You are on page 1of 7

Three dimensional computational analysis of fatigue crack propagation

in functionally graded materials
Baris Sabuncuoglu
a,⇑
, Serkan Dag
a
, Bora Yildirim
b
a
Department of Mechanical Engineering, Middle East Technical University, Ankara 06800, Turkey
b
Department of Mechanical Engineering, Hacettepe University, Ankara 06800, Turkey
a r t i c l e i n f o
Article history:
Available online xxxx
Keywords:
Functionally graded materials
Fracture mechanics
Fatigue crack propagation
Finite element method
Stress intensity factors
a b s t r a c t
This article proposes a new finite elements based three dimensional method developed to study the phe-
nomenon of fatigue crack propagation in functionally graded materials (FGMs). The particular problem
examined in detail is that of an initially-elliptical crack located in a functionally graded medium, sub-
jected to mode I cyclic loading. The crack is modelled by employing three dimensional finite elements;
and the stress intensity factors (SIFs) around the crack front are computed by the application of the dis-
placement correlation technique (DCT). Fatigue crack propagation calculations are based on the Paris–
Erdogan crack growth law. The developed procedure makes it possible to generate the crack front profiles
corresponding to given numbers of loading cycles. Proposed methods are validated by examining the
behaviours of both stationary and propagating cracks. Numerical analyses conducted for an initially-
elliptical crack lying in a graded medium demonstrate that the proposed model can effectively capture
the evolution of the crack front morphology. Hence, the methodology presented in this article can be used
along with fracture toughness data to assess the fatigue lives of functionally graded components contain-
ing cracks that possess arbitrary front profiles.
Ó 2011 Published by Elsevier B.V.
1. Introduction
Functionally graded materials (FGMs) are multiphase compos-
ites that possess spatial variations in the volume fractions of the
constituents. These variations are intentionally introduced so as
to take advantage of desirable properties of the constituents, such
as the high thermal resistance of ceramics and superior fracture
toughness of metals. FGMs have been used successfully in a num-
ber of technological applications including thermal barrier coat-
ings, solid oxide fuel cells, high performance cutting tools, and
biomedical materials. The spatial variations in the volume fractions
of the constituents of FGMs require that certain physical properties
be represented by continuous functions of spatial coordinates for
analytical and computational purposes.
Fracture mechanics of FGMs has been studied extensively in or-
der to develop an understanding of the influence of the continuous
gradation on the cracking behaviour. Although great bulk of this
work is based on the two-dimensional fracture mechanics theory,
in recent years there have been attempts to build computational
models for three dimensional fracture mechanics analysis of func-
tionally graded materials. Among the computational methods
proposed for three dimensional fracture analysis of FGMs, we can
mention the domain integral method [1], the displacement corre-
lation technique (DCT) [2,3], the method of enriched finite ele-
ments [4], the interaction integral approach [5], and the
boundary domain integral method [6]. Note that in all these stud-
ies on three dimensional computational fracture mechanics of
FGMs, the focus is on stationary cracks, and there seems to be no
work on fatigue behaviour of FGMs.
Present study is directed towards developing a three dimen-
sional computational analysis technique to study the phenomenon
of fatigue crack propagation in functionally graded structures. Fa-
tigue crack growth computations are carried out by considering
an embedded crack in the examined functionally graded medium.
The initial front profile of the embedded crack is assumed to be in
the shape of an ellipse. Elliptical cracks or voids are known to be
the source of various structural failures especially under repeated
loading conditions. Fatigue crack growth calculations involving
elliptical cracks require the computation of the stress intensity fac-
tors (SIFs) around the crack border. We use the displacement cor-
relation technique [2,3] in conjunction with the finite element
method to evaluate the stress intensity factors. The subsequent
front profiles of the initially-elliptical cracks subjected to cyclic
loading are generated by adopting the fatigue crack growth analy-
sis methods proposed by Paris and Erdogan [7], and Joseph and
Erdogan [8]. The main outcome of this study is the finite elements
based three dimensional computational analysis tool, which can be
0927-0256/$ - see front matter Ó 2011 Published by Elsevier B.V.
doi:10.1016/j.commatsci.2011.06.010

