse by off-center shots. Fox et al. [8]
results for both cases are given in Fig. 14. In the case where target compared centered to quarter-off-center shots for flat, v angle
 
floor height is uniform, it is noted that the v has a lower impulse 13 , and v angle 21 . It was shown that impulse was increased by
than the flat. This result suggests that adding a v-shaped underbody 19%, 32%, and 19%, respectively.
mass to the target underbody is better at reducing impulse than not Two types of off-axis scenarios are examined in this work: off-
adding a mass or and a flat underbody structure. This behavior was axis static and off-axis dynamic. Off-axis static is the case where
also seen in Anderson [13] and may challenge the notion that the target is not moving and the underbody of the target is loaded
increasing standoff (moving the structure away from the charge) is beneath the front passenger side. In the off-axis dynamic case, the
the only way to reduce target impulse. An examination of u impulse target is loaded beneath the front passenger side while traveling at
vs. flat shows that u is higher than flat by a small amount. This a speed of 13 m/s. In the numerical simulation, the velocity is
result is reasonable since u is closer to the blast. In comparing u and applied at time t ¼ 0, and the impulse load is applied shortly
flat to v, it is seen that v produces the lowest impulse because the thereafter. To allow for comparison with previously generated
slanted edges produce a reduction in impulse. The w target pro- centered buried target impulse data, the vertical positions of the
duces a much larger impulse than flat, v, and u because its nadirs targets are the same as the targets with uniform underbody center
are closets to blast. Thus, the relative impulses of w, u, flat, and v of gravity (see Fig. 9). The horizontal positions are below the front
appear to be reasonable. The inverse u target produces a smaller passenger side wheel and represent the case where the passenger
impulse than the flat because its underbody peak point is further wheel applies pressure to a buried charge. The results in Fig. 17 are
away from the blast. for the static and dynamic off-axis loading cases shown in Fig. 16.
In the case where target standoff is uniform (Fig. 13), standoff is The results indicate that there is no significant difference in the
approximately 41 cm. The total distance from the charge to the amount of impulse the target receives between the two scenarios.
nadir of the underbody hull is approximately 46 cm. The results The v target has the highest impulse, which seems reasonable since
show that the flat has the highest impulse and the v target has the one half of the underbody is directly facing the mine. It must be
lowest impulse. It was also shown that the impulse magnitudes are noted that experimental work from Barker et al. and Fox et al. show
much lower than in the previously considered target positions. This that off-axis shots produce an increase in target impulse. However,
result is caused by the increase in distance from the charge to the the results in Fig. 17 show a decreased impulse. For example, the flat
structure with respect to the previously considered positions. target is decreased from 1 (see result in Fig. 11) to about 0.5. This
Target loading in theater is stochastic and can come from many result may indicate a limitation of *IIM to properly handle off-
different underbody locations. One feasible location is the off-axis center loading. In addition, results from Barker et al. and Fox
front driver or passenger side. This type of loading could occur et al. are for quarter shots, which are different than the current
when the front driver or passenger wheel applies pressure to a loading. Further investigation is required to determine *IIM ability

buried charge. Barker et al. [12] conducted off-center shots for a 20 to accurately apply off-center load to finite element structures.

Fig. 16. The targets are positioned so that the center of gravity of the target underbodies are vertically aligned. The charge is located off-axis.
238 T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240

