Struct Multidisc Optim

DOI 10.1007/s00158-012-0845-x

INDUSTRIAL APPLICATION

Process for design optimization of honeycomb core sandwich

panels for blast load mitigation
S. K. Nayak A. K. Singh A. D. Belegundu C. F. Yen

Received: 25 February 2012 / Revised: 14 August 2012 / Accepted: 15 August 2012

Springer-Verlag 2012

Abstract A general process for optimization of a sandwich which increases panel stiffness. For acceleration minimiza-
panel to minimize the effects of air blast loading is presented tion, results again produce a stiffer front face plate, but
here. The panel geometry consists of two metal face plates accompanied by a sufficiently soft core. The mechanism of
with a crushable honeycomb or other type of core. Opti- lowering the backface acceleration is by absorbing energy
mization is necessary as there is strong coupling between with low transmitted stress. A clear cut comparison between
the several variables and the physics, which makes para- monolithic metal plates and sandwich plates, for the same
metric studies relatively ineffective. Virtual testing is used loading and failure criteria, is presented here.
to develop a homogenized model for the stressstrain curve
of the honeycomb core, which can be readily applied to Keywords Honeycomb Homogenization Blast
other types of cellular core. The homogenized model has Optimization Sandwich plates Virtual testing
been validated by comparison to existing results as well
as to results from detailed finite element (FE) models. A
design of experiments (DOE) based response surface opti- 1 Introduction
mization method in combination with LS-DYNA is used to
minimize dynamic deflection or acceleration of the back Sandwich panels, such as a honeycomb core with two metal
face plate. Constraints on total mass and on plastic strain facing plates, are finding increasing use over monolithic or
in the face plates are imposed. The mechanism of lower- solid plates in structural design to withstand intense short
ing the backface deflection is by increasing front face plate duration pressure loads. Applications include protection of
thickness which effectively distributes the blast load to a land vehicles, ships or other structures. The cellular core
larger area of the core and avoids local concave deforma- has the ability to absorb the impact energy of the pres-
tion of the front face plate. Further, core depth is increased sure pulse by undergoing large plastic deformation at almost
constant nominal stress. This characteristic of the cellular
core results in significant reduction in the backface acceler-
ation and hence mitigates the damage causing potential of
S. K. Nayak A. K. Singh
the blast impulse. Though metal sandwich panels have been
The Pennsylvania State University, University Park, PA, USA
used for a long time in aircraft and other light weight struc-
S. K. Nayak tures to maximize the bending stiffness per unit density,
Brookhaven National Laboratory, Upton, NY, USA only recently have researchers investigated the possible use
of sandwich panels for blast protection. In sandwich panels,
A. D. Belegundu (B)
Mechanical and Nuclear Engineering, The Pennsylvania State studies involving one or at most two parameters at-a-time
University, University Park, PA, USA have been carried out. However, optimization with several
e-mail: [email protected] variables is necessary to capture interacting physics which
C. F. Yen
include: (1) too thin a front face plate will result in sig-
Army Research Laboratory - WRMD, nificant concave deformation under load that will increase
Aberdeen Proving Ground, MD, USA the momentum on the structure which is detrimental, (2)
 Form Approved
Report Documentation Page OMB No. 0704-0188

Public reporting burden for the collection of information is estimated to average 1 hour per response, including the time for reviewing instructions, searching existing data sources, gathering and
maintaining the data needed, and completing and reviewing the collection of information. Send comments regarding this burden estimate or any other aspect of this collection of information,
including suggestions for reducing this burden, to Washington Headquarters Services, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, Arlington
VA 22202-4302. Respondents should be aware that notwithstanding any other provision of law, no person shall be subject to a penalty for failing to comply with a collection of information if it
does not display a currently valid OMB control number.

1. REPORT DATE 3. DATES COVERED

2. REPORT TYPE
2012 00-00-2012 to 00-00-2012
4. TITLE AND SUBTITLE 5a. CONTRACT NUMBER
Process for design optimization of honeycomb core sandwich panels for 5b. GRANT NUMBER
blast load mitigation
5c. PROGRAM ELEMENT NUMBER

6. AUTHOR(S) 5d. PROJECT NUMBER

5e. TASK NUMBER

5f. WORK UNIT NUMBER

7. PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION

REPORT NUMBER
Army Research Laboratory - WRMD,Aberdeen Proving
Ground,MD,20783
9. SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10. SPONSOR/MONITORS ACRONYM(S)

11. SPONSOR/MONITORS REPORT

NUMBER(S)

12. DISTRIBUTION/AVAILABILITY STATEMENT

Approved for public release; distribution unlimited
13. SUPPLEMENTARY NOTES
Structural and Multidisciplinary, Optimization, vol 46, no. 6, December 2012
14. ABSTRACT
A general process for optimization of a sandwich panel tominimize the effects of air blast loading is
presented here. The panel geometry consists of two metal face plates with a crushable honeycomb or other
type of core. Optimization is necessary as there is strong coupling between the several variables and the
physics, which makes parametric studies relatively ineffective. Virtual testing is used to develop a
homogenized model for the stress?strain curve of the honeycomb core, which can be readily applied to
other types of cellular core. The homogenized model has been validated by comparison to existing results
as well as to results from detailed finite element (FE) models. A design of experiments (DOE) based
response surface optimization method in combination with LS-DYNA is used to minimize dynamic
deflection or acceleration of the back face plate. Constraints on total mass and on plastic strain in the face
plates are imposed. The mechanism of lowering the backface deflection is by increasing front face plate
thickness which effectively distributes the blast load to a larger area of the core and avoids local concave
deformation of the front face plate. Further, core depth is increased which increases panel stiffness. For
acceleration minimization results again produce a stiffer front face plate, but accompanied by a sufficiently
soft core. The mechanism of lowering the backface acceleration is by absorbing energy with low
transmitted stress. A clear cut comparison between monolithic metal plates and sandwich plates, for the
same loading and failure criteria, is presented here.
15. SUBJECT TERMS

16. SECURITY CLASSIFICATION OF: 17. LIMITATION OF 18. NUMBER 19a. NAME OF
ABSTRACT OF PAGES RESPONSIBLE PERSON
a. REPORT b. ABSTRACT c. THIS PAGE Same as 16
unclassified unclassified unclassified Report (SAR)
Standard Form 298 (Rev. 8-98)
Prescribed by ANSI Std Z39-18
 S.K. Nayak et al.

front and back face plate thicknesses must ensure proper performed similar tests on aluminum foam core sandwich
load transfer to the core to enable crushing, (3) the core panels with similar results. Main and Gazonas (2008) inves-
itself must be stiff enough to minimize peak backface dis- tigated the uniaxial crushing of a cellular sandwich plate
placement and soft enough to crush and absorb energy to subjected to air blast. The study showed that the shock trans-
minimize acceleration, (4) total mass and space or envelope mission can be reduced by suitably distributing the mass
constraints must be satisfied, and (5) face plates must be among the face plates and the core for a given mass of
thick enough to maintain integrity. the sandwich. The study considered the effect of face plate
In this paper, a process is presented which enables simul- thicknesses and core depth but did not consider parame-
taneous consideration of several variables relating to core ters related to the core such as core density. They showed
and face plate geometry. A key step in the process is a that prior to densification, the core provides structural sup-
validated approach to generating a homogenized model for port to the front plate and regulates the stress transferred
the honeycomb core. A design of experiments (DOE) based to the back face plate. Once the onset of core densification
response surface optimization method in combination with starts, higher stresses are transferred to the back face plate
LS-DYNA is used to minimize dynamic deflection or accel- (Main and Gazonas 2008; Chi et al. 2010). Zhu et al. (2008,
eration of the back face plate. Constraints on mass and 2009) presented a limited optimal design study of the hon-
on plastic strain in the face plates are imposed. Further, a eycomb sandwich panel and showed that there exists an
clear cut comparison between monolithic metal plates and optimum core density and core depth to minimize the sand-
sandwich plates for the same loading and failure criteria is wich deflection. Karagiozova et al. (2009) states that the
presented here. optimum sandwich configuration depends upon the applied
The shape of a monolithic aluminum plate to reduce blast load, and an optimum structure compromises between
dynamic displacement under blast was optimized in Argod energy absorption of the core and the load transfer to the
et al. (2010) and Belegundu et al. (2008). For this, a back face plate of sandwich. For the same load, the acceler-
differential evolution optimizer was coupled to LS-DYNA ation of the back face plate also depends upon the mass of
to minimize the plates peak displacement subject to limits the back face plate.
on mass and on peak plastic strain; the shaped plate per- It is relevant to also cite papers that address the stress
formed very well compared to a flat plate of equal mass. strain response of honeycomb sandwich panels, as these
Xue and Hutchinson (2004) compared the performance of have helped to validate portions of the homogenization pro-
sandwich panels (such as pyramidal truss core, square hon- cess in this paper. Yamashita and Gotoh (2005) studied
eycomb and folded plate) to a monolithic plate of equal the impact behavior of honeycomb cells through numerical
weight for blast resistance. They found that square hon- simulations and experiments. Numerical simulation using
eycomb and folded plate outperformed the pyramidal truss a single Y cross-sectional unit cell model predicted the
core, but all three sandwich panels were capable of offering crush behavior quite well compared to experiments with
higher blast resistance compared to the monolithic plate. drop hammer velocity of 10 m/s. Highest crush strength
Their optimization study did not consider material failure, per unit mass was obtained when cell shape is a regular
shape and certain core parameters. Fleck and Deshpande hexagon. Wierzbicki (1983) derived a simple formula from
(2004) developed an analytical methodology to analyze the the basic principles of material continuity and plasticity for
dynamic response of metallic sandwich beams subject to calculating the mean crush strength of metal honeycombs
both air and water blasts. Their finding on the basis of sim- in terms of the cell diameter, foil thickness and the flow
ple analytical formulas matched well with the result from stress. The derivation is given for a general shape, and is
Xue and Hutchinsons (2004) three-dimensional FE calcu- then specified for a regular hexagon cell. The result from
lations. Yen et al. (2005) carried out both experimental and this analytical solution is well matched to the experimen-
computational analyses to study the effect of honeycomb tal results. Zhang and Ashby (1992) analyzed the collapse
crush strength on the dynamic response of a honeycomb behavior of the honeycomb under both axial compression
core sandwich pendulum system. The result indicated that and in-plane shear load. Buckling, debonding and fracture
total impulse of the system increased due to concave defor- are identified as possible collapse mechanisms. For flexible
mation of the front face plate, and that significant reduction honeycombs such as those made from Nomex, buckling and
in maximum stress amplitude propagating within the core fracture are dominant mode of failure in simple axial com-
can be achieved by suitable selection of honeycomb mate- pression test, but for rigid-plastic honeycombs (made from
rial with proper crush strength. Further, suitable shape of aluminum), buckling and plastic yielding dominates. Depth
the front face plate can reduce the concave deformation of the honeycomb has no effect and cell angle has little
and hence the total blast impulse. Numerical analyses car- effect on out-of-plane strengths (compressive and shear).
ried out in LS-DYNA using ConWep air blast function These strengths are highly sensitive to the density of the
validated the experimental result. Hanssen et al. (2002) honeycomb. It is also found that out-of-plane loading has
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

little effect on in-plane failure and vice versa. Wu and Jiang and a FE model of the cell is developed, (2) geometrical
(1997) performed both quasi-static and high speed impact parameters associated with the unit cell are identified, (3)
(up to 28.14 m/s) crush test on six types honeycomb cel- nonlinear virtual testing is carried out for different values of
lular structure. They mentioned that small cell size, short the geometrical parameters, followed by curve fitting which
height honeycomb made from high strength material has parametrizes the stressstrain curve in terms of the geomet-
high energy absorbing capacity. rical parameters, (4) validation of the homogenized model
both by detailed FE modeling of the unit cell and/or of a test
specimen and by comparison to any existing results in the
literature. Since the actual blast panel undergoes compres-
2 Homogenization of honeycomb structure sion as well as bending, validation must include different
via nonlinear virtual testing types of virtual tests. Following this, optimization of the
geometry of both the honeycomb core and of the face plates
It is necessary to homogenize the honeycomb core, as a is carried out. These four steps are described below.
detailed finite element model of a sandwich panel with a
core will require a high density mesh to capture the cyclic
plastic buckling or folding deformation of the core accu- 2.1 Unit cell
rately, entailing enormous computing time for each analysis.
Moreover, optimization involves iterative analysis. Homog- Figure 1 shows the hexagonal cell structure, its unit cell, and
enization will allow the honeycomb to be replaced by a 3-D its FE model. The simplest repeating unit in this structure is
continuum structure which can then be modeled using, say, a Y shape, which is taken to be the unit cell (Yamashita
eight-noded hexahedral elements. A process is given here and Gotoh 2005). The regular hexagon cell, branch angle =
which can be used to homogenize any structural concept for 120 , is considered here as it gives the highest crush strength
the core. The main steps are: (1) a unit cell is identified per unit mass (Yamashita and Gotoh 2005). The unit cell

(a) (b) (c)

(d) (e)
Fig. 1 Numerical model of honeycomb core (a) Honeycomb cell geometry, (b) Unit cell, (c) Boundary conditions on the unit cell, (d) FEM in
LSDYNA and (e) Adhesive in model
 S.K. Nayak et al.

Table 1 Material properties of

the honeycomb unit cell model Material Density Youngs modulus Yield Stress Tangent modulus Poisons
(kg/m3 ) (GPa) (MPa) (MPa) ratio

Foil-AL5052 2680 72 300 50 0.34

Adhesive 2000 5 30 0 0.3
Drop Hammer-Steel 288 105 200 0.24

has one double wall and two single walls. The double wall hammer is used to crush the honeycomb. Material proper-
consists of two layers of foil glued together by an adhesive. ties are given in Table 1. To replicate the actual test, the
LS-DYNA is used to carry out the virtual simulation. The rigid surface is modeled using the rigid shell element and
foil is modeled by quadrilateral Belytschko-Tsay shell ele- the mechanical properties are defined as that of steel but
ments, and the 0.01 mm thick layer of adhesive at the double with a high fictitious density. AL5052 aluminum alloy with
wall is modeled by solid elements. Symmetric boundary bilinear isotropic-hardening elastoplastic material model is
conditions are applied along all the edges of the foil, bottom used for the foil. Since the yield and ultimate strength of
areas are fixed and displacement load (crushing) is applied the AL5052 foil are very close, bilinear elastoplastic mate-
to an external rigid surface via a drop hammer which hits rial model with very low tangent modulus is a reasonable
the top areas and moves with them. The main role of the approximation. The adhesive is modeled as perfectly plas-
top and bottom face plate is to contain the crushed hon- tic. Automatic single surface contact is applied to the model
eycomb foil. In the actual mechanical test, a heavy steel with sliding and sticking frictional coefficients equal to 0.2

(a) (b) (c) (d)

Densification
D = 9.52 mm, t = 0.06 mm, h = 20mm stress/strain
6.0
Nominal Compressive stress (MPa)

5.0 Peak Simplified

stress/strain Actual
4.0

3.0

Mean crush
2.0 stress

1.0

0.0
0.0 0.2 0.4 0.6 0.8 1.0
Volumetric Strain
(e)

Fig. 2 Honeycomb core crushing (a)(d) Different stages of honeycomb unit cell crushing and (e) obtained load curve and its different parameters
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

25
and 0.3, respectively. These mechanical properties of the
foil and the adhesive, and friction coefficient values are
20
obtained from the literature (Yamashita and Gotoh 2005).
Crush Stress (MPa)

All the tests are carried out at a hammer speed of 80 mm/s

15 Virtual test . along the depth of the honeycomb, which can be consid-
fit
ered as quasi-static tests since impact tests involve very high
10 hammer speeds in the order of m/s.
Crush strength = 2296(t/D)1.49

5
2.2 Geometrical parameters related to unit cell
0
0.00 0.01 0.02 0.03 0.04 0.05 The following parameters of the honeycomb cell geometry
t/D
were considered: foil thickness t, cell size D which equals
Fig. 3 Variation of crush strength with t/D obtained from virtual the distance between opposite walls of the honeycomb cell,
testing
and core depth, h. Preliminary FE runs show that the stress
strain or load curve does not show any visible change with
core depth h (Zhang and Ashby 1992), provided that it is
not too small so as not to allow folds to occur during crush-
ing. Further the ratio t/D is the defining parameter that
35.0 characterizes the load curve. This has also been reported
30.0 in Wierzbicki (1983). The density is linearly proportional to
Crush Stress (Pa)

25.0 Virtual test

t/D, as will be given below.
Wierzbicki
20.0

15.0 2.3 Nonlinear virtual testing and parametrization of the

stressstrain curve
10.0

5.0 Figure 2 shows a typical crushing phenomenon and load

0.0 curve obtained from the FE test. As the hammer travels,
0.00 0.01 0.02 0.03 0.04 0.05 buckling (Yamashita and Gotoh 2005; Zhang and Ashby
t/D 1992) of the foil starts from near the impact edge and propa-
Fig. 4 Validation of the crush strength obtained from the virtual test gates downward. Compressive stress is obtained by dividing
the reaction force experienced by the hammer by the trian-
gular unit cell area, and volumetric strain is calculated from
change in core depth divided by its original value. The core

Fig. 5 Detailed model of test specimen

 S.K. Nayak et al.

5
resists buckling until the peak stress point. Onset of buck-
4.5
ling causes a sudden drop in the compressive stress (Fig. 2a).
Compressive stress drops until the first folding of the cell 4

Crush stress (MPa)

3.5 No face plate-large scale (1.51)
wall is complete (Fig. 2b) and then stress increases. This Face plate (1.47)
goes on, although the peaks are very small in comparison 3
to the first peak, until the whole core is folded. The crush 2.5
strength (i.e. the plateau stress level) is the average of the 2
oscillatory stress during the cyclic collapse of the foil. Once 1.5
the entire core is folded, densification starts resulting in very 1
high compressive stress. Large amount of energy gets dis- 0.5
sipated through the plastic deformation of the cell wall at 0
each folding. The energy absorbed per unit area of the core 0 0.2 0.4 0.6 0.8 1
Strain
is essentially equal to the area under the load curve multi-
plied by core depth. Although sufficient care has been taken Fig. 6 Comparison of load curves from unit cell and detailed model
in approximating the load curve, it is not possible to define on a virtual test specimen. Numbers in the parenthesis is the average
crush strength
the crush start and end strain very accurately.
By repeated virtual tests for different values of t/D, and
using curve fitting (Fig. 3 shows this for crush strength), we
obtain the following equations that parameterize the load The analytically derived formula in (2) shows good agree-
curve in terms of honeycomb geometry. Below, f = den- ment with the corresponding equation obtained from virtual
sity of foil material in kg/m3 and S y = yield stress of the foil testing in (1) as shown in Fig. 4. Secondly, the load curve
material (aluminum, here). Mass density refers to the mass obtained from the model in Fig. 1 is compared to that
per unit cell area and per unit depth. Yield stress and density obtained from a detailed model of a test specimen (Fig. 5).
refer to that of AL5052 material. Crush start and end strain The load curves match well (Fig. 6). Here, various bound-
do not change much with t/D and corresponding strain val- ary conditions have been virtually tested including random
ues are taken as 0.009 and 0.744 respectively. Final strain is imperfections in geometry, all producing good agreement
fixed at 0.85. with the baseline test reported above.
Thirdly, since the homogenized model is based on crush-
Mean crush stress (MPa) = 2296 (t/D)1.49 ing, and the blast panel is subject to both bending and
Peak strain = 0.0626 (t/D) + 0.0035 crushing, validation via both bending and crushing virtual
Peak stress (MPa) = 845.8 (t/D) 0.959 were performed. A 3-point bend test comparing the homog-
enized model above with a detailed virtual test specimen
Youngs modulus = Peak stress/Peak strain was carried out (Fig. 7). Provisions of ASTM standards D
Crush start strain = 0.009 7250 and C 393 are used (ASTM 2006a, b). For this test,
t/D = 0.00754 is used. The response matches well for
Crush end strain = 0.744
this bending application as well (Fig. 8). Comparison of a
Densification strain = 0.85 beam with span of beam perpendicular to the ribbon was
Densification stress (MPa) = 35592 (t/D)1.69 also carried out with good agreement. Further, energy cal-
culations were also in good agreement. Many details related
8S y
Tensile stress cutoff (MPa) = (t/D)
3
  8
f
Mass density kg/m3 = (t/D) (1)
3
Relations in (1) provide a homogenized model for the
honeycomb core.

2.4 Validation of the homogenized model

Several validation techniques are discussed here. Firstly,

Wierzbicki (1983) showed that the mean crush strength may
be given by
 5/ Fig. 7 Honeycomb ribbon orientation parallel to span of sandwich
Mean cr ush str ength = 16.565 y t/D 3 (2) beam-detailed model
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

Fig. 8 Maximum deflection of

the sandwich vs. contact load
when honeycomb ribbon is
parallel to the span of sandwich
beam

to loading, boundary conditions and material properties in Two separate objectives are considered for minimization:
this beam test are omitted for brevity (Singh 2011), as the one is the peak displacement magnitude and the other is the
focus is on showing a process by which we can be confident rigid body acceleration, of the back face plate. Constraints
of using the homogenized model for optimization. are: limits on total mass and on plastic strain limit in the
face plates, the latter to ensure structural integrity. Thus, we
have the problem

3 Optimization problem formulation Minimize max b (case 1)

max b (case 2)
Optimal design of a square honeycomb core sandwich panel subject to pj pmax for each element j in the
subjected to air blast loading is considered (Fig. 9). The (3)
face plates
panel is freely suspended. The role of the stiffener at the top M Mmax mass limit
is to impose high inertia to the back face plate and hence xL x xU bounds on variables
the sandwich, permitting compression of the core. This set
of boundary conditions are motivated by design problems in In case-1, since the sandwich model is not constrained, the
vehicle protection and approximate commonly used exper- maximum Z -deflection of the back face plate is obtained
imental fixtures. A study of different boundary conditions by subtracting the rigid body displacement of the stiffener
on monolithic plates, in contrast with sandwich panels as as b = Z _ max stiffener . Displacements along the x-
considered here, was previously conducted in Argod et al. and y- direction are not significant and are not considered.
(2010). The displacement is a function of time, and the peak value
In general, design variables include thicknesses of face is monitored. In case-2, the absolute maximum rigid body
plates, core depth, core t/D or equivalently core density, acceleration of back face plate along z-direction is consid-
and bulge magnitudes of the face plates (Fig. 10). Thus, x = ered. Like displacement, accelerations along the x- and y-
{tb , t f , h, t/D, sb , s f }. direction are not significant. Rigid body acceleration refers

Fig. 9 (a) Honeycomb

sandwich panel (b) Exploded
view of the honeycomb core
sandwich model used for
optimization study

(a) (b)
 S.K. Nayak et al.

maintaining equal spacing, to result in the desired thick-

nesses. For variables sb and s f , which are the amplitudes
of basis shapes that correspond to bulges in the face plates
as shown in Fig. 10, a procedure is needed to generate qi .
The velocity fields qi are generated here by first applying
a dummy load on each of the plate surfaces and using ana-
Fig. 10 Optimization parameters for the sandwich optimization lytical expressions in Timoshenko and Woinowsky-Krieger
(1959), Eq. 145, page 142, to obtain the corresponding
nodal displacements on all nodes on the surface. Then, equal
spacing in the z-direction is used to complete the definition
of qi . A square portion (1.016 m 1.016 m) at the cen-
to net force divided by net mass. The plastic strains in the
ter of the face plates is taken as the domain for applying
face plates increase with simulation time until a plateau is
velocity fields. Asymmetric shape variations are not nec-
reached and this saturated or maximum value is considered.
essary in this problem owing to centrally located charge.
The maximum value of the all elements in the face plates is
Lower bounds sb and s f are zero implying that only con-
monitored. The mass M is the sum of the mass of the front
vex (outward) bulges are allowed for shape changes, as this
and back face plates and of the core as M = M f + Mb + Mc .
has found to be beneficial in deflecting the waves in the
Mass of the stiffener remains constant during optimization
monolithic plate (Argod et al. 2010; Belegundu et al. 2008).
and is not included in M.
As variables x = {tb , t f , h, t/D, sb , s f } are changed
Optimization is carried out in the following sequence: (I)
for each sampling point, the grid point coordinates G, the
optimization for case-1 considering only {tb , t f , h, t/D}
load curve from (1), and thickness values are updated, an
as variables, (II) optimization for case-1 considering all six
LS-DYNA input file is then written, and an analysis is car-
variables {tb , t f , h, t/D, sb , s f }, and (III) optimization for
ried out to evaluate the various functions in the optimization
case-2 considering only {tb , t f , h, t/D}.
problem. The process is schematically given as:
For brevity in the figure captions, we refer to {tb , t f ,
h, t/D}-optimized panel as size-optimized panel, and {tb ,
Use current values of {tb , t f , h, sb , s f } at sampling point
t f , h, t/D, sb , s f }-optimized panel as, size+shape opti-
xk , construct G using {qi }; from current value of {t/D},
mized panel.The variable t/D is a parameter that defines
construct the stressstrain curve data using (1) Write
the stressstrain curve which is input into LS-DYNA. The
input file and perform LS-DYNA analysis Evaluate
remaining five variables {tb , t f , h, sb , s f } require a
objective and constraint functions Create response
change in the coordinates of the nodes in the 3D finite
equations Run optimizer
element model. That is, they affect the shape of the struc-
ture. The key equation to implement shape optimization is
(Belegundu and Rajan 1988) The DOE response surface method is used to solve this
problem, implemented using Design Expert, a commer-

Ndv cially available software. Using the response equations,
G (x) = Goriginal + xi qi (4) optimization is carried out by fmincon, a gradient based
i=1 optimizer in MATLAB optimization toolbox. Central com-
posite face centered (CCF) method is adopted to create
where G is a grid point coordinates vector, representing x-, design points. It uses three levels for each factor. Lower
y-, z- coordinates of all nodes in the model. Each xk repre- design limits are decided in such a way that FE evaluation
sents the amplitude of a permissible shape change vector is feasible at all sampling points and to avoid high aspect
or what is commonly called a velocity field vector qk . ratios in the hexahedral elements. Also, appropriate upper
Velocity fields have nothing to do with actual velocities of
the model under loading. Vectors {qk } are generated just
once in the optimization procedure. Goriginal is the current
(flat) shape. Visualization of a {qi } is identical to visual- Table 2 Design limits for {tb , t f , h, t/D}-optimization, min. b
ization of a displacement field in finite elements: {qk } is
Design limits Lower limit Upper limit
multiplied by a magnification scalar and added to the cur-
rent grid to obtain a displaced grid, except that here the Front face plate thickness (mm), t f 4.4 18
displaced grid represents a new shape and is called a basis Core depth (mm), h 280 500
shape. Foil thickness/cell size, t/D 0.018 0.046
Velocity fields associated with variables tb , t f , h are Back face plate thickness (mm), tb 4.4 10
straightforward: nodes are moved in the +/ z-direction,
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

Table 3 Optimization results for {tb , t f , h, t/D}-optimization with

varying mass limits, min. b (units: mm) back face plate and the stiffener are defined using *CON-
TACT features and *TIED_SURFACE_TO_SURFACE_ID
Sandwich Optimized parameters pmax card. After a mesh convergence study a 38 38
mass (kg)
tf h tb (t/D) b (Obj) 2 and 28 28 3 element mesh is taken in x-y
plane for the face plates and the core, respectively. The
130 8.07 300.6 4.4 0.0252 24.3 0.0384 *MAT_PLASTIC_KINEMATIC material model is used for
140 7.90 359.9 4.4 0.0238 19.96 0.0373 the face plates and *MAT_CRUSHABLE_FOAM material
150 7.55 401.6 4.4 0.0240 16.92 0.033 model is used for the core. Material properties used for the
160 8.4 390.2 4.4 0.0263 14.65 0.023 face plate are same as that of the foil used for honeycomb
170 9.12 388 4.4 0.0282 12.43 0.017 virtual testing (Table 1). The mechanical properties of the
core are defined in (1), which are functions of t/D. Poissons
ratio is taken as zero for the core. The *LOAD_BLAST
input parameters used are equivalent TNT mass (8 kg), type
design limits are taken so that better response fit is possi- of blast (air blast-spherical charge), load curve, charge loca-
ble and at the same time giving more design space to the tion (0, 0, 0.4064 m). The exposed bottom surface of the
optimizer to find the minima. Design limits for the different plate is defined as the area on which blast load applied. A
cases are given in Tables 2, 3 and 4 below. Some results low value of damping 1e-9 and a reduced tssfac help in
have been cross-checked by using the differential evolution smoother convergence in LS-DYNA.
(DE) technique which is computationally more expensive.

5 Optimization results

4 FE modeling using LS-DYNA Results are discussed here for different mass limits for the
sandwich, while detailed response is given for the 150 kg
In this study, aluminum AL5052 is considered for the entire mass limit. The optimization study is carried out for a fixed
sandwich such that results can be easily compared with amount of charge, viz. 8 kg TNT. When the DOE response
AL5052 monolithic plate considered earlier (Argod et al. surface optimizer provides an optimum set of design vari-
2010; Belegundu et al. 2008). In the monolithic plate prob- ables, LS-DYNA is executed, and the resulting response
lem, various boundary conditions were considered giving values are used in the tables below.
similar optimum shapes (Argod et al. 2010). The model is
free to move in space, which approximates to some extent 5.1 Optimization results for minimum backface
commonly used experimental fixtures. As a consequence of displacement, b
this, the back face plate is not restrained and can deform
freely without creating unrealistically high plastic strain. We note that b refers to the peak in the displacement-time
The role of the stiffener at the top is to impose high inertia response of the backface relative to the stiffener. Table 5
to the back face plate and hence the sandwich, permitting shows the {tb , t f , h, t/D}-optimized sandwich panel param-
compression of the core. Mass of the stiffener is 1,850 kg. eters for different mass limits. The pmax is always at
High fictitious density is defined for the stiffener. The con-
tacts between the face plate and the core, and between the

Table 5 Optimization results for {tb , t f , h, t/D, sb , s f }-optimization

with varying mass limits, min. b (units: mm)
Table 4 Design limits for {tb , t f , h, t/D, sb , s f }-optimization, min.
b Sandwich Optimized parameters pmax
mass (kg)
Design limits Lower limit Upper limit tf h tb (t/D) sf sb b
(Obj)
Front face plate thickness (mm), t f 4.4 10
Core depth (mm), h 150 500 130 4.4 150 4.4 0.0364 42.41 0 20.76 0.031
Foil thickness/cell size, t/D 0.018 0.046 140 4.4 150 4.4 0.0371 52.69 0 17.04 0.020
Back face plate thickness (mm), tb 4.4 10 150 4.4 150 4.4 0.0377 62.96 0 14.4 0.014
Front face plate bulge (mm), s f 10 100 160 4.4 150 4.4 0.0384 73.24 0 12.25 0.012
Back face plate bulge (mm), sb 0 100 170 4.4 150 4.4 0.039 83.52 0 10.73 0.001
 S.K. Nayak et al.

Table 6 Design limits for {tb , t f , h, t/D}-optimization, min. ab

Fig. 12a, plastic strains in Fig. 12b and Z -momentum in
Design limits Lower limit Upper limit Fig. 12c. Compared to the uniformly thick aluminum panel,
the sandwich panel shows significant reduction. Shape
Front face plate thickness (mm), t f 15 25 optimization of the sandwich adds to this improvement.
Core depth (mm), h 120 400 However, a shape-optimized all-aluminum panel scored a
Foil thickness/cell size, t/D 0.00754 0.025 victory as compared to the sandwich (11.5 mm to 14.4 mm
Back face plate thickness (mm), tb 8 25 deflection). A physical explanation for this is given subse-
quently. In the sandwich, the mass fraction in the front face
plate, core and the back face plate are 0.20, 0. 68 and 0.12,
respectively, for the {tb , t f , h, t/D}-optimized panel and
0.48, 0.40 and 0.12, respectively, for the {tb , t f , h, t/D, sb ,
the center of the front face plate. With increase in mass, s f }, optimized panel.
b decreases. The optimizer produces thicker and denser
(higher t/D) core. Mass is always active and the plastic
strain is active only for 130 kg and 140 kg.
Table 6 shows the {tb , t f , h, t/D, sb , s f }-optimized sand- 6 Mechanism causing the improvement in backface
wich panel parameters for different mass limits. The pmax deflection
is always in the back face plate. With increase in mass, b
decreases. The optimizer produces a denser or higher t/D In the {tb , t f , h, t/D}-optimized panel, the optimizer
core. The tb is always at lower limit (4.4 mm) for the rea- has chosen a larger core depth which increases the overall
son mentioned in Section 6. The t f is always at lower limit stiffness of the sandwich and lowers b . The tb is always
(4.4 mm) as the front face plate thickness re-emerges in the at lower limit (4.4 mm) as it is sufficient to keep pmax
bulge. The bulge also reduces the effective blast load on the below the strain limit (0.038). The t f is decided based on
panel by deflecting the blast. The bulge also provides higher three things, viz. to keep pmax below the strain limit, to
stiffness to the panel, compensating for the loss in stiffness provide sufficient stiffness for the front face plate to trans-
due to lower core depth. With increase in mass, t/D and s f fer the blast pressure load to a larger area of the core, and
increase consistently. Mass is always active and the plastic to reduce local concave deformation of the front face plate
strain is always inactive. which will increase the impulse.
Now, focusing on the 150 kg mass limit, Fig. 11 shows Compared to the {tb , t f , h, t/D}-optimized panel above,
the performance of the optimized sandwich panel and the {tb , t f , h, t/D, sb , s f }-optimized panel has further
how it compares with an all-aluminum monolithic plate. reduced the deflection by 14.9 %. This may be attributed
A non-optimized uniformly thick aluminum panel is also to the following mechanisms. A lesser core depth has been
included in the comparison for reference purposes. The cor- produced but with a bulge in the front face plate. The core
responding backface deflection-time responses are shown in is denser. The bulge deflects the blast wave and reduces

Fig. 11 Optimized sandwich

panel and shape optimized
monolithic plate of 150 kg mass
for minimizing b (stiffener not
shown)
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

Fig. 12 Comparison for 60 Uniformly thick solid plate

minimum b for 150 kg mass
Shape optimized solid plate
limit: (a) b , (b) pmax and (c) 50
total Z-momentum. Note: by Size optimized sandwich panel
solid plate is meant a monolithic Size and shape optimized sandwich panel
40
all-metal plate

b (mm)
30

20

10

-10
0 2 4 6 8 10
Time (ms)
(a)

Uniformly thick solid plate

Shape optimized solid plate
Size optimized sandwich panel
0.05 Size and shape optimized sandwich panel
Failure strain limit
0.04

0.03
pmax

0.02

0.01

0
0 2 4 6 8 10
Time (ms)
(b)

6000

5000
Total Z-momentum (Ns)

4000
Uniformly thick solid plate
3000 Shape optimized solid plate
Size optimized sandwich panel
2000 Size and shape optimized sandwich panel

1000

0
0 2 4 6 8 10
Time (ms)
(c)

the impulse imparted to the structure, and it also increases plate stiffer, which helps to transfer the blast load to a large
the moment of inertia of the sandwich at its center where area of the core thereby reducing local deformation at the
blast load is maximum. The bulge also makes the front face center.
 S.K. Nayak et al.

Table 7 Optimization results for {tb , t f , h, t/D}-optimization with

varying mass limits, min. ab (units: mm, m/s2 )
Focusing on the 150 kg mass limit, Fig. 13 shows the
performance of the optimized sandwich panel and how it
Sandwich Optimized parameters pmax compares with an all-aluminum monolithic plate. The corre-
mass (kg)
tf h tb (t/D) ab (Obj) sponding acceleration-time responses are shown in Fig. 14a.
{tb , t f , h, t/D}-optimized sandwich panel results in 84 %
130 17.94 194.93 10.77 0.00754 2.77 104 0.0355 (f) reduction in ab compared to a uniformly thick monolithic
140 17.79 196.84 13.4 0.00754 2.22 104 0.0360 (f) plate of equal mass. The location of the pmax is at the center
150 17.63 199.22 16.02 0.00754 1.84 104 0.0371 (f) of the front face plate.
160 17.45 202.16 18.64 0.00754 1.57 104 0.0388 (f) Now, comparing the optimized sandwich to an optimized
170 17.45 205.2 21.1 0.00754 1.37 104 0.0388 (f) all-aluminum monolithic panel, the sandwich scores a clear
victory (compare 1.84 104 to 8.88 104 m/s2 ). Recall
that the victor was reversed for b minimization. The mono-
lithic plate has no mechanism for energy absorption. The
mass fraction in the front face plate, core and the back face
In the all-aluminum monolithic plate, more mass is avail- plate are 0.47, 0.11 and 0.42 respectively.
able to go in terms of bulge, whereas in the sandwich,
mass is distributed between face plates and core. Higher
the bulge height towards the blast, greater is the reduc-
tion in impulse and hence greater the reduction in back- 7 Mechanism causing the improvement in backface
face deflection. This aspect and higher rigidity makes the acceleration
monolithic plate slightly better than the sandwich panel as
shown in Fig. 11a. A paper by Avalle et al. (2001) explains the relation between
energy absorption and transmitted force. We note that ab is a
6.1 Optimization results for minimum backface direct measure of the transmitted force, and that core density
acceleration, ab and t/D are linearly
 related. In Fig. 15, the area under
the three curves d are the same. The area represents
Table 7 summarizes the {tb , t f , h, t/D}-optimized results energy per unit volume, and b denotes the corresponding
for different mass of the sandwich. With increase in mass, stress value, which is here the compressive stress in the core
ab decreases. It shows that both h and tb increases with elements in contact with the back face plate. For low density
sandwich mass, but t f does not vary much. The core density core 1 , the deformation is high and the stressstrain state is
(proportional to t/D) always remains at the lower bound. in the densification zone with high stress. For high density
The pmax is very close to the limiting value of 0.038 for core 3 , the energy is absorbed with low deformation and
all the mass limits. It can be observed that with increase in high stress. The medium density foam 2 is optimum as b
sandwich mass, the optimizer adds mass more to the back is lowest. In fact, an efficiency index has been defined in
face plate than to the core; ab not only depends upon the (Avalle et al. 2001) as
force transmitted to the back face plate but also on the back 
face plate mass (Main and Gazonas 2008; Karagiozova et al. d
= (5)
2009). b

Fig. 13 Optimized sandwich

panel and shape optimized
monolithic plate of 150 kg mass
for minimizing ab (stiffener not
shown)
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

Fig. 14 Comparison for 1.E+05

minimum ab for 150 kg mass
limit: (a) ab , (b) pmax and (c) 1.E+05
total Z-momentum Uniformly thick soild plate
8.E+04 Shape optimized solid plate
Size optimized sandwich panel
6.E+04

ab (m/s2)
4.E+04

2.E+04

0.E+00

-2.E+04

-4.E+04
0 2 4 6 8 10
Time (ms)
(a)

Uniformly thick solid plate

0.05 Shape optimized solid plate
Size optimized sandwich panel
Plastic strain limit
0.04

0.03
pmax

0.02

0.01

0
0 2 4 6 8 10
Time (ms)
(b)
6000

5000
Total Z-momentum (Ns)

4000
Uniformly thick solid plate
Shape optimized solid plate
3000 Size optimized sandwich panel

2000

1000

0
0 2 4 6 8 10
Time (ms)
(c)

Higher efficiency implies that the core absorbs a given t/D and core depth have been adjusted so that even though
amount of energy with lesser transmitted force. A look at the maximum strain in front face elements have gone to
the results show that optimization has exploited this physics: the densification region (strain > 0.744), the strain in the
 S.K. Nayak et al.

distributes the blast load to a larger area of the core and

avoids increase in impulse stemming from local concave
3 2
3 > 2 > 1 1 deformation of the front face plate. Further, core depth
is increased which increases panel stiffness. In the sand-
wich, the mass fraction in the front face plate, core and
b
the back face plate are 0.20, 0. 68 and 0.12, respectively,
for the {tb , t f , h, t/D}-optimized panel. Interestingly, for
the same mass, the shape-optimized monolithic panel is
more effective than the optimized honeycomb core sand-
wich panel (11.5 mm vs. 14.4 mm). In the all-aluminum
monolithic plate, more mass is available to go in terms of
bulge, whereas in the sandwich, mass is distributed between
face plates and core. Higher the bulge height towards the
blast, greater is the reduction in impulse and hence greater
Fig. 15 Stress (at the backface) for different density foams corre- the reduction in backface deflection. This aspect and higher
sponding to same energy absorption
rigidity makes the monolithic plate slightly better than the
sandwich panel as shown in Fig. 11a.
Considering acceleration minimization, results produce a
stiffer front face plate, which helps to distribute the crush-
backface elements are well within the crushing zone avoid- ing load to a wider region of the core, and a soft core by
ing complete densification. The t/D has been pushed to its reducing t/D to its minimum value. The mechanism of low-
lower bound. Further, a comparison of between the opti- ering the backface acceleration is by absorbing energy with
mized results for ab vs. b minimization (Table 8) clearly low transmitted stress. In this case, honeycomb core sand-
shows that the optimizer has done its task: when the objec- wich panel proves to be significantly more effective than an
tive is ab , a more efficient core is designed. Of course, optimized monolithic panel (1.84e4 m/s2 vs. 8.88e4 m/s2 ).
optimization also adjusts the mass distribution between face With increase in sandwich mass, the optimizer adds mass
plates and core, subject to constraints on sandwich mass, more to the back face plate rather than the core; accelera-
plastic strain in plates and lower bounds on t/D. tion not only depends upon the force transmitted to the back
face plate but also on the back face plate mass. In the sand-
wich, the mass fraction in the front face plate, core and the
8 Conclusions back face plate are 0.47, 0. 11 and 0.42, respectively.
As noted earlier, Yen et al. (2005) had carried out an
A process for optimizing sandwich panels to mitigate blast experimental and computational study, and concluded that
loading is presented. While honeycomb core has been used significant reduction in maximum stress amplitude propa-
in the study, the process is also applicable to other types gating within the core can be achieved by suitable selection
of cellular core. Two independent design objectives, back- of honeycomb material with proper crush strength. This
face deflection of the plate and backface acceleration are observation has been borne out by the results here, as
minimized, subject to mass and plastic strain constraints. optimization with LS-DYNA achieves a proper balance,
The optimization is carried out using DOE response surface accounting for different interacting physics. A general and
methodology. LS-DYNA is used for finite element simu- efficient design process has been presented.
lations. Virtual testing is used to develop a homogenized Fracture of the face plates themselves has been avoided
model for the stressstrain curve of the honeycomb core, by imposing a limit on maximum plastic strain during opti-
and this model has been validated by comparison to existing mization. However, fracture causing debonding between the
results as well as to detailed FE model results. core and the face plates has not been considered in this
The mechanism of lowering the backface deflection is study. Finally, the work here can be extended to include
by increasing front face plate thickness which effectively strain rate sensitivity in the homogenization model.

Table 8 Comparison of 
honeycomb core energy Case t/D d (Pa) b (average) (Pa)
absorption efficiency
acceleration_ min (ab = 1.84 104 ms2 ) 0.00754 702377.6 1.58 106 44.5 %
deflection_min (ab = 7.25 104 ms2 ) 0.024 569514.2 5.59 106 10.2 %
Process for design optimization of honeycomb core sandwich panels for blast load mitigation

Acknowledgments This material is based upon work partly sup- Hanssen AG, Enstock L, Langseth M (2002) Close-range blast loading
ported by the Army Research Office, Proposal Number 50490-EG, of aluminium foam panels. Int J Impact Eng 27:593618
monitored by Dr. Bruce LaMattina. Partial financial and computational Karagiozova D, Nurick GN, Langdon GS (2009) Behaviour of sand-
support from the High Performance Computing Group at Penn State wich panels subject to intense air blasts-parts 2: numerical
under Mr. Vijay Agarwala is gratefully acknowledged. simulation. Compos Struct 91:442450
Main JA, Gazonas GA (2008) Uniaxial crushing of sandwich plates
under air blast: influence of mass distribution. Int J Solid Struct
45:22972321
Singh AK (2011) MS thesis. Department of Mechanical and Nuclear
References Engineering, The Pennsylvania State University, University Park,
PA 16802
Argod V, Nayak SK, Singh AK, Belegundu AD (2010) Shape opti- Timoshenko SP, Woinowsky-Krieger S (1959) Theory of plates and
mization of solid isotropic plates to mitigate the effects of air blast shells. Engineering societies monographs, 2nd edn. McGraw-Hill,
loading. Mech Base Des Struct Mach 38:362371 New York
ASTM D 7250/D 7250M-06 (2006a) Standard practice for determining Wierzbicki T (1983) Crushing analysis of metal honeycombs. Int J
sandwich beam flexural and shear stiffness Impact Eng 1(2):157174
ASTM C 393/C 393M-06 (2006b) Standard test method for core shear Wu E, Jiang W-S (1997) Axial crushing of metallic honeycomb. Int J
properties of sandwich constructions by beam flexure Impact Eng 19:439456
Avalle M, Belingardi G, Montanini R (2001) Characterization of Xue Z, Hutchinson JW (2004) A comparative study of impulse-
polymeric structural foams under compressive impact loading resistant metal sandwich plates. Int J Impact Eng 30:12831305
by means of energy-absorption diagram. Int J Impact Eng 25: Yamashita M, Gotoh M (2005) Impact behavior of honeycomb struc-
455472 tures with various cell specifications - numerical simulation and
Belegundu AD, Rajan SD (1988) A shape optimization approach experiment. Int J Impact Eng 32:618630
based on natural design variables and shape functions. J Comput Yen CF, Skaggs R, Cheeseman BA (2005) Modeling of shock mit-
Methods Appl Mech Eng 66:87106 igation sandwich structures for blast protection. In: The 3rd
Belegundu AD, Argod V, Rajan SD, Krishnan K (2008) Shape first international conference on structural stability and dynamics,
optimization of panels subject to blast loading modeled with Kissimmee, Florida, 1922 June 2005
LS-DYNA. In: 49th AIAA/ASME/ASCE/AHS/ASC structures, Zhang J, Ashby MF (1992) Buckling of honeycombs under in-plane
structural dynamics, and materials conference, paper AIAA biaxial stress. Int J Mech Sci 34:491509
20082285, Schaumburg, IL Zhu F, Zhao L, Lu G, Wang Z (2008) Deformation and failure of blast-
Chi Y, Langdon GS, Nurick GN (2010) The influence of core height loaded metallic sandwich panels-experimental investigations. Int
and face plate thickness on the response of honeycomb sandwich J Impact Eng 35:937951
panels subjected to blast loading. Mater Des 31:18871899 Zhu F, Wang Z, Lu G, Zhao L (2009) Analytical investigation and opti-
Fleck NA, Deshpande VS (2004) The resistance of clamped sandwich mal design of sandwich panels subjected to shock loading. Mater
beams to shock loading. ASME J Appl Mech 71:386401 Des 30:91100