Dynamic CDynamic Crack Propagation Based On Loss of Hyperbolicity and A New Discontimuous Enrichmentrack Propagation Based On Loss of Hyperbolicity and A New Discontimuous Enrichment

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING
Int. J. Numer. Meth. Engng 2003; 58:18731905 (DOI: 10.1002/nme.941)

Dynamic crack propagation based on loss of hyperbolicity
and a new discontinuous enrichment
Ted Belytschko
,
, Hao Chen
1, 2
, Jingxiao Xu
1
and Goangseup Zi
1
1
Department of Mechanical Engineering, Northwestern University, 2145 Sheridan Road,
Evanston, IL 60208-3111, U.S.A.
2
Livermore Software Technology Corporation, 7374 Las Positas Road, Livermore, CA 94550, U.S.A.
SUMMARY
A methodology is developed for switching from a continuum to a discrete discontinuity where the
governing partial dierential equation loses hyperbolicity. The approach is limited to rate-independent
materials, so that the transition occurs on a set of measure zero. The discrete discontinuity is treated by
the extended nite element method (XFEM) whereby arbitrary discontinuities can be incorporated in
the model without remeshing. Loss of hyperbolicity is tracked by a hyperbolicity indicator that enables
both the crack speed and crack direction to be determined for a given material model. A new method
was developed for the case when the discontinuity ends within an element; it facilitates the modelling of
crack tips that occur within an element in a dynamic setting. The method is applied to several dynamic
crack growth problems including the branching of cracks. Copyright ? 2003 John Wiley & Sons, Ltd.
KEY WORDS: nite element method; fracture mechanics; dynamic fracture; loss of hyperbolicity;
cohesive crack model; extended nite element method
1. INTRODUCTION
It is well known that the loss of hyperbolicity, i.e. a change of type of the momentum
equations for a rate-independent material leads to a localization of deformation to a set of
measure zero. In Bazant and Belytschko [1], a closed form solution was obtained for the one-
dimensional case for a rate-independent material and it was shown that the strain becomes a
Dirac delta function at the rst point where hyperbolicity is lost, i.e. at the point where the
tangent modulus loses its positive slope; material behaviour with a negative tangent modulus
is often called strain softening. As a consequence, the displacement eld develops a discon-
tinuity, i.e. the material separates, at this point. Although this change of type of the partial
dierential equation is often called ill-posedness, this characterization is controversial since a
Correspondence to: Ted Belytschko, Department of Mechanical Engineering, Northwestern University, 2145
Sheridan Road, Room A212, Evanston, IL 60208-3111, U.S.A.
E-mail: tedbelytschko@northwestern.edu
Contract}grant sponsor: Oce of Naval Research
Received 14 February 2003
Revised 12 March 2003
Copyright
?
2003 John Wiley & Sons, Ltd. Accepted 3 July 2003
1874 T. BELYTSCHKO ET AL.
solution can be obtained after loss of hyperbolicity. However, the continuation of the solution
requires additional models because the displacement discontinuity is developed without the
dissipation of any energy [1, 2]. Thus if we consider the separation process to be fracture,
modelling it by a rate-independent material with strain softening without transition to an in-
terface model that can dissipate energy is inappropriate as a model for the physical process of
cracking.
In this paper, we consider the treatment of the fracture process by introducing a discrete
discontinuity wherever the partial dierential equation changes type. Across this discontinuity,
a traction-displacement law, i.e. a cohesive law, is imposed. The energy dissipation across
the discontinuity is chosen to match the energy of fracture. The direction and velocity of the
propagation of fracture are directly determined by the loss of hyperbolicity criterion. Thus the
method is capable of providing the crack speed and direction from the material model. The
transition from continuum to discrete separation in an element is implemented through the
extended nite element method (XFEM), which permits arbitrary discontinuities to be added
to a nite element mesh without remeshing via discontinuous partitions of unity.
Another class of fracture methods are the inter-element models with cohesive laws, Xu and
Needleman [3], Camacho and Ortiz [4] and Ortiz and Pandol [5]; (these are often called
cohesive models but the name is misleading since cohesive fracture laws can be implemented
in many methods, such as the XFEM implementation described here). These models have
similar capabilities to the method described here, since the cohesive law can be triggered
wherever an adjacent element loses hyperbolicity. However, the directions of the crack are
limited to the element edges, so the crack paths are limited to specied directions. Furthermore,
element edges separate one by one, so the crack growth is not smooth. On the other hand,
these methods are simple and robust and some very good results have been obtained.
A third class of fracture methods are the embedded discontinuity methods, where the crack
is represented by a band of high strain in the element. Embedded discontinuity methods are
given in Belytschko et al. [6], Simo et al. [7], Dvorkin et al. [8] and many other papers;
see Jirasek [9] for a comprehensive study of these methods. In these methods, the unstable
material behaviour is conned to a narrow band, which can be arbitrarily aligned within the
element. However, the bands are added element-wise, so the discontinuity must propagate in
increments no smaller than the element size.
Discontinuous partitions of unity enrichments were rst used to model cracks by Belytschko
and Black [10], who used the discontinuous near tip eld to model the entire crack for elasto-
static problems. In Moes et al. [11] and Dolbow et al. [12], a step function enrichment was
developed for elements completely cut by the crack; they called the method the extended nite
element method (XFEM). In Belytschko et al. [13], the approach was generalized to arbitrary
discontinuities, including discontinuities in derivatives and tangential values of displacement.
Wells et al. [14] have modelled static cohesive cracks by this method. They dealt with a
time-independent material model and introduce the discontinuity when the stress exceeds the
tensile strength. Remmers et al. [15] have introduced an interesting, simpler variant of this
methodology where the discontinuity is introduced independently in three contiguous elements
when the principal tensile stress exceeds the strength of the material in an element.
In the extended nite element method proposed by Moes et al. [11], the enrichment changes
character when the crack passes from the interior of an element to the next element. When
the cracktip is within the element, the enrichment in Reference [11] is the Westergaard neartip
solution; once the cracktip passes out of the element, the enrichment is the Heaviside step
Copyright ? 2003 John Wiley & Sons, Ltd. Int. J. Numer. Meth. Engng 2003; 58:18731905
DYNAMIC CRACK PROPAGATION BASED ON LOSS OF HYPERBOLICITY 1875
function. While this causes no diculties in linear equilibrium solutions (i.e. elasto-static
problems), it is not readily incorporated in methods for time-dependent solutions.
Therefore two approaches were taken in this paper:
1. The cracktip was restricted to crossing one element at a time; i.e. passing from edge to
edge; the enrichment for that situation can be treated with the step function enrichment.
2. A new enrichment was developed for crack tips within an element that is compatible
with the step function enrichment.
Both of these approaches yield similar crack paths. The second approach yields smoother
results since the crack propagates smoothly through each element.
The path and velocity of the crack were determined by loss of hyperbolicity. For this
purpose, a hyperbolicity indicator was dened as the minimum of the scalar projection of
the acoustic tensor. The minimum denes the direction of the surface of the developing
discontinuity (i.e. the crack) and by progressing to the point where the hyperbolicity indicator
vanishes, the cracktip speed is determined. Since the hyperbolicity indicator is a function of
the stress and strain state in an element, it is only piecewise continuous. In fact, for the
constant strain triangular element used in this study, it is piecewise constant. Therefore, in
order to permit a smooth passage of the crack through an element, the hyperbolicity indicator
was projected on a piecewise continuously dierentiable eld. With this approach, it becomes
possible to consistently determine the velocity of a cracktip within each element as the crack
progresses. Loss of hyperbolicity was previously used in dynamic crack propagation by Gao
and Klein [16]. Loss of ellipticity has been used to drive cracks in equilibrium solutions
by Oliver et al. [17]. Peerlings et al. [18] have discussed the relationship between crack
formation and loss of ellipticity in damage models.
The paper is organized as follows. Section 2 describes the governing equations, the crack
representation and the motion in the presence of a crack. Section 3 describes the nite element
approximation of the motion by XFEM. Then we introduce the treatment of the loss of
hyperbolicity and the cohesive laws in Section 4. In Section 5 we give the weak form and
the discretized equations. Then we will describe several numerical studies in Section 6. Finally,
Section 7 provides a summary and some concluding remarks.
2. GOVERNING EQUATIONS AND MOTION
Consider a body
0
in the reference conguration as shown in Figure 1. The material co-
ordinates are denoted by X and the motion is described by x =(X, t) where x are the spatial
co-ordinates. In the current conguration the image of
0
is . We dene a crack surface
implicitly by [
1
(x) =0. To specify the edge of the crack we construct another set of implicit
functions [
2
(x) whose gradient is orthogonal to the gradient of [
1
(x). The crack edge is given
by [
1
(x) =[
2
(x) =0; [
i
(x) is only dened in a subdomain about the crack denoted by
c
. We
also dene [
0
1
(X) and [
0
2
(X, t) where [
0
i
(X) is the image of [
i
(x) in the initial conguration.
The surface [
0
1
(X) =0 corresponds to the surface where hyperbolicity has been lost, i.e.
the crack surface, and is also denoted by
0
c
. We arbitrarily choose one side of the crack to
be the direction of the normal to the crack, n
0
and choose the sign of [
0
1
so that
n
0

X
[
0
1
0 (1)
c
c
c
c
( , ) X
t
f =0
f >0
n
1
1
1
f <0
0
n
0
f =0
f >0
f <0
n
0
1
1
1
0
0
0
0
0
u
0
0
Figure 1. A two-dimensional body with a crack and its representation in the initial and current domains.
f =0
2
f <0
2
f >0
2
f >0
f <0
1
1
1
f =0
crack
Figure 2. A two-dimensional crack represented by two sets of implicit function [
1
and [
2
.
The function [
0
2
(X, t) is dened so that [
0
2
(X, t)0 on the crack surface and vanishes on
the crack edge. The crack edge (i.e. the crack front) is dened by the curve given by the
intersections of the surfaces [
0
1
(X) =0 and [
0
2
(X, t) =0. In two dimensions the crack front is
the point where the zero isolevel curves of these functions intersect, as shown in Figure 2.
The strong form consists of the following:
cP
)i
cX
)
+ j
0
b
i
j
0
u
i
=0 in
0
(2)
n
0
)
P
)i
=

t
0
i
on
t
0
(3)
n
0
)
P
)i
=n
0
)
P
+
)i
=t
0
i
on
c
0
(4)
u
i
= u
i
on
u
0
(5)
where P is the nominal stress, j
0
the initial density and b the body force per unit mass, B
0
is the cohesive traction across the crack,

t
0
is the applied traction on the Neumann boundary
t
and u is the applied displacement on the Dirichlet boundary
u
;
u
t
=,
u
t
=0.
Indicial notation is used for any lower case indices, so repeated subscripts imply summations.
We can also write the momentum equation and the traction boundary condition in terms of
the Cauchy stress as follows
co
i)
cx
)
+ jb
i
j u
i
=0 in (6)
n
i
o
i)
=

t
i
on
t
(7)
n
)
o
)i
=n
)
o
+
)i
=t
i
on
c
(8)
u
i
= u
i
on
u
(9)
where

t
i
is the prescribed traction on the current surface, and
c
is the image of
0
c
in the
current conguration.
We consider only rate-independent materials. The Truesdell rate of the Cauchy stress can
then be related to the rate of deformation D
k!
by
o
i)
=C
i)k!
D
k!
(10)
where the tangent modulus C may depend on the stress A and other state variables.
We dene the tensor A by
A
i)k!
=C
i)k!
+ o
i!
o
)k
(11)
The momentum equation with a constitutive law given by Equation (10) is then hyperbolic
when
n
i
h
)
A
i)k!
n
k
h
!
0 n and h ,= 0 (12)
The above conditions are identical to the conditions for ellipticity of the equations of
equilibrium. They can also be viewed as a condition for material stability, since the inequality
implies the stable response of an innite medium in a uniform state of stress and strain when
subjected to the perturbation u =he
ict+knx
(see Belytschko et al. [19]). The hyperbolicity
conditions are equivalent to the requirement that the material admit a plane progressive wave
(see Marsden and Hughes [20]). It also implies that the momentum equation is a hyperbolic
partial dierential equation.
Extrapolating from the results of Bazant and Belytschko [1], it can be inferred that when
the hyperbolicity condition is violated for a rate-independent material, then the strain localizes
to a set of measure zero on which the strain becomes innite. It must be stressed that these are
empirical inferences and that little appears to be known about the morphology and evolution
of such surfaces in multi-dimensions. For example, it is possible that loss of hyperbolicity in
a body in R
3
occurs on lines, on arrays of points, fractals, etc. However, we will assume that
hyperbolicity is lost on a smooth surface.
Let the displacement eld be decomposed into a continuous part and a discontinuous part,
as in Armero and Garikipati [21] and Oliver [22].
u =u
cont
+ u
disc
(13)
The discontinuous part of the displacement eld is then given by
u
disc
=(t, X)H([
0
1
(X))H([
0
2
(X, t)) (14)
where H is the Heaviside function given by
H(x) =
_
1 x0
0 x0
(15)
and (X, t) is a continuous function that satises the following conditions:
1. That vanish on the zero isolevel of [
0
2
(X, t), i.e.
(t, X) =0 for X[
0
2
(X, t) =0 (16)
2. That be non-negative on the zero isolevel of [
0
1
(X, t), i.e.
(t, X)0 for X[
0
1
(X, t) =0 (17)
The rst condition enforces continuity of the crack opening displacement at the cracktip. The
second condition precludes overlap of the crack surfaces.
The function [
0
1
(X) is not a function of time. This construction of [
0
1
is motivated by the
fact that once hyperbolicity is lost on a surface, we can never revert to a continuum treatment
on that surface and therefore the surface must be xed in the reference conguration. In other
words, once a crack developes, the image of that crack in the reference domain is xed, i.e.
any crack surface that has developed is frozen. In the computational method, this implies that
the crack surface must be considered separately from the continuum for the remainder of the
analysis. The function [
0
2
(X, t) is a function of time to reect the growth of the crack surface.
The discontinuous part of deformation gradient F is obtained from its denition
F
disc
i)
=
cu
disc
i
cX
)
=
c
i
cX
)
H([
0
1
)H([
0
2
) +
i
o([
0
1
)
c[
0
1
cX
)
H([
0
2
) +
i
H([
0
1
)o([
0
2
)
c[
0
2
cX
)
(18)
where o() is the Dirac delta function. We note that
n
0
i
=
c[
0
1
cX
i
(19)
since (X, t) =0 at the crack edge, the last term vanishes. Therefore, the deformation gradient
is given by
F
disc
i)
=
c
i
cX
)
H([
0
1
)H([
0
2
) + m
0
)
i
o([
0
1
)H([
0
2
) (20)
Thus the deformation gradient is innite on the surface of lost hyperbolicity.
To treat the mechanics on this surface, we then switch from a continuum description of the
constitutive behaviour Equation (10) to jump relations. We write these relations in rate form
in terms of the tractions on the current surface of discontinuity:
t
i
=D
i)
< u
)
= for [
0
1
=0, [
0
2
0 (21)
where D
i)
is a constitutive matrix for the interface and t
i
is an objective rate of the traction
(see Reference [19, p. 585]). In the numerical examples in this paper, we neglect the tangential
components of the traction, i.e. only the normal component is considered.
3. FINITE ELEMENT APPROXIMATION
In this section, we describe the enriched nite element approximation that models the discon-
tinuity across the crack interface. Two types of models are considered:
1. Element-by-element progression where the crack progresses one complete element at a
time, i.e. whenever it grows, it advances to an edge of an adjacent element.
2. Continuous progression where the crack can grow continuously within an element but
only in a xed direction; the crack change its direction as it enters a new element.
In element-by-element progression, the displacement approximation is
u
h
(X) =
n
I =1
N
I
(X)u
I
+

I S
N
I
(X)
I
(X)q
I
(22)
where q
I
are enrichment variables and
I
is given as
I
(X) =H([(X)) H([(X
I
)) (23)
where [(X) [
0
1
(X) =0 denes the crack interface as shown in Figures 1 and 2 (we drop
the subscript and superscript on [), S is the set of enriched nodes, and n is the total number
of nodes.
We will also use the following compact expression for the displacement:
u
h
(X) =
n
I =1
N
I
(X)u
I
+

I S
I
(X)q
I
(24)
where
I
(X) =N
I
(X)
I
(X) (25)
We next describe the procedure for selecting the set of nodes S to be enriched for a mesh
of constant strain triangular elements. The procedure depends on whether we are modelling
the nucleation of a discontinuity, say from loss of hyperbolicity in an element not contiguous
to an existing crack, or the propagation of a discontinuity. We assume here that we have
a constitutive relation that models both crack nucleation and crack propagation by loss of
hyperbolicity.
We rst describe the method for the case when the crack grows one element at a time;
consider a mesh of three-node, constant strain triangular elements. The nodes to be enriched
are selected as follows. Let the nodes of a generic element e
1
be I , J, K. Let the normal
to side JK be denoted n
I
; it is the normal to the side opposite to node I . Suppose that the
material in element e
1
loses hyperbolicity according to Equation (12) with a normal n and
element e
1
is not contiguous to any existing cracked elements. The specic node I to be
(a)
(b) (c)
Figure 3. Crack formation due to loss of hy-
perbolicity in element e
1
in (a); the enriched
nodes depend on the direction of the crack as
shown in (b) and (c).
Figure 4. Enrichment scheme for crack initiating
at the edge of a body.
enriched is then given by
I =arg
_
max
I
(n
I
n)
_
(26)
i.e. the node I is the node opposite to the side that is most closely aligned with the normal
to the incipient crack n according to Equation (12).
When the enrichment is added to node I , the displacement elds in the adjacent two
elements also become discontinuous, as illustrated in Figure 3. In the two elements denoted
as e
2
and e
3
, the discontinuity varies linearly and vanishes at the edge of the elements as
shown in Figure 3.
Note that the enrichment is not strictly a partition of unity, but is instead a local partition
of unity; see Chessa et al. [23]. Only the set of elements around the crack are enriched. On
the elements adjacent to these, the set of functions [
I
(X), I S is a blending function. So,
multiplying the step function by the shape function simply provides a localization for the step
function.
If the material rst fails (i.e. there is a loss of hyperbolicity) in an element adjacent to the
edge of the body, the node to be enriched is selected by Equation (26) in terms of a virtual
element with one node outside the body as shown in Figure 4. The normal n
I
is dened as
before for this virtual element in evaluating (26). The magnitude of the discontinuity varies
linearly in the element and vanishes at the interior edge of the element.
The procedure for selecting the subsequent nodes to be enriched depends on whether or
not loss of hyperbolicity occurs in an element adjacent to an element with a discontinuity.
When the element is contiguous to a failed element, the node to be enriched depends on the
direction of the crack, i.e. the normal obtained by the hyperbolicity indicator. The next node
to be enriched is selected as follows. From the normal to the crack, we know the direction of
propagation and hence which edge of the adjacent element the crack will impinge. The node
Figure 5. Schematic of the enrichment scheme for cracks ending within an element.
to be enriched, i.e. the node added to the set S, is then the node opposite to that edge; this
is illustrated for two cases in Figure 3.
When the crack is allowed to grow continuously within an element, the crack is restricted to
be rectilinear within the element. The enrichment for the partially cut element is constructed
as follows. Consider an element in which a crack will grow to point A as shown in Figure 5.
For clarity, we consider the particular node numbers indicated in Figure 5. The procedure
consists of the following.
1. Determine the element edge that the crack will intersect when it completely cuts the
element, i.e. edge 3-1 in Figure 5, and the edge already cut by the crack, i.e. edge 1-2
in Figure 5.
2. Find the common node of those two edges. In the case shown in Figure 5, it is node 1.
3. Construct a virtual triangle by connecting node 1 to the opposite edge so that the crack
tip lies on side 1-3
; designate the nodes of the virtual triangle by 1, 2, 3
. The shape
functions for the virtual element are the triangular co-ordinates of the element (1, 2, 3
),
which are denoted by N
I
(^
), where ^
are the triangular co-ordinates of (1, 2, 3
); the
node to be enriched is node 2.
The displacement approximation is then
u
h
(X) =
n
I =1
N
I
(^)u
I
+

I S
I
(X)q
I
(27)
where
I
(X) =
_
N
I
(^)(H([(X)) H([(X
I
))) for a fully cut element
N
I
(^
)(H([(X)) H([(X
I
))) for a partially cut element
(28)
We next show that this displacement approximation is continuous across edge 1-2. The
displacement along edge 1-2 in element e
1
in Figure 5 is
u
h
(X) =N
1
u
1
+ N
2
u
2
+ N
4
u
4
+ N
2
2
q
2
+ N
4
4
q
4
=N
1
u
1
+ N
2
u
2
+ N
2
2
q
2
(29)
The displacement along edge 1-2 in element e
2
is
u
h
(X) =N
1
u
1
+ N
2
u
2
+ N
3
u
3
+ N
2

2
q
2
=N
1
u
1
+ N
2
u
2
+ N
2

2
q
2
(30)
where N
I
=N
I
(
I
). The last step in the above two equations use the condition that N
4
and
N
3
vanish along edge 1-2. Since the modied shape function N
2
is equivalent to the standard
shape function N
2
on edge 1-2, i.e. N
2
(^
) =N
2
(^), the displacement eld is continuous across
edge 1-2. The partially enriched element developed here for linear triangular elements has been
extended to quadratic triangular elements by Zi and Belytschko [24].
4. COHESIVE LAW
The mechanics of the discontinuity that is created by loss of hyperbolicity, i.e. the crack,
will be treated by a cohesive model. In a cohesive model, the energy dissipated due to crack
propagation is modelled by a tractiondisplacement jump relationship at the crack surface; see
Barenblatt [25], Dugdale [26], Hillerborg et al. [27], Planas and Elices [28], etc.
We only consider cohesive laws involving the normal component of the traction. The
displacement jump o
N
is dened by
o
N
=n <u= =n

I S
N
I
(X)q
I
(31)
The last term in the above is derived as follows from (27):
<u= =u(X
+
) u(X
)
=
n
I =1
N
I
(X
+
)u
I
+

I S
N
I
(X
+
)
I
(X
+
)q
I
I =1
N
I
(X
)u
I

I S
N
I
(X
)
I
(X
)q
I
=

I S
N
I
(X)(
I
(X
+
)
I
(X
))q
I
=

I S
N
I
(X)q
I
(32)
Figure 6. Two dierent cohesive models, in which the area under a cohesive law is the fracture energy
G
F
: (a) linear cohesive law; and (b) bilinear cohesive law.
In the above, X
represents X cn
0
where c 0 and n
0
is the normal to the crack surface
in the reference conguration. In the second line of the above, we have used the continuity
of the shape functions across the crack, i.e. that N
I
(X
+
) =N
I
(X
).
The normal traction is dened by
t
N
=n B (33)
We will use traction laws of the form depicted in Figure 6 with a radial return algorithm as
in References [5, 29, 30]. When the material switches from continuous to discontinuous the
traction at the interface takes on the current value of the traction in the element. The cohesive
strength t
max
is not a xed parameter of the cohesive law. Otherwise, the tractions can be
suciently discontinuous to lead to noise (see Papoulia et al. [31]).
Given the shape of a cohesive law, the cohesive crack model is completely dened by the
fracture energy G
F
and the cohesive strength t
max
. Therefore, the critical crack opening o
max
is calculated from G
F
and t
max
. For example, for the linear cohesive model shown in Figure
6(a), the critical crack opening o
max
is
o
max
=
2G
F
t
max
(34)
The traction law for the case shown in Figure 6(a) is formulated as an incremental law
with radial return by the following algorithm
if o
n+1
N
60 go to end
t
trial
N
=t
n
N
+ E
n
(o
n+1
N
o
n
N
)
if t
trial
N
[(o
n+1
N
) go to end
t
n+1
N
=[(o
n+1
)
E
n+1
=t
n+1
N
}o
n+1
N
end
Note that if o
n+1
N
60, i.e. whenever the crack surfaces close, the tractions t
N
are computed by
the equations of motion; the cohesive law is only active when the crack surfaces are separated.
The relation between traction and the jump in displacement does not suce to provide
closure to the problem. In other words, the cohesive law and the strong form given by
y

n
h
x
Figure 7. n and h in two dimensions. Figure 8. Crack increment and the hyperbolicity
indicator e.
Equations (7)(10) do not provide a complete statement of the problem of crack propagation.
Closure is provided by the loss of hyperbolicity condition, which can be viewed as a crack
propagation law. Similarly, a maximum tensile stress criterion can provide closure.
It is interesting to observe that in the inter-element approach of References [3, 5], a crack
propagation law is not needed. Since the crack advances one edge at a time (or has the
possibility of advancing partially through all edges when the cohesive model joins all element
as in Xu and Needleman [3]), there is no need for a propagation law. Thus the inter-element
approach achieves closure by the discretization.
The following describes our methodology for using the loss of hyperbolicity criterion to
propagate the crack. The methodology described here is only applicable to rate-independent
materials. This method also allows a new crack to initiate and determines its direction. Thus
there is no need for a crack growth law: the crack growth, i.e. the cracktip velocity and the
direction of the crack, are determined by the constitutive equation.
In essence, the method uses the polarization of the plane wave emanating from the hyper-
bolicity condition to set the direction of the crack and the value of the hyperbolicity condition
to set the speed. Let
e = min
n, h
(n
i
h
)
A
i)k!
n
k
h
!
) (35)
where A is dened in (11). We call e the hyperbolicity indicator. The momentum equations
remain hyperbolic as long as e0. If the inequality e0 is violated in the neighborhood
of a cracktip, we then choose the normal of the surface increment (line increment in two
dimensions) to be the vector n that minimizes e. The direction of the cracktip is denoted by
a unit vector s, and it is given by
s n =0 (36)
For example, in two dimensions, we can express n and h by two angles 0 and [, i.e.
n
1
= cos 0, n
2
= sin 0, h
1
= cos [, h
2
= sin [; see Figure 7. The hyperbolicity indicator cor-
responds to the minimum of the right hand side of (35), so
ce
c0
=0 (37)
ce
c[
=0 (38)
Substituting Equation (35) into the above equations, we get two equations for 0 and [:
dn
i
d0
h
)
A
i)k!
n
k
h
!
+ n
i
h
)
A
i)k!
dn
k
d0
h
!
=0 (39)
n
i
dh
)
d[
A
i)k!
n
k
h
!
+ n
i
h
)
A
i)k!
n
k
dh
!
d[
=0 (40)
For the material laws we have used, n and h are approximately parallel to each other. This
means that the incipient instability is associated with a displacement normal to the surface,
which corresponds to crack opening.
The hyperbolicity indicator is a piecewise continuous function, i.e. a C
1
function, since it
depends on the tangent modulus and the stresses, which are C
1
. This prevents the smooth
motion of the cracktip in an element. Therefore, we project e onto a C
0
eld. The continuous
values of the hyperbolicity indicator around the crack tip are obtained by a least-square t to
the piecewise continuous indicator values (e.g. see Reference [7]). Let e =N
I
(X)e
I
C
0
, and
e
h
C
1
be a piecewise continuous eld. The least-square t for the hyperbolicity indicator
is obtained by projecting the piecewise continuous eld onto a C
0
eld. This is done by min-
imizing the square of the dierence between the C
0
projection e =
I
N
I
e
I
and the piecewise
continuous data e
h
obtained from the solution:
min
1
2
_
(e
h
e)
2
d (41)
To nd the minimum of the above, we set its derivative with respect to the e
J
to zero:
c
2ce
J
_
(e
h
e)
2
d=0 (42)
From above we can obtain the equation for e
I
:
_
(e
h
N
J
e
I
N
I
N
J
) d=0 (43)
The least-square t is only applied to several elements surrounding the crack tip, because its
purpose is to obtain the direction and velocity of the crack tip. From experience, a subdomain
with a radius of 3 to 4 elements is enough.
Since the hyperbolicity indicator should always vanish at the crack tip, it follows that at
the crack tip.
ce
ct
+ t
c
i
e
, i
=0 (44)
where t
c
i
is the velocity of the crack tip. The cracktip velocity can be expressed as
t
c
i
=t
c
s
i
(45)
where t
c
is the crack speed and s is dened in (36). Substituting this denition into (44)
gives
ce
ct
+ t
c
s
i
e
, i
=0 (46)
from which we obtain
t
c
=(s
i
e
, i
)
1
ce
ct
=(s _e)
1
ce
ct
(47)
The above form entails the evaluation of the rate of the hyperbolicity indicator. To avoid that,
we have used
s
)
=e
n
(x
n
c
)s
)
(s
i
e
, i
)
1
or s =e
n
(x
n
c
)s(s e)
1
(48)
where x
n
c
is the position of the cracktip at time step n. Thus the hyperbolicity indicator provides
both the direction and the size of the cracktip displacement increment (see Figure 8). While
the hyperbolicity indicator should be negative only over sets of measure zero, in numerical
solutions it becomes negative in small neighbourhoods around the cracktip.
There are some diculties in the application of the hyperbolicity criterion. Because of the
high frequency noise in explicit computations, the cracktip direction and speed when directly
determined by the preceding procedure often become rather erratic. Therefore, unless a large
change in direction is indicated, we smooth the direction by using an average of the current
and several preceding directions.
5. WEAK FORM AND DISCRETIZED EQUATIONS
The trial and test functions reside in the following spaces, respectively:
U=u(X, t)[u(X, t) C
0
, u(X, t) = u(t) on
u
0
, u discontinuous on crack surface (49)
U
0
=ou(X, t)[ou(X, t)C
0
, ou(X, t)=0 on
u
0
, ou discontinuous on crack surface (50)
The weak form of the momentum equation is: for u U and u U
0
,
_
0
ou
i
j
0
u
i
d
0
=
_
0
ou
i
j
0
b
0
i
d
0
+
_
t
ou
i
t
i
d
_
c
<ou
i
=t
i
d
_
0
\
c
0
c(ou
i
)
cX
)
P
)i
d
0
(51)
(note we have both integrals in the current and reference congurations for convenience).
The discrete equations are obtained by standard BubnovGalerkin procedures. We let the
trial displacement eld (24) or (27) be
u
i
(X) =
I
N
I
(X)u
iI
(t) +

I S
I
(X)q
iI
(t) (52)
where
I
(X) have been dened in Equations (25) and (28). The test displacement eld is
ou
i
(X) =
I
ou
iI
(t)N
I
(X) +

I S
oq
iI
(t)
I
(X) (53)
In taking any derivatives, we consider
0
the open set with
0
c
excluded. Furthermore, we
carefully distinguish between
+
c
and
c
in evaluating the surface integrals of the crack. The
derivatives of the trial displacement eld are then given by
cu
i
cX
)
=
I
N
I, )
u
iI
+

I S
I, )
q
iI
(54)
A similar expression holds for the derivatives of the test displacement eld cou
i
}cX
)
.
Substituting the above into the weak form Equation (51) gives
I
[ou
iI
([
int
iI
+ M
u
IJ
u
iJ
+ M
uq
IJ
q
iJ
[
ext
iI
) + oq
iI
(Q
int
iI
+ M
q
IJ
q
iJ
+ M
uq
JI
u
iJ
Q
ext
iI
)] =0 (55)
for all ou
iI
and oq
iI
that are not constrained by displacement boundary conditions or the
inequality (17); the two parentheses give the discrete equations. In the above,
[
int
iI
=
_
0
\
c
0
cN
I
cX
)
P
)i
d
0
[
ext
iI
=
_
t
0
N
I

t
0
i
d
0
+
_
0
N
I
j
0
b
i
d
0
(56)
Q
int
iI
=
_
0
\
c
0
c
I
cX
)
P
)i
d
0
Q
ext
iI
=
_
t
0
I

t
0
i
d
0
+
_
I
j
0
b
i
d
0

_
c
N
I
t
i
d (57)
M
u
IJ
=
_
0
N
I
N
J
d
0
M
uq
IJ
=
_
0
N
I
J
d
0
M
q
IJ
=
_
J
d
0
(58)
Note that the cohesive traction only eects the nodal force Q
ext
iI
, and that it is evaluated
in the current conguration. As noted in Reference [19], the nodal forces can be evaluated
in either the current or the reference congurations; it is often convenient to evaluate surface
integrals over the current conguration, volume integrals over the initial conguration.
2, 4
f( )=0: discontinuity
3
1
interface
f( )>0 : domain a
enriched node
f( )<0: domain b
X
X
X
Figure 9. A 3-node triangle element with a discontinuous displacement eld.
It might be noted that the change of the enrichment during a time step is neglected for sim-
plicity. Otherwise, the inertia term in (55) would be non-linear. Although this may introduce
some inaccuracy, this simplication was also adopted in meshless methods [32].
The mass matrix corresponding to the standard degrees-of-freedom u
iI
will be lumped, i.e.
diagonalized, by a row-sum procedure. The consistent mass matrix is retained for any terms
linked to the enrichment parameters q
iI
, because we have been unable to determine an eective
diagonal mass matrix for these degrees-of-freedom.
5.1. Quadrature
For elements containing the crack, the integrals in Equations (56)(58) cannot be evaluated
by standard quadrature methods since the integrand is discontinuous. The following quadrature
procedure was used, similar to Moes et al. [11]. For an element completely cut by the crack,
we rst divided the element into three triangular subdomains by a Delaunay triangulation
for ve pointsthe three nodes and the two crossing points of the crack. Then for each
triangular subdomain, we used a standard quadrature method to evaluate the integrals over
the subdomain. The integrals were then obtained by summing over the three subdomains.
For an element containing a crack tip, such as element e
2
in Figure 5, the integrals were
obtained by summing the integral of over the virtual element 1
and the subdomain

1 3
3. The rst integral was obtained as described in the previous paragraph, the second
by standard methods, since the integrands are continuous in the subdomain 1 3
3.
5.2. Constant strain triangle with one enriched node
To illustrate how the additional degrees of freedom are handled, we consider a 3-node constant
strain triangle with one enriched node. This enriched node is treated as 2 nodes, so we add
one additional node. A generic constant strain triangle is shown in Figure 9. It has 3 nodes
associated with the nodal displacements u
iI
, I =1 to 3, and an additional node 4 where q
i1
is
stored. The nodal displacements associated with the element are then stored in a displacement
matrix d
e
given by
d
T
e
=
_
u
x1
u
x2
u
x3
q
x1
u
,1
u
,2
u
,3
q
,1
_
(59)
The corresponding nodal internal forces are
_
f
int
e
_
T
=
_
[
x1
[
x2
[
x3
Q
x1
[
,1
[
,2
[
,3
Q
,1
_
(60)
with a similar expression for the external nodal forces.
The displacement eld is
_
u
x
u
,
_
=
_
u
x1
u
x2
u
x3
q
x1
u
,1
u
,2
u
,3
q
,1
_
_
_
N
1
N
2
N
3
1
_
_
d
T
e
N (61)
The displacement gradient is
_
_
cu
x
cX
cu
x
cY
cu
,
cX
cu
,
cY
_
_
=
_
u
x1
u
x2
u
x3
q
x1
u
,1
u
,2
u
,3
q
,1
_
_
_
N
1, X
N
1, Y
N
2, X
N
2, Y
N
3, X
N
3, Y
N
1, X
1
N
1, Y
1
_
_
(62)
Note that we do not include the terms N
1
1, X
=N
1
o([(x))[
, X
or N
1
1, Y
=N
1
o([(x))[
, Y
,
since the continuum is restricted to the open set
0
c
0
.
The internal nodal forces are given by
f
int
e
=
_
_
[
x1
[
,1
[
x2
[
,2
[
x3
[
,3
Q
x1
Q
,1
_
_
int
=
_
0
\
c
0
_
_
N
1, X
N
1, Y
N
2, X
N
2, Y
N
3, X
N
3, Y
N
1, X
1
N
1, Y
1
_
_
_
P
11
P
12
P
21
P
22
_
d
0
(63)
The cohesive tractions are treated as external nodal forces that are given later. For an element
completely cut by a crack the stress is piecewise constant and will be denoted by P
a
i)
and P
b
i)
for subdomain a and b, respectively. We denote the areas of the two subdomain by A
a
0
and
A
b
0
. The nodal forces are then given by
f
int
e
=
A
a
0
2A
0
_
_
Y
23
X
32
Y
31
X
13
Y
12
X
21
0 0
_
_
_
P
a
11
P
a
12
P
a
21
P
a
22
_
+
A
b
0
2A
0
_
_
Y
23
X
32
Y
31
X
13
Y
12
X
21
2Y
23
2X
32
_
_
_
P
b
11
P
b
12
P
b
21
P
b
22
_
(64)
where X
IJ
=X
I
X
J
, etc. This corresponding expression in terms of the updated co-ordinates
and the Cauchy stress is
f
int
e
=
A
a
2A
_
_
,
23
x
32
,
31
x
13
,
12
x
21
0 0
_
_
_
o
a
11
o
a
12
o
a
21
o
a
22
_
+
A
b
2A
_
_
,
23
x
32
,
31
x
13
,
12
x
21
2,
23
2x
32
_
_
_
o
b
11
o
b
12
o
b
21
o
b
22
_
(65)
In the absence of body forces and boundary tractions, the external nodal forces are:
f
ext
e
=
_
_
[
x1
[
,1
[
x2
[
,2
[
x3
[
,3
Q
x1
Q
,1
_
_
ext
=
_
c
_
_
0 0
0 0
0 0
N
1
N
1
_
_
_
t
x
0
0 t
,
_
d (66)
5.3. Implicitexplicit time integration
Explicit time integration was not used over the entire mesh because the critical time step in
elements traversed by the crack can be very small. We examined the relationship between
discontinuity locations and critical time steps for linear one dimensional elements. We found
that the critical time step tends to zero as the discontinuity approaches the element nodes.
So we adopt an implicitexplicit time integration scheme [33]) in our computation. For
elements with enriched nodes, we use implicit integration while for elements without any
enrichment we use explicit integration. It is necessary to solve a set of non-linear equations
due to the non-linearity of the constitutive model. Since the extension of the implicitexplicit
scheme to consider non-linearities is straightforward, the procedure is not described here.
6. NUMERICAL EXAMPLES
6.1. Edge-cracked plate under impulsive loading
6.1.1. Problem description In experiments of Kaltho [34], a plate with two edge notches is
impacted by a projectile as shown in Figure 10. By modifying the projectile speed and the
acuity of the notch tip, two dierent failure modes were observed. At higher strain rates, a
shear localization mode due to the formation of a shear band ahead of the notch at a negative
angle of about 10
was observed. On the other hand, at lower strain rates, a brittle fracture
mode with a propagation angle of 70
was observed. In this study, we focused on the

brittle failure.
Twofold symmetry is used in modelling the problem. A symmetry condition was applied
on the bottom edge of the model (u
,
=0), and a step velocity on the cracked edge for
06,625 mm; the other edges are traction-free. We assume that the projectile has the same
elastic impedance as the plate section that is impacted, so the impact velocity is approximately
one-half of the projectile speed [35]. The impact velocity is chosen as t
0
=16.5 m}s, which is
100mm
v
0
50mm
50mm
100mm
100mm
75mm
20 30 40 50 60 70 80
0
500
1000
1500
2000
2500
3000
time (s)
c
r
a
c
k

p
r
o
p
a
g
a
t
i
o
n

s
p
e
e
d

(
m
/
s
)
Crack speed
Rayleigh wave speed
Figure 10. Experimental set-up for crack growth
in an edge-cracked plate; the experimental results
were reported by Kaltho [34].
Figure 11. Crack propagation speed for the sim-
ulation of Kalthos [34] experiment with loss
of hyperbolicity criterion.
a representative projectile speed for which the brittle failure mode was observed in Kalthos
experiments [34].
The material properties are representative of maraging steel type 18Ni1900 [36]: j=8000
kg}m
3
, E =190 GPa, v =0.30. The critical stress intensity factor is taken as K
I c
=68 MPa
m,
so G
F
=2.217 10
4
J}m
2
. The dilatational wave speed is c
d
=5654.3 m}s; the shear wave
speed c
s
=3022.4 m}s; the Rayleigh wave speed c
R
=2799.2 m}s.
6.1.2. XFEM with loss of hyperbolicity criterion. In this example, we use the loss of hy-
perbolicity criterion to advance the crack. The cohesive energy is taken to be mode I energy
release rate G
F
=2.217 10
4
J}m
2
, the critical crack opening displacement o
max
=5.378 10
5
m. Lemaitres damage model (summarized in Appendix C) was used with the following dam-
age parameters: A=1.0, B=200 and the damage threshold c
D
0
=3 10
3
. This is a toy
problem because the applicability of Lemaitres damage model to this material has not been
established. We were primarily interested in whether a damage model qualitatively reproduces
the major features of the experiment.
In our numerical simulations, we advanced the crack until it completely went through the
specimen. The simulation time is around 90 s. A 40 40 cross triangular mesh was used with
implicitexplicit time integration and a Courant number of 0.4 for the uncracked elements.
Figure 12. Crack evolution for Kalthos [34] experiment by XFEM with loss of hyperbolicity criterion:
(a) t =30.06 s; (b) t =45.59 s; (c) t =65.08 s; and (d) t =85.57 s.
As shown in Figure 11, the crack begins to propagate at 26.17s. The initial cracktip speed
averages about 75% of the Rayleigh wave speed. The crackspeed then slowly diminishes.
After 50 s, the crack decelerates more rapidly and near the end of the simulation is below
1000 m}s. The overall crack propagation angle with the horizontal axis is around 58
which
compares well with the observed angle of 70
; see Figure 12. The crack path is nearly straight.

Some areas of high damage also occur at the right bottom of specimen, and propagate toward
the original crack position due to the reection of the wave; there was no reported evidence
of this damage in the experiment.
6.1.3. XFEM with tensile stress criterion. We also studied this problem with a maximum
tensile stress criterion given in Appendix A with element-by-element progression of the crack.
The simulation time is around 90 s and the Courant number is 0.4. We used a linear co-
hesive model with a tensile strength of 844 MPa (1}225 of Youngs Modulus) and a frac-
ture energy of 2.217 10
4
J}m
2
; this corresponds to a critical crack opening displacement
o
max
=5.245 10
5
m.
Numerical simulations with three dierent meshes were carried out: 40 40 and 80 80
uniform meshes and an irregular mesh of 9445 elements. The crack paths are shown in Figures
1315. It should be noted that the crack speed is almost independent of element size once
a sucient renement is achieved, in contrast to Lin et al. [37].
0 0.05 0.1
0
0.05
0.1
0
0.05
0.1
0
0.05
0.1
0
0.05
0.1
0 0.05 0.1
(b) (a)
0 0.05 0.1
(c)
0 0.05 0.1
(d)
Figure 13. Crack evolution for 40 40 cross-triangular mesh for the simulation of Kalthos
[34] experiment by XFEM with the maximum tensile stress criterion: (a) t =40.50 s;
(b) t =53.94 s; (c) t =66.14 s; and (d) t =85.65 s.
We can see from Figures 13 to 15 that the crack paths obtained by the three dierent
meshes are almost the same. The overall crack propagation angle is about 65
, which agrees
well with the experimental value of 70
.
The crack propagation velocities are shown in Figure 16. Crack propagation is initiated at
approximately 27 s. The crack velocity for the coarse mesh is substantially slower than for
the ne mesh, this may be due to the element-by-element progression. From 2.8 to 6 s, the
crack velocity averages about 1800 m}s which agrees with the damage model results. It is
noteworthy that for both the unstructured and structured meshes, the crack velocities agree
quite well.
6.1.4. Inter-edge cohesive model. We also solved this problem with the inter-element method
of Xu and Needleman [3]; the cohesive law is described in Appendix B. In our numerical
simulations, the parameters in Equation (B2) are: r =0, q =1, [
n
=[
t
=2.217 10
4
J}m
2
.
The cohesive strength o
max
=1.73 GPa is 1}110 of the Youngs modulus.
n
=
t
=4.72 m,
t
max
is then 4.03 GPa. The simulation time was 100 s and the Courant number was 0.02.
Numerical simulations with two dierent meshes were carried out: 40 40 and 80 80
uniform cross-triangular mesh. The crack paths are shown in Figures 17 and 18. The crack
propagation speeds for dierent meshes are shown in Figure 19.
Figure 14. Crack evolution for 80 80 cross-triangular mesh for the simulation of
Kalthos [34] experiment by XFEM with tensile stress criterion: (a) t =35.19 s;
(b) t =46.34 s; (c) t =57.48 s; and (d) t =80.38 s.
For both meshes, the overall crack angle is around 55
. The crack rst moves up vertically

in the ne mesh, then grows primarily along the 45
edges with a few spurts upward. The

crack path and overall response change to a signicant degree with renement. For the ner
mesh, the right of the crack begins to contact near the end of the simulation, whereas for
the coarse mesh the piece above the crack moves quite a bit to the left. We can see that the
mesh constrains the crack propagation angle in both cases. As shown in Figure 19, the crack
begins to propagate at around 30 s for both meshes. But the crack speed is greater in the
coarse mesh than in the ne mesh. After an initial spike, the crack speed is almost constant
at 1800 m}s, which agrees quite well with our results in Figure 16. The results for the crack
velocity are smoother than for the XFEM results, which is surprising and may be due to the
smaller time step.
6.2. Energy release rate
In this section, we examine the accuracy of energy release rate predicted by the continuous
growing crack model. We compare our numerical results for the energy release rate for an edge
crack moving at a constant speed in an innite two-dimensional body to the analytical results
0
0 0.05
0.05
0.1
0.1
0
0.05
0.1
0
0.05
0.1
0
0.05
0.1
(a)
0 0.05 0.1
(c)
0 0.05 0.1
(d)
0 0.05 0.1
(b)
Figure 15. Crack evolution for an irregular mesh of 9445 triangular elements for the simu-
lation of Kalthos [34] experiment by XFEM with tensile stress criterion: (a) t =38.29s;
(b) t =48.64 s; (c) t =61.10 s; and (d) t =82.68 s.
Figure 16. Crack propagation speed for the simulation of Kalthos [34] experiment
with tensile stress criterion.
given by Freund [38]. The crack is assumed to be propagating at a specied velocity so crack
propagation criteria play no role. We tested two elements: three-node triangles and four-node
quadrilaterals.
0 0.05 0.1 0 0.05 0.1
0
0
0.05
0.05
0.1
0.1
0
0.05
0.1
0
0.05
0.1
0
0.05
0.1
0 0.05 0.1
(a)
(c) (d)
(b)
Figure 17. Crack evolution for 40 40 cross-triangular mesh for the simulation by XFEM
of Kalthos [34] experiment with the inter-element crack model of Xu and Needleman
[3]: (a) t =40 s; (b) t =60 s; (c) t =80 s; and (d) t =100 s.
A schematic of the problem is shown in Figure 20; plane strain was assumed. A uniform
traction of magnitude o
0
is applied to the top edge. The plate dimensions are 10.0 m4.0 m,
and the initial crack length is 5.0 m. Since the specimen is nite, we stopped the simulation
before the reected wave from the edge reaches the crack tip at t =2.5 H}c
d
, where H =2 m
is one half of the plate width. The material properties are: j=7800 kg}m
3
, E =211 GPa, and
v =0.30. The body is loaded by a tensile stress perpendicular to the crack face. The traction
o
0
=1000 Pa was applied as a step function in time. We use a domain of 1.5 m1.5 m to
calculate the energy release rate. The problems were run over a time corresponding to 2.5H}c
d
.
The wave reaches the crack at H}c
d
, so no spurious reections reach the crack region by the
end of the simulation. A cross-triangular mesh generated from a 100 41 quadrilateral mesh
was used.
The normalized energy release rate versus normalized time is shown in Figures 23 and 24
for three dierent cases: stationary cracks, moving cracks and stationary cracks that start
moving at a constant velocity. We used ve point averaging on the results for moving
cracks and stationary}moving cracks to reduce numerical noise. The results for the station-
ary crack are quite smooth and oscillate about the analytic solution. The results for the
moving crack are substantially more noisy. The oscillations coincide with the number of el-
ement edges that are crossed: the cracktip crosses about six edges of the cross-triangular
mesh.
0
0 0.05
0.05
0.1
0.1
0
0.05
0.1
0
0.05
0.1
0
0.05
0.1
(a)
0 0.05 0.1
(c)
0 0.05 0.1
(d)
0 0.05 0.1
(b)
Figure 18. Crack evolution for 80 80 cross-triangular mesh for the simulation by XFEM
of Kalthos [34] experiment with the inter-element crack model of Xu and Needleman
[3]: (a) t =40 s; (b) t =60 s; (c) t =80 s; and (d) t =100 s.
2 3 4 5 6 7 8 9 10
x 10
-5
0
500
1000
1500
2000
2500
3000
Time (s)
C
r
a
c
k

P
r
o
p
a
g
a
t
i
o
n

S
p
e
e
d

(
m
/
s
)
40x40 mesh
80x80 mesh
5.0 m
0
2.0 m
10.0 m
4.0 m = 2 H
Figure 19. Crack propagation speed for dier-
ent meshes for the simulation by XFEM of
Kalthos [34] experiment with the inter-element
crack model of Xu and Needleman [3].
Figure 20. Finite discrete model used for mod-
elling the innite plate with a semi-innite crack.
Figure 21. Energy release rates for 100 41
cross-triangular mesh. The dashed lines are
analytical solutions; the solid lines are the
numerical solutions.
cross-triangular mesh. The crack remains station-
ary until t =1.5H}c
d
and then begins moving at
t =0.4c
s
. The dashed line is the analytical solu-
tion, and the solid line is the numerical solution.
We also solved the problem with a 100 41 mesh of quadrilateral elements; the ele-
ment is described in Chen [39]. The normalized energy release rate versus normalized time
is shown in Figures 21 and 22 for three dierent cases: stationary cracks, moving cracks
and stationary}moving cracks. We used a ve point lter on results for moving cracks and
stationary}moving cracks to reduce numerical noise.
The results in Figure 22 for a crack moving at constant velocity and at rest compare
very well with the analytical solution, although the latter shows some downward drift toward
the end. Overall the solutions are relatively free of noise. When the crack suddenly starts
moving, as can be seen from Figure 24, there is substantial overshoot at the time of the
initiation of crack motion. This has also been the case for meshfree solutions [32]. However,
the quadrilateral element solutions on the whole are substantially better than the triangular
element solutions.
6.3. Crack branching by loss of hyperbolicity criterion
This example concerns the propagation of a crack in a prenotched specimen shown in Figure
20 but by the loss of hyperbolicity criterion described in Section 4. The rectangular specimen
is of length of 0.1 m and width 0.04 m. Initially a horizontal crack spans from the left edge
to the centre of the plate. Tensile tractions (o =1MPa) are applied on the top and the bottom
edges.
Material constants are j=2450 kg}m
3
, E =32 GPa, v =0.2. We used the Lemaitre [40]
damage model (see Appendix C). The damage parameters are A=1.0, B=7300 and c
D
0
=8.5
10
5
. The initial dilatational wave speed (with D=0) is c
d
=3809.5 m}s; the shear wave
speed c
s
=2332.8 m}s; the Rayleigh wave speed c
R
=2119.0 m}s. A 50 21 cross triangular
mesh was used with implicitexplicit time integration with a Courant number of 0.4. The total
simulation time is 61 s.
Figure 23. Energy release rates for crack moving
at constant velocity for a 100 41 quadrilateral
mesh. The dashed lines are the analytical solu-
tions of Freund [38], the solid lines are numerical
solutions by XFEM.
quadrilateral mesh. The crack remains station-
ary until t =1.5H}c
d
and then begins moving
at t =0.4c
s
. The dashed line is the analyti-
cal solution and the solid line is the numerical
solution by XFEM.
In the linear cohesive law that is imposed after loss of hyperbolicity, the cohesive energy
is 3 J}m
2
. The cohesive spall strength t
max
is about 3.08 N}m
2
(it depends on the state of
stress on loss of hyperbolicity). The critical crack opening displacement o
max
=1.95 10
6
m
for the 3 J}m
2
cohesive energy.
The evolution of the crack is shown in Figure 25, and the crack speed is shown in
Figure 26. The crack begins propagate at 11.34s and then speeds up along a straight line.
During this initial phase of crack propagation, the hyperbolicity indicator at each time step is
negative at the crack tip after each update of the stress eld and its minimum corresponds to
a vector n in (35) normal to the existing crack path. The direction of n is almost constant
in the elements in the subdomain about the cracktip. The crack speed increases with time
and reaches a maximum at 32.34 s. At that point, the vectors n in elements ahead of the
crack tip are no longer constant, in fact, the n vectors take on markedly dierent values in
the elements ahead of the cracktip. We used these vectors to redirect the crack direction at
this point, which resulted in branching. The ratio of the maximum crack speed (just before
branching) to the dilatational wave speed is 2050}3809.5 =0.54, which is signicantly above
the values given by Yoe [41]; the dierences may result from dierence in the material
model.
After the crack branches the crack speed decreases and its velocity becomes almost constant,
with a speed markedly less than the Rayleigh wave speed. Towards the end of the simulation,
small branches of high damage develop around the crack. These may be secondary cracks.
They are dicult to treat with our algorithm, so we have not introduced discontinuities in
these small branches. The overall crack growth pattern agrees qualitatively with what has been
observed in experiments. We repeated the calculation with a cross-triangular mesh of 60 25
renement and found similar results.
Figure 25. Evolution of crack and damage for crack branching: (a) t =30.03s;
(b) t =46.00 s; and (c) t =60.27 s.
10 20 30 40 50 60 70
0
500
1000
1500
2000
2500
time (s)
c
r
a
c
k

p
r
o
p
a
g
a
t
i
o
n

s
p
e
e
d

(
m
/
s
)
Crack speed
Rayleigh wave speed
Figure 26. Crack speed for crack branching problem.
At the point of branching, three cracks occur in one element (the original crack and two
new cracks). This causes diculties in the enrichment inside that element. Therefore, we
simply deleted that element.
It should be kept in mind that the algorithm for the continuum}crack transition assumes a
high degree of the smoothness in the path of the crack. This precludes any roughness from
developing in the crack path. Thus the branches in the damage zones may indicate that small
areas of instability develop at intervals around the crack path. These may correspond to the
roughening that has been observed in experiments, e.g. in Ravichandran and Knauss [42].
7. CONCLUSIONS
A methodology has been developed for treating dynamic crack growth by the extended nite
element method (XFEM). In this method, the discontinuity of the crack is represented in the
nite element approximation by a discontinuous partition of unity enrichment. This enables the
method to treat arbitrary crack growth without remeshing. To extend XFEM to dynamic crack
propagation, a new technique for treating elements that contain crack tips was developed. In
this new technique, the velocity eld of the element containing the crack tip can change
smoothly from a partially cut element to a fully cut element. By contrast, in previous XFEM
enrichments for cracks, the motion of a partially cut element was not consistent in the limit
with a fully cut element, i.e. in the previous XFEM methods, the enrichment resulting from
moving the cracktip to an edge of the element did not correspond with the enrichment of a
fully cut element. This new technique has been applied to higher order elements by Zi and
Belytschko [24].
Results were obtained for several problems. The results matched certain features of exper-
iments quite well and overall seem quite promising. Substantially less mesh dependence was
observed than in the inter-element crack model of Reference [3]. We were not entirely satised
with the quality of the computations with regards to smoothness and quantitative measures.
As could be seen from comparison with the analytic energy release rates, the passage of
the wave through element edges still introduces substantial oscillations. These oscillations are
also evident in the cracktip velocity in the simulations of the Kalthos experiment and in
the prenotched specimen under tension. Some of this might stem from the explicitimplicit
integration, but this facet requires further research.
In comparing the method to existing methods such as inter-element crack methods, the
relative complexity of the methods should be borne in mind. Overall, this method is some-
what more complex than inter-element crack methods; the stable time step of the latter is
not eected by the passage of the crack and the elements are never modied. Thus the
increased accuracy of the present method comes at the cost of a moderate increase in
complexity.
The loss of hyperbolicity criterion for propagating the crack appears to be a promising
direction. This criterion is very attractive for modelling crack growth because it ts naturally
with a material model. When hyperbolicity is lost in a rate-independent material model, then
a computation cannot continue with a continuum model, since the strain localizes to a set of
measure zero occurs. Even with a rate-dependent model, loss of hyperbolicity of the principal
part of the partial dierential equation results in local exponential growth of strain which
would require exorbitant mesh renement to resolve. Thus the transition to a discontinuous
eld is essential in a rate-independent model, and advantageous, though not essential, in a
rate-dependent model.
Furthermore, the transition to a discontinuity by the loss of hyperbolicity criterion provides
both the crack speed and the crack direction. Thus it provides a natural closure for the dynamic
crack propagation problem. The crack propagation speeds we observed in the calculations
reported here were quite consistent with experimental results: they were always less than the
Rayleigh wave velocity for the material. Similarly, the qualitative pattern of the crack paths
agreed with experimental observations.
This suggests a new paradigm for crack and material modelling: the development of material
models that not only predict the right behaviour in the stable regime but correctly predict the
transition to material instability, i.e. cracking or shear banding. It should be noted that driving
a crack by loss of hyperbolicity diers from a cohesive law. A cohesive law can govern certain
aspects of crack propagation, but, as pointed out in the paper, it does not provide closure for
the crack propagation problem unless a discretization is introduced. Material models have so
far been little tested for their validity in both the stable and unstable regimes. In fact, crack
models are usually driven independent of the material model, for example by stress intensity
factors, energy release rates or cohesive stress criteria. The ability of constitutive models to
predict the correct crack velocity and direction to our knowledge has not been examined in
any depth.
APPENDIX A: MAXIMUM TENSILE STRESS CRITERION
The maximum tensile stress criterion determines if a crack should propagate by comparing
the maximum tensile stress value in the element ahead of the crack tip to a prescribed pa-
rameter t
max
. Once the maximum tensile stress exceeds the tensile strength t
max
, the cracktip
is advanced into the next element normal to the maximum tensile stress direction until the
crack tip reaches the element boundary. In practice, we found that the crack angle provided
by this criterion sometimes leads to erratic crack paths. Therefore we used the average value
of the stress elds in several elements ahead of crack tip to determine the crack direction.
APPENDIX B: INTER-EDGE COHESIVE MODEL
In the computations with the inter-element crack method, the cohesive model of Xu and
Needleman [3] was used. In this model the traction across the cohesive surface is given by
T=
c[
cT
(B1)
where T is the displacement jump across the cohesive interface and [ is the cohesive potential
[3, 43]:
[(T) =[
n
+ [
n
e
(o
n
}
n
)
__
1 r +
o
n
n
_
1 q
r 1

_
q +
r q
r 1
o
n
n
_
e
(o
2
t
}
2
t
)
_
(B2)
where the subscripts n and t denote the normal and tangential components, respectively. In
Equation (B2),
[
n
=eo
max
n
[
t
=
_
e}2t
max
t
(B3)
where [
n
and [
t
are the normal and tangential separation work, respectively, and o
max
and t
max
are the cohesive strength in the normal and tangential directions, respectively, and
n
and
t
are the corresponding characteristic lengths.
q =
[
t
[
n
r =
o
n
(B4)
In the above o
n
is the value of o
n
after complete shear separation. The cohesive tractions are
then obtained by
1
n
=
[
n
n
e
(o
n
}
n
)
_
o
n
n
e
(o
2
t
}
2
t
)
+
1 q
r 1
[1 e
(o
2
t
}
2
t
)
]
_
r
o
n
n
__
1
t
=
[
n
n
_
2

n
t
_
o
t
t
_
q +
r q
r 1
o
n
n
_
e
(o
n
}
n
)
e
(o
2
t
}
2
t
)
(B5)
APPENDIX C: LEMAITRES DAMAGE MODEL
The damage model we used was developed by Lemaitre [40] for an isotropic linear elastic
virgin material. The stressstrain law is given as
o
i)
=(1 D)C
i)k!
c
k!
D [0, 1] (C1)
where D is a scalar that represents the extent of damage. The damage evolution law is
D( c) =1 (1 A)c
D
0
c
1
Ae
B( cc
D
0
)
(C2)
where c is the eective strain dened by
c =
i=1
c
2
i
H(c
i
) (C3)
where H(x) is the Heaviside function dened in Equation (15), A and B are characteristic
parameters of the material; c
D
0
is the initial damage threshold strain.
ACKNOWLEDGEMENT
The authors would like to acknowledge the support of the Oce of Naval Research.
REFERENCES
1. Bazant ZP, Belytschko T. Wave propagation in a strain-softening bar: Exact solution. Journal of Engineering
Mechanics (ASCE) 1985; 111(3):381389.
2. Belytschko T, Wang X-J, Bazant ZP, Hyun Y. Transient solutions for one-dimensional problems with strain
softening. Journal of Applied Mechanics (ASME) 1987; 54(3):513518.
3. Xu X-P, Needleman A. Numerical simulations of fast crack growth in brittle solids. Journal of the Mechanics
and Physics of Solids 1994; 42:13971434.
4. Camacho GT, Ortiz M. Computational modeling of impact damage in brittle materials. International Journal
of Solids and Structures 1996; 33:28992938.
5. Ortiz M, Pandol A. Finite-deformation irreversible cohesive elements for three-dimensional crack-propagation
analysis. International Journal for Numerical Methods in Engineering 1999; 44:12671282.
6. Belytschko T, Fish J, Englemann B. A nite element method with embedded localization zones. Computer
Methods in Applied Mechanics and Engineering 1988; 70:5989.
7. Simo JC, Oliver J, Armero F. An analysis of strong discontinuities induced by strain-softening in rate-
independent inelastic solids. Computational Mechanics 1993; 12:277296.
8. Dvorkin EN, Cuitino AM, Gioia G. Finite elements with displacement interpolated embedded localization lines
insensitive to mesh size and distortions. International Journal for Numerical Methods in Engineering 1990;
30:541564.
9. Jirasek M. Comparative study on nite elements with embedded discontinuities. Computer Methods in Applied
Mechanics and Engineering 2000; 188:307330.
10. Belytschko T, Black T. Elastic crack growth in nite elements with minimal remeshing. International Journal
for Numerical Methods in Engineering 1999; 45(5):601620.
11. Moes N, Dolbow J, Belytschko T. A nite element method for crack growth without remeshing. International
Journal for Numerical Methods in Engineering 1999; 46(1):133150.
12. Dolbow J, Moes N, Belytschko T. Discontinuous enrichment in nite elements with a partition of unity method.
Finite Elements in Analysis and Design 2000; 36(3):235260.
13. Belytschko T, Moes N, Usui S, Parimi C. Arbitrary discontinuities in nite elements. International Journal for
Numerical Methods in Engineering 2001; 50(4):9931013.
14. Wells GN, de Borst R, Sluys LJ. A consistent geometrically non-linear approach for delamination. International
Journal for Numerical Methods in Engineering 2002; 54:13331355.
15. Remmers JJC, de Borst R, Needleman A. A cohesive segments method for the simulation of crack growth.
Computational Mechanics 2003; 31:6977.
16. Gao H, Klein P. Numerical simulation of crack growth in an isotropic solid with randomized internal cohesive
bonds. Journal of the Mechanics and Physics of Solids 1998; 42(6):187218.
17. Oliver J, Huespe AE, Pulido MDG, Samaniego E. On the strong discontinuity approach in nite deformation
settings. International Journal for Numerical Methods in Engineering 2003; 56:10511082.
18. Peerlings RHJ, de Borst R, Brekelmans WAM, Geers MGD. Localisation issues in local and nonlocal continuum
approaches to fracture. European Journal of Mechanics A}Solids 2002; 21:175189.
19. Belytschko T, Liu WK, Moran B. Nonlinear Finite Elements for Continua and Structures. Wiley: Chichester,
2000.
20. Marsden JE, Hughes TJR. Mathematical Foundations of Elasticity. Prentice-Hall, Dover: Englewood Clis,
NJ, 1983.
21. Armero F, Garikipati K. An analysis of strong discontinuities in multiplicative nite strain plasticity and their
relation with the numerical simulation of strain localization in solids. International Journal of Solids and
Structures 1996; 33(2022):28632885.
22. Oliver J. Modelling strong discontinuities in solid mechanics via strain softening constituitive equations, part 1:
fundamentals. part 2: numerical simulation. International Journal for Numerical Methods in Engineering 1996;
39:35753624.
23. Chessa J, Wang H, Belytschko T. On the construction of blending elements for local partition of unity enriched
nite elements. International Journal for Numerical Methods in Engineering 2003; 57(7):10151038.
24. Zi G, Belytschko T. New crack-tip elements for xfem and applications to cohesive cracks. International Journal
for Numerical Methods in Engineering 2003; 57:22212240.
25. Barenblatt G. The mathematical theory of equilibrium of cracks in brittle fracture. Advances in Applied Fracture
1962; 7:55129.
26. Dugdale D. Yielding of steel sheets containing slits. Journal of the Mechanics and Physics of Solids 1960;
8:100108.
27. Hillerborg A, Mod eer M, Peterson PE. Analysis of crack formation and crack growth in concrete by means of
fracture mechanics and nite elements. Cement and Concrete Research 1976; 6:773782.
28. Planas J, Elices M. Nonlinear fracture of cohesive materials. International Journal of Fracture 1991; 51:
139157.
29. Pandol A, Guduru PR, Ortiz M. Three dimensional cohesive-element analysis and experiments of dynamic
fracture in c300 steel. International Journal of Solids and Structures 2000; 37:37333760.
30. Pandol A, Krysl P, Ortiz M. Finite element simulation of ring expansion and fragmentation: The capturing
of length and time scales through cohesive models of fracture. International Journal of Fracture 1999; 95:
279297.
31. Papoulia KD, Sam CH, Vavasis SA. Time continuity in cohesive nite element modelling. International Journal
for Numerical Methods in Engineering 2003; 58:679701.
32. Belytschko T, Tabbara M. Dynamic fracture using element-free Galerkin methods. International Journal for
Numerical Methods in Engineering 1996; 39:923938.
33. Hughes TJR, Liu WK. Implicitexplicit nite elements in transient analysis: implementation and numerical
examples. Journal of Applied Mechanics (ASME) 1978; 45:375378.
34. Kaltho JF, Winkler S. Failure mode transition at high rates of shear loading. International Conference on
Impact Loading and Dynamic Behavior of Materials 1987; 1:185195.
35. Lee YJ, Freund LB. Fracture initiation due to asymmetric impact loading of an edge cracked plate. Journal of
Applied Mechanics (ASME) 1990; 57:104111.
36. Decker RF. Source Book on Maraging Steels. American Society for Metals: Cleveland, OH, 1979.
37. Liu Y, Murakamim S, Kanagawa Y. Mesh-dependence and stress singularity in nite element analysis of creep
crack growth by continuum damage mechanics approach. European Journal of Mechanics A}Solids 1994;
13:395417.
38. Freund LB. Crack propagation in an elastic solid subjected to general loading. 3. stress wave loading. Journal
of the Mechanics and Physics of Solids 1973; 21(2):4761.
39. Chen H. Enriched nite element methods and its applications. Ph.D. Thesis, 2003.
40. Lemaitre J. Evaluation of dissipation and damage in metal submitted to dynamic loading. Proceedings ICM 1,
1971.
41. Yoe EH. The moving Grith crack. Philosophical Magazine 1951; 42:739750.
42. Ravichandran G, Knauss WG. A nite elastostatic analysis of bimaterial interface cracks. International Journal
of Fracture 1989; 39(13):235253.
43. Xu X-P, Needleman A. Void nucleation by inclusion debonding in a crystal matrix. Modelling and Simulation
in Materials Science and Engineering 1993; 1:111132.

Dynamic CDynamic Crack Propagation Based On Loss of Hyperbolicity and A New Discontimuous Enrichmentrack Propagation Based On Loss of Hyperbolicity and A New Discontimuous Enrichment

Încărcat de

Informații document

Titlu original

Drepturi de autor

Formate disponibile

Partajați acest document

Partajați sau inserați document

Opțiuni de partajare

Vi se pare util acest document?

Este necorespunzător acest conținut?

Drepturi de autor:

Formate disponibile

Dynamic CDynamic Crack Propagation Based On Loss of Hyperbolicity and A New Discontimuous Enrichmentrack Propagation Based On Loss of Hyperbolicity and A New Discontimuous Enrichment

Încărcat de

Drepturi de autor:

Formate disponibile

INTERNATIONAL JOURNAL FOR NUMERICAL METHODS IN ENGINEERING

Int. J. Numer. Meth. Engng 2003; 58:18731905 (DOI: 10.1002/nme.941)

; designate the nodes of the virtual triangle by 1, 2, 3

are the triangular co-ordinates of (1, 2, 3

and the subdomain

was observed. In this study, we focused on the

; see Figure 12. The crack path is nearly straight.

. The crack rst moves up vertically

edges with a few spurts upward. The

S-ar putea să vă placă și