Table 4 number of iterations) was studied. The results from metamodel-


This is a summary of the shape study. based optimization are given in Table 5. Three error measures are
Type, loading Highest impulse Lowest impulse used to assess the solution accuracy e maximum metamodel error
uucg, cl Inverse u w Flat u v
εm using 90 check points, solution prediction error εp (percentage
utcg, cl w u Flat Inverse u v error between the predicted and computed value at the predicted
ugmc, cl Flat Inverse u w u v solution) and actual solution error εa (percentage error between
ustf, cl w u flat Inverse u v metamodel value at the predicted solution and the computed value
uucg, ol, s v u w Flat Inverse u
at the best known solution using 2150 direct GA runs). It should be
uucg, ol, d v u w Flat Inverse u
noted that the check points comprise a separate set of samples that
uucg e uniform underbody c.g.
are not used for training. Check point-based metamodel error was
utcg e uniform total c.g.
ugmc e uniform standoff.
not calculated for SDR strategies. See Section 5.2.2 for more details
ustf e uniform height of target's floor. on the best known solution.
cl e center load. In the SI strategy cases (cases 1e3), the metamodel was created
ol e offset load. from a set of design points that were generated in one iteration
s e static.
(non-recursive, fixed variable ranges). This approach was imple-
d e dynamic.
mented for three total sample size values, as shown in column 6,
A summary of results from the previous studies is given in rows 1e3, of Table 5. The results show that, in cases 1e3, the ac-
Table 4. The results have shown that *IIM produces reasonable re- curacy of the developed metamodels is increased as the number of
sults for the selected methods of comparison. In the cases where a points used to create the metamodels is increased. In particular,
center load is considered, the v target produced the lowest impulse. increasing the number of points used to create the metamodel,
An interesting conclusion is that the flat underbody produced the from 120 to 360, resulted in a 70% increase in accuracy at the
lowest impulse in the case where the loading is off-center and predicted solution and a 38% decrease in the global metamodel
adding forward velocity to the target does not greatly influence error εm. The solution error εa is also decreased from 12.92% to 7.13%
target impulse. While it is seen that the proposed methods of as the number of points is increased. It must be noted, that as the
aligning targets are legitimate for impulse comparison, the method number of points is increased from 240 to 360, the εa increases
of aligning targets to have uniform floor height will be used in despite a decrease in εm. This result indicates a locally inaccurate
further study. The method of aligning targets to have uniform floor metamodel in critical regions of the design space (i.e., close to the
height was chosen because it represents the case of adding retrofit optimum). Because the strategy is non-adaptive for these cases, a
kits to vehicles and it simplifies the optimization algorithm by lower global error may not necessarily correspond to a better so-
fixing target vertical alignment. lution. It should be noted that the predicted objective function
value for all three single iteration cases is lower than the actual best
5.2. Optimization results solution value of 2158.17. This clearly indicates that all three met-
amodels underpredict the objective function value at the solution.
In this section, the results from the implementation of optimi- To enhance the accuracy of the optimal solution, the SDR
zation to minimize impulse (Equation (4)) are presented. The value strategy [27] is used, which focuses on sampling only within an
of threshold deflection d0 is set to 3 cm. Two optimization ap- iteratively updated “critical region” in the design space defined
proaches were implemented: metamodel-based (recommended) around the previous iteration optimal solution. Thus, instead of
and direct GA (for comparison). spacing samples globally, it locally refines the metamodel around
the optimum. The cases for optimization using SDR were parti-
5.2.1. Metamodel-based optimization tioned into three categories that are organized with respect to the
Two different strategies of metamodel-based optimization were total number of points that were simulated: 120, 240, and 360.
implemented: single iteration (SI) and sequential with domain Because columns 5 and 6 of Table 5 required whole numbers, Cases
reduction (SDR). The effect of varying the optimization parameters 6 and 9 were offset in the number of total points evaluated. In
(the total sample size, the number of samples per iteration and the addition, for each total number of samples with the SDR strategy,

Table 5
These are the results from metamodel-based optimization.

Case Strategy Load type No. of points No. of iter. Total no. Metamodel Predicted optimal Computed Predict. Actual solu.
per iter. of points error (%) εm solution (kg-m/s) (kg-m/s) error (%) εp error (%) εa

1 SI Center 120 52 1342 2437 81 12.92


2 SI Center 240 18 1904 2280 19 5.65
3 SI Center 360 14 2081 2312 11 7.13
4 SDR Center 12 10 120 e 2280 2288 0.4 6
5 SDR Center 24 5 120 e 2610 2635 0.9 22.1
6 SDR Center 36 3 108 e 2331 2333 0.1 8.11
Average 2419
7 SDR Center 12 20 240 e 2274 2272 0.09 5.3
8 SDR Center 24 10 240 e 2289 2300 0.5 6.6
9 SDR Center 36 7 252 e 2260 2265 0.2 5
Average 2271
10 SDR Center 12 30 360 e 2272 2270 0.09 5.2
11 SDR Center 24 15 360 e 2201 2214 0.006 2.6
12 SDR Center 36 10 360 e 2202 2201 0.05 2
Average 2228
13 SDR Off-center static 36 10 360 e 854 849 0.5 e

Single iteration (SI).


Sequential with domain reduction (SDR).
T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240 239

three sets of results were noted. As the number of points per iter-
ation was increased, the number of iterations was decreased so that
the total number of points evaluated was nearly the same. For a
total of 120 points, the SDR strategy gives lower errors than the
corresponding single iteration optimization in two out of three
cases. The solution errors are more than 5% in all of these cases,
suggesting that the solution is yet to converge. A significantly high
error is obtained in the case with 24 starting points and 5 iterations.
This may indicate a local optimum or simply a slow rate of
convergence with five iterations being too low for the problem. In
the cases with 240 total points, two of the three errors using SDR
are lower than the SI error and the third one is slightly higher. The
prediction error is quite low, however, in all the cases, and the
slightly higher error in the case with 24 points per iterations and 10
iterations may be due to a local optimum. The mean εa and εp errors
using the SDR strategy are lower than the SI strategy, as expected
from an iterative strategy that refines critical regions. The advan-
tage of SDR over SI in providing an accurate solution is most
prominent in the cases with 360 total samples, as the iterative
solution approaches convergence. The errors in cases 11 and 12 are
significantly lower than with the SI strategy with an equal number
of points. The error in case 10 is also lower than with the SI, but the
difference is not as remarkable. This fact may be because of the low
number of points per iteration (and therefore low starting points) Fig. 18. Figure (a) represents the optimal solution from metamodel-based optimiza-
in this case, which can lead to a local optimum. Overall, it can be tion center-loaded target (the optimal design variables are a1 ¼ 0.0514, a2 ¼ 0.0386,
stated that, in general, the optimum solution obtained using an SDR a3 ¼ 0.006, a4 ¼ 0.0034, a5 ¼ 0.0020, b1 ¼ 0.002579, b2 ¼ 0.0057, b3 ¼ 0.004,
b4 ¼ 0.001, b5 ¼ 0.003); (b) rotated view.
strategy is likely to be more accurate than the SI strategy with the
same number of points. It is, however, possible that the SDR solu-
tion may provide a local optimum. However, the accuracy of the optimal solution is dependent on the
The optimal solution using SDR was compared to the impulse number of points used to generate the metamodel and the associated
obtained for the v target. The results show that for a center-loaded optimization strategy. In this work, the GA population size and
target, the lowest computed minimum impulse (2201 kg-m/s) is number of generations are 30. The total number of simulations is 900.
produced using the SDR strategy, with 360 points used to develop The optimization history is given in Fig. 20. The load case evaluated is
sequential metamodels. The optimal values of design variables center, and the targets have uniform floor height. The optimization
a1ea4 for this solution are active. In comparison to impulse of a v- history plot shows that after 30 iterations, the minimum impulse is
shaped target (2610 kg-m/s), the SDR strategy has produced a so- 2377 kg-m/s. This result is 6% greater than the minimum solution
lution with a 16% reduction in crew compartment (Part 1) impulse. found using metamodel-based optimization. This demonstrates that
The optimal underbody shape produced from metamodel-based the solution is not fully converged. When compared to the
optimization is given in Fig. 18. The result shows that a v-shaped metamodel-based optimization Case 4 of Table 5, it is seen that the
profile, with edges that are slightly concaved, is ideal when the metamodel-based approach produces a better solution after only 113
target is centered above the blast. The concaved edges may be a simulations. This number is eight times fewer simulations than with
result of the standoff effect. The underbody shape profile is crude GA optimization, demonstrating that the metamodel-based approach
due to the number of terms used in the Fourier series equation. The
profile could have easily been made smoother by adding more
Fourier terms, but this would increase the design spaces and
computational time required to find the optimal solution. This
demonstrates that there is a tradeoff in the optimization scheme.
The optimal solution for the off-center loading case is given in
case 13 of Table 5 and shown in Fig. 19. The optimization strategy
used to obtain the optimal solution is SDR. The optimal shape,
shown in Fig. 19, is double-concaved. A comparison to the basic
inverse u underbody shape shows that the double-concaved sec-
tions have higher peaks. The double-concaved peaks allow the
target to be further away from the blast load, thus minimizing the
impulse delivered to the target.

5.2.2. Direct optimization


For the purpose of comparison with metamodel-based optimiza-
tion, direct optimization was also conducted. The direct optimization
algorithm that was used is the GA. This approach is different than a
metamodel-based approach in that each design is directly simulated
using the FEA solver, whereas in the metamodel-based approach, the Fig. 19. Figure (a) represents the optimal solution from metamodel-based optimiza-
tion off-center-loaded target. (The optimal design variables are a1 ¼ 0.0514,
design solutions were approximated using the developed meta- a2 ¼ 0.037, a3 ¼ 0.006, a4 ¼ 0.0034, a5 ¼ 0.0013, b1 ¼ 0.005, b2 ¼ 0.004,
models. The metamodel-based approach is able to find an optimal b3 ¼ 0.005, b4 ¼ 0.0009, b5 ¼ 0.003); figure (b) shows an overlay of optimal shape
solution to a given problem much quicker than with a direct approach. with basic inverse u shape.
240 T.E. Johnson, A. Basudhar / International Journal of Impact Engineering 75 (2015) 229e240

2520 Pecora of the U.S. Army Research Laboratory for useful discussions
2500
about target response that results from buried blast loading. The
authors also thank the SURVICE Engineering Company for support
Momentum (kg-m/s)

2480 provided in the finally stages of this work.


2460
2440 References

2420 [1] Bird Roy. Protection of vehicles against landmines. J Battlef Technol 2001;4:
14e7.
2400 [2] Ramasamy A, Hill AM, Hepper AE, Bull AM, Clasper JC. Blast mines: physics,
2380 injury mechanisms and vehicle protection. J R Army Med Corps 2009;155(4):
258e64.
2360 [3] Gurumurthy Ganesh. Blast mitigation strategies for vehicles using shape
0 5 10 15 20 25 30 optimization methods. Diss. Cambridge, Massachusetts: Massachusetts Insti-
tute of Technology; 2008. p. 13e68.
Number of Iterations [4] Chung Kim Yuen S, Langdon GS, Nurick GN, Pickering EG, Balden VH.
Response of V-shape plates to localised blast load: experiments and numerical
Fig. 20. This figure shows the genetic algorithm optimization history. simulation. Int J Impact Eng 2012;46:97e109.
[5] Genson Kevin William. Vehicle shaping for mine blast damage reduction. Diss.
2006.
can be a more efficient approach to satisfying an objective. In addition, [6] Taylor LC, Fourney WL, Leiste U, Genson K. Geometrical shaping of vehicles for
reducing impulse from buried explosions. In: Joint classified bombs/warheads
to run with a population size and number of iterations equal to 30, a & ballistics symposium. Monterey, CA: Department of Defense; 2008. p. 1e18.
direct optimization with a population size and number of iterations [7] Benedetti Robert. Mitigation of explosive blast effects on vehicle floorboard.
equal to 50 was also run. The solution converged after 43 iterations, ProQuest; 2008.
[8] Fox DM, Huang X, Jung D, Fourney WL, Leiste U, Lee JS. The response of small
with an objective function value of 2158 kg-m/s. scale rigid targets to shallow buried explosive detonations. Int J Impact Eng
2011;38(11):882e91.
6. Conclusions [9] Gupta AD. Effect of target bottom shapes on mine-blast loading from deto-
nation of an explosive on a rigid surface. ARL-TR-3152. Aberdeen Proving
Ground, MD: U.S. Army research laboratory; February 2004.
A shape optimization methodology has been developed to [10] Huang X, Yen Chian-Fong, Hoppel CP. Modeling characterizations of
determine the optimal underbody shape for underbody center and charge distance effect on a v-target structure with crews and seats. ARL-
TR-4476. Aberdeen Proving Ground, MD: U.S. Army Research Laboratory;
off-center shallow-buried loaded targets. Specifically, the meth-
June 2008.
odology consisted of iteratively developing finite element models [11] Tan H, Goetz JA, Tovar JE, Renaud. Simultaneous topography optimization of a
using a command text file. The command file was coupled with vehicle hull and topology optimization of the assembly interface for blast
LSTC's LS-OPT metamodel-based executable to implement shape mitigation. In: Proceedings of the 2010 ground vehicle systems engineering
and technology symposium. Troy, Michigan: GVSETS Conference; August
optimization. Shallow-buried blast was simulated by using a newly 2010. p. 1e12.
developed, fast empirical model *INITIAL_IMPULSE_MINE. But [12] Barker C, Howle D, Holdren T, Koch J, Ciappi R. Results and analysis from mine
before shape optimization was conducted, an examination of target impulse experiments using stereo-digital image correlation (No. ARL-RP-
0370). Army Research Lab Aberdeen Proving Ground MD Survivability/
position with respect to the blast load was conducted. It was shown Lethality Analysis Directorate. 2012.
that target vertical position affects impulse response comparison. [13] Anderson Jr Charles E, Behner Thilo, Weiss Carl E. Mine blast loading exper-
As such, it is important that a uniform target position is selected iments. Int J Impact Eng 2011;38(8):697e706.
[14] Argod V, Nayak SK, Singh AK, Belegundu AD. Shape optimization of solid
before target impulse is compared. It was also shown that for isotropic plates to mitigate the effects of air blast loading. Mech Based Des
center-buried loads, the v-shaped underbody target produced a Struct Mach 2010;38(3):362e71.
lower impulse than the u, w, flat, and inverse u targets. Finally, the [15] Argod V, Belegundu AD, Aziz A, Agrawala V, Jain R, Rajan D. MPI-enabled
shape optimization of panels subjected to air blast loading. Int J Simul Mul-
developed methodology was used to conduct shape optimization. tidiscip Des Optim 2008;2(4):273e82.
The results showed that a v-like shaped underbody target, with [16] Jain Rajeev. Blast mitigation solutions via FEM-based design optimization.
concaved edges, produces the lowest crew compartment impulse Diss. Arizona State University; 2009.
[17] Nayak Sumanta Kumar. Optimization of honeycomb core sandwich panel to
for center-buried loads. It was also shown that for an off-center
mitigate the effects of air blast loading. Diss. The Pennsylvania State Univer-
load, a double-concave-like target produces the lowest impulse. A sity; 2010.
comparison of optimization approaches showed that the [18] Nayak SK, Singh AK, Belegundu AD, Yen CF. Process for design optimization of
metamodel-based approach produced a more optimal solution in a honeycomb core sandwich panels for blast load mitigation. Struct Multidiscip
Optimi 2013;47(5):749e63.
shorter time than with direct optimization. Thus, a metamodel- [19] Goetz JC, Tan H, Renaud JE, Tovar A. Structural topology optimization for blast
based or similar approach is also ideal for higher-fidelity prob- mitigation using hybrid cellular automata. In: Proceedings of the 2009 ground
lems that require long runtimes. vehicle systems engineering and technology symposium (GVSETS); 2009
August. p. 1e9.
The results presented in this work were for the case where [20] Goetz J, Tan H, Renaud J, Tovar A. Two-material optimization of plate armour
targets were loaded by a centered shallow-buried underbody blast. for blast mitigation using hybrid cellular automata. Eng Optim 2012;44(8):
However, the vehicles in theater experience stochastic loading. The 985e1005.
[21] Hofstetter Jr Dwight, Gupta Rahul, Bitting Robert. Constraint-based hybrid
bottom of the vehicle could be loaded by a charge from any position cellular automaton topology optimization for advanced lightweight blast
or buried depth. The charge could be coupled with additional ob- resistant structure development. No. ARL-TR-5820. Aberdeen Proving Ground,
jects that might further enhance loading. The vehicle may also be MD: U.S. Army Research Laboratory; 2011.
[22] Israel, Joshua J. and Andre s Tovar. Investigation of plate structure design
loaded by multiple charges. These complexities should be consid-
under stochastic blast loading.
ered in optimizing the underbody shape for shallow-buried blast [23] Westine PS, Morris BL, Cox PA, Polch EZ. Development of computer program
loading. Addressing these complexities will be the focus of opti- for floor plate response from land mine explosions. Contract report 1345.
1985. p. 13e68.
mization efforts in future work.
[24] Tremblay JE. Impulse on blast deflectors from a landmine explosion. No. DREV-
TM-9814. Valcartier (Que bec): Defence Research Establishment; 1998. p. 1e24.
Acknowledgments [25] Belegundu Ashok D, Chandrupatla Tirupathi R. Optimization concepts and
applications in engineering. Cambridge University Press; 2011.
[26] Barthelemy JFM. Function approximation. In: Kamat MP, editor. Structural
The authors would like to thank Len Schwer of Schwer Engi- optimization: status and promise; 1993. p. 51e70.
neering & Consulting Services and Christopher Cummins and Collin [27] LS-OPT user's manual. CA, Livermore: August 2013. p. 5e666. Version 4.3.

You might also like