University of Nebraska - Lincoln

DigitalCommons@University of Nebraska - Lincoln

Mechanical & Materials Engineering Faculty Mechanical & Materials Engineering, Department
Publications of

2013

Impact damage on a thin glass plate with a thin

polycarbonate backing
Wenke Hu
University of Nebraska-Lincoln

Yenan Wang
University of Nebraska-Lincoln

Jian Yu
U.S. Army Research Laboratory, [email protected]

Chian-Fong Yen
U.S. Army Research Laboratory

Florin Bobaru Ph.D.

University of Nebraska at Lincoln, [email protected]

Follow this and additional works at: http://digitalcommons.unl.edu/mechengfacpub

Hu, Wenke; Wang, Yenan; Yu, Jian; Yen, Chian-Fong; and Bobaru, Florin Ph.D., "Impact damage on a thin glass plate with a thin
polycarbonate backing" (2013). Mechanical & Materials Engineering Faculty Publications. 92.
http://digitalcommons.unl.edu/mechengfacpub/92

This Article is brought to you for free and open access by the Mechanical & Materials Engineering, Department of at DigitalCommons@University of
Nebraska - Lincoln. It has been accepted for inclusion in Mechanical & Materials Engineering Faculty Publications by an authorized administrator of
DigitalCommons@University of Nebraska - Lincoln.
 International Journal of Impact Engineering 62 (2013) 152e165

Contents lists available at SciVerse ScienceDirect

International Journal of Impact Engineering

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

Impact damage on a thin glass plate with a thin polycarbonate backing

Wenke Hu a, Yenan Wang a, Jian Yu b, Chian-Fong Yen b, Florin Bobaru a, *
a
Department of Mechanical and Materials Engineering, University of Nebraska-Lincoln, Lincoln, NE 68588, United States
b
U.S. Army Research Laboratory, Aberdeen Proving Grounds, MD 21005, United States

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

Article history: We present experimental and computational results for the impact of a spherical projectile on a thin
Received 5 September 2012 glass plate with a thin polycarbonate backing plate, restrained in a metal frame, or in the absence of the
Received in revised form frame. We analyze the dependence of the damage patterns in the glass plate on the increasing impact
27 June 2013
velocities, from 61 m/s to 200 m/s. Experimental results are compared with those from peridynamic
Accepted 1 July 2013
Available online 13 July 2013
simulations of a simplified model. The main fracture patterns observed experimentally are captured by
the peridynamic model for each of the three projectile velocities tested. More accurate implementation
of the actual boundary conditions present in the experiments will likely further improve modeling of
Keywords:
Glass
brittle damage from impact on a multi-layered system. The peridynamic computational model sheds
Multi-layer light into the early stages of the complex brittle damage evolution in the glass layer, and the influence of
Brittle fracture boundary conditions on the dynamic fracture process.
Impact Ó 2013 Elsevier Ltd. All rights reserved.
Peridynamics

1. Introduction Sun [6] studied the impact of small steel spheres on a range of
glasses and ceramics. They give tables of the threshold velocities at
Due to their low cost of manufacturing, light weight, and high which different damage mechanisms operate. For instance, in
performance, glass laminates are widely used in skylight glazing glasses, single cone cracks can be formed up to 40 m/s, while at
and in windshields for automobile, train, and military vehicle. In higher impact speeds, stress wave damage rather than a ‘quasi-
order to design and improve glass laminates for protective appli- static’ response is obtained. Their research ([4e6]) also identified
cations, it is important to be able to predict the damage and fracture important differences in the response depending on whether the
behavior of glass plates under impact. ratio of the hardness of the projectile to the target was greater or
Experiments have been conducted in the past (see Refs. [1e8]) less than 1.The thickness of the glass plate in relation to the impact
to analyze the behavior of a single glass plate or glass laminates velocity, and the angle of impact are also important factors affecting
subject to low or high impact velocity. Knight et al. [1] found that the progression of damage in glass (see Refs. [1,7]). The picture of
the Hertzian cone angle decreases with increasing impact velocity damage progression is further complicated in multi-layered sys-
for a glass block under small steel spheres impact. Ball and tems due to the stress wave reflections from the various interfaces.
Mckenzie [2] performed a series of impact tests on circular float Multi-layer glass laminates subject to high velocity impact display
glass plates with thicknesses between 3 mm and 12 mm with very complex damage pattern by dynamic brittle fracture through
various impact speed (10 - 50 m/s). They identified a number of the layers. Bless et al. [8] studied the morphology of damage from
failure mechanisms in the plates and constructed a fracture map high velocity impact (1120 m/s) onto a seven-layer glass laminate
which incorporates the effect of plate thickness and impact veloc- with a polycarbonate backing.
ity. Important contributions to experimental analysis of damage in Developing models that can correctly capture all the observed
glass from impact and understanding the mechanisms of brittle features of dynamic fracture and fragmentation from impact on
failure from solid particle impact have been made over several glass plates and laminates has been extremely challenging.
decades by the group at Cavendish laboratory at University of A number of numerical studies based on the finite element method
Cambridge (see Ref. [3]). For example, Field ([4,5]) and Field and (FEM) or molecular dynamics [9e12] have been performed to
compute the response of glass plates and laminates under impact
loading. In the work of Timmel et al. ([9]), glass laminates with a
* Corresponding author. PVB-interlayer under low impact velocity are modeled using LS-
E-mail address: [email protected] (F. Bobaru). DYNA shell elements and a failure criterion based on the

0734-743X/$ e see front matter Ó 2013 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.ijimpeng.2013.07.001

This article is a U.S. government work, and is not subject to copyright in the United States.
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 153

principal strain reaching a critical value. Element deletion is used peridynamic results for impact damage on a glass cylinder from a
as the elements fail. The fracture behavior is highly mesh depen- spherical projectile has been studied in Ref. [23].
dent and certain meshes lead to unrealistic results. Finite elements Here, we employ the simplest peridynamic constitutive model
with element deletion method (sometimes called “vanishing for a brittle material (linear micro-elasticity with brittle damage)
element technique” or “element erosion method”) are also used in and observe that we capture most of the damage morphology seen
Refs. [10], with a stress-based continuum damage model for in the experiments, including the changes induced by the higher
describing the constitutive behavior of glass. The contribution of impact speed on the damage patterns. Furthermore, the trend for
different cracking systems is explicitly taken into account in this the projectile rebound speed compares favorably with the experi-
work. The authors of [10] study the initiation and propagation of mental results. The boundary conditions used in the simulations
the ring/cone crack in a glass plate under impact based on this are for a more rigid setting than those present in the experiments;
model. They obtained multi-ring cracks when using a higher crit- most of the differences between the computational results and the
ical stress value in contrast with extensive crushing and ejected experimental observations can be attributed to the rigidity of the
debris when the critical stress value used was lower. Since a 2D boundary conditions used here to approximate the holding metal
axisymmetric model is used in Refs. [10], only a limited set of frame employed in the experiments. The evolution of damage and
cracks are obtained: for instance, radial cracks cannot be simulated its connection to the stress waves’ propagation in the two-layer
with this approach. system is described in detail.
Motivated by experimental observation of glass failure, the au- The paper is organized as follows: in Section 2, the experimental
thors of reference [11] proposed a model based on the idea that a tests are described and the investigation of damage patterns
critical energy value must be reached over a finite region before the dependence on the impact speeds is presented; in Section 3, we
glass can fail, independent of the calculated stress. This is a nonlocal briefly review the basic formulation for peridynamics and the
damage model. A finite element model using shell elements im- damage model used; in Section 4, we present the numerical results
plements this nonlocal criterion for a compound glass-PVB-glass from the peridynamic model of the impact on the G-PC system and
system and uses element deletion once failure is determined to compare them with the experimental results. We also include a
occur. While the nonlocal damage model improves the simulation description of the early stages of damage evolution in the glass
results compared with local damage models, the obtained damage layer, based on the computational model results. Conclusions are
morphology differs in significant ways with the experimentally given in Section 5.
observed one.
At the other end of the spectrum in terms of modeling failure of 2. Experiments of impact on the glass-polycarbonate system
glass, atomistic models have been used recently (see e.g Ref. [12].)
to study the fracture of silica glass from hypervelocity impact. The 2.1. Experimental setup
authors of [12] report cracks initiate and propagate from the bot-
tom to the top of the 24 nano-meter thick glass sheet. Molecular The two layer glass-polycarbonate (G-PC) system tested at the
dynamics (MD) models, however, cannot be used to model macro- ARL lab consists of a square soda-lime glass plate with dimensions
scale structures, while multiscale models coupling MD with FEM 10.16 cm
10.16 cm
0.33 cm, and a backing polycarbonate layer
models, for example, need to take into account the well-known fact of 10.16 cm
10.16 cm
0.3 cm. An aluminum frame holds the G-
(see e.g Ref. [13].) that stress waves that propagate and reflect from PC plates together and has the same outer dimensions as the G-PC
the boundaries strongly affect the evolution of fracture in a brittle laminate, but has an opening of 5.08 cm
5.08 cm (see Fig. 1). The
system. Therefore, to accurately capture the evolution of damage in inside surface of the frame is lined with a 1 mm thick rubber gasket.
brittle systems, the entire structure needs to be considered in the Tests were also performed without rubber gasket but no major
simulations and the dynamics of the stress waves, in the presence differences are observed between these two conditions in terms of
of evolving damage and failure fronts, needs to be solved for damage patterns. The frame, along with the G-PC system, is clam-
correctly. ped on the edge to a target fixture for impact testing. A spherical
In the present paper, we report on results from experimental projectile (440C stainless steel) with 0.556 cm diameter and 0.692 g
and numerical studies using peridynamics on impact on a two- mass is shot from a gas gun normal to the center of the glass plate at
layer square laminate composed of a thin (33 mm) glass plate different velocities: 61, 100, and 200 m/s.
and a thin (30 mm) polycarbonate backing plate with sides of about
10 cm. The two-layer system is placed in a metallic frame and
impacted with a spherical projectile at speeds ranging from 61 m/s
to 200 m/s. We will refer to this glass-polycarbonate system as the
G-PC system. The numerical studies conducted with peridynamics,
using the code EMU (see Ref. [14]) with some modifications of the
spatial integration detailed in Ref. [15] are performed for a model
which only approximates the presence of the actual aluminum
frame used in the experiments. The recently introduced nonlocal
continuum model, peridynamics [16], is well suited for solving
dynamic fracture problems (see e.g Refs. [17e22].). In order to
overcome mathematical inconsistencies in the classical continuum
mechanics models of problems in which cracks are initiated and
evolve in time, peridynamics uses an integral of forces (per volume-
squared) over a nonlocal region (called horizon) around a point to
replace the divergence of the stress tensor in the equations of
motion. The method has been shown to correctly predict phe-
nomena in dynamic fracture in brittle materials and composite
materials, including crack branching and the crack propagation Fig. 1. Experimental setup. The glass and polycarbonate plates are sandwiched be-
speed [16e20].The role of the “influence function” on the tween two metal frames and clamped together at the outer edge.
154 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165

2.2. Experimental results and discussion features seem to associate primarily with cracks that split the plate
into four quadrants, which are called quadrant cracks in Ref. [8]. In
Fig. 2 shows the post-mortem samples (glass plate only) the 100 m/s case, we also observe the formation of cracks that do
impacted at three different velocities. The polycarbonate plate is not connect to the central region, thus they could grow from the
not damaged at the lower two impact velocities and at 200 m/s only boundaries, as seen in the triangular regions. These cracks are
some minor plastic deformation is observed near the impact. The harder to distinguish in the 200 m/s case due to the severe frag-
signature of the metal frame opening is visible from the cracks mentation that had taken place. Corner cracks cutting the corners
enclosed in the rectangular regions marked on the figures in each of at 45 (see square regions) from the impact velocity of 200 m/s at
the three impact speed cases. Note that clamping is not at the three of the four corners. These types of cracks are absent at impact
opening’s edges but on the outer edges of the frame, thus some speeds of 61 m/s and 100 m/s.
separation between the glass and PC layers, and the metal frame is At the two low impact speeds (61 m/s and 100 m/s), no visible
possible and likely takes place during the impact. This possible damage is observed in the polycarbonate plate, aside from some
separation explains the non-uniformity of the cracks in the frame superficial surface scratches. At these impact speeds, the projectile
opening area. Radial cracks emanating from the impact crater are rebounds from the target with speeds of about 3 m/s and 8 m/s,
observed in all cases, but with some differences in terms of the respectively. At a high impact speed of 200 m/s, plastic deformation
length and number of these radial cracks, which depend on the near the impact site is observed in the polycarbonate plate but no
impact velocity. Circumferential cracks or ripple cracks (see ellip- cracks or penetration. At this impact speed the projectile rebounds
tical regions in Fig. 2) become denser as the impact speeds increase. from the G-PC system faster than the rebound from the two other
Some of the through-thickness cracks are not perpendicular to the two impact speeds, at about 33 m/s.
plane of the plates, instead they are tilted, as they can be seen from Some glass fragments around the impact crater were lost during
the light reflections in the diamond regions. For impact speeds of the impact on the strike face. Tape was used to hold the pieces
100 m/s and 200 m/s, some of the radial cracks split (or branch) as together after the tests. At the highest impact speed, the predom-
seen inside the circular regions. If these cracks form from the edges, inant damage pattern outside of the central region, where the fine
however, then they are coalescing, instead of branching. These and dense radial and circumferential cracks are formed, is

Fig. 2. Experimental results: damage patterns in the glass layer under various impact speeds. The signature of the frame opening is indicated by rectangular regions, diamonds
highlight through-thickness titled cracks, ellipses contain ripple cracks, circles surround splitting cracks, some boundary cracks are inside the triangles, and squares show corner
cracks.
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 155

fragments of glass that have some resemblance to intersecting kx þ hk  kxk

radial and circumferential cracks. However, in this case, the “radial” s ¼
kxk
and “circumferential” cracks are discontinuous and not well orga-
nized, the picture showing mostly irregular “islands” of unbroken The corresponding pairwise force becomes
glass. Cracks that split near the boundaries, or that coalesce if they
vuðh; xÞ vkx þ hk
are actually starting from the boundaries, are observed as well on fðh; xÞ ¼ ¼ cðkxkÞs
each of the four sides. A unique feature observed at the high impact vh vh
speed and not seen at 61 or 100 m/s, is the presence of corner The function c(jjxjj) is called the micromodulus function and
cracks. represents the bond elastic stiffness. For isotropic materials, the
micromodulus function version is computed by matching the per-
3. Review of the peridynamic formulation idynamic strain energy density to the classical strain energy density
Z
Introduced by Silling [16], the peridynamic theory is a non-local 1 cðkxkÞs2 kxk
Wclassical ¼ dVbx
formulation that extends the classical continuum mechanics 2 2
H
formulation. The term “peridynamics” comes from the Greek roots
for “nearby” and “force” [16]. In peridynamics, every material point The value for the 3D constant micromodulus function is then
is connected to the other points inside a certain “horizon” region derived as (see Ref. [25])
through peridynamic bonds. Instead of the stress divergence term
in the classical equations of motion for a continuum, an integral of 18k
c ¼
forces per volume-squared in the bonds at a point, over the current pd4
point horizon region is used. The peridynamic equations of motion
are: where k is the bulk modulus.
Z The damage model in peridynamics consists in breaking the
ru€ ðx; tÞ ¼ b  xÞdV þ bðx; tÞ
fðuðx; tÞ  uðx; tÞ; x peridynamic bonds connecting any two nodes when the relative
b
x
change in distance between them exceeds a certain prescribed
H
value s0 [16]. This critical relative elongation parameter s0 is ob-
where f is the pairwise force function in the peridynamic bond that tained by equating the work, per unit fractured area, required to
connects point b x to x, and u is the displacement vector field. r is the break all the bonds across a surface, to the fracture energy required
density and b(x, t) is the body force. The integral is defined over a for complete separation along the surface (see Ref. [25])
region H, the compact supported domain of the pairwise force
Zd Z2p Zd cos1
Zðz=kxkÞ
function around a point x, called the “horizon region”. The region is h i
2
taken here to be a sphere of radius d. The peridynamic horizon may G0 ¼ cðkxkÞs20 kxk=2 kxk sin4d4dxdqdz
be viewed as an “effective” interaction distance or an “effective 0 0 z 0
length-scale” of a continuum [21]. In principle, the exact size and
shape of the horizon could be found from wave dispersion curves For the constant micro-modulus given above, the critical rela-
for a specific material under specific dynamic conditions (see tive elongation is obtained as
Refs. [16,24]). sffiffiffiffiffiffiffiffiffiffiffi
b  x be the relative position in the reference configu-
Let x ¼ x 10G0
s0 ¼
ration and h ¼ u b  u the relative displacement between two ma- pcd5
terial points, x and bx . From the definition of the horizon, we have
We impose the same s0 over the entire region, which results in
an effectively slightly weaker material in the regions near the
kxk > d0fðh; xÞ ¼ 0
boundary or where damage already happened, since the integra-
tion region is reduced compared to a location in the undamaged
A micro-elastic material is defined as one for which the pairwise
bulk. This is similar to the skin effect discussed in Ref. [15].
force derives from a micropotential u:
A damage-dependent s0 can be introduced in order to reduce the
effectively weaker bond strength near the boundary or in damaged
vuðh; xÞ regions (see Refs. [25,18]). Especially in impact problems where
fðh; xÞ ¼
vh many new boundaries are formed as a result of a multitude of
newly created fracture surfaces, a constant critical relative elon-
The strain energy density at a given point is defined as: gation may result in the over prediction of the number of fragments
Z formed. In this paper we choose to “strengthen” the material in
1
W ¼ uðh; xÞdVx damaged regions by allowing s0 to depend monotonically on the
2
H amount of damage at that particular node. One could attempt an
exact evaluation of the damage-dependent (variable) s0. However,
The factor 1/2 factor is present because the elastic energy in a here we resort to a simpler way, namely an approximation pro-
bond is shared by the two nodes connected by the bond. A linear posed in Ref. [14] which is described by the following formula
micro-elastic potential, which leads to a linear relationship be-
tween the bond force and the relative elongation of the bond, is (  
s0
min g; 1 þ b
Da
1D ; if D > a
obtained if we take s ¼
s0 ; otherwise
cðkxkÞs2 kxk
uðh; xÞ ¼ where the damage index D is defined as the ratio of the number of
2
broken bonds to the number of initially bonds. s0 and s are the
where s is the bond relative elongation initially given critical relative elongation and the computed critical
156 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165

relative elongation, respectively. In the simulations that follow, we

use the values a ¼ 0.35, b ¼ 1, and g ¼ 2. More details can be found
in Ref. [18].
A validation of the peridynamic micro-elastic model against a
finite element model has been performed in Refs. [22], where a
comparison of the elastic bending deformation between the
nonlocal solution for the problem of elastic impact, in which no
damage is allowed to initiate, with the horizon size of about 2.7 mm
and a grid spacing less than 1 mm, and a dynamic (explicit) finite
element solution obtained by using Abaqus. The results for the
bending deformation in the problem of elastic impact of a spherical Fig. 4. Through-thickness view (along the x-z plane passing through the center of the
rigid projectile onto a two-layer glass backed by a polycarbonate plate) of the boundary conditions for the glass-polycarbonate system.
layer given by a converged finite element solution with solid ele-
ments and the peridynamic model that uses a grid of comparable
speed where some plastic deformation is visible. As a consequence,
size were shown to match very well. The time-evolution for the
in the peridynamic simulations of the polycarbonate layer we
projectile speed also matches well between the two different
prescribe “no-fail” conditions, by using a sufficiently large critical
models (see Ref. [22]).
relative elongation that is never met during the simulation. The
For the model with damage, the comparison of the peridynamic
material properties for the polycarbonate layer are: density
results will be made with the experimental results discussed in
Section 2.
r ¼ 1200 kg/m3, Young’s modulus E ¼ 2 GPa. In bond-based peri-
dynamics, the version used in this paper, the Poisson’s ratio is fixed
and equals 0.25 in 3D (close to the soda-lime glass Poisson ratio of
4. Peridynamic results for impact on the framed thin-plate 0.22), and 1/3 in 2D, since the assumption is that material points
glass-polycarbonate system interact only through a pair-potential. The Poisson ratio limitation
is removed in the state-based formulation of peridynamics (see
4.1. Problem setup Ref. [26]), however, here we use the bond-based for simplicity. We
note that, in dynamic fracture problems of the type considered in
The peridynamic model for the two layer G-PC system uses the this work, the Poisson ratio value does not have a significant in-
geometry and boundary conditions shown in Fig. 3. The geomet- fluence on the crack propagation speed or the shape of the crack
rical dimensions are the same as in the experiments, but the path (see Refs. [17,26]).
boundary conditions are slightly different from those present in the In the simulations, we use a spherical rigid projectile of the same
experiments. For example, in the experiment, two metal frames dimensions and mass as the steel projectile used in the experi-
with clamps are used in order to hold the specimen. Since a model ments and described in Section 2. The projectile is placed very close
that includes the metal plate, the clamps, and the holding fixture to the glass plate so that impact happens in the first few time steps.
would lead to a much larger computational simulation, and since The use of a deformable projectile is left for the future.
the exact pressure applied on the clamps is not known with pre- We perform the simulations using the same impact speeds as
cision, we choose here to approximate the actual boundary con- those used in the experiments. Based on a convergence study
ditions with the following: zero out-of-plane displacements (z- performed in Refs. [27], we select to use a horizon size d ¼ 1 mm
direction in Fig. 3) for the nodes on the top face of the glass plate and m ¼ d/Dx ¼ 4, where Dx is the grid spacing used for the dis-
and bottom face of the polycarbonate layer, over the frame region cretization of the domain integral in the peridynamics equations of
(see Fig. 4). We expect such conditions to lead to cracking along the motion. The selection is based on the damage patterns results and
inner boundary of the frame due to prevention of out-of-plane the projectile speed after perforation of a single glass plate (see
bending, but this issue will be analyzed in detail in Section 5 Ref. [27]). Discussions about the dependence or independence of
where we conduct tests with almost no boundary conditions. the crack propagation speed on the horizon size are given in
In the experimental samples, there is no adhesive layer between Refs. [21] and [22]. The 3D discrete model has a total of over 6
the glass and the polycarbonate. In the computational model, we million nodes (about 20 million degrees of freedom). The total
“disconnect” all the peridynamic bonds between the two different simulation time is 77 ms with a time step size of about 27.5 ns. We
material regions (glass and polycarbonate) and use simple non- use 100 processors on 2.2 GHz/64 bit Opteron Linux cluster with
penetrating contact conditions to describe contact conditions. 70 Gb memory.
The material properties used in the model for soda-lime glass
are: density r ¼ 2440 kg/m3, Young’s modulus E ¼ 72 GPa, facture 4.2. Damage maps for different impact velocity: results and
energy G0 ¼ 8 J/m2. In the experimental results, no damage is discussions
observed in the polycarbonate layer, except for the highest impact
The peridynamic results for the damage of the glass plate are
presented in terms of nodal damage index maps. The damage index
d at a node is defined as:

nbroken
d ¼
n

where nbroken and n are the number of broken bonds and the
number of initial bonds at a node, respectively. The damage index is
a number between 0 and 1 (0 means no bonds connected at the
node are broken, and 1 means all the bonds are broken). Note that a
damage index of around 0.4e0.5 may indicate that a fracture sur-
Fig. 3. Dimensions for the two-layered glass-polycarbonate system. face exists, if damage is localized along a surface (line). Diffuse
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 157

damage regions may indicate a “shattered” zone or an area of induced by impact on the G-PC system: in the diamond-shaped
densely packed cracks that are too fine to resolve with a certain zones, a quadrant radial crack kinks just before reaching the
discretization/horizon (see Ref. [20]). The legend shown in Fig. 5 is boundary and the photograph of the experiment shows that the
used in all damage plots. through-thickness crack is not perpendicular onto the plane of the
In Figs. 5e7, the damage maps for the top face (xey plane), plate, but it is rather tilted. A similar pattern is displayed by the
bottom face (xey plane), and cross-sectional cut (xez plane) peridynamic results, in which two quadrant cracks kink before
through the center of the glass plate are shown at about 77 ms. We reaching the boundary and the slight difference between the
compare our simulation results with experimental results for the damage paths on the strike face and the back face of the glass plate
three impact speeds, 61, 100, and 200 m/s. means that these cracks are tilted through the thickness. Stress
waves propagating through the G-PC system are responsible for
4.2.1. Results for impact speed of 61 m/s these features.
From Fig. 5 we observe that radial cracks extend from the central By examining the cross-sectional view of the peridynamics so-
crater region along the quadrant directions (towards the middle of lution in Fig. 5(c), we observe a Hertz-type cone crack transitioning
the sides of the square plate) and the diagonal directions (towards into a transverse crack (parallel to the strike face) form in sequence
the plate corners). For the region nearby the crater, radial cracks as a result of the impact and wave reflections from the poly-
appear to be straight (see damage on the bottom face in Fig. 5(b)), carbonate backing plate.
and the diagonal cracks continue to be straight whereas the The time-evolution of the system of cracks for this impact ve-
quadrant cracks have more turns. These features can also be locity is discussed in Section 4.4.
distinguished in the experimental result in Fig. 5(d). Circumferen-
tial or ripple cracks (see the highlighted elliptical areas) form in the 4.2.2. Results for impact speed of 100 m/s
computational model at about the same distance from the center as At the impact speed of 100 m/s, the mark of the frame opening is
in the experiments. The mark left by the frame opening is much now stronger, with cracks showing on both the strike and the back
more prominent in the peridynamic simulations than in the face of the glass plate (see Fig. 6). As before, due to the more
experiment (see the highlighted rectangular areas) because in the compliant response of the actual metal frame than the rigidly
computations we impose rigid, zero displacements conditions over imposed boundary in our model, the experimental results indicate
where the metal frames are in the experimental setup. The actual the presence of the frame opening in a much reduced way (see
aluminum frame deforms slightly (in bending) as the G-PC system highlighted rectangular regions). Ripple cracks (marked by ellipses)
bends due to impact. The imperfect contact between the frame’s now show farther away from the center of the plate than with the
inner boundary and the glass leads to some cracking. The damage low impact speeds and they are not continuous but appear to be
maps also indicate another remarkable characteristic of fracture interrupted by radial cracks. The distinct difference between the

Fig. 5. Damage maps for impact speed of 61 m/s. The glass plate at about 77 ms: the strike face in (a), the bottom face in (b), and the central through-thickness cross section in (c),
where displacements are enhanced by a factor of two, to better observe separation, and thickness direction is enlarged; experimental result shown in (d).
158 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165

Fig. 6. Damage maps for impact speed of 100 m/s. The glass plate at about 77 ms: the strike face in (a), the back face in (b), and the central through-thickness cross section in (c),
where displacements are enhanced by a factor of two, to better observe separation, and thickness direction is enlarged; experimental result shown in (d).

damage on the strike face and back face of the glass plate near the can be considered as the signature of the boundary conditions that
center of the plate can be better understood from the cross- mimic the frame opening, and the rest of the plate in which mostly
sectional view in Fig. 6(c). It is interesting to remark that the large and some small size fragments are formed. These regions seen
blurry region, between the center of the plate and the first major from the experiments are well captured by the peridynamic
ripple crack observed in the experimental results in Fig. 6(d), is an simulation, except for the signature cracks related to the frame
indication of the conical crack we observed in the simulations opening: it is quite possible that under such impact energy, there is
(Fig. 6(c)). The back face damage is extensive and diffuse damage partial separation between the holding frame and the glass-
(shattered glass) whereas on the front face, outside of the crater polycarbonate plates and due to this, the frame opening signature
region, localized damage develops in the form of crack lines. Some is hardly visible (see the highlighted rectangular area in Fig. 7(d)) at
cracks appear to start from the boundary of the structure in both this impact speed. In the simulations, considerable amount of
the experiments and simulations (see zones marked by triangles). damage is seen around the opening area and this is attributable to
The quadrant cracks branch out (see regions marked by circles) as the rigid boundary conditions imposed there. The difference be-
they approach the boundaries. In most cases, the corresponding tween the damage seen on the strike face and the back face in this
crack paths on the strike face and back face of the glass plate are not area is due to the presence of the backing polycarbonate plate that
identical, demonstrating that such cracks have tortuous through- is in contact with the back face of the glass layer, whereas the top
thickness surfaces. This is observed in the experiments as well. face of the glass layer has the imposed zero-displacements outside
This particular feature is seen in the photograph of the post- of the frame opening area. Since there is no energy dissipation
mortem sample as a darker and thicker crack line, due to the mechanism in the model other than via breaking of bonds, stress
light reflection (see Fig. 6(d)). waves create more damage than in the experiment where some of
the energy is transferred to the metal frame and the holding fixture.
4.2.3. Results for impact speed of 200 m/s Notice also that through-thickness corner cracks (highlighted
When the G-PC system is subject to impact speed of 200 m/s, square regions) are a new feature seen in the experiments, not
the damage obtained by the peridynamic model is extensive and observed at the lower impact speed. The peridynamic simulations
most of the glass layer is fragmented. Three different regions can be capture the corner cracks very well. For these corner cracks, the
identified from Fig. 7 (a) and (b): a central region with finely spaced damage maps of the strike and back faces are similar, meaning that
radial cracks on which a few ripple cracks superpose, a region that these cracks are straight through-thickness cracks, and not tilted.
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 159

Fig. 7. Damage maps for impact speed of 200 m/s. The glass plate at about 77 ms: the strike face in (a), the back face in (b), and the central through-thickness cross section in (c),
where displacements are enhanced by a factor of two, to better observe separation, and thickness direction is enlarged; experimental result shown in (d).

Some quadrant cracks still branch as they approach the boundaries

(circular zones) and quite a few more cracks start from the
boundaries and propagate inwards.
At this impact speed some plastic deformation is observed in the
experiments in the polycarbonate layer, but no cracks. In the per-
idynamic solution, no damage (the only type allowed here would
be crack formation, not plasticity) forms in the polycarbonate layer.
Since the peridynamic model used in this work does not include
plastic deformation, and additional means to reduce the energy in
the system, such as dissipation through the boundary conditions,
are not present in our model, we observe a somewhat larger
amount of damage in the glass layer in the computations, compared
with the experimental results.

4.3. Trends with increasing projectile impact speed

The projectile speed profiles are shown in Fig. 8. Shortly after

the projectile hits the target, it rebounds, thus the negative sign of
the projectile speed, for all three cases tested. This coincides with
the experimental observation. The magnitude of the rebound ve-
locity, however, is slightly different from the experimental tests: at
61 m/s impact velocity the model gives a rebound speed of 13 m/s,
while in the experiments it measures 3 m/s. At 100 m/s impact
speed, the computations show a rebound speed of 15 m/s, while
experimentally it measures about 8 m/s. For the 200 m/s case, the Fig. 8. Evolution of the projectile speed at different impact velocities.
160 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165

computations give 17 m/s while the measured value is 33 m/s. energy between the deformable projectile and the PG-PC sys-
Notice also that the projectile time-in-contact with the target, tem takes place;
estimated from the velocity plots in Fig. 8 from the moment the  Differences between the boundary conditions on the two-layer
speed starts to decrease until the moment the velocity becomes system that affect the bending structural response and thus
negative, is in the range of 6e7 ms. These contact times are signif- the projectile rebound: in the experiments a metal frame
icantly higher than when small projectiles (1 mm diameter) impact and holding fixture are used, while in the simulations these
a glass block, measured to be between 1 and 2 ms (see Ref. [7]). The conditions are replaced by imposed zero out-of-plane
difference is due to the bending deformation of the G-PC system displacements.
used in our work and the larger projectile size (5.56 mm diameter).
In Refs. [7], the glass block is thick enough (2.5 cm) so that the stress Nevertheless, the fact that the trend of increasing rebound
waves generated at the time of impact did not return to the impact speed with an increased impact speed is well captured by this
site after reflecting from the edges of the glass block, during the simple peridynamic model and that the computed values are in a
contact time. In our case (thickness of 0.33 cm), stress waves similar range with those measured experimentally are very
interact with the projectile. encouraging.
While the trend observed in the experiments, of increased
rebound velocity with an increasing impact speed, is matched by 4.4. The computed time-evolution of damage
the peridynamic computations, two possible reasons are believed
to contribute to the differences between the measured values and In what follows we analyze the evolution of damage in the glass
the peridynamic results: layer in the first 25 ms from impact. The 61 m/s impact speed case is
discussed here.
 The use of a rigid model for the projectile: in reality the pro- From the results shown in Fig. 9 for the median cross-section
jectile deforms elasto-plastically and a complex transfer of through the thickness, we observe that the cracks start from the

Fig. 9. Evolution of damage and nodal velocities from the impact at 61 m/s. The central cross-section of the glass plate (over a 6.5 cm length) at 4.4, 5.2, 6, 11, and 25 ms (from top to
bottom, see movie 1 in supplementary material). Nodal velocities are plotted using black arrows (quiver plot) over the damage maps.
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 161

bottom face of the glass plate. These initial cracks are radial cracks the inner bounds of the metal frame region. These surface cracks
in the form of an eight-spoke wheel. They propagate radially but continue to grow as they propagate both towards the boundaries of
also grow upwards through the thickness. This confirms the results the sample and along the frame cracks. Radial cracks grow and
from molecular dynamic simulations presented in Ref. [12]. The arrest before reaching the frame cracks, but eventually continue
cracking from bottom to top has also been observed in computa- towards them, and cross them merging and getting ahead of the
tions of high velocity impact on a seven-layer glass system (see surface cracks. Radial cracks are primarily driven along the bottom
Ref. [22]). The first cracks on the top face form in the shape of a ring face and they cut through to the top face as they propagate, but not
crack (see Figs. 9 and 10), around 5.2ms from the moment of impact. always perpendicular to the faces. The non-planar radial cracks and
Supplementary video related to this article can be found at their tortuosity is a direct consequence of interactions between the
http://dx.doi.org/10.1016/j.ijimpeng.2013.07.001. variety of stress waves reflecting from the boundaries and the
Driven by the stress waves, the radial cracks from the bottom and propagating crack.
the ring crack on the top face move towards one another and deflect The spokes in the 45-degree direction eventually arrest due to
from each other, before they coalesce. The crack interactions and the interactions with reflected waves, and only the so-called quadrant
continuing propagation and reflection of stress waves in the system cracks (the horizontal and vertical spokes) continue to propagate,
then lead to the formation of other ring cracks on the top face, both but their paths and speed of propagation are also perturbed by
inside and outside of the original ring crack of the top face. A interactions with stress waves. Eventually, new diagonal cracks
Hertzian cone-crack is fully formed around 11 ms. The reflections appear near the corners of the opening region and on the top face
from the boundaries (the frame region) further complicate the first, initiated by the interaction of waves that reflect from the sides
evolution of the damage, as can be seen from Fig. 10. The presence of of the “frame opening”. These cracks grow in both directions (to-
the boundary conditions that approximate the existence of the wards the corners and the center) in the in-plane direction, and
metal frame holding fixture generates straight cracks that propagate also grow through the thickness to give rise to diagonal cracks on
on the top face only. We call these “frame cracks”. These are mainly the bottom face. At that stage, the diagonal cracks on the bottom
the features that are located where the frame opening is. In the face merge with the original diagonal cracks that started at the
simulations, since the top surface of the glass is held in place, surface center.
cracks (chips and flakes) are produced, while the constraint on the The vector plots of the nodal velocity (see Figs. 9 and 10) su-
displacements of the bottom face of the PC layer is responsible for perposed onto the damage index plots clearly indicate the propa-
inducing the through thickness cracks. In the experiments, these gation of the surface waves (Rayleigh waves) on the top and
“frame cracks” happen only where the PC and the glass happen to bottom faces of the glass plate. The complex interaction among
come in contact with the frame. In the experiments, the clamps are longitudinal, shear, and surface waves, with the reflected waves
on the outer edges of the frame, not on the inner edges and due to from the boundaries and the polycarbonate interface, depict a
the impact the glass plate loses contact with the top metal frame. If complex picture of how damage is generated in the two-layer
direct contact between the PC plate and the inner part of the frame system at this impact velocity. Plans for the future include exper-
takes place then that gives rise to conditions that initiate cracks that imental setups capable of recording the evolution of damage in a
trace portions of the frame opening (see cracks in the rectangular two-layer system.
contours on the pictures in Fig. 2). To further analyze the influence of boundary conditions on our
The computational results show that on the top face, wave in- simulation results, in the next section we present experiments and
teractions create conditions for glass chips to form in the middle of simulations for a setup that does not involve a frame.

Fig. 10. Evolution of damage and nodal velocities from the impact at 61 m/s for a central region of 6.5 cm by 6.5 cm. Top row: the top face of the glass plate at about 6, 11, and 25 ms
(left to right, see movie 2 in supplementary material). Bottom row: the bottom face of the glass plate at about 6, 11, and 25 ms (see movie 3 in supplementary material). Nodal
velocities are plotted using black arrows (quiver plot) over the damage maps.
162 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165

Fig. 11. Experimental results in the absence of the frame (plates are taped together along their edges prior to impact, and, post-mortem, on the front glass face and back PC face).
View from the impact face. At the higher impact speeds some glass fragments from the glass back face have moved and are stuck between the plates.

5. Analysis of influence of boundary conditions: experiments results (see Refs. [17,18,21]) that demonstrated that, in the presence
and peridynamic simulations of stress waves, a horizon size of about 1 mm or smaller results in
matching observed crack propagation speeds, while in cases in
We conduct a new set of experiments that does not involve a which the propagation of the crack is not influenced by incoming
frame. In this case the plates are suspended (by tape), and, to be stress waves reflected from the boundaries, any horizon size gives
able to recover glass fragments in the experiments, the two plates the observed propagation speed.
are taped together along their boundaries only. The front face of the Due to the new experimental conditions (suspended plates), the
glass plate is taped over only after the experiment to prevent projectile does not hit the center of the plate, but about 0.7e1 cm
further loss of fragments from handling. The computational model closer to one of the boundaries of the plate (see Fig. 11). In the
for these tests has no boundary conditions (mimicking the sus- computations we use the approximate impact location measured
pended plates from the experiment), so the two plates are in con- from the post-mortem samples. The experimentally measured
tact but otherwise free. We do not simulate the taping of the two impact velocities just before impact are: 58 m/s, 100 m/s, and
plates along their edges used in experiments. The new results, 150 m/s. The PC plate is not damaged at these impact speeds.
together with the original ones, should provide a better under- The following differences between the experimental results with
standing of the role and influence of boundary conditions in dy- the frame and without the frame can be observed (compare Figs. 2
namic brittle fracture of glass in multilayered configurations. and 11):
We again focus on the types and shapes of the cracks observed
since they are an indication of the dynamics of the fracture process.  The cracks present around the frame opening in Fig. 2 (see
A kink and tilt in a crack at a certain location in the plate is caused rectangular contours) are now absent at the lower impact
by interaction with the stress waves reflected from the boundaries speed, confirming the conclusion about the origin of
or the material surfaces. Obtaining such features in the simulations these cracks discussed before. At the higher speeds,
at similar locations as in the experiments indicates that the however, cracks that run near the boundaries and parallel to
computed cracks move with the same speeds as in the experiments. them are observed (see rectangular contours in Fig. 11).
In crack branching events, the authors have previously published These cracks are likely induced by stress waves reflected
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 163

Fig. 12. Damage maps computed by peridynamics as viewed from the strike face (left column) and back face of the glass plate (right column). The rows, from top to bottom,
correspond to the impact speeds of 58 m/s, 100 m/s, and 150 m/s at about 100 ms after impact, respectively.

from the edge boundaries. The location of these cracks,  There are only a few cracks that emanate from a boundary and
relative to the impact location, appears to reproduce the would not connect with other radial cracks emanating from the
asymmetry of the impact location relative to the center of center. Thus, many boundary cracks are a result of the presence
the plate. of the confining frame.
 Corner cracks are not present in the absence of the frame,  A set of very fine circumferential “cracks”1 around the coarse
leading us to conclude that these cracks were induced by stress circumferential cracks in the Hertzian cone region (about
waves generated from the interaction between the plates and 1.5e2 cm away from the impact point) are clearly visible in
the frame. the 100 m/s and 150 m/s cases. These “cracks” do not feel to
 The branching cracks that were near the boundaries in Fig. 2,
are still present and their prevalence increases with the
increased impact speed. 1
The reason for the quotation marks will become clear in what follows.
164 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165

the touch on the top glass surface (see also the zoom in
Fig. 11).
 Some of the glass fragments from the Hertzian cone region on
the back face of the glass plate have shifted from their original
locations and are trapped between the two plates.

The total simulation time is 100 ms, and the initial location of the
rigid projectile is just above the glass plate and therefore impact
happens after the first few time steps. The peridynamic computa-
tions are performed using the parameters as before: horizon size
d ¼ 1 mm and a grid spacing Dx defined by m ¼ d/Dx ¼ 4; the time
step is still 27.5 ns.
Fig. 13. Damage progression seen along a cross-section near the impact location for
The results in terms of the damage maps, at about 100 ms
the 100 m/s impact speed case. Observe the formation of the transverse cracks (parallel
after impact, viewed from the strike side and from the back-face with the strike face of the plate). Waves reflected from the boundaries ripple over this
of the glass plate are shown in Fig. 12. While the bulk of crack region and create roughness which shows as the fine circumferential “cracks” observed
growth has ended by end-time of these simulations, a few cracks in Fig. 11.
are still growing, and more cracks would form if we had the PC
and glass plates taped along their edges in our computations,
light-reflecting ridges. A more detailed analysis of these inter-
due to the elastic rebound of the PC plate onto the cracked glass
esting damage features will be provided in a future publication.
plate. In experiments, the rebound of the PC plate is likely the
While the impact problem on the two-layer system is signifi-
cause for which we now see less cracks in the simulations than
cantly more difficult than the dynamic branching of a single crack
in the experiment with the 58 m/s test conditions. The elastic
treated in Refs. [17,18,21], the results shown here demonstrate how
rebound is not captured in these simulations due to the absence
peridynamics can aid in understanding the dynamics of the process
of the edge taping between the plates, and the short simulation
of fracture and damage in difficult scenarios.
time. Moreover, in the experiments, the forced contact at the
ends of the plates due to the edge-tape certainly increases wave
reflections and elastic energy transfer between the plates, 6. Conclusions
possibly leading to extra damage compared to a case with no
edge-tape. In experiments, the tape is needed so that the glass This paper presents experimental and numerical simulation
fragments are recovered. The only difference in the boundary results of the impact damage in a thin glass layer with a thin pol-
conditions used for the computational model compared to the ycarbonate backing subject to impact from a small spherical pro-
experimental conditions is the absence of edge tape that keeps jectile at different speeds. Significant changes in the damage
them together. patterns that develop in the glass plate at three different impact
At the higher impact velocity, the computational results match speeds take place in the experimental tests. These changes are
very closely the observed damage patterns. predicted by the peridynamic model in both the case with a metal
For the 100 m/s impact speed case, damage progression hap- frame holding the two plates together, or without the metal frame.
pens as follows: the impact creates ring cracks on the top surface Some very specific damage and crack features observed in experi-
(due to deformation at the impact site) while on the bottom, stress ments are, remarkably, reproduced by the computational model, at
waves create hoop stresses that result in radial cracks. These two about the same location as in the experiments. This allows us to
damage fronts interact with each other in forming a shallow conclude that, the cracks computed by the peridynamic model
Hertzian cone, which is deflected into a transverse crack by stress move at a similar speed as those in the experiments.
waves. The transverse crack splits the glass parallel to the strike The numerical results are obtained here using the simplest
plane. Some of the radial cracks growing on the bottom face peridynamic model for micro-elastic brittle materials. We can
eventually arrest, and only a few continue to grow. conclude that at these levels of loading, modeling the glass plate as
A particular “crack” system seen in both the experiments and a linear micro-elastic brittle material is sufficient to explain the
the computations is the set of fine circumferential “cracks” sur- damage structure from impact in the thin glass plate with a thin
rounding the Hertzian cone damage area on the bottom face of the polycarbonate backing plate. The simplified boundary conditions
glass (see Fig. 12). These cracks are not complete circumferential used in the computational model is the most important factor for
cracks, meaning they appear late in the damage process (fact the differences between the experiments and the computations in
confirmed by our simulations), and are much finer than other this problem.
circumferential cracks seen on either the top or the bottom sur- In both the experiments and the peridynamic simulations,
faces. It is interesting to notice that the damage lines on the bottom ejection of the projectile from the target is observed, and the trend
face of the glass plate obtained by the peridynamic model show of increasing rebound velocity with increased impact speed
damage levels of about 20e30%, indicating that these are not fully observed in experiments is reproduced by the peridynamic model.
developed crack surfaces on the bottom face of the glass. Analysis of The computational results allow for an understanding of the
the computational results through a transverse cross-section near complex evolution of damage in the glass layer induced by stress
the impact point reveals that the fine lines are the signature of waves propagation and reflections. In the peridynamic model, no
“steps” or roughness from the transverse crack that splits the glass special crack initiation and propagation criteria and/or special
parallel to the strike face (see Fig. 13). These ridges where found on meshing techniques are necessary.
the fragments from the experiments also, and, as in our computa- The results presented here lead us to conclude that, for pre-
tions, the bottom glass surface, where these ridges appear, is intact. dicting dynamic brittle fracture, determining factors are: the elastic
The conclusion is that these fine features appear on the surface of properties of the material (since stress waves play such an impor-
the transverse crack that grows from the deflected Hertz cone tant role in the evolution of fracture), the fracture energy of
crack, and as the layer is very thin, stress waves bounce rapidly the material, and a model that correctly traces the bouncing of
from the bottom of the glass plate and create these micron-size stress waves from surfaces and interfaces, and which allows for
 W. Hu et al. / International Journal of Impact Engineering 62 (2013) 152e165 165

unrestricted crack evolution and feedbackeloop interaction be- [6] Field JE, Sun O. Townsend ballistic impact of ceramics. Inst Phys Conf Ser
1989;102:387e93.
tween advancing crack fronts and stress waves. The peridynamic
[7] Chaudhri MM, Walley SM. Damage to glass surfaces by the impact of small
model used here, the simplest one can think of, can be extended to glass and steel spheres. Philos Mag A 1978;37:153e65.
incorporate, for example, effects such as toughening of glass (via [8] Bless SJ, Chen T. Impact damage in layered glass. Int J Fract 2010;162:151e8.
compressive residual stresses at the glass surface). [9] Timmel M, Kolling S, Osterrieder P, DuBois PA. A finite element model for
impact simulation with laminated glass. Int J Impact Eng 2007;34:1465e78.
Plans for the future also include more accurate description of the [10] Ismail J, Zairi F, Nait-Abdelaziz M, Azari Z. How cracks affect the contact
holding fixtures/boundary conditions. In dynamic fracture and characteristics during impact of solid particles on glass surface: a computa-
damage of brittle materials, boundary conditions are important tional study using anisotropic continuum damage mechanics. Int J Impact Eng
2012;40-41:10e5.
factors that influence stress waves propagation in the system, that, [11] Pyttel T, Liebertz H, Cai J. Failure criterion for laminated glass under impact
in turn, control the evolution of failure fronts and the growth and loading and its application in finite element simulation. Int J Impact Eng
propagation of damage in the system. Experimental setups moni- 2011;38:252e63.
[12] Holmstrom E, Samela J, Nordlund K. Atomistic simulations of fracture in sillica
toring the time-evolution of damage and comparisons with the glass through hypervelocity impact. Europhys Lett 2011;96. 16005epp.1e
damage progression obtained from the peridynamic models are 16005-p.5.
also planned. [13] Ravi-Chandar K. Dynamic fracture. Amsterdam: Elsevier; 2004. p. 208e16.
[14] Silling SA. Fragmentation modeling with EMU. Albuquerque, NM: Sandia
National Laboratories; 2005. Technical report. p. 6e7.
Acknowledgment [15] Bobaru F, Ha YD. Adaptive refinement and multiscale modeling in 2D peri-
dynamics. Int J Multiscale Comput Eng 2011;9:635e60.
[16] Silling SA. Reformulation of elasticity theory for discontinuities and long-
The authors are thankful for the financial support offered range forces. J Mech Phys Solids 2000;48:175e209.
through research contracts between UNL and the ARO (Dr. Larry [17] Ha YD, Bobaru F. Studies of dynamic crack propagation and crack branching
Russell), and ARL (project coordinators Dr. C.F. Yen), ARO award with peridynamics. Int J Fract 2010;162:229e44.
[18] Ha YD, Bobaru F. Characteristics of dynamic brittle fracture captured with
number 58450 EG. The last author acknowledges partial support peridynamics. Eng Fract Mech 2011;78:1156e68.
from Callahan Research of New Zeeland (project coordinator Dr. [19] Hu W, Ha YD, Bobaru F. Modeling dynamic fracture and damage in fiber-
Biswajit Banerjee). The computations in this paper have been reinforced composites with peridynamics. Int J Multiscale Comput Eng
2011;9:707e26.
completed using facilities at the Holland Computing Center of the [20] Hu W, Ha YD, Bobaru F. Peridynamic simulations of dynamic fracture in
University of Nebraska-Lincoln. unidirectional fiber-reinforced composites. Comput Meth Appl Mech Eng
2012;217-220:247e61.
[21] Bobaru F, Hu W. The meaning, selection, and use of the peridynamic horizon.
References Int J Fract 2012;176:215e22.
[22] Bobaru F, Ha YD, Hu W. Damage progression from impact in multilayered
[1] Knight CG, Swain MV, Chaudhri MM. Impacts of small steel spheres on glass glass modeled with peridynamics. Cen Eur J Eng 2012;2:551e61.
surfaces. J Mater Sci 1977;12:1573e86. [23] Seleson P, Parks M. On the role of the influence function in the peridynamic
[2] Ball A, Mckenzie HW. On the low velocity impact behaviour of glass plates. theory. Int J Multiscale Comput Eng 2011;9:689e706.
J Phys IV 1994;4:783e8. [24] Weckner O, Silling SA. Determination of nonlocal constitutive equations from
[3] Walley SM, Field JE. The contribution of the Cavendish laboratory to the un- phonon dispersion relations. Int J Multiscale Comput Eng 2011;9:623e34.
derstanding of solid particle erosion mechanisms. Wear 2005;258:552e66. [25] Silling SA, Askari E. A meshfree method based on peridynamic model of solid
[4] Field JE. Dynamic fracture: its study and application. Contemp Phys 1971;12: mechanics. Comput Struct 2005;83:1526e35.
1e31. [26] Silling SA, Epton M, Weckner O, Xu J, Askari E. Peridynamic states and
[5] Field JE. Investigation of the impact performance of various glass and ceramic constitutive modeling. J Elasticity 2007;88:151e84.
systems. Cambridge, U.K.: European Research Office of the US Army; 1988. [27] Hu W. Peridynamic models for dynamic brittle fracture. Lincoln, NE:
Report No. R&D 5087-MS-01. p. 14e5. University of Nebraksa-Lincoln; 2012. Ph.D. dissertation.