Corresponding author. Present address: Interuniversity Microelectronics Center
(IMEC), Kapeldreef 75, B-3001 Leuven, Belgium. Tel.: +44 (0) 1509 227520.
E-mail addresses: Baris.Sabuncuoglu@imec.be (B. Sabuncuoglu), sdag@metu.e-
du.tr (S. Dag).
Computational Materials Science xxx (2011) xxx–xxx
Contents lists available at ScienceDirect
Computational Materials Science
j our nal homepage: www. el sevi er . com/ l ocat e/ commat sci
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010
used to predict the fatigue lives of functionally graded compo-
nents. Presented numerical results illustrate the influence of mate-
rial property gradation on the fatigue propagation characteristics
of cracks located in cyclically loaded ceramic–metal FGMs.
2. The computational procedure
In this section, we provide the details of the three dimensional
computational analysis methodology developed to examine fatigue
crack growth in functionally graded materials. The geometry of the
particular problem considered is depicted in Fig. 1. An elliptical
crack resides in an elastic functionally graded medium, whose
material properties are functions of the x-coordinate. The major
and minor radii of the elliptical crack are denoted by c and a,
respectively. The crack is loaded through the normal stress r ap-
plied over the remote bounding planes. Since, the material is as-
sumed to possess property gradation in x-direction, one-quarter
of the medium is modelled by considering symmetries about xy-
and xz-planes. The finite element model for the problem depicted
in Fig. 1 is constructed by using the general purpose finite element
analysis software ANSYS [9]. Fig. 2 shows the quarter model gener-
ated and close-up view of the region around the crack front. The
dimensions were taken such that the bounding planes would have
no effect on the behaviour of the elliptical crack. In order to simu-
late the square-root singularity of the strain field near the crack
border, the volume around this border is modelled by utilizing col-
lapsed 20-noded isoparametric brick elements, whereas the do-
main excluding this volume is discretized by means of brick and
tetrahedral elements. Smooth spatial variations of the material
properties of the graded medium are taken into account in the fi-
nite element model by computing the properties of each finite ele-
ment at its centroid. The properties are uniformover each element;
however they vary from element to element according to the posi-
tion of the centroid. This is the so called homogeneous finite ele-
ment approach and known to lead to highly accurate numerical
results so long as there is appropriate mesh refinement in the finite
element model [2,3].
If the displacement field for the cracked functionally graded
medium is known, the mode I stress intensity factor distribution
around the crack front can be evaluated by using one of the avail-
able computational methods. In the present study, we use the dis-
placement correlation technique in the evaluation of the SIFs. In
this technique, the expression of the stress intensity factor is de-
rived by considering a point P on the crack front. Fig. 3 shows such
a point P lying on the border of the crack. The origin of a Cartesian
coordinate system is also located at point P. This coordinate system
comprises tangential (t), normal (n), and binormal (b) directions.
Fig. 4 depicts the deformed shape of the crack in the normal plane
passing through point P. The normal plane is formed by n and b
Fig. 1. (a) An elliptical crack in an infinite volume and (b) a point P on the elliptical crack front.
Fig. 2. (a) The quarter model and (b) the crack front region.
2 B. Sabuncuoglu et al. / Computational Materials Science xxx (2011) xxx–xxx
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010
axes. The nodes 1–3 in Fig. 4 are considered to be at the edge of a
collapsed 20-noded brick element. Under mode I loading condi-
tions, the asymptotic displacements of these nodes in b-direction
can be found from [3]
u
b
ðnÞ ¼
4ð1 Àðm
P
Þ
2
Þ
E
P
ffiffiffiffiffiffiffi
jnj
2p

K
I
; ð1Þ
where K
I
is the mode I stress intensity factor; E
P
and m
P
are the mod-
ulus of elasticity and the Poisson’s ratio computed at point P,
respectively; and the superscript P implies that the corresponding
material property should be calculated at point P. The mode I SIF
at P is defined as [2,3]
K
I
¼
ffiffiffiffiffiffiffi
2p
p
E
P
4ð1 Àðm
P
Þ
2
Þ
lim
jnj!0
u
b
ðjnjÞ
ffiffiffiffiffiffi
jnj

¸
: ð2Þ
Then, using Eqs. (1) and (2) and the displacements of the nodes 2
and 3, it is possible to express the mode I stress intensity factor
in the following form:
K
I
¼
ffiffiffiffiffiffiffi
2p
p
E
P
4ð1 Àðm
P
Þ
2
Þ
R
3=2
3
u
b2
ÀR
3=2
2
u
b3
ffiffiffiffiffi
R
2
p ffiffiffiffiffi
R
3
p
ðR
3
ÀR
2
Þ
¸ ¸
: ð3Þ
R
2
and R
3
in Eq. (3) are the distances of nodes 2 and 3 with respect
to the origin.
The fatigue crack growth calculations for the configuration
shown in Fig. 1 are based on the mathematical models proposed
by Paris and Erdogan [7], and Joseph and Erdogan [8]. Suppose that
the stress applied to the graded medium r varies cyclically in time
with a constant stress range Dr and a constant mean stress. Then,
the growth rate at a point on the crack front can be calculated
using Paris–Erdogan law [7,8], which is expressed as:
dq
dN
¼ CDK
n
I
; ð4Þ
where q is the radius of curvature at the considered point on the
crack front, N is the number of cycles, DK
I
is the stress intensity fac-
tor range, and C and n are material parameters. In the finite element
implementation, first, the increment for a particular point on the
crack front is specified as Dq
0
. The number of cycles corresponding
to this increment is computed using
DN
0

Dq
0
C
0
DK
n
0
I0
: ð5Þ
The increments at other points around the crack front are then cal-
culated by substituting Eq. (5) into Eq. (4):
Dq
i

C
i
DK
n
i
Ii
C
0
DK
n
0
I0
Dq
0
; ð6Þ
where the subscript i stands for a generic point on the crack front. A
sufficiently small value needs to be used for Dq
0
to be able to obtain
accurate results through Eqs. (5) and (6).
After calculating the crack growth increments for a specified
number of points on the crack front, the new crack front profile
is generated by connecting the new locations by splines. The pro-
cedure is then repeated to form the crack front profile correspond-
ing to the next specified increment.
3. Verification of the developed procedure
We validated the methods described in the previous section on
problems involving stationary and propagating cracks. For the sta-
tionary cracks, the application of the displacement correlation
technique is verified by considering two different problems, the
first of which is the problem of an elliptical crack in a homoge-
neous medium, and the second is that of a surface crack in a func-
tionally graded plate. As for the propagating cracks, the validation
study is conducted by examining fatigue growth in a homogeneous
medium. In what follows below, we present the results generated
for these validation problems.
By considering the elliptical crack problem shown in Fig. 1, Ir-
win [10] derived a simple closed form expression for the mode I
stress intensity factor. In this derivation, the material was assumed
to be homogeneous. We first focus on this problem solved by Irwin
[10] and provide comparisons of the mode I SIFs computed by the
DCT developed in the present study to those obtained through Ir-
win’s formula. These comparisons are given in Table 1. In the cal-
culation of these results, the medium shown in Fig. 1 is assumed
to be homogeneous and made of Ti–6Al–4V. The material proper-
ties are taken from Refs. [3,11] and provided in Table 2. The minor
and major radii of the crack are respectively set as 0.5 cm and
1.5 cm. The normalized mode I stress intensity factor provided in
Table 1 at various locations around the crack front is defined as
K
In
¼
K
I
r
ffiffiffiffiffiffi
pa
p : ð7Þ
Examining the results given in Table 1, it is seen that the nor-
malized mode I stress intensity factors computed by the DCT are
in excellent agreement with those obtained through Irwin’s for-
Fig. 3. Crack front and the local coordinate system at point P.
Fig. 4. Deformed shape of the crack surface in the normal plane.
Table 1
Comparisons of the normalized stress intensity factors computed by the displacement
correlation technique to those obtained by the formula given by Irwin [10].
2Ø/p Irwin [10] DCT % Difference
1 (minor axis) 0.8979 0.8977 0.02
0.875 0.8902 0.8898 0.05
0.75 0.8671 0.8675 0.04
0.5 0.7752 0.7755 0.03
0.375 0.7075 0.7068 0.10
0.25 0.6293 0.6270 0.37
0.125 0.5540 0.5523 0.31
0 (major axis) 0.5184 0.5180 0.09
B. Sabuncuoglu et al. / Computational Materials Science xxx (2011) xxx–xxx 3
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010
mula. Hence, it is possible to compute the mode I stress intensity
factors to within a high degree of accuracy by employing the dis-
placement correlation technique developed in the present study.
To be able to demonstrate the applicability of the developed
method for functionally graded materials, we provide additional
comparisons by considering the static surface crack problem stud-
ied by Walters et al. [1]. The problem geometry is shown in Fig. 5.
The figure depicts a semi-elliptical surface crack located in an iso-
tropic FGM plate, that is under the effect of remote normal stress r
at the ends y = ±L. The modulus of elasticity of the plate is repre-
sented by an exponential function, and given by
EðxÞ ¼ E
1
expðbxÞ; ð8Þ
where b is a nonhomogeneity constant. Poisson’s ratio is assumed
to be constant and equal to 0.25. L and B are taken as sufficiently
large by Walters et al. [1]; and as a result the stress field around
the crack front is not affected by the boundaries at y = ±L and those
at z = ±B. Table 3 tabulates the normalized mode I stress intensity
factors computed by using the displacement correlation technique
developed in the present study and those given by Walters et al.
[1], which are evaluated by means of a three dimensional domain
integral method. The normalized mode I SIF is defined by
K
In
¼
K
I
r
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
pa
1þ1:464ðc=aÞ
1:65
: ð9Þ
The results obtained by the developed displacement correlation
technique are in very good agreement with those provided by
Walters et al. [1], which is indicative of the high level of accuracy
achieved through the application of this technique. Percent differ-
ence at the point where the crack front intersects the free surface,
i.e. at / = 0, is relatively larger compared to the percent differences
computed at the other points. The degree of singularity at this
point is different from 0.5 and depends on the Poisson’s ratio
[12]. Both in the article by Walters et al. [1] and in the present
study, the special singular behaviour at the free surface is not con-
sidered, and the results are obtained by assuming that there is a
square-root singularity at all points around the crack front includ-
ing the point of intersection at the free surface. The change in the
singularity at this single point could be the main reason behind the
difference between the results computed for / = 0. However,
although the percent difference is relatively larger at this point,
it is less than 3% and the agreement is still acceptable. The finite
element mesh used in the calculation of the results tabulated in Ta-
ble 3 is shown in Fig. 6.
Next, we consider fatigue crack propagation in a homogeneous
medium subjected to cyclic loading for further verification. The ini-
tial geometry of the crack is as depicted in Fig. 1; and the medium
is assumed to be subjected to a cyclically varying stress with a con-
stant stress range given as Dr = 100 MPa and a constant mean
stress of 50 MPa. The material is homogeneous Ti–6Al–4V. The ini-
tial minor and major radii of the elliptical crack are taken as 0.5 cm
and 1.5 cm, respectively. Fatigue growth behaviour of this crack is
examined by using two different methods, the first of which is the
method elucidated in the previous section. The second method is
based on the assumption that the initially-elliptical crack retains
Table 2
Properties of Ti–6Al–4V [3,11].
Modulus of
elasticity (GPa)
Poisson’s
ratio
Paris–Erdogan law
coefficient (C)
Paris–Erdogan law
exponent (n)
105.8 0.298 5.2 Â 10
À12
3.17
Fig. 5. (a) Surface cracked FGM plate problem considered by Walters et al. [1] and
(b) plan view of the semi-elliptical surface crack.
Table 3
Comparisons of the normalized mode I stress intensity factors computed by the
displacement correlation technique to those given by Walters et al. [1]. a/h = 0.5, a/
c = 2, E(h)/E
1
= 5.
2Ø/p Walters et al. [1] DCT % Difference
1 (major axis) 0.580 0.587 1.21
0.875 0.595 0.600 0.84
0.75 0.629 0.631 0.32
0.5 0.679 0.686 1.03
0.375 0.685 0.693 1.17
0.25 0.677 0.684 1.03
0.125 0.656 0.664 1.22
0 (minor axis) 0.596 0.613 2.85
Fig. 6. Finite element model used in the solution of the semi-elliptical crack
problem considered by Walters et al. [1].
4 B. Sabuncuoglu et al. / Computational Materials Science xxx (2011) xxx–xxx
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010
its elliptical form during propagation. Hence, in the second method
it is possible to generate the crack front profile at each step by con-
sidering only two points on the crack border. In our calculations
pertaining to this second method, we used the points on the minor
and major axes to generate the crack front data. In fact, it is previ-
ously shown that, for homogeneous materials it is sufficient to use
just two points along the crack front for accurate prediction of the
crack shape and the fatigue life [13]. The crack front profiles gener-
ated by these two different approaches are shown in Fig. 7. The cir-
cular symbols and the full lines in this figure represent the crack
front profiles formed by using the fatigue crack growth analysis
procedure described in Section 2, whereas the dashed lines are
for the results obtained through the numerical procedure based
on the invariant elliptical crack front assumption. The number of
cycles corresponding to each crack front profile is provided in Table
4. As can be seen from Fig. 7, the results obtained using the two dif-
ferent approaches are in very good agreement, which is a valida-
tion of the proposed finite elements based crack growth analysis
algorithm.
4. Fatigue crack propagation in functionally graded materials
The numerical results regarding fatigue crack propagation in
FGMs are generated for the initially-elliptical crack depicted in
Fig. 1. Two different types of gradation profiles are considered in
the analyses. In the first case, material properties are assumed to
vary in the direction of the minor axis as shown in Fig. 8a, whereas
in the second case the variation is along the major axis as illustrated
in Fig. 8b. In both examples, the total length His taken as 15 cm, and
the initial major and minor radii are set as 1.5 cm and 0.5 cm,
respectively. The applied stress range is 100 MPa and the mean
stress is 50 MPa. The medium is 100% metal at x = ÀH/3 and 100%
ceramic at x = 2H/3. The particular metallic and ceramic compo-
nents used are Ti–6Al–4V and zirconia (ZrO
2
). Material properties
of Ti–6Al–4V are given Table 1 and the properties of zirconia are
tabulated in Table 5.
The spatial variation of each of the material properties is repre-
sented by the following relation:
AðxÞ ¼ A
cr
exp
2
3
À
x
H

ln
A
m
A
cr

; ð10Þ
where A symbolizes any of the material parameters E, m, C, and n; and
the subscripts cr and m refer to the 100% ceramic and 100% metallic
material properties, respectively. Notice that A(ÀH/3) = A
m
and
A(2H/3) = A
cr
.
Three different types of finite elements are used to simulate fa-
tigue crack propagation in functionally graded materials. The vol-
ume around the crack border is modelled by utilizing collapsed
20-noded isoparametric brick elements, whereas the domain
excluding this volume is discretized by means of brick and tetrahe-
dral elements. Note that remodelling is required to compute stress
intensity factors, for the crack front morphology changes under re-
peated loading. The number of finite elements used at a given step
in the finite element computations is different from those used at
the other steps. Extensively refined finite element meshes are used
throughout the computations so as to assure the generation of con-
vergent results. For example, in the initial configuration of the prob-
lem in which the direction of elastic gradation coincides with the
minor axis, a total of 71,514 finite elements are used; while for
FGMs with elastic gradation in the direction of the major axis total
initial element number is 75,635. For each case, 960 collapsed sin-
gular elements are utilized in the modelling of the crack front re-
gion. The element number is determined through a mesh
refinement study. After successive mesh refinements, the mesh
resulting in convergent results is chosen to compute the stress
intensity factors. Since the examined medium is functionally
graded, the material properties vary from element to element. In
our analyses, we followed the homogeneous finite element method,
and computed the material properties of a given finite element at
its centroid. The approach of utilizing homogeneous finite elements
leads to the generation of highly accurate results so long as there is
sufficient mesh refinement in the model. The validation study we
describe in Section 3 indicates that, highly accurate fracture analy-
ses can be conducted through the use of the homogeneous finite
element method employed in the present study. Fig. 7. Crack front profiles for an elliptical crack in a homogeneous medium.
Table 4
Cycle numbers corresponding to the crack front profiles.
Profile Number of cycles (N)
1 178,943
2 307,326
3 409,383
4 494,869
Fig. 8. (a) Material property gradation in the direction of the minor axis and (b)
material property gradation in the direction of the major axis.
B. Sabuncuoglu et al. / Computational Materials Science xxx (2011) xxx–xxx 5
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010
Results illustrating the crack front profiles in FGMs subjected to
cycling loading are presented in Figs. 9 and 10. The numbers of cy-
cles corresponding to the crack front profiles are provided in Tables
6 and 7. From Figs. 9 and 10, it can be observed that the elliptical
shape of the crack is distorted in functionally graded materials due
to the existence of the material property gradation. In both cases,
the crack grows more rapidly towards the 100% ceramic surface.
This is the expected result since the exponent of the Paris–Erdogan
law for ceramic materials is much larger compared to that for
metallic materials. Due to the brittle nature of ceramics, crack
growth in these materials occurs at much larger rates. Moreover,
crack growth in the direction orthogonal to the direction of grada-
tion is seen to be significant when the orthogonal direction coin-
cides with the minor axis and suppressed when the orthogonal
direction coincides with the major axis. The results given in Figs.
9 and 10 and Tables 6 and 7 indicate that the method presented
in this article could prove useful in keeping the track of crack front
morphology and computing the fatigue lives of functionally graded
structures.
5. Concluding remarks
The main objective in the present study is to develop a compu-
tational method capable of generating useful data regarding fati-
gue crack propagation behaviour of functionally graded
materials. For this purpose, we considered the general configura-
tion of an initially-elliptical crack in an elastic functionally graded
medium. The crack and the graded medium are modelled by means
of a three dimensional finite element approach. The square-root
singular behaviour of the strain field around the crack front is inte-
grated into the finite element model by employing singular finite
elements. The smooth spatial variations in the material properties
of FGMs are taken into account by specifying material properties at
the centroid of each finite element. Mode I stress intensity factors
around the crack front are computed through the application of the
displacement correlation technique. Fatigue crack propagation
algorithm is based on the Paris–Erdogan law [7] and can be consid-
ered as an extension of the technique proposed by Joseph and
Erdogan [8]. The comparisons of SIFs given in Section 3 display
the high level of accuracy achieved in the applications of the three
dimensional displacement correlation technique for both homoge-
neous and functionally graded materials. Moreover, results on fati-
gue crack growth provided in Section 3 illustrate that, the elliptical
crack front form, that occurs during propagation in homogeneous
materials, can be captured quite accurately by means of the pro-
posed computational algorithm.
In Section 4, we present results pertaining to fatigue crack
growth in functionally graded materials, which demonstrate the
distortion of the elliptical shape of the crack due to the influence
of the material property gradation. Under cyclic loading, the crack
is seen to grow much more rapidly towards the 100% ceramic
plane; and crack growth increments are generally relatively small
towards the 100% metallic plane. Furthermore, crack growth rates
in the direction orthogonal to the direction of gradation are larger
when the orthogonal direction coincides with the minor axis than
when the orthogonal direction coincides with the major axis. It is
concluded that the computational method proposed in this study
is an effective way of generating crack front profiles and computing
fatigue lives of functionally graded structures. This method is
deemed to be among the first in the literature dealing with the
phenomenon of fatigue in functionally graded materials. It can fur-
ther be extended to study the propagation of mixed-mode cracks
in FGMs. This requires the computation of the modes I–III stress
intensity factors as well as the utilization of a relation that corre-
lates crack growth rates with mixed-mode stress intensity factor
ranges.
References
[1] M.C. Walters, G.H. Paulino, R.H. Dodds Jr., International Journal of Solids and
Structures 41 (2004) 1081–1118.
[2] O. Inan, S. Dag, F. Erdogan, Materials Science Forum 492–493 (2005) 373–378.
Table 5
Properties of ZrO
2
[3,14].
Modulus of
elasticity (GPa)
Poisson’s
ratio
Paris–Erdogan law
coefficient (C)
Paris–Erdogan law
exponent (n)
116.4 0.333 2.7 Â 10
À16
19
Fig. 9. Crack front profiles generated by considering property variations along the
minor axis.
Fig. 10. Crack front profiles generated by considering property variations along the
major axis.
Table 6
Numbers of cycles corresponding to the crack front
profiles shown in Fig. 9.
Crack profile Number of cycles
1 5359
2 8529
3 10,472
4 11,863
Table 7
Numbers of cycles corresponding to the crack front
profiles shown in Fig. 10.
Crack profile Number of cycles
1 5349
2 8512
3 10,442
4 11,768
6 B. Sabuncuoglu et al. / Computational Materials Science xxx (2011) xxx–xxx
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010
[3] B. Yildirim, S. Dag, F. Erdogan, International Journal of Fracture 132 (2005)
369–395.
[4] A.O. Ayhan, International Journal of Solids and Structures 46 (2009) 796–
810.
[5] H. Yu, L. Wu, L. Guo, H. Wu, S. Du, International Journal of Solids and Structures
47 (2010) 2178–2189.
[6] Ch. Zhang, M. Cui, J. Wang, X.W. Gao, J. Sladek, V. Sladek, Engineering Fracture
Mechanics 78 (2011) 585–604.
[7] P. Paris, F. Erdogan, ASME Journal of Basic Engineering 85 (1963) 528–534.
[8] P.F. Joseph, F. Erdogan, International Journal of Fracture 41 (1989) 105–131.
[9] ANSYS, ANSYS Basic Analysis Procedures Guide, Release 5.4, ANSYS Inc.,
Canonsburg PA, USA, 1997.
[10] G.R. Irwin, ASME Journal of Applied Mechanics 29 (1962) 651–654.
[11] R.O. Ritchie, D.L. Davidson, B.L. Boyce, J.P. Campbell, O. Roder, Fatigue and
Fracture of Engineering Materials and Structures 22 (1999) 621–631.
[12] A.O. Ayhan, H.F. Nied, International Journal for Numerical Methods in
Engineering 54 (2002) 899–921.
[13] J.M. Martinez-Esnaola, A. Martin-Meizoso, International Journal of Fracture
109 (2001) L17–L22.
[14] J. Acala, M. Anglada, Materials Science and Engineering A 232 (1997) 103–109.
B. Sabuncuoglu et al. / Computational Materials Science xxx (2011) xxx–xxx 7
Please cite this article in press as: B. Sabuncuoglu et al., Comput. Mater. Sci. (2011), doi:10.1016/j.commatsci.2011.06.010

The origin of a Cartesian coordinate system is also located at point P. normal (n). the material is assumed to possess property gradation in x-direction. / Computational Materials Science xxx (2011) xxx–xxx used to predict the fatigue lives of functionally graded components. we provide the details of the three dimensional computational analysis methodology developed to examine fatigue crack growth in functionally graded materials. The major and minor radii of the elliptical crack are denoted by c and a. we use the displacement correlation technique in the evaluation of the SIFs.06.commatsci. The normal plane is formed by n and b Fig.010 . Fig. the mode I stress intensity factor distribution around the crack front can be evaluated by using one of the available computational methods. 4 depicts the deformed shape of the crack in the normal plane passing through point P. The geometry of the particular problem considered is depicted in Fig. respectively.2011. 1. however they vary from element to element according to the position of the centroid. If the displacement field for the cracked functionally graded medium is known. Mater. 1 is constructed by using the general purpose finite element analysis software ANSYS [9]. The dimensions were taken such that the bounding planes would have no effect on the behaviour of the elliptical crack. a point P lying on the border of the crack. Since. The computational procedure In this section. Please cite this article in press as: B. The properties are uniform over each element. The finite element model for the problem depicted in Fig. 2 shows the quarter model generated and close-up view of the region around the crack front. Sci. Fig.3]. Smooth spatial variations of the material properties of the graded medium are taken into account in the finite element model by computing the properties of each finite element at its centroid.1016/j. and binormal (b) directions.2 B. Sabuncuoglu et al. (a) An elliptical crack in an infinite volume and (b) a point P on the elliptical crack front. 1. doi:10. In this technique. 3 shows such Fig. the expression of the stress intensity factor is derived by considering a point P on the crack front. (2011). In order to simulate the square-root singularity of the strain field near the crack border. This is the so called homogeneous finite element approach and known to lead to highly accurate numerical results so long as there is appropriate mesh refinement in the finite element model [2. This coordinate system comprises tangential (t). Presented numerical results illustrate the influence of material property gradation on the fatigue propagation characteristics of cracks located in cyclically loaded ceramic–metal FGMs. The crack is loaded through the normal stress r applied over the remote bounding planes. An elliptical crack resides in an elastic functionally graded medium. the volume around this border is modelled by utilizing collapsed 20-noded isoparametric brick elements. one-quarter of the medium is modelled by considering symmetries about xyand xz-planes.. Comput. 2. (a) The quarter model and (b) the crack front region. whose material properties are functions of the x-coordinate. 2. In the present study. whereas the domain excluding this volume is discretized by means of brick and tetrahedral elements. Sabuncuoglu et al. Fig.

2Ø/p 1 (minor axis) 0.5523 0. Sci. Under mode I loading conditions. and C and n are material parameters. EP and mP are the modulus of elasticity and the Poisson’s ratio computed at point P. The procedure is then repeated to form the crack front profile corresponding to the next specified increment. it is seen that the normalized mode I stress intensity factors computed by the DCT are in excellent agreement with those obtained through Irwin’s for- Table 1 Comparisons of the normalized stress intensity factors computed by the displacement correlation technique to those obtained by the formula given by Irwin [10].5184 DCT 0. 1 are based on the mathematical models proposed by Paris and Erdogan [7]. 3.6293 0.7075 0. The fatigue crack growth calculations for the configuration shown in Fig.05 0. We first focus on this problem solved by Irwin [10] and provide comparisons of the mode I SIFs computed by the DCT developed in the present study to those obtained through Irwin’s formula.10 0. Then. 3. doi:10.6270 0. we present the results generated for these validation problems. which is expressed as: Examining the results given in Table 1.8977 0. (2011).375 0. the medium shown in Fig. the increment for a particular point on the Please cite this article in press as: B.B. it is possible to express the mode I stress intensity factor in the following form: # 3=2 3=2 R3 ub2 À R2 ub3 pffiffiffiffiffipffiffiffiffiffi : KI ¼ R2 R3 ðR3 À R2 Þ 4ð1 À ðmP Þ2 Þ pffiffiffiffiffiffiffi P 2pE " We validated the methods described in the previous section on problems involving stationary and propagating cracks.5180 % Difference 0. A sufficiently small value needs to be used for Dq0 to be able to obtain accurate results through Eqs. the validation study is conducted by examining fatigue growth in a homogeneous medium. The nodes 1–3 in Fig.8979 0. and Joseph and Erdogan [8]. Sabuncuoglu et al.8898 0. / Computational Materials Science xxx (2011) xxx–xxx 3 crack front is specified as Dq0. In this derivation.8675 0. the asymptotic displacements of these nodes in b-direction can be found from [3] ub ðnÞ ¼ 4ð1 À ðmP Þ2 Þ EP rffiffiffiffiffiffiffi jnj KI. The minor and major radii of the crack are respectively set as 0. and the second is that of a surface crack in a functionally graded plate. The mode I SIF at P is defined as [2. 4 are considered to be at the edge of a collapsed 20-noded brick element. 4.11] and provided in Table 2.02 0. Comput. Verification of the developed procedure Fig.125 0 (major axis) Irwin [10] 0.5 cm and 1. (3) are the distances of nodes 2 and 3 with respect to the origin. and the superscript P implies that the corresponding material property should be calculated at point P. axes. (5) into Eq.37 0.8]. 2p ð1Þ where KI is the mode I stress intensity factor. (1) and (2) and the displacements of the nodes 2 and 3.. These comparisons are given in Table 1. [3. As for the propagating cracks.04 0.3] ) ub ðjnjÞ KI ¼ lim pffiffiffiffiffiffi : jnj 4ð1 À ðmP Þ2 Þ jnj!0 pffiffiffiffiffiffiffi P 2pE ( ð2Þ Then. For the stationary cracks.25 0. first. respectively. C i DK Iii Dq0 . 1 is assumed to be homogeneous and made of Ti–6Al–4V. The normalized mode I stress intensity factor provided in Table 1 at various locations around the crack front is defined as ð3Þ KI K In ¼ pffiffiffiffiffiffi : r pa ð7Þ R2 and R3 in Eq. 1.commatsci. Suppose that the stress applied to the graded medium r varies cyclically in time with a constant stress range Dr and a constant mean stress. In the calculation of these results. Irwin [10] derived a simple closed form expression for the mode I stress intensity factor.010 .31 0. using Eqs.2011. In the finite element implementation. By considering the elliptical crack problem shown in Fig. the growth rate at a point on the crack front can be calculated using Paris–Erdogan law [7.5540 0.7752 0. I dN ð4Þ where q is the radius of curvature at the considered point on the crack front.7068 0.8671 0.03 0.8902 0. (5) and (6).7755 0. (4): Dqi ffi Fig. C 0 DK n 0 I0 n ð6Þ where the subscript i stands for a generic point on the crack front. N is the number of cycles.1016/j. In what follows below.875 0. Mater.5 0. DKI is the stress intensity factor range. The number of cycles corresponding to this increment is computed using DN 0 ffi Dq0 : C 0 DK n0 I0 ð5Þ The increments at other points around the crack front are then calculated by substituting Eq. the new crack front profile is generated by connecting the new locations by splines. The material properties are taken from Refs. Sabuncuoglu et al. the first of which is the problem of an elliptical crack in a homogeneous medium.06.09 dq ¼ C DK n . the application of the displacement correlation technique is verified by considering two different problems. Crack front and the local coordinate system at point P. After calculating the crack growth increments for a specified number of points on the crack front. Deformed shape of the crack surface in the normal plane.75 0. the material was assumed to be homogeneous.5 cm.

693 0. although the percent difference is relatively larger at this point.5 and depends on the Poisson’s ratio [12].298 B. ð8Þ where b is a nonhomogeneity constant. and as a result the stress field around the crack front is not affected by the boundaries at y = ±L and those at z = ±B.595 0. 5..2 Â 10À12 Paris–Erdogan law exponent (n) 3. Finite element model used in the solution of the semi-elliptical crack problem considered by Walters et al. that is under the effect of remote normal stress r at the ends y = ±L. Percent difference at the point where the crack front intersects the free surface.631 0.629 0. 6. 6.875 0. which is indicative of the high level of accuracy achieved through the application of this technique. The modulus of elasticity of the plate is represented by an exponential function. Poisson’s ratio is assumed to be constant and equal to 0. The figure depicts a semi-elliptical surface crack located in an isotropic FGM plate. 2Ø/p 1 (major axis) 0. and the medium is assumed to be subjected to a cyclically varying stress with a constant stress range given as Dr = 100 MPa and a constant mean stress of 50 MPa. is relatively larger compared to the percent differences computed at the other points. the first of which is the method elucidated in the previous section. [1].84 0. [1]. The initial minor and major radii of the elliptical crack are taken as 0.5 0. doi:10.580 0. [1]. E(h)/E1 = 5. it is less than 3% and the agreement is still acceptable.684 0. Please cite this article in press as: B. The problem geometry is shown in Fig. The change in the singularity at this single point could be the main reason behind the difference between the results computed for / = 0. [1] and in the present study. (2011). Comput.010 . Fig.75 0. at / = 0.679 0. The normalized mode I SIF is defined by The results obtained by the developed displacement correlation technique are in very good agreement with those provided by Walters et al. To be able to demonstrate the applicability of the developed method for functionally graded materials. 1.03 1.22 2. The initial geometry of the crack is as depicted in Fig.32 1.17 mula.613 % Difference 1. the special singular behaviour at the free surface is not considered. [1] 0.677 0. and given by KI K In ¼ qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi : r pa 1þ1:464ðc=aÞ1:65 ð9Þ EðxÞ ¼ E1 expðbxÞ.686 0. 5. Sabuncuoglu et al. Both in the article by Walters et al. Mater. i. respectively. (a) Surface cracked FGM plate problem considered by Walters et al.587 0. Sabuncuoglu et al.4 Table 2 Properties of Ti–6Al–4V [3.e. Fatigue growth behaviour of this crack is examined by using two different methods.25.685 0. However. [1]. Table 3 tabulates the normalized mode I stress intensity factors computed by using the displacement correlation technique developed in the present study and those given by Walters et al. Modulus of elasticity (GPa) 105. / Computational Materials Science xxx (2011) xxx–xxx Table 3 Comparisons of the normalized mode I stress intensity factors computed by the displacement correlation technique to those given by Walters et al. The finite element mesh used in the calculation of the results tabulated in Table 3 is shown in Fig.2011. which are evaluated by means of a three dimensional domain integral method.25 0.664 0. The degree of singularity at this point is different from 0.600 0.596 DCT 0. and the results are obtained by assuming that there is a square-root singularity at all points around the crack front including the point of intersection at the free surface. we provide additional comparisons by considering the static surface crack problem studied by Walters et al.5 cm.21 0. it is possible to compute the mode I stress intensity factors to within a high degree of accuracy by employing the displacement correlation technique developed in the present study.17 1.656 0.125 0 (minor axis) Walters et al. [1].5.8 Poisson’s ratio 0.11].5 cm and 1.commatsci. we consider fatigue crack propagation in a homogeneous medium subjected to cyclic loading for further verification. Sci.1016/j. Next.03 1. The material is homogeneous Ti–6Al–4V. [1] and (b) plan view of the semi-elliptical surface crack. Hence. a/h = 0.375 0. L and B are taken as sufficiently large by Walters et al.85 Paris–Erdogan law coefficient (C) 5.06. a/ c = 2. [1]. The second method is based on the assumption that the initially-elliptical crack retains Fig.

The validation study we describe in Section 3 indicates that. which is a validation of the proposed finite elements based crack growth analysis algorithm. 7. In the first case. In both examples. The crack front profiles generated by these two different approaches are shown in Fig. ð10Þ dral elements. Mater. Please cite this article in press as: B. material properties are assumed to vary in the direction of the minor axis as shown in Fig. whereas the domain excluding this volume is discretized by means of brick and tetrahe- Fig. and computed the material properties of a given finite element at its centroid.326 409.010 . 8a. Sci. Two different types of gradation profiles are considered in the analyses. Extensively refined finite element meshes are used throughout the computations so as to assure the generation of convergent results. Notice that A(ÀH/3) = Am and A(2H/3) = Acr. the results obtained using the two different approaches are in very good agreement. we used the points on the minor and major axes to generate the crack front data. In our analyses. the mesh resulting in convergent results is chosen to compute the stress intensity factors. 8.635. C. where A symbolizes any of the material parameters E.1016/j. for homogeneous materials it is sufficient to use just two points along the crack front for accurate prediction of the crack shape and the fatigue life [13]. the material properties vary from element to element. Fatigue crack propagation in functionally graded materials The numerical results regarding fatigue crack propagation in FGMs are generated for the initially-elliptical crack depicted in Fig. 4.383 494.06. Material properties of Ti–6Al–4V are given Table 1 and the properties of zirconia are tabulated in Table 5. The volume around the crack border is modelled by utilizing collapsed 20-noded isoparametric brick elements. For each case. Profile 1 2 3 4 Number of cycles (N) 178. Note that remodelling is required to compute stress intensity factors.commatsci. In fact. / Computational Materials Science xxx (2011) xxx–xxx 5 its elliptical form during propagation. The number of cycles corresponding to each crack front profile is provided in Table 4. a total of 71. Hence. we followed the homogeneous finite element method.2011. the total length H is taken as 15 cm. À 3 H Acr   Fig.5 cm and 0. The number of finite elements used at a given step in the finite element computations is different from those used at the other steps. In our calculations pertaining to this second method. 1. respectively. while for FGMs with elastic gradation in the direction of the major axis total initial element number is 75.. doi:10. 7. After successive mesh refinements. whereas in the second case the variation is along the major axis as illustrated in Fig. Since the examined medium is functionally graded. 7. The particular metallic and ceramic components used are Ti–6Al–4V and zirconia (ZrO2). Crack front profiles for an elliptical crack in a homogeneous medium. Sabuncuoglu et al. for the crack front morphology changes under repeated loading. Comput. respectively.869  AðxÞ ¼ Acr exp  2 x Am ln . For example. it is previously shown that. The circular symbols and the full lines in this figure represent the crack front profiles formed by using the fatigue crack growth analysis procedure described in Section 2. 960 collapsed singular elements are utilized in the modelling of the crack front region.5 cm. (2011).514 finite elements are used. and n. whereas the dashed lines are for the results obtained through the numerical procedure based on the invariant elliptical crack front assumption. m. (a) Material property gradation in the direction of the minor axis and (b) material property gradation in the direction of the major axis. Three different types of finite elements are used to simulate fatigue crack propagation in functionally graded materials. As can be seen from Fig. The applied stress range is 100 MPa and the mean stress is 50 MPa. and the subscripts cr and m refer to the 100% ceramic and 100% metallic material properties. Sabuncuoglu et al. and the initial major and minor radii are set as 1.B. 8b. in the second method it is possible to generate the crack front profile at each step by considering only two points on the crack border. The medium is 100% metal at x = ÀH/3 and 100% ceramic at x = 2H/3.943 307. The approach of utilizing homogeneous finite elements leads to the generation of highly accurate results so long as there is sufficient mesh refinement in the model. The spatial variation of each of the material properties is represented by the following relation: Table 4 Cycle numbers corresponding to the crack front profiles. in the initial configuration of the problem in which the direction of elastic gradation coincides with the minor axis. The element number is determined through a mesh refinement study. highly accurate fracture analyses can be conducted through the use of the homogeneous finite element method employed in the present study.

can be captured quite accurately by means of the proposed computational algorithm. 5.768 neous and functionally graded materials. Under cyclic loading. [1] M. Crack profile 1 2 3 4 Number of cycles 5359 8529 10. 9 and 10.06. Paulino. International Journal of Solids and Structures 41 (2004) 1081–1118. (2011). Please cite this article in press as: B. which demonstrate the distortion of the elliptical shape of the crack due to the influence of the material property gradation. The crack and the graded medium are modelled by means of a three dimensional finite element approach. Materials Science Forum 492–493 (2005) 373–378. we considered the general configuration of an initially-elliptical crack in an elastic functionally graded medium. 9. The numbers of cycles corresponding to the crack front profiles are provided in Tables 6 and 7...1016/j. Crack front profiles generated by considering property variations along the major axis.2011. Moreover. The comparisons of SIFs given in Section 3 display the high level of accuracy achieved in the applications of the three dimensional displacement correlation technique for both homoge- Fig. Furthermore. This requires the computation of the modes I–III stress intensity factors as well as the utilization of a relation that correlates crack growth rates with mixed-mode stress intensity factor ranges. The smooth spatial variations in the material properties of FGMs are taken into account by specifying material properties at the centroid of each finite element. Sci. It can further be extended to study the propagation of mixed-mode cracks in FGMs.472 11. crack growth in these materials occurs at much larger rates. Dag.442 11. This method is deemed to be among the first in the literature dealing with the phenomenon of fatigue in functionally graded materials. The results given in Figs.4 Poisson’s ratio 0. Crack front profiles generated by considering property variations along the minor axis. we present results pertaining to fatigue crack growth in functionally graded materials. 9 and 10 and Tables 6 and 7 indicate that the method presented in this article could prove useful in keeping the track of crack front morphology and computing the fatigue lives of functionally graded structures. Sabuncuoglu et al. Moreover. the crack grows more rapidly towards the 100% ceramic surface.H. and crack growth increments are generally relatively small towards the 100% metallic plane. it can be observed that the elliptical shape of the crack is distorted in functionally graded materials due to the existence of the material property gradation. Modulus of elasticity (GPa) 116. that occurs during propagation in homogeneous materials. G. From Figs. Erdogan. / Computational Materials Science xxx (2011) xxx–xxx Paris–Erdogan law coefficient (C) 2. Due to the brittle nature of ceramics. crack growth rates in the direction orthogonal to the direction of gradation are larger when the orthogonal direction coincides with the minor axis than when the orthogonal direction coincides with the major axis. crack growth in the direction orthogonal to the direction of gradation is seen to be significant when the orthogonal direction coincides with the minor axis and suppressed when the orthogonal direction coincides with the major axis. S.14]. Mode I stress intensity factors around the crack front are computed through the application of the displacement correlation technique.6 Table 5 Properties of ZrO2 [3. [2] O.333 B. In Section 4. It is concluded that the computational method proposed in this study is an effective way of generating crack front profiles and computing fatigue lives of functionally graded structures. R. Inan.010 . Table 6 Numbers of cycles corresponding to the crack front profiles shown in Fig. Comput. F. Sabuncuoglu et al.863 Table 7 Numbers of cycles corresponding to the crack front profiles shown in Fig. doi:10. results on fatigue crack growth provided in Section 3 illustrate that. 10. Walters. For this purpose.commatsci. This is the expected result since the exponent of the Paris–Erdogan law for ceramic materials is much larger compared to that for metallic materials. Fatigue crack propagation algorithm is based on the Paris–Erdogan law [7] and can be considered as an extension of the technique proposed by Joseph and Erdogan [8]. In both cases. The square-root singular behaviour of the strain field around the crack front is integrated into the finite element model by employing singular finite elements. Crack profile 1 2 3 4 Number of cycles 5349 8512 10. 9. Dodds Jr. 9 and 10.C. the elliptical crack front form. Concluding remarks The main objective in the present study is to develop a computational method capable of generating useful data regarding fatigue crack propagation behaviour of functionally graded materials. References Fig. 10. the crack is seen to grow much more rapidly towards the 100% ceramic plane.H.7 Â 10À16 Paris–Erdogan law exponent (n) 19 Results illustrating the crack front profiles in FGMs subjected to cycling loading are presented in Figs. Mater.

X. