Comput Mech (2008) 42:579591

DOI 10.1007/s00466-008-0283-1
ORIGINAL PAPER
Fracture strength assessment and aging signs detection in human
cortical bone using an X-FEM multiple scale approach
Elisa Budyn Thierry Hoc Julien Jonvaux
Received: 30 April 2007 / Accepted: 18 March 2008 / Published online: 22 April 2008
Springer-Verlag 2008
Abstract We present a multiple scale approach for mod-
eling multiple crack growth in human cortical bone under
tension. The Haversian microstructure, a four phase com-
posite, is discretized by a classical nite element method
fed with the morphological and mechanical characteristics,
experimentally measured, to mimic human bone heterogene-
ity at the micro scale. The fracture strength of human bone,
exhibiting aging signs, is investigated through tensional per-
colation simulations in statistical microstructures. The cracks
are initiated at the micro scale at locations where a criti-
cal elastic-damage strain-driven criterion is met. The cracks,
modeled by the eXtended Finite Element Method, are then
grown until complete failure when a critical stress intensity
factor criterion is attained. The model provides the fracture
strength and the global response at the material scale and
the stressstrain elds at the microscopic level. The model
creates a constitutive law at the material scale and empha-
sizes the inuence of the microstructure on bone failure and
fracture risk assessment. These results are validated against
experiments.
Keywords Cortical bone Multiple cracks Failure
X-FEM Multiple scale
E. Budyn (B) J. Jonvaux
Department of Mechanical and Industrial Engineering,
University of Illinois at Chicago, 842 West Taylor Street,
Chicago, IL 60607, USA
e-mail: ebudyn@uic.edu
T. Hoc
Department of Material Science LMSSMat,
Ecole Centrale Paris, Grande Voie des Vignes,
92295 Chatenay Malabry, France
1 Introduction
We present a statistical X-FEM multiple scale method for
modeling failure of human cortical bone under tension. One
goal of this model creates a constitutive law at the (mac-
roscopic) scale of the material that can later be coupled to
a higher and/or lower scale(s). The model will assess the
fracture strength of human bone microstructures and their
overall behavior during the fracturing process. The model
focuses on the inuence of the microstructure aging signs
on its mechanical response. The construction of the model
falls into two parts: the modeling of random human Haver-
sian bone microstructures as a multi-phase composite and the
modeling of crack growth inside the heterogeneous micro-
structures. Therefore the model provides the global response
at the material (macro) scale and the local (micro) stress and
strain elds inside the material microstructure.
Haversian cortical bone microstructure has been
extensively studied as a factor inuencing the mechanical
behavior of bone during failure [14]. At the micro scale,
cortical bone is composed of densely packed concentric
lamellar structures (osteons) that are embedded in an intersti-
tial matrix (Fig. 1). An interface called a cement line around
each osteon isolates the sub-microstructures from the inter-
stitial matrix. Bone porosity is principally due to Haversian
canals. Bone is also a living tissue with the unique ability to
adapt both its structure and its architecture to its mechanical
environment [5]. This self-healing process termed remod-
eling produces a heterogeneous material highly variable in
osteon bone-mineral density that is related to the level of
maturation of the newly-formed tissue.
A mechanistic understanding of the role of this complex
microstructure and its aging evolution is highly relevant to
predicting the risk of fracture associated with age and dis-
eases. From a statistical point of view, several studies have
123
580 Comput Mech (2008) 42:579591
demonstrated that cortical bone fracture toughness can be
related to osteon density and porosity [69].
At the microstructural scale, considerable work has been
developed to study the initiation and growth of microcracks,
which have been shown to act as a stimulus for bone remod-
eling. In in vivo conditions, cement lines act as a microstruc-
tural barrier for the majority of cracks [10]. In contrast, in
transverse loading the underlying microstructure does not
have any significant inuence on the crack propagation [11].
Bone mechanical behavior and fracture mechanisms have
been very difcult to predict due to their complex micro-
structural hierarchy. Because Haversian cortical bone can be
compared to a composite where discontinuities within the
material may generate mechanical stress concentration sites
for crack initiation, it is essential to include the microstruc-
tural aspects into our model. Several models using hierarchi-
cal homogenizationtheoryhave beendevelopedtopredict the
macroscopic behavior [3, 12, 2] without resolving the spatial
distribution of local strains. Self-consistent models for multi-
coated cylinders [13, 14] have also been applied to obtain
more precise bounds. Thus attempts to model larger unit
cells [15] have not included the geometrical and mechan-
ical local anisotropy and the local fracturing process. The
objective of this work is to construct a constitutive model
for human Haversian cortical bone based on a multi-scale
approach. The model is intended to capture the inuence of
the microstructure and its pathological aging alterations on
the strain localization pattern inside the material, and the
fracturing process under tension at the scale of the material,
which is strongly relevant in the mechanism of bone fracture
and its prediction.
The fracture process is incorporated into the proposed
model by the eXtended Finite Element Method (X-FEM), a
numerical method for treating arbitrary discontinuities with-
out remeshing [16, 17]. In this method, cracks are initiated at
the micro scale at locations where a critical elastic-damage
strain-driven criterion is met. The cracks can be randomly
located within the microstructure. In the paper, the human
Haversian microstructure is described as a four phase com-
posite (Haversian canal, osteons, cement line and matrix)
that is discretized by a classical nite element approxima-
tion. The morphological and mechanical bone parameters
are obtained by experiments and are randomly distributed to
mimic the heterogeneous nature of bone at the micro scale.
After initiation, the discontinuities are represented using
an X-FEM formulation by adding to the standard nite ele-
ment approximation space the following enrichments: the
Heaviside function for the crack edges and the Westergaard
eld for the crack tip that are suitable for cracks in the frame-
work of linear elastic fracture mechanics. X-FEMis an appli-
cation of the partition of unity [18] introduced in [16, 19, 20]
and adapted for multiple cracks in [21, 22]. The crack topol-
ogy is represented by vector level sets originally developed
by [2326] for problems of interface tracking and the
evolution of curves. Our model uses a narrow banded vec-
tor level set method [27], in order to simplify the process of
freezing the existing level sets for cracks [17] and allowing
cracks to curve.
The cracks are then grown in heterogeneous linear elas-
tic media when a critical stress intensity factor criterion is
reached. The cracks with the maximum stress intensity fac-
tors that are computedbymeans of aninteractionintegral [28]
are grown and a load parameter is adjusted so that these
cracks remain approximately at the critical stress intensity.
The explicit solution method follows a crack length con-
trol algorithm. In the case of competition between cracks,
a stability analysis determines the crack conguration path
that leads to the maximum decrease in the potential energy.
The outline of this paper is as follows. Section 2 presents
the description of the nite element algorithmapplied to gen-
erate statistically equivalent human Haversian bone micro-
structures. Section 3 explains the X-FEM for initiation and
multiple crack growth problems until coalescence of cracks
and percolation. Section 4 presents the results for microstruc-
tures with different aging effects and are compared against
the experience. Conclusions are given in Sect. 5.
2 Continuum model of statistically equivalent human
cortical bone microstructures
We consider three-dimensional (3D) unit cells of human cor-
tical bone of 1.52 mm width denoted L and 0.1 mm thick-
ness. Note that the thickness is chosen arbitrarily small for
computation efciency but sufciently thick to ensure the 3D
effect and the presence of several elements in the thickness
discretization. Cortical bone microstructures are represented
as four phase composites composed by an interstitial matrix
and osteonal bers (hollow ellipses in two dimensions, hol-
low elliptic tubes in three dimensions) coated by thin cement
lines and hollowed by Haversian canals considered as free
boundaries where invivocytoplasmic uidwouldowfreely,
see Fig. 2.
2.1 Geometrical description
All geometrical parameters are experimentally measured
from light microscope pictures of eight women bone spec-
imens. The paper concentrates on the specic cases of the
three women as in Fig. 1: (a) denoted w02_07, (b) denoted
w03_07 and (c) denoted w10_07 for their distinctive mor-
phologies. Our morphological measurements include global
parameters such as the osteonal volume fraction denoted F
and porosity denoted p in Table 1. Local parameters such as
the osteonal density
ost
is also measured. Note that w02_07
and w03_07 have similar porosities p but w03_07 has a
123
Comput Mech (2008) 42:579591 581
Fig. 1 a Light microscope
human bone observation w02
0
7.
b Light microscope human bone
observation w03
0
7. c Light
microscope human bone
observation w10
0
7
OR
l
l
1
2

A
B
x
y

t
t = t

l l
3 4

4
cr
cr

3
cr

2
cr

u
u = u

Fig. 2 Schematic model of the cortical unit cell

higher osteonal fraction F and a lower osteonal density
ost
thus larger osteons that can be observed in the case of low
remodeling activity and also in patients treated for osteopo-
rosis. w10_07 displays osteoporosis symptoms with a sig-
nificantly higher porosity and an average osteonal fraction.
The thickness of cement lines is taken as 5 m on aver-
age from experimental observations. The distributions of the
major and minor axis and their angle of the elliptic osteons
and the Haversian canals are statistically measured from the
microscopic observations of each woman and an example is
given in Fig. 3. These distributions are denoted H
o
for the
osteons and H
c
for the Haversian canals.
The random microstructure is constructed using a hard
sphere-type scheme for which details can be found in
Torquato [29]. In this scheme each osteon i is represented in
two dimensions as an ellipse and dened by a center P
i
(x, y)
and a major axis B
i
, a minor axis A
i
and an angle
i
. We
assume the osteons are impenetrable. The construction pro-
cess can be outlined as follows:
1. The position P
i
for one osteon is randomly chosen inside
the unit cell.
2. A major axis B
i
is randomly chosen within the H
oB
distribution.
3. A minor axis A
i
is randomly chosen within the H
oA
distribution.
4. An angle
i
is randomly chosen within the H
o
distribution.
5. If the osteon volume is included in the unit cell without
overlapping another osteon, it is accepted; otherwise, it
is rejected.
6. The procedure is repeated until the surface osteonal
fraction covers more than specic patient osteonal frac-
tion F.
Finally, a major axis and a minor axis for the Haversian
canal for each osteon are randomly chosen within the H
cB
and H
cA
distributions in Fig. 3 and oriented with the same
i
angle. To reproduce microstructures statistically equivalent
to the patient cortical bone specimen, additional constraints
are added to limit the density of small, large or very distorted
osteons or Haversian canals. The constraint locations within
the morphological distributions are chosen so that the global
osteonal density is respected within 10% of the experimen-
tally observed osteonal density. This procedure leads to a rst
approximated porosity that is very close to the experimen-
tally measured porosity p. This rst porosity is then adjusted
exactly to the patient porosity p through an algorithm that
slightly increase Haversian canals that are too small for large
osteons or reduce Haversian canals that are too large for small
osteons.
The algorithm to construct the microstructure is progra-
med in Fortran 90 and then encoded in a Python script. This
geometry is rst discretized using a commercial 3D FEM
code [30] and linear tetrahedron elements for their efciency
and adaptability to complicated geometries. A preliminary
local strain eld analysis is performed to determine the loca-
tions where the cracks initiate. The same geometry is then
discretized in a 2D X-FEM code using Matlab [31] and
Gmsh [32] and quadratic triangles. The longitudinal axis of
the osteons is parallel to the z-direction.
The boundary conditions shown in Fig. 2 are chosen to
mimic the experience and are applied in the 3D model on
the top and bottom surfaces and later in the 2D model on the
top and bottom edges to be compatible. We rst construct
a 3D anisotropic continuum model to initiate the cracks as
precisely as possible despite the fact that the effect of hole
123
582 Comput Mech (2008) 42:579591
Table 1 Morphological parameters of w02_07, w03_07 and w10_07
Specimen w02_07 w03_07 w10_07
L (mm) 1.5 1.5 2.0
F (%) 41.6 57.6 56.6
p (%) 4.6 6.0 17.7

ost
48.6 39.4 89.6
is greater in 2D than 3D. Other 2D numerical and physical
criteria are under study before constructing a 3D X-FEM
model.
2.2 Mechanical properties
Inthis studythe material properties are assignedineachphase
based on experimental characterizations shown in Table 2.
The four phases of the interstitial matrix, the Haversian
canals, the cement lines and the osteons can be described by
four different constitutive laws. Note that the bone remodel-
ing process produces an heterogeneous material with a high
variability in osteon bone-mineral densities and geometries.
Therefore, the local mechanical properties of the osteons and
the cement lines will be different in each osteon.
The osteons are chosen transversally isotropic in the 3D
FEM Abaqus model used for preliminary strain eld analy-
sis and isotropic in the 2DX-FEMMatlab/Gmsh nal model
representing the transversal plane with respect to the osteon
longitudinal axis. Nanoindentationmeasurements of the local
Youngs modulus are performed in dry conditions for over
a hundred osteons in the longitudinal osteonal axis for each
patient. It is shown that local Youngs moduli and the bone-
mineral content are reasonably correlated (r
2
= 0.75) [33,
34]. We assumed the Youngs moduli to be directly related
to the mineral content in our model. Humid nanoindentation
measurement was also performed in the transversal direction.
The obtained experimental distributions of the Youngs mod-
uli are shown in Fig. 4 denoted H
T
in the transversal direc-
tion and H
L
in the longitudinal direction [22]. The micro-
scopic average transversal moduli in the osteonal (E
ost
2
)
nano
and matrix (E
mat
2
)
nano
phase are shown in Table 2. A matrix
stiffness-hardening ratio is denoted .
Gaussian distributions for local longitudinal and trans-
versal Poissons ratio H

L
and H

T
are constructed from
microextensometry measurements on millimetric bone sam-
ples [35]. The transverse shear modulus G
T
is described by a
Gaussian distribution H
G
of mean value and standard devia-
tiontakenfrom[36]. The attributionof the material properties
for each osteon i can be outlined as follows: a probability
p
i
is randomly chosen to which corresponds a longitudi-
nal Youngs modulus within H
L
distribution, a transversal
Youngs modulus within H
T
and the shear modulus within
H
G
. For the Poissons ratio, the probability is xed to 1 p
i
0 100 200 300 400 500
0
20
40
60
80
100
b1,b2 (micrometres)
P

(
%
)
Osteons
Havers canals
(a)
0 1 2 3 4 5
0
20
40
60
80
100
b1/a1, (b2/a2)/(b1/a1)
P

(
%
)
Osteons
Havers canals
(b)
0 50 100 150 200
0
20
40
60
80
100
(degree)
P

(
%
)
(c)
Fig. 3 Example of w03_07 experimental distribution of osteonnal and
Haversian morphological parameters: a major axis, b minor axis ratio,
c angle
123
Comput Mech (2008) 42:579591 583
Table 2 Microscopic average Youngs moduli measured by nanoin-
dentation in w02_07, w03_07 and w10_07
Specimen w02_07 w03_07 w10_07
(E
ost
2
)
nano
(GPa) 13.5 14.3 13.9
(E
mat
2
)
nano
(GPa) 14.6 14.6 15.4
(%) 8.23 2.01 10.94
10 15 20 25 30 35
0
20
40
60
80
100
E (GPa)
P

(
%
)
Experimental E2
Experimental E3
Fig. 4 Example of w03_07 experimental distributions of the Youngs
moduli of the osteons measured by nanoindentation in the transversal
plane (E2) and along the axis of the osteons (E3)
in order to obtain the longitudinal and transversal Poissons
ratio
i
L
and
i
T
. In our model, we imposed a larger Poissons
ratio to osteons with lower Youngs moduli considering that
less mineralizedosteons containmore collagen[37]. The crit-
ical stress intensity factors of the cement lines are linearly
correlated to their Youngs moduli by a factor 10
4
to t in
the range of 0.72 MPa [38, 39]. Tables 2 and 3 summarizes
the local and macroscopic material properties of women 2, 3
and 10.
Lakes has shown cement lines present a specic chemical
composition which gives an isotropic viscoelastic behav-
ior [40]. This composition confers different elastic proper-
ties compared to the osteons that are encircled by the cement
lines. The Youngs modulus is taken 25% lower than the
Youngs modulus of the osteon it encircles as experimen-
tally measured [34]. The Poissons ratio is chosen to be
25% higher [41]. Although only the elastic properties of
the cement lines are used in this study, the critical stress
intensity factors of the cement lines are linearly correlated to
their Youngs moduli by a factor 10
4
and increased by 40%
as they act as crack barriers [10]. The Haversian canals are
modeled by free boundaries and assume no viscous contact
between the cytoplasmic uid and the canal walls.
The interstitial matrix is considered homogeneous and
transversally isotropic with Youngs moduli equal to the
Table 3 Average macroscopic Youngs moduli over 10 statistical
microstructures of w02_07, w03_07 and w10_07
Specimen w02_07 w03_07 w10_07
Transversal modulus (GPa) 12.86 12.59 8.69
Longitudinal modulus (GPa) 21.64 22.97 19.29
average of the nanoindentation measurements performed in
this phase. The matrix is usually stronger than the mean val-
ues of the osteonal moduli by a percentage that is spe-
cic to each patient and can indicate healthy or pathological
conditions. These observations are consistent with medical
observations [42] and the nanoindentation tests performed by
Rho et al. [43], which reveal usually higher Youngs moduli
in the matrix than in the osteons. The Poissons ratio are cho-
sen with the corresponding percent lower than the mean
value of the osteonal Poissons ratio.
3 Multiple crack growth by X-FEM
3.1 Elastic-damage strain driven criterion to initiate cracks
A linear elastic analysis using a FEM discretization with a
commercial code Abaqus is applied to each patient cortical
bone microstructure. The transversal Young modulus E2 can
be determined and compared to experimental measurements
before any damage occurs in Table 3. As in many biological
materials, bone failure can be described by an elastic-damage
strain driven criterion. Thus we determine the initial crack
locations in the microstructure by using the critical maximum
principal strain. When a critical value of 0.4 % [44, 45] has
been reached in the elastic calculation, cracks are initialized
for the XFEM calculation.
Tension tests are performed on the microstructures such
as in Fig. 5 in the y-direction to determine regions where the
critical yield strain is reached. Note that the strain localiza-
tion zone direction depends on the loading direction [22]. The
FEM analysis shows that cracks can be initiated perpendic-
ularly to the direction of maximum principal stress, mode I
in tension. Under vertical tension loading, initial cracks are
placed horizontally starting from the edge of Haversian
canals. Note that we progressively increase the displacement
at the top of the face until local zones with a strain above
0.4%appear; Fig. 5shows the local maximumprincipal strain
elds when the top face of the bone cell is submitted to 0.4%
for patient 3.
3.2 Description of the problem
For our X-FEM multiple crack growth problem, we consider
a 2D elastic unit cell with boundary , n
c
cracks with
surfaces
cr
= {

cr
, = 1 to n
c
}, and n
t
crack tips as
123
584 Comput Mech (2008) 42:579591
Fig. 5 Example of w03_07 unit cell under 0.3% strain tension on the
top face where the local region over 0.4% are in light grey near the
Haversian canals
shown in Fig. 2. The normal to a surface is denoted by n.
The cracks are assumed to be traction free.
Prescribed tractions t (if present) are imposed on the
boundary
t
and prescribed displacements are imposed on

u
. We previouslyinvestigatedconsistencybetweenthe aver-
age local stress eldwhena displacement is appliedonthe top
edge and the average local strain eld when the correspond-
ing traction load to the rst displacement boundary case is
applied to the top of cells containing defects [46]. This aspect
had also been studied for cells of uniform materials contain-
ing defects and the global responses was extremely similar
in both cases when load BC or corresponding displacement
BC were imposed to the top of the cells [21]. For heteroge-
neous cells without defect it is also possible to recover the
same macro eld by applying displacement BC or the trac-
tion BC corresponding to the rst load BC case. According
to Sect. 3.1 in the framework of linear elasticity and in an
explicit algorithm, the boundary conditions can be either to
prescribe a uniform traction t or a uniform displacement u
along the top edge of the cell. Along the bottom edge, the
displacement is xed in the y-direction and the displacement
in the x-direction of node A is also xed in Fig. 2. For sta-
bility reason, prescribing a uniform displacement u along
the top edge usually provided smoother global response than
a prescribed traction. The former are the chosen boundary
conditions to mimic the experimental settings.
The implementation is limited to linear elastic fracture
mechanics in n
p
multi-phase unit cells. The equilibrium
equation and boundary conditions are:
= 0 in (1)
n = t on
t
(2)
u = u on
u
(3)
where is the stress, u(x) is the displacement eld, u
T
=
_
u
x
, u
y
_
; u is the prescribed displacement on
u
.
Each phase of material is considered elastic governed by
C, the elastic moduli tensor, under small deformation. The
length of each crack is denoted
i
; the set of crack lengths is
represented by a matrix = {
i
}, i = 1 to n
t
. The imposed
top displacement u depends linearly on a scalar parameter
called the load factor : u = u
o
, where u
o
is a reference dis-
placement eld. (If a traction t is imposed instead, it depends
linearly on a scalar parameter called the (traction) load factor

t
: t =
t
t
o
, where t
o
is a reference traction eld.
t
is not
necessarily exactly equal to due to numerical approxima-
tion).
For a preliminary study of multiple crack growth in cor-
tical bone modeling, the crack propagation is described by
linear elastic fracture mechanics for brittle multi phase body
by the following Lagrangian form [21]:
L(, u) = W(, u) +
n
t

i =1
_

i
cr
G
i
c
d
i
(4)
where W(, u) is the potential energy of the system, G
i
c
is
the critical energy release rate at crack tip i . G
i
c
is a con-
stant parameter characteristic of each phase and a function
of (x, y). If we dene n
act
as the number of active crack tips,
the second term in the right hand side of Eq. (4) is the energy
dissipated during the growth of the n
act
active crack tips. The
potential energy W(, u) can be decomposed into the strain
energy W
int
and the load potential W
ext
:
W(, u) = W
int
(, u) W
ext
(u) (5)
where
W
int
(, u) =
1
2
_
\
cr
(, u) : C : (, u) d (6)
W
ext
(u) =
t
_

t
t
o
u d (7)
Note W
ext
(u) is not present when we impose a displacement
along the top edge of the cell.
The equilibrium states of the body correspond to the
stationary points of Eq.(4) or points on the boundary of the
feasible domain, so the solution u U corresponds to the
minimization:
L=
u
W(, u)u +
n
t

i =1
_
W(, u)

i
+ G
i
c
_

i
0,
u U
o
,
i
> 0 (8)
where
u
W(, u) is the variation of W with respect to u, u
is the variation of the displacement and d
i
is a crack length
123
Comput Mech (2008) 42:579591 585
differential and:
U = {uuH
1
(\
cr
) , [[u n]] 0 on

cr
, u= u on
u
}
(9)
U
o
= U {uu=0 on
u
} (10)
where n is the normal to the crack and H
1
is the Hilbert space
of functions with square integrable derivatives. Note that the
displacement eld is discontinuous across the crack and that
the jump [[]] in the normal displacement is required to be
non-negative.
The above gives:

u
W(, u) = 0 (11)
W(, u)

i
+ G
i
c
= G
i
+ G
i
c
0, i {1, . . . , n
t
}
(12)
where the energy release rate is dened as:
G
i
=
W(, u)

i
(13)
Equation (11) is the equilibrium and Eq. (12) is the Grifth
criterion for each crack tip i . As it is often common to express
the crack growth law in terms of stress intensity factors, the
energy release rate at tip i can be rewritten K
i
=

E

G
i
where E

is the effective Youngs modulus. At each tip i ,

we compute the equivalent stress intensity factor K
i
eq
[47]
expressed as the norm of the contribution in mode I and
mode II of the interaction integrals [48, 16] as a surface inte-
gral [21]:
K
i
eq
=
_
_
K
i
I
_
2
+
_
K
i
II
_
2
(14)
The crack growth law in linear elastic fracture mechanics of
Eq. (12) can be written:
_
if 0 < K
i
eq
< K
i
c

i
= 0 (no growth)
if K
i
eq
= K
i
c

i
0 (growth)
(15)
where
i
is the crack growth increment of tip i . We do not
consider any closure of the cracks, though we note that
i
<
0 is not a valid solution. If the crack closes, the inequality
[[u n]] 0 must be enforced. The cracks are grown in the
direction of the maximum hoop stress,
i
:

i
= 2 arctan
_
1
4
_
K
i
I
/K
i
II

_
_
K
i
I
/K
i
II
_
2
+ 8
__
(16)
The above gives two directions; we choose the angle that
corresponds to the positive maximum hoop stress. The stress
intensity factors K
i
I
and K
i
II
are computed by an interaction
integral. The equivalent stress intensity factor is computed
by Eq. (14). The expressions for the J-integral and the inter-
action integral as domain integrals can be found in [28].
An explicit algorithm is used to satisfy both Eqs. (11) and
(12). The Equilibrium (11) is discretized and solved for a
prescribed top displacement u
o
and then the load parameter
is adjusted to satisfy Eq. (12), so that the crack with the
maximumstress intensity factor met its critical value. (When
a traction t
o
is prescribed along the top edge, the load param-
eter
t
is adjusted to satisfy Eq. (12), so that the crack with
the maximum stress intensity factor met its critical value).
The crack growth increments are set at the beginning of each
step based on a crack length control scheme so that the
evolution is controlled by a monotonically increasing func-
tion of the total crack length
tot
, i.e. the sum of all active
crack lengths [49]. In some cases, several cracks are close
to their critical stress intensity factors, the number of n
act
active crack to grow is determined by stability analysis and
a set of competitive crack tips N
comp
is determined as in
Budyn et al. [21]. A stability analysis based on a second var-
iation of the Lagrangian form (4) determines the most stable
crack conguration evolution corresponding to that with the
minimum energy dissipation [21, 22]. The derivatives of the
energy release rate with respect to the crack length [50, 51],
are computed by a generalized X-FEM formulation [21] of
the FEM formulation developed by several authors [5254].
All subdeterminants of this matrix [G
i
/
j
] are computed
at time t
n1
and the maximum subdeterminant gives the set
of tips N
act
that will grow at time step t
n
determined by:
N
act
=
_
i N
comp
max
{i, j }N
comp
det
_

2
L(, u)

j
__
=
_
i N
comp
max
{i, j }N
comp
det
_
(G
i
(
o
))

j
_
0
_
(17)
The variation of G
i
c
in Eq. (17) vanishes as we consid-
ered strong discontinuities in bone material properties and
G
i
c
constant in each material phase. When multiple cracks
grow at the same time, it can be noted that the Grifth crite-
rion (12) is exactly satised at the crack when the maximum
stress intensity factor meets its critical value, and approxi-
mately satised at the other cracks.
3.3 Formulation of the X-FEM displacement eld
The approximated displacement eld is discretized by the
eXtended Finite Element Method (X-FEM). The method
describes the cracks with a step function enrichment to repro-
duce the discontinuity of the interior of a crack and uses the
asymptotic near-tip displacement eld enrichment at the tips
of the cracks [16, 2022]. Figure 6 displays the enrichment
scheme. In a mesh with a set of nodes I, all corner nodes of
elements crossed by a crack will be enriched. J
n
contains
the set of corner nodes of the elements cut by the crack n
enriched by the step function of crack n (circled nodes in
Fig. 6). K
m
contains the set of corner nodes of the element
that contains the crack tip m enriched by the branch function
123
586 Comput Mech (2008) 42:579591
(a) (b)
Fig. 6 X-FEM representation of a crack. The circled nodes are
enriched by the step function of crack n and the square nodes are
enriched with the tip enrichment of tip m. The function f gives the
distance of a point to the crack
(squared nodes in Fig. 6). N
c
is the set of cracks and N
t
the
set of crack tips in the entire model. The crack geometry of
crack n is described by narrow banded level set functions
interpolated by signed distance function f
n
(x) [27].
The displacement eld is based on modied functions [55]
adapted for multiple crack problems [21] is as follows:
u
h
(x) =

I I
N
I
(x)u
I
+
n
c

n=1

JJ
n

N
J
(x)a
n
J

H
n
J
(x)
+
n
t

m=1

KK
m

N
K
(x)
_
4

l=1
b
m
l K

F
m
l K
(x)
_
(18)
where N
I
(x) are the shape functions for the continuous dis-
placement eld;

N
J
(x) are the shape functions applied to
the enrichment eld. This choice of shape functions for the
enrichment eld is explained in [55, 56]. a
n
J
are the additional
unknowns for the modied step enrichment

H
n
J
(x) of crack
n and b
m
l K
are the additional unknowns for the tip enrichment
of tip m for the modied lth branch function

F
m
l K
(x); and x
J
is the position of node J. The modied enrichment functions

H
n
J
and

F
m
l K
are given in Budyn et al. [21].
3.4 Crack tip reaching a free boundary or another crack
Within each step, the length of each active crack tip is incre-
mented in the direction of the maximum hoop stress as fol-
lows:

i
=

tot
n
act
(19)
where n
act
is the number of active cracks that grow at the
beginning of step n; n
act
is determined by a stability analysis
at the end of step n 1 described in Eq. (17).
Three possibilities can arise before percolation: (a) a crack
can reach the external free boundary, (b) a crack can reach
the free boundary of an Haverse canal or (c) a crack can coas-
lesce with another crack. In each case, the crack tip is annihi-
lated and its near enrichment removed. This tip is replaced by
either a simple step function for free boundaries or a junc-
tion enrichment for bridging cracks derived in Budyn et al.
[21, 22]. For each active crack tip i , we rst consider a virtual
increment
i
virt
that is set to
i
virt
=
tot
when approach-
ing free boundaries.
tot
is the total crack growth per step. A
more restrictive criterion is chosen to
i
virt
= max(
i
, r
i
)
when approaching cracks.
i
is the increment of growth of
Eq. (19) and r
i
is the radius of the domain of computation
of the interaction integral at tip i ; the radius r
i
being about
twice the element length. The length of the added increment

i
is adjusted when the growing crack encounters a free
boundary or another crack so that the tip dies exactly on the
boundary edge.
The direction and location where to join boundaries or
another crack is usually determined by the maximum hoop
stress direction when the mesh is ne. However the direc-
tion is chosen equal to the minimal principal strain direction
in front of the crack tip when the interaction integral cal-
culation does not hold anymore. This situation can occur if
many cracks are interacting or when the domain of calcu-
lation of the interaction integral is largely truncated by an
internal (Haverse canals) or external boundary or contains
another crack (and other phases) in a region where only a
very small amount of brittle material remains. In these cases
the calculation of K
I
and particularly K
II
might be impre-
cise and the direction of the maximum hoop stress incorrect.
However a very small amount of brittle material subjected
to a tensile stress will break and therefore join the bound-
ary or another crack [5761]. Therefore the local minimal
principal strain will provide the correct growth direction. A
junction enrichment is applied when a crack joins another
crack. An extreme case can occur when a crack joins another
crack in the same element where the tip of the connected
crack is located also. In this special junction case, the near
tip enrichment of the connected crack should be removed as
well and changed into a step enrichment to ensure a perfect
strong discontinuity within this entire element that contains
the junction. Whenonlyanother phase encounters the domain
of calculation of the J-integral, the correct direction of crack
propagation is still obtained because the gradient between
the moduli of the phases is not too large.
4 Results
We consider square unit cells of cortical bone of specic
width L given in Table 1. In 3-D a thickness of 0.1 mm was
chosen. Five statistical microstructures were generated for
each patient to attach statistics to the results of the compu-
tations. After initiation the sample geometries are given in
Table 4 and Fig. 7. The cells are loaded under tension in the
y-direction; a positive displacement is prescribed at the top
edge and the displacement in the y-direction is prescribed
123
Comput Mech (2008) 42:579591 587
Table 4 Geometrical parameters of statistical samples of w02_07,
w03_07 and w10_07
Case 1 Case 2 Case 3 Case 4 Case 5
w02_07
Phase number 89 93 95 95 95
Crack number 54 60 66 61 66
w03_07
Phase number 75 75 71 77 83
Crack number 49 50 46 55 54
w10_07
Phase number 167 157 151 159 157
Crack number 49 38 27 43 45
Fig. 7 Cracks are initiated by an elastic-damage criterion (0.4%
max
prime
strain [44]) in the gray regions of Fig. 5. Initiated cracks in blue; crack
growths in red using the X-FEM in w02_07 (a, d, g, j, m), w03_07 (b,
e, eh, k, n)) and w10_07 (c, f, i, l, o)
equal to zero at the bottom edge. The displacement of the
bottomleft corner node in the x-direction is prescribed equal
to zero (see Fig. 2). The material properties are random and
representative of each patient cortical bone microstructure as
described previously.
The forcedeection curves, expressed as a nominal stress
versus nominal strain curve for ve samples of each women
2, 3 and 10 are given in Fig. 8. Our analysis is static. The
load deection curves of the three women (Fig. 8) typically
display three phases: in the rst few steps the cracks are
growing inside the osteons, which produces various degree
of strainhardening; after reachinga peak, one dominant crack
is able to cross through the cement line of the osteon it origi-
natedfrom, anddamage the matrix. Duringthis perioda strain
softening is observed with various degree of severity as the
matrix looses progressively its structural integrity. Finally in
the last steps, the structure has lost most of its strength and
integrity and the cracks will grow towards complete perco-
lation. In the case of w10_07, which is very osteoporotic,
the load deection curve displays a short hardening phase
(Table 5).
For each woman the loaddeection curves appear rela-
tively consistent in Fig. 8 and Table 6 with a relatively nar-
row standard deviation. Note that for the crack initiation, a
mesh renement would not affect their initial locations but
slightly rene their initial lengths. The effect of the mesh
renement for the continuum and the cracks would smooth
the load-defection contour and the crack paths; this phenom-
enon has been demonstrated in a convergence study in Budyn
et al. [21].
Table 6 and Fig. 8 showw02_07 fracture strength is about
one third higher than w03_07 fracture strength despite these
two samples exhibit similar macroscopic moduli in Table 3.
w03_07 has a higher osteonal fraction and slightly higher
porosity but a lower osteonal density than w02_07. There-
fore w03_07 has larger osteons. In Table 2, we also note
that w02_07 has a higher osteons/matrix modulus contrast
and appears more heterogeneous. This would correlate of the
observation of the benecial structural function of cement
lines on the fracture strength in Budyn et al. [46]. In the for-
mer study it was noticed that in the absence of cement lines
the osteons tend to transfer more deformation to the matrix.
With cement lines, the osteons remain conned and exhibit a
wider range of deformations. This phenomenon can partially
explain the hardening phase observed in the global behav-
ior due to primary growth of cracks inside more numerous
osteons before damaging the matrix. We observe that more
inhomogeneity makes the response richer and contribute to
strengthen the structure. Therefore cement lines and numer-
ous osteons appear as critical elements in the protection of
bone against fracture as some experimental observation of
the effect of cement lines as osteonal barriers can be seen
in Mohsin et al. [10]. However w02_07 exhibits a steeper
123
588 Comput Mech (2008) 42:579591
0 1 2 3 4 5
0
20
40
60
80
100
120
140
160
nominal strain (%)
n
o
m
i
n
a
l

s
t
r
e
s
s

(
M
P
a
)
w02t1_07
w02t2_07
w02t3_07
w02t4_07
w02t5_07
(a)
0 1 2 3 4 5
0
20
40
60
80
100
120
140
160
nominal strain (%)
n
o
m
i
n
a
l

s
t
r
e
s
s

(
M
P
a
)
w03t2_07
w03t3_07
w03t4_07
w03t5_07
w03t6_07
(b)
0 1 2 3 4 5
0
20
40
60
80
100
120
140
160
nominal strain (%)
n
o
m
i
n
a
l

s
t
r
e
s
s

(
M
P
a
)
w03t2_07
w03t3_07
w03t4_07
w03t5_07
w03t6_07
(c)
Fig. 8 Global stressstrain response in w02_07, w03_07 and w10_07
softening phase than w03_07. Despite having a lesser num-
ber of phases, w03_07 osteonal volume fraction is higher
than in w02_07 and might play a role in slowing down the
Table 5 Fracture strengths of the statistical samples of w02_07,
w03_07 and w10_07

u
(MPa) Case 1 Case 2 Case 3 Case 4 Case 5
w02_07 109.2 144.5 152.6 131.3 127.0
w03_07 112.7 116.7 81.6 77 95.9
w10_07 56.2 50.9 47.9 70.9 49.4
Table 6 Average fracture strengths of w02_07, w03_07 and w10_07
Specimen w02_07 w03_07 w1_07
Numerical strength (MPa) 132.7 101.7 55.4
Standard deviation (MPa) (14.98) (12.52) (8.24)
Experimental strength (MPa) 92 95 low
softening process. On the other hand, w10_07 displays an
significantly higher porosity than the other two women,
which contributes to a lower modulus (Table 3) and lower
fracture strength(Tables 5, 6; Fig. 8). Despite a relativelyhigh
osteonal volume fraction and a high osteon density, w10_07
exhibits almost no hardening but a relatively slow softening.
The numerical fracture strength of w03_07 is in close agree-
ment with the experimental measurements in Table 6. The
numerical fracture strength of w02_07 agrees to some extend
with the experimental measurement performed on this highly
variable biological tissue for which more experimental tests
are scheduled.
5 Conclusion
The results of this preliminary human study present a direct-
simulation multiple scale approach to describe fracture in
Haversian cortical bone microstructures under tension. The
human Haversian microstructure is discretized by a statisti-
cal nite element model to investigate the inuence of the
local patient morphological and mechanical parameters of
the microstructure on its mechanical behavior. The majority
of these parameters were based on physical data, however
some limitations can be given. For the geometrical descrip-
tion the osteons are modeled by disjoint elliptical tubes,
which is an idealized geometry and does not allow them
to connect to each other through Volkman canals [62]. We
study bone at the osteonal level and do not include osteocyte
lacunae [15, 63]. We also modeled interstitial matrix as
homogeneous material. Our study focused on the effect of the
heterogeneities density and size ratio on the microstructure
fracture strength. Other heterogeneities such as cement line
thickness, matrix heterogeneities (remodeling), etc., inu-
ence the crack paths and will be studied in further work.
123
Comput Mech (2008) 42:579591 589
Improvement in the implementation of the boundary
conditions might also impact the smoothness of the global
response.
The X-FEM is particularly suitable for crack growth in
heterogeneous media because remeshing is avoided. For
accuracy purposes, higher order elements, which are qua-
dratic for the standard displacement eld and linear for the
enrichment, have been applied. In contrast to boundary
element methods, the method easily handles microstructure
heterogeneities. The method applied to static crack growth
is explicit and satises exactly both the equilibrium (11) and
the Grifth criterion (12) at each step of the load deec-
tion curve. Astability analysis based on energy consideration
enables us to solve the case of competitive crack growth. The
response of the unit cell is tracked until almost complete fail-
ure when the cracks have joined and reached the free bound-
aries and almost percolated the cell. In the present model we
triedtoimplement boundaryconditions close toexperimental
testing.
The model accesses two different scales: the material
(macro) scale at the unit cell level and the (micro) scale inside
the material for the straineldinside the osteons, ina mimetic
osteonal microstructure. At the macroscopic scale, the model
provides the overall response and thus the fracture strength
of the material. The present multi-scale model shows the
inuence of the local geometrical and mechanical properties
of the microstructure on the macroscopic properties of cor-
tical bone [34], which are the only accessible information in
clinical medical studies [64, 65]. The model shows in particu-
lar howtwo patients (2 and 3) exhibiting similar macroscopic
stiffness, display different fracture strength and failure paths
in tension. The model helps to build a constitutive stress
strain law at the material scale that can be coupled later on to
a higher and/or lower scale(s) for more complex multi-scale
approaches.
At the microscopic level, even if the matrix is homoge-
neous its strain eld is not. Our results show how morpho-
logical and mechanical heterogeneities inuence the local
strain eld and on the macroscopic response. Previously
in [22] cement lines, less stiff but tougher, have been shown
to play an important role in isolating the osteons from the
matrix, explaining only partially why cement lines are impli-
cated with energy in fracture processes in the deection of
crack propagation by slowing it down [66] or debonding the
osteon [67]. Our present results suggest how critical Haver-
sian porosity is source of localization and fracture nucle-
ation for transversal tension loading. The model also shows
how an increase in morphlological (high osteon density) and
mechanical (harder matrix) heterogeneities tend to increase
the hardening phase and the fracture strength. Therefore low
remodeling activity that reduces osteon density lowers the
macroscopic fracture strength. The model shows how high
porosity and large osteons in osteoporosis contribute to lower
the macroscopic moduli and the fracture strength and shorten
strain hardening. Finally the model tends to suggest that high
osteonal volume fractionseems toextendthe softeningphase.
However further investigations on more patients are sched-
uled to rene the understanding between the effects of osteon
density and osteonal volume fraction on the load deection
response.
In summary, this model presents preliminary results that
characterize the local properties of human cortical bone and
emphasizes the importance of the microstructural constitu-
ents (porosity, osteonal fraction and density, cement lines,
etc.) to prevent the progression of localized damage zones
and cracks in healthy bone. Any pathological modications
of these constituents alters the integrity of the microstruc-
ture. While some of the alterations are detectable at the mac-
roscopic scale such as a loss in stiffness, however degra-
dation of the fracture strength are not detectable by conven-
tional clinical tools. Our model proposes alternative means to
investigate such properties and diagnose pathologies. More
experimental validations will be developed for this type of
numerical approach. Further investigations include the phys-
iological and mechanical properties of the cement lines and
the matrix to feed this model with more experimental damage
and fracture parameters in tension and compression loading
for a more complete multiscale study of bone fracture in an
entire bone.
Acknowledgments The authors are grateful for the research support
of the University of Illinois at Chicago and the CNRS.
References
1. Katz JL (1976) Hierarchical modeling of compact haversian bone
as a ber reinforced material. Adv Bioeng ASME 1718
2. Aoubiza B, Crolet JM, Meunier A (1996) On the mechanical char-
acterization of compact bone structure using the homogenization
theory. J Biomech 29(12):15391547
3. Gottesman G, Hashin Z (1980) Analysis of viscoelastic behaviour
of bones on the basis of microstructure. J Biomech 13:8996
4. Hogan HA (1992) Micromechanics modeling of haversian cortical
bone properties. J Biomech 25(5):549556
5. Currey JD (2003) The many adaptations of bone. J Biomech
36:14851495
6. Wright TM, Hayes WC (1977) Fracture mechanics parameters for
compact boneeffects of density and specimen thickness. J Bio-
mech 10:419430
7. Hui SL, Slemenda CM, Johnston CC (1988) Age and bone mass as
predictors of fracture in a prospective study. J Clin Invest 81:1804
1809
8. Schafer MB, Burr DB (1988) Stiffness of compact bone effect of
porosity and density. J Biomech 21:1316
9. OBrien FJ, Taylor D, Lee TC (2005) The effect of bone micro-
structure on the initiation and growth of microcracks. J Orthop Res
23:475480
10. Mohsin S, OBrien FJ, Lee TC (2006) Osteonal crack barriers in
ovine compact bone. J Anat 208:8189
123
590 Comput Mech (2008) 42:579591
11. Nalla RK, Kinney JH, Ritchie RO (2003) Mechanistic fracture cri-
teria for the failure of human cortical bone. Nat Mater 2:164168
12. Katz JL (1971) Hard tissue as composite materiali. Bounds on
the elastic behavior. J Biomech 4(5):455473
13. Dong XN, Zhong XN, Huang YY, Guo XE (2005) A generalized
self-consisitent estimate for the effective elastic moduli for ber-
reinforced composite materials with multiple transversely isotropic
inclusions. Int J Mech Sci 47(6):922940
14. Dong XN, Guo XE (2006) Prediction of cortical bone elastic con-
stants by a two-level micromechanical model using a generalized
self-consistent method. J Biomech Eng Trans ASME 128(3):
309316
15. Prendergast PJ, Huiskes R (1996) Microdamage and osteocyte-
lacuna strain in bone: a microstructural nite element analysis.
J Biomech Eng Trans ASME 118:240246
16. Belytschko T, Black T (1999) Elastic crack growth in nite
elements with minimal remeshing. Int J Numer Methods Eng
45(5):610620
17. Belytschko T, Mos N, Usui S, Parimi C (2001) Arbitrary dis-
continuities in nite elements. Int J Numer Methods Eng 50(4):
9931013
18. Melenk JM, Babuska I (1996) The partition of unity nite ele-
ment method: basic theoryandapplications. Comput Methods Appl
Mech Eng 39:289314
19. Dolbow J, Mos N, Belytschko T (2001) An extended nite ele-
ment method for modeling crack growth with frictional contact.
Comput Methods Appl Mech Eng 190(5152):68256846
20. Mos N, Dolbow J, Belytschko T (1999) A nite element method
for crack growth without remeshing. Int J Numer Methods Eng
46(1):131150
21. Budyn E, Zi G, Mos N, Belytschko T (2004) A method for multi-
ple crackgrowthinbrittle materials without remeshing. Int J Numer
Methods Eng 61(10):17411770
22. Budyn E, Hoc T (2007) Multiple scale modeling of cortical bone
fracture in tension using X-FEM. Revue Europenne de Mcanique
Numrique (Eur J Comput Mech) 16:213236
23. Osher S, Sethian JA (1988) Fronts propagating with curvature
dependent speed: algorithms based on hamilton-jacobi formula-
tions. J Comput Phys 79(1):1249
24. Sethian JA (1999) Level sets methods & fast marching methods:
evolving interfaces in computational geometry, uid mechanics,
computer vision and materials science. Cambridge University
Press, Cambridge
25. Burchard P, Cheng LT, Merriman B, Osher S (2001) Rmotion of
curves in three spacial dimensions using a level set approach. Int
J Numer Methods Fluids 170:720741
26. Osher S, Cheng LT, Kang M, Shim Y, Tsai YH (2002) Geometri-
coptics in a phase-space-based level set and eulerian framework.
J Comput Phys 179(2):622648
27. Ventura G, Budyn E, Belytschko T (2003) Meshfree and particle
methods and their applications. Int J Numer Methods Eng 58:
15711792
28. Moran B, Shih CF (1987) A general treatment of crack tip contour
integrals. Int J Fract 35:295310
29. Torquato S (2000) Randomheterogeneous mediamicrostructure
and macroscopic propertiesinterdisciplinary applied mathemat-
icsmechanics and materials. Springer, New York
30. ABAQUS (2004) Users manual. Hibbit, Version 6.3, Karlsson&
Sorensen, Providence
31. MATLAB7 (2006) Users manual. MATLAB, Version 7, The
MathWorks, USA
32. GMSH (2006) Users manual. GMSH, Version 62, Geuzaine &
Remacle, Cleveland USA; Louvain Belgium
33. Fan Z, Swadener E, Rho JY, Roy ME, Pharr GM (2002) Aniso-
tropic properties of human tibial cortical bone as measured by nan-
oindentation. J Orthop Res 20:806810
34. Hoc T, Henry L, Verdier M, Aubry D, Sedel L, Meunier A
(2006) Effect of microstructure on the mechanical properties of
haversian cortical bone. Bone 38:466474
35. Henry L, Hoc T, Budyn E (2007) Microextensometry and mechan-
ical behaviour of haversian cortical bone. Eur J Comput Mech
(submitted)
36. Reilly DT, Burstein AH (1975) The elastic and ultimate properties
of compact bone tissue. J Biomech 8(6):393405
37. Katz JL, Ukraincik K (1971) On the anisotropic elastic properties
of hydroxyapatite. J Biomech 29(4):221227
38. Shelton DR, Martin RB, Stover SM, Gibeling JC (2003)
Transverse fatigue crack propagation behavior in equine cortical
bone.
J Mater Sci 38:35013508
39. Hazenberg JG, Taylor D, Lee TC (2006) Mechanisms of short
crack growth at constant stress in bone. Biomaterials 27:2114
2122
40. Lakes R, Saha S (1979) Cement line motion in bone. Nature
204:501503
41. Sabelman EE, Koran P, Diep N, Lineaweaver WC (1997) Colla-
gen/hyaluronic acid matrices for connective tissue repair. In: First
Smith & Nephew international symposium: advances in tissues
engineering and biomaterials
42. Sumner DR (2006) Private communications
43. Rho JY, Zioupos P, Currey JD, Pharr GM (2002) Microstructural
elasticity and regional heterogeneity in human femoral bone of
various ages examined by nano-indentation. J Biomech 35:189
198
44. Pattin CA, Calet WE, Carter DR (1996) Cyclic mechanical prop-
erty degradation during fatigue loading of cortical bone. J Biomech
29:6979
45. Bayraktar HH, Morgan EF, Niebur GL, Morris GE, Wong EK,
Keaveny TM (2004) Comparison of the elastic and yield properties
of human femoral trabecular and cortical bone tissue. J Biomech
37:2735
46. Budyn E, Hoc T (2006) Multiple scale modeling for crack growth
in cortical bone under tension using the extended nite element
method. Revue Europenne de Mcanique Numrique (Eur J Finite
Element Methods), Special Issue (submitted)
47. Paris PC, Sih SC (1965) Stress analysis of cracks. In: Fracture
toughness and testing and its applications. Society for Testing and
Materials, Philadelphia, pp 3083
48. Yau Y, Wang S, Corten H (1980) A mixed-mode crack analysis of
isotropic solids using conservation laws of elasticity. J Appl Mech
47:335341
49. Bocca P, Carpinteri A, Valente S (1991) Mixed mode fracture of
concrete. Int J Solids Struct 27:11391153
50. Nguyen QS, Stolz S (1985) ENDOMAGEMENT, FATIGUE,
RUPTUE.sur le problme en vitesse de propagation de ssure
et de dplacement en rupture fragile ou ductile. note de nguyen
quoc son et claude stolz, prsente par paul germain. Comptes-
Rendus de lAcadmie des Sciences de Paris, t. 301, Srie II(10):
661664
51. Nguyen QS, Stolz C, Debruyne G (1990) Energy methods in frac-
ture mechanics: stability, bifurcation and second variations. Eur
J Mech A/Solids 2:157173
52. Suo XZ, Combescure A(1989) Sur une formulation mathmatique
de la drive seconde de lnergie potentielle en thorie de la rup-
ture fragile. Comptes-Rendus de lAcadmie des Sciences de Paris,
t. 308, Srie II:11191122
53. Suo XZ, Combescure A (1992) Double virtual crack extension
method for crack growth stability assessment. Int J Fract 57:127
150
54. Suo XZ, Valeta MP (1998) Second variation of energy and an asso-
ciated line independent integral in fracture mechanics. ii numerical
validations. Eur J Mech A/Solids 17(4):541565
123
Comput Mech (2008) 42:579591 591
55. Stazi FL, Budyn E, Chessa J, Belytschko T (2002) An extended
nite element method with higher-order elements for curved
cracks. Comput Mech 31:3848
56. Chessa J, Wang HW, Belytschko T (2003) On construction of
blending elements for local partition of unity enriched nite ele-
ments. Int J Numer Methods Eng 57:10151038
57. Kienzler R, Herrmann G (2002) Fracture criteria based on local
propertiesof the eshelby tensor. Mech Res Commun 29(6):
521527
58. Rubinstein A (1996) Macrocrack-microdefect interaction. J Appl
Mech 53:505510
59. Demir I, Zbib HM, Khaleel M (2001) Microscopic analysis of
crack propagation for multiple cracks, inclusions and voids. Theor
Appl Fract Mech 36:147164
60. Deng H, Nemat-Nasser S (1992) Bridging multi-scale method for
localization problems. Mech Mater 13:1536
61. Nemat-Nasser S, Hori M (1987) Toughening by partial or full
bridging of cracks in ceramics and ber reinforced composites.
Mech Mater 6:245269
62. Jones AC, Sheppard AP, Sok RM, Arns CH, Limaye A,
Averdunk H, Brandwood A, Sakellariou A, Senden TJ, Milthorpe
BK, Knackstedt MA (2004) Three-dimensional analysis of corti-
cal bone structure using X-ray micro-computed tomography. Phys
A 339:125130
63. Qiu S, Rao DS, Fyhrie DP, Palnitkar S, Partt AM(2005) The mor-
phological association between microcracks and osteocyte lacunae
in human cortical bone. Bone 37:1015
64. Bloebaum RD, Liau DW, Lester DK, Rosenbaum TG (2006)
Dual-energy x-ray absorptiometry measurement and accuracy of
bone mineral after unilateral total hip arthroplasty. J Arthroplasty
21(4):612622
65. Cooper D, Turinsky A, Sensen C, Hallgrimsson B (2007) Effect of
voxel size on 3d micro-ct analysis of cortical bone porosity. Calcif
Tissue Int 80:211219
66. Advani SH, Lee TS, Martin RB (1987) Analysis of crack arrest
by cement lines in osteonal bone. In: Proceedings of ASME Annu
Winter Meeting, vol 3, pp 57581
67. Guo XE, He MY, Goldstein SA (1995) Analysis of viscoelastic
behaviour of bones on the basis of microstructure. Bioeng Conf
ASME, BED-29:112117
123