Journal of Physics: Conference

Series

OPEN ACCESS You may also like

- Mesoscale strain and damage sensing in
Peridynamics for multiscale materials modeling nanocomposite bonded energetic
materials under low velocity impact with
frictional heating via peridynamics
To cite this article: E Askari et al 2008 J. Phys.: Conf. Ser. 125 012078 Krishna Kiran Talamadupula, Stefan J
Povolny, Naveen Prakash et al.

- Two-dimensional peridynamic simulation

of the effect of defects on the mechanical
behavior of Bi2Sr2CaCu2Ox round wires
View the article online for updates and enhancements. Q V Le, W K Chan and J Schwartz

- Peridynamic Simulation of Cracking

Behavior in Thermal Barrier Coatings
H Wang, ZW Cai, WZ Wang et al.

This content was downloaded from IP address 69.248.246.204 on 18/07/2024 at 20:52

SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

Peridynamics for multiscale materials modeling

E Askari1 , F Bobaru2 , R B Lehoucq3 , M L Parks4 , S A Silling5 , O
Weckner6
1,6
Applied Math, The Boeing Company, Seattle, WA 98124, USA
2
Department of Engineering Mechanics, University of Nebraska-Lincoln, Lincoln, NE,
68588-0526, USA
3,4
Applied Mathematics and Applications, Sandia National Laboratories Albuquerque, NM
87185, USA
5
Multiscale Dynamic Materials Modeling, Sandia National Laboratories, Albuquerque, NM
87185, USA
E-mail:
[email protected],[email protected],[email protected],[email protected],
[email protected],[email protected]

Abstract. The paper presents an overview of peridynamics, a continuum theory that employs
a nonlocal model of force interaction. Specifically, the stress/strain relationship of classical
elasticity is replaced by an integral operator that sums internal forces separated by a finite
distance. This integral operator is not a function of the deformation gradient, allowing for a more
general notion of deformation than in classical elasticity that is well aligned with the kinematic
assumptions of molecular dynamics. Peridynamics effectiveness has been demonstrated in
several applications, including fracture and failure of composites, nanofiber networks, and
polycrystal fracture. These suggest that peridynamics is a viable multiscale material model
for length scales ranging from molecular dynamics to those of classical elasticity.

1. Introduction
The response of materials to the environments and loads occurring in practice all require
an understanding of mechanics at disparate spatial and temporal scales. Such “multiscale”
understanding is a fundamental challenge for next generation materials modeling; see [1]
for information and references to the existing literature. A currently popular multiscale
approach couples two or more well-known models, for example, molecular dynamics and classical
elasticity [2], each of which is useful at different scales. An alternative approach is to develop
a single multiscale material model that remains valid and useful over a wide range of temporal
and spatial scales.
The onerous practical limitations of molecular dynamics and the limited validity of classical
elasticity has led to generalized continuum theories purporting to supply a single multiscale
material model. Such theories are motivated by introducing a length-scale absent in classical
elasticity. These generalized continua are certainly not new; see [3, 4, 5] for overviews and
citations to the abundant literature. However, these theories develop material models with
a length-scale dependence by augmenting the displacement field with supplementary fields
(e.g., rotations) that provide information about fine-scale kinematics, or by using higher-order
gradients of the displacement field, or by averaging local strains and/or stresses.

c 2008 IOP Publishing Ltd 1
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

This paper presents an overview of peridynamics [6, 7], a generalized continuum theory that
employs a nonlocal model of force interaction. Specifically, the stress/strain relationship of
classical elasticity is replaced by an integral operator that sums internal forces separated by a
finite distance. This integral operator is not a function of the deformation gradient, allowing
for a more general notion of deformation than classical elasticity that is well aligned with the
kinematic assumptions of molecular dynamics.

2. Peridynamic primer
This section reviews peridynamic fundamentals including kinematics, balance of angular and
linear momentum, length scale, a relationship with Newtonian mechanics, and discretization.
We refer the reader to [8] demonstrating that linearized peridynamics can reproduce nonlinear
dispersion relationships for short wavelengths, in contrast to linear elasticity.

2.1. Kinematics and internal force density

The spatial point y (in the deformed configuration) is related to the material point x (in the
reference configuration) via y(x, t) = x+u(x, t) where u(x, t) is the displacement of x ∈ B ⊂ R3 .
We say the body B undergoes deformation if the displacement u(x, t) is nonzero for some x ∈ B.
Deformation represents the mechanism through which continuum mechanics relates deformation
to the internal forces of a body. Classical elasticity assumes that the displacement field is
continuously differentiable at every x ∈ B so that

u(x0 , t) − u(x, t) = ∇x u(x, t)(x0 − x) + O(kx0 − xk2 ),

where ∇x denotes the gradient operator with respect to the material point x. If F(x, t) :=
I + ∇x u(x, t) denotes the deformation gradient, then

y(x0 , t) − y(x, t) ≈ F(x, t)(x0 − x).

In words, the body deformation y(x0 , t) − y(x, t) can be approximated by the deformation
gradient acting on the bond x0 − x. By relying on the true deformation y(x0 , t) − y(x, t),
peridynamics avoids assumptions on the smoothness of the displacement field, in contrast to
classical elasticity. The mapping from a bond x0 − x to y(x0 , t) − y(x, t), or equivalently
u(x0 , t) − u(x, t), is in general nonlinear. The use of the deformation gradient is tantamount to
approximating this nonlinear map with a local linear approximation.
Denote by y0 , y the values y(x0 , t), y(x, t), respectively, and consider the integral operator
Z
k y0 − y, x0 , x dx0


(1)
B

representing the (internal) force density. The integral operator is nonlocal because material
points x0 6= x are involved. The constitutive equation relating deformation to internal force
density is contained in the kernel k that has units of force per volume squared. As an example,
consider a proportional microelastic material [9] with the peridynamics kernel (derivable from a
stored energy function)

y0 − y ky0 − yk − kx0 − xk
k y0 − y, x0 , x = cs 0


, s= (2)
ky − yk kx0 − xk,

where c/kx0 − xk > 0 is the stiffness per unit volume squared. The corresponding force density
operator of classical elasticity is given by

C(x, t) := ∇x · P (F(x, t)) , (3)

2
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

where P denotes the Piola-Kirchoff stress tensor and ∇x · the divergence with respect to x.
The operator C is local because its value at a point x depends, via the deformation gradient,
only on the value of the stress tensor at that same point. The Piola-Kirchoff stress tensor is
a constitutive equation, for example, a mapping from strain to stress tensors incorporating the
material behavior.
The peridynamics force density operator (1) obviates the need for the stress/strain
relationship implicit in (3). Such relationships are based on the idea of contact forces
(interactions between material in direct contact) and follow from Cauchy’s postulate [10, pp. 60–
61]. Peridynamics is fundamentally different in that material separated by a finite distance may
exert forces on each other. The paper [11] establishes conditions under which a peridynamic
stress tensor ν(x, t) exists such that
Z
k y0 − y, x0 , x dx0 .


∇x · ν(x, t) =
B

But even in this case, peridynamics is fundamentally different from classical elasticity in that
the peridynamics stress tensor represents a sum of forces per unit area through x instead of at
x, as is the case for the Piola-Kirchoff stress tensor P.
The bond-based peridynamics model originally introduced in [6] assumed that k(y0 − y, x0 , x)
is collinear with y0 − y. This assumption results in an effective Poisson ratio = 1/4 for isotropic,
linear, microelastic materials. The paper [6, § 15] also briefly described how the bond-based
peridynamics kernel may be augmented so that a Poisson ratio 6= 1/4 can be achieved. Recently,
in [7], the bond-based peridynamics model of [6] was generalized to that of a state-based
peridynamics model for which k(y0 − y, x0 , x) may be a force state; for example, it may depend
on the collective behavior at x and x0 . Peridynamics states is the continuum equivalent of the
multibody potentials of classical particle mechanics that go beyond central force interactions (or
the extension discussed in [6, § 15].)

2.2. Balance of linear and angular momentum

The peridynamic global balance of linear momentum is
Z Z Z Z
ρ(x)ü(x, t) dx = k(y0 − y, x0 , x) dx0 dx + b(x, t) dx, (4)
Ω Ω B/Ω Ω

where ρ(x) and b(x, t) are the mass and body force densities, respectively. This balance
holds over any Ω ⊂ B if and only if the peridynamic kernel is antisymmetric, for example,
k(y0 − y, x0 , x) = −k(y − y0 , x, x0 ). The first term on the right-hand side of (4) is the analogue
of the contact force in classical elasticity; and when Ω = B, this force is zero implying that
external (to B) contact forces play no role in peridynamics. We may define this first term to
represent the force that the material in B/Ω exerts on the material in Ω.
The local/nonlocal distinctions drawn can be given a mechanical interpretation. This force is
nonlocal precisely because material inside and outside of Ω interacts. In contrast, the classical
theory of elasticity restricts the force interaction to the surface of Ω. The peridynamic global
balance of angular momentum is
Z Z Z Z
0 0 0
y × ρ(x)ü(x, t) dx = y × k(y − y, x , x) dx dx + y × b(x, t) dx (5)
Ω Ω B Ω

for Ω ⊂ B. The R first

R term on the right-hand side of (5) represents the torque of Ω and may
be replaced by Ω B/Ω y × k(y0 − y, x0 , x) dx0 dx if and only if k(y0 − y, x0 , x) is collinear with
y0 − y as in bond-based peridynamics.
3
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

In classical elasticity, the locality of force interactions also implies that there is no length scale
independent of specific material behavior (e.g., outside of material inhomogeneities.) On the
other hand, the nonlocal force density operator (1) does contain a length-scale, the peridynamic
horizon—a nonzero volume over which material exerts forces. For instance, let the horizon be
given by
H(x) = {x0 ∈ B | 0 ≤ kx0 − xk < δx } ⊂ B
so that the first terms on the right-hand side of (5) and (4) are replaced with Ω H(x) y × k(y0 −
R R

y, x0 , x) dx0 dx and Ω H(x)/Ω k(y0 − y, x0 , x) dx0 dx, respectively, for Ω ⊂ H(x). The horizon
R R

may vary with x and/or the forces about x so that the largest horizon over the body defines a
maximum length scale.
The recently accepted paper [12] explains how the general state-based peridynamic material
model converges to the classical elastic material model as the length scale, or horizon, vanishes,
(assuming that the underlying deformation is sufficiently smooth). Building on the analysis of
[12] and [11], a collapsed peridynamics stress tensor ν 0 (x, t) is derived that is a function of the
deformation gradient at x and so is a Piola-Kirchoff stress tensor. This limiting Piola-Kirchhoff
stress tensor is differentiable and its divergence represents the force density due to internal
forces. The limiting, or collapsed, stress-strain model satisfies the conditions in the classical
theory for angular momentum balance, isotropy, objectivity, and hyperelasticity, provided the
original peridynamic constitutive model satisfies appropriate conditions.
A body composed of discrete particles (e.g., atoms) can be represented as a peridynamic
body. For example, suppose a set of discrete particles is given with reference positions xi and
masses mi , i = 1, 2, . . . , n. Let the force exerted by particle j on particle i P after deformation of
the system be denoted by F j,i (t). Define a peridynamic body by ρ(x) = i mi δ(x − xi ) and
k(y0 − y, x0 , x) = 0 − x )δ(x − x ) for all x, x0 in R3 , where δ(x) denotes
P P
i F
j6=i j,i (t)δ(x j i
the Dirac delta function or density (with units of per volume). Global conservation of linear
momentum (4) where Ω is a volume enclosing only xi reduces to
n
X
mi ü(xi , t) = Fj,i (t), i = 1, 2, . . . , n,
j6=i

which is the familiar statement of Newton’s second law in the particle mechanics setting.
Replacing the Dirac delta densities with more general probability densities provides a mechanism
for converting the set of discrete particles into a (continuum) peridynamics material model.

2.3. Discretization
Let x1 , . . . , xn be distinct points in B, where B = ni=1 Ωi and xi ∈ Ωi ⊂ H(xi ). The spatial
S
discretization of (4) results in
X
ρ(xi )ü(xi , t) = k(yj − yi , xj , xi )Vj + b(xi , t), i = 1, . . . , n, (6)
j6=i

where Vj is the volume of Ωj . Note that the sum is not over n − 1 but rather over the
number of xj ∈ H(xi ). An advantage of quadrature-based approaches is that they lead to
three-dimensional, meshless Lagrangian implementations. A time integration scheme is then
applied to (6); see [9, 13] for details. For a uniformly spaced xi , the midpoint quadrature rule
gives a quadratic spatial rate of convergence. See [14, 15] for extensive verification studies in
one and two dimensions (including adaptivity and comparisons with a commercial finite-element
code), respectively. The papers [16, 17] discuss the use of more sophisticated quadrature for the
internal force calculation, and [18] discusses an implementation within an engineering finite
element analysis code.
4
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

3. Applications
We review three successful multiscale applications of peridynamics within engineering analyses.
All of the results discussed here were obtained with the Emu [9] or the molecular dynamics
code LAMMPS [19, 13]. Both implementations are meshless three-dimensional Lagrangian
codes—a mesh generator is not required. The examples (including validation where available)
provide concrete evidence that peridynamics has been successfully used for three-dimensional
engineering analyses that strain, perhaps go beyond, the limits of classical elasticity, by exploiting
the nonlocality of force central to peridynamics. This nonlocality allows peridynamics to model
complex material behavior possessing an intrinsic length-scale. The three applications presented
deal with fracture problems—a quintessential multiscale phenomena. A compelling advantage
of peridynamics is the ability to deal with fracture in a straightforward fashion. This is
accomplished by allowing bonds (e.g., between x0 and x) to break when stretched beyond some
predetermined limit (see [9, 13] for a discussion) so enabling cracks initiate and propagate when
and where it is energetically favorable—unguided crack growth and complex interaction among
cracks is the rule. This is in contrast to standard fracture mechanics approaches; see [20] for a
review. The reader is referred to [21, 22] for applications of peridynamics not discussed in our
paper.

3.1. Polycrystal fracture

Brittle fracture is a catalyst for failure in ceramic materials subject to appropriate mechanical
loading. Predicting the performance and structural integrity of ceramics requires a fundamental
understanding of brittle fracture in polycrystalline ceramics at the grain-size level. Such an
understanding can then be used to design advanced structural ceramic systems with increased
impact resistance, higher thermal shock resilience, and so forth. Here we present a novel
approach for simulating fracture in polycrystalline ceramics, which has been difficult to do
in the past, in no small measure because of the complexity of the phenomenon.
Combined transgranular (cracks pass through the grains) and intergranular (cracks propagate
between the grains) fracture can take place in brittle fracture of ceramics. A transition between
the trans- and intergranular fracture is observed under certain loading. Experiments reported
in [23] show that, while the main fracture mode is intergranular, local transgranular fracture
also appears because of the existence of crystalline phase at grain junctions. In addition, crack
deflection and crack bridging mechanisms were observed because of the presence of rodlike grains.
Further evidence of the importance of accurate modeling of inter- and transgranular fracture in
polycrystalline ceramics is presented in [24].
Models for simulating brittle fracture in polycrystalline ceramics that can include combined
trans- and intergranular crack propagation have severe limitations, including the inability of
modeling propagating cracks that naturally coalesce and/or branch; limitation to modeling only
a single or a small number of cracks; and complicated algorithms that cannot extend to three
dimensions.
The finite-element method applied to equations of classical elasticity has been the preferred
tool used in numerical models of polycrystalline materials (see [25, 26, 27, 28]). Fracture in
polycrystalline materials has been modeled by using cohesive-zone models initiated in [29, 30, 31].
Recently, some difficulties regarding time discontinuities in cohesive zone models have been
pointed out in [32]. Bias of the crack propagation path and possible remedies are discussed
in [33]. Mathematical and physical limitations, and constraints on cohesive laws are discussed
in [34]. The most common approach for simulating brittle fracture of polycrystalline ceramics
is to consider only intergranular fracture: cracks are restricted to grain boundaries (see, e.g.,
[35, 36, 37], and finite elements with a “soft-kill method” are used to propagate the crack along
grain boundaries instead of using a cohesive zone model. Lattice spring models have been used
to study brittle fracture and damage in polycrystalline materials by, for example, [38, 39]. A
5
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

transition from intergranular to transgranular fracture with increasing grain boundary toughness
is observed in [38]; however, these studies are limited to 2D quasi-static analysis.
Compared to the methods discussed above, peridynamics for analyzing crack initiation,
propagation, and fragmentation in a rate-dependent mechanically loaded Voronoi polycrystalline
ceramics has important advantages. These include inter- and transgranular fracture are direct
consequences of the computations and they do not have to be postulated via ad hoc assumptions
as is the case for the classical approach; mode-transition and mode-mixing of crack propagation
is naturally captured by the peridynamic formulation; fracture at triple-junction points is not
controlled by ad hoc assumptions but by the actual loading conditions in a region surrounding
the triple-junctions; and the meshfree discretization of peridynamics eliminates the need for
complicated meshing algorithms for Voronoi polycrystals that represents an impediment to three
dimensional computations.
We now summarize some results for simulating polycrystalline fracture in silicon ([40] will
contain more details). Single-crystal silicon has a cubic structure, with the elastic moduli
c11 = 166 GPa, c44 = 80 GPa, and c12 = 64.5 GPa. Extensions to other types of crystals is
immediate and do not require special considerations. We generate a Voronoi polycrystal over a
square with dimensions of 1 cm by 1 cm. Using 120 randomly distributed x and y coordinates for
the Voronoi cells seeds, we obtain a grain-size distribution that is close to a Weibull distribution.
This number of cells in the Voronoi structure gives the same grain-size distribution for different
realizations (see [40]). For each Voronoi cell we assign a certain, random, orientation angle (from
a uniform distribution). We define the micro-stiffness for the peridynamic bonds to vary with
their orientation inside the grain and match the effective stiffness. Thus, the micro-stiffness
depends on the orientation of each grain and the orientation of the bond inside the grain.

Figure 1. Fracture dependence on parameter β: β=0.25 (left), β=1 (center), and β=4(right).

The peridynamic bond relative elongation is given by s defined in (2), and the critical relative
elongation at which a bond p breaks is associated with the fracture energy and the peridynamic
horizon via (see [9]) s0 = (5G0 )/(9κδ), where G0 is the material’s (single crystal silicon)
fracture energy and κ is the bulk modulus. Bonds that connect different grains have special
properties that may be related to the properties of the grain boundaries. We select the micro-
stiffness for such bonds to be the average of the corresponding directions in the two grains.
For the critical relative elongation of the bonds that have ends in different grains we define an
“interface strength coefficient” β = sGB GB is the critical relative elongation of the
0 /s0 , where s0
bonds that pass over a grain boundary, and s0 is the value defined above for the single-crystal
material. As we vary β from subunitary to superunitary numbers, we change the strength of
the grain boundaries from lower than the single crystal to higher than the single crystal.
The tests are performed using EMU [9] on a two-dimensional grid of 100 nodes in
each direction. Extensions to three-dimensional simulations are immediate and does not
require special considerations (except for the increased computer storage and floating point
computation). We apply velocity boundary conditions on two sides for stretching along the
6
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

horizontal direction. The top and the bottom sides are free. The applied strain rate is constant
equal to 50/s and the initial velocity distribution is consistent with this strain rate. We simulate
32 µs to reach an effective strain of 0.2%. We do not introduce initial cracks.

Figure 2. Evolution of damage and cracks in time for the β = 1 case. The top and bottom
rows show the polycrystals boundaries and damage indices, respectively.

The plots in figure 1 show the polycrystal after 22 µs of the simulation for different values
of the interface strength coefficient β. As β increases from 0.25 (left figure) to 1 (center) to 4
(right figure), we transition between purely intergranular fracture to pure transgranular fracture,
passing through combined trans- and inter-granular fracture observed for average values of β. To
better observe the ability of the peridynamic method to simulate crack propagation and crack
interaction in polycrystalline ceramics we monitor the damage time-evolution for the case when
β = 1 (see figure 2). We note that there are cracks that begin from the top and bottom edges,
but also cracks that start in the interior of the sample, grow, branch, and finally join together.

3.2. Composite fracture and failure

The prediction of the spontaneous formation of a crack and its subsequent growth in load-
carrying structures such as the wing or fuselage of an airplane presents a long-standing
problem in computational solid mechanics. In a complex loading situation such as a bird
strike, multiple cracks may occur simultaneously and interact. The path along which a crack
propagates in a three-dimensional structure is generally not known in advance and must be
determined as part of the solution of a computer simulation and ultimately depends on the
underlying material. Recently anisotropic composite materials (such as CFRP) are replacing
more traditional isotropic materials (such as high-strength aluminum alloys) in part because of
their larger specific strength promises significant weight savings. A typical composite material
used within the aircraft industry is called “prepreg” and consists of individual plies stacked on
top of each other and bonded together in an autoclave oven. Each ply consists of extremely stiff
fibers that are embedded in a relatively soft matrix. Such a material exhibits many different
failure mechanisms that can be active at the same time: delamination of adjacent plies or
microbuckling of individual fibers are examples of common failure mechanisms in composites
that have no counterpart in metals. Resolving the myriad failure mechanisms is essential in a
simulation so that the correct path along which a crack propagates is computed. Peridynamics is
ideally suited for describing damage propagation in composite materials because of the ability to
handle crack propagation when the crack path is not known a priori, the natural correspondence
7
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

between fibers and peridynamic bonds, and the (infinitely) larger number of degrees of freedom
in a peridynamic deformation state (in comparison to the classical local deformation gradient)
that can capture a greater range of failure phenomena.
Anisotropy is introduced by allowing bonds in different directions to have different stiffness
and failure properties. In the peridynamic model of a unidirectional fiber-reinforced lamina, the
bonds in the direction of the fibers represent the fiber properties, while the bonds in all other
directions represent the matrix properties. The fiber and matrix bonds have properties that are
independent of each other, including both elastic response and failure properties.
The peridynamic model of a lamina is illustrated in figure 3. The stiffness of matrix and fiber
bonds, cm , cf , are chosen to match the effective laminate stiffness E11 and E22 . The critical
ft
strains for bond breakage in tension, smt0 and s0 , are used to match the critical failure strains
in a 0 and 90 degree lamina. In compression, the matrix and fiber bonds are assumed to fail
at the same critical strain, denoted sc0 . This assumption is made because the microbuckling
mechanism for lamina failure in compression involves simultaneous localization in both the resin
and fibers; hence it is not possible to identify separate failure criteria for each in compression.
A model of a laminate is constructed from a stack of lamina models, each with a particular
fiber direction. The laminas interact with each other through special peridynamic bonds that
capture the through thickness stiffness E33 as well as the interlayer shear stiffness G13 and the
corresponding energy release rates for mode I and II that can be measured in an DCB and ENF
tests. For more detailed descriptions see [41, 42, 43, 44].
Using this simple model, a wide variety of problems involving damage propagation in CFRP
panels were studied. Figure 4 shows a stiffened notched composite panel under compressive
loading. Only half of the panel is shown because of symmetry. Note that in addition to the
notch in the skin, the middle stringer has also been cut. The aim of such as simulation is to
predict the damage tolerance capabilities in an lower wing structure of a composite airplane.

3.3. Nanofiber networks

There is much interest in polymer nanofibers because of their desirable structural mechanical
properties. In particular, they possess a large surface area to volume ratio, as well as high
axial strength and extreme flexibility. The use of nanofibers is being explored in such diverse
areas as composites, protective clothing, catalysis, electronics, biomedicine (including tissue
engineering, implants, membranes, and drug delivery), filtration, and agriculture [45]. Recently,
polymer nanofibers have also been proposed as a filters for chemical warfare defense [46]. The
ability to tailor the properties of nanofiber networks and membranes for specific applications
requires that an understanding of the mechanical properties of these networks and membranes
through modeling and simulation. A peridynamic model for simulating the mechanical behavior
and strength and toughness properties of three-dimensional nanofiber networks under imposed
stretch deformation was proposed in [47].
In this numerical model, the peridynamic prototype microelastic brittle (PMB) [9] was
augmented with a Lennard-Jones potential to mimic the effects of van der Waals forces. This
augmented potential was applied to nanofiber networks similar to that shown in figure 5(a).
The model domain is 400 nm wide by 400 nm high by 10 nm thick. The top, bottom, left,
and right ends have a constant velocity boundary condition applied to induce a biaxial strain
that eventually leads to the failure of the nanofiber network. Simulations performed with the
PDLAMMPS code [13] are shown in figure 5. Detailed numerical simulation in [47] led to
the observations that van der Waals interactions are important in increasing the strength and
toughness of the nanofiber network, and that heterogeneity in the strength of bonds in the
nanofiber network can increase its toughness and ductility. Additionally, reorientation due to
breakage and deformation and a Poisson effect due to van der Waals induced accretion are the two
main mechanisms that control the dynamic deformation and damage of nanofiber networks [47].
8
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

Acknowledgment
Sandia is a multiprogram laboratory operated by Sandia Corporation, a Lockheed Martin
Company, for the United States Department of Energy under contract DE-AC04-94AL85000.
The authors thanks Gail Pieper and Lori Diachin for their helpful comments.

References
[1] de Pablo J J and Curtin W A (eds) 2007 Multiscale Modeling in Advanced Materials Research (MRS Bulletin
vol 32) (Material Research Society)
[2] Miller R E and Tadmor E B 2007 Multiscale Modeling in Advanced Materials Research (MRS Bulletin vol 32)
ed de Pablo J J and Curtin W A pp 920–926
[3] Bažant Z P and Jirásek M 2002 J. of Eng. Mech. 128 1119–1149
[4] Chen Y, Lee J D and Eskandarian A 2004 Int. J. of Solids and Structures 41 2085–2097
[5] Eringen A C 2002 Nonlocal continuum field theories (Springer-Verlag NewYork, Inc)
[6] Silling S A 2000 J. Mech. Phys. Solids 48 175–209
[7] Silling S A, Epton M, Weckner O, Xu J and Askari E 2007 J. Elasticity 88 151–184
[8] Weckner O and Abeyaratne R 2005 J. Mech. Phys. Solids 53 705–728
[9] Silling S A and Askari E 2005 Comp. Struct. 83 1526–1535 EMU available at www.sandia.gov/emu/emu.htm
[10] Ciarlet P 1988 Finite Element Method for Elliptic Problems (North-Holland, Amsterdam)
[11] Lehoucq R B and Silling S A 2008 J. Mech. Phys. Solids 56 1566–1577
[12] Silling S A and Lehoucq R B 2008 J. Elasticity Article in Press, available online
[13] Parks M L, Lehoucq R B, Plimpton S J and Silling S A 2008 Comp. Phys. Comm. Article in Press, available
online
[14] Bobaru F, Yang M, Alves L F, Silling S A, Askari A and Xu J 2008 Int. J. Num. Meth. Eng. Submitted
[15] Bobaru F, Yang M, Alves L F, Silling S A and Askari A 2008 Int. J. Num. Meth. Eng. In preparation
[16] Emmrich E and Weckner O 2007 Math Mod Analysis 12 17–27
[17] Emmrich E and Weckner O 2005 Math and Mech of Solids 12 363–384
[18] Macek R and Silling S 2007 Fin. Elem. Anal. Design 43 1169–1178
[19] Plimtpon S J 1995 J. Comp. Phys. 117 1–19 LAMMPS available at http://lammps.sandia.gov
[20] Ingraffea A R 2007 Encyclopedia of Computational Mechanics (Solids and Structures vol 2) ed Stein E,
de Borst R and Hughes T J R (Wiley) chap 11
[21] Gerstle W, Sau N and Silling S 2007 Nuc. Eng. Des 237 1250–1258
[22] Silling S A and Bobaru F 2005 Int. J. Non-Linear Mech. 40 395–409
[23] Lee B T, Han B D and Kim H D 2003 Materials Letters 58 74– 79
[24] Sun X, Li G J, Guo S, Xiu Z, Duan K and Hu X 2005 J of the American Ceramic Society 88 1536–1543
[25] Ghosh S and Mukhopadhyay S 1993 Comp Meth Appl Mech Eng 104 211–247
[26] Ghosh S and Mallett R 1994 Comp. Struct. 50 33–46
[27] Zavattieri P, Raghuram P and Espinosa H 2001 J. Mech. Phys. Solids 49 27–68
[28] Zhang D, Wu M and Feng F 2001 International Journal of Plasticity 21 801–834
[29] Barenblatt G 1959 Applied Mathematics and Mechanics (PMM) 23 622636
[30] Dudgale D 1960 J. Mech. Phys. Solids 8 100–104
[31] Barenblatt G 1962 Advances in Applied Mechanics 7 55–129
[32] Papoulia K, Sam C and Vavasis S 2003 Int. J. Num. Meth. Eng. 58 679–701
[33] de Borst R 2003 Eng. Frac. Mech. 70 1743–1757
[34] Jin Z and Sun C 2005 Int. J. Frac. 134
[35] Grah M, Alzebdeh K, Sheng P, Vaudin M, Bowman K and Ostoja-Starzewski M 1996 Acta Materialia 44
4003–4018
[36] Holm E 1998 Journal of the American Ceramic Society 81 455–459
[37] Maiti S, Rangaswamy K and Geubelle P 2005 Acta Materialia 53 823 – 834
[38] Yang W, Srolovitz D, Hassold G and Anderson M 1990 In Simulation and Theory of Evolving Microstructures
(The Metallurgical Society: Warrendale, PA) chap Microstructural effects in the fracture of brittle
materials, pp 277–284
[39] Rinaldi A, Krajcinovic D, Peralta P and Lai Y C 2008 Mechanics of Materials 40 17–36
[40] Bobaru F and Silling S 2008 in preparation
[41] Askari A, Xu J and Silling S A 2006 Proc. 44th AIAA Aerospace Sciences Meeting and Exhibit AIAA 2006-88
(Reno, NV)
[42] Silling S A, Askari A, Nelson K, Weckner O and Xu J 2008 Proc. SAMPE Fall Technical Conference 2008
[43] Xu J, Askari A, Weckner O and Silling S A 2008 J. Aero. Eng. Article in Press

9
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

[44] Xu J, Askari A, Weckner O, Razi H and Silling S A 2007 Proc. 48th AIAA/ASME/ASCE/AHS/ASC
Structures, Structural Dynamics, and Materials Conference (Honolulu, HI)
[45] Dzenis Y 2004 Science 304 1917–19
[46] Ramaseshan R, Sundarrajan S, Liu Y, Barhate R, Lala N L and Ramakrishna S 2006 Nanotechnology 17
2947–2953
[47] Bobaru F 2007 Modelling Simul. Mater. Sci. Eng. 15 397–417

10
SciDAC 2008 IOP Publishing
Journal of Physics: Conference Series 125 (2008) 012078 doi:10.1088/1742-6596/125/1/012078

Figure 3. Peridynamic lamina model

Localization Panel bends downward

near notch

Axial displacement Out-of-plane displacement during Damage at late time shows

during
Figure 4. (Left todamage
right)growth
Axial displacement during
damage damage growth;
growth out under
disbond of plane displacement
stringer
(displacements exaggerated)
growth; damage at a later time shows disbonding.

Figure 5. Peridynamic simulation of polymer nanofiber network under biaxial strain. (Left to
right) Undeformed configuration; Biaxial strain of 17.6% (30 ns); Biaxial strain of 29.4% (50
ns). Colors indicate individual fibers.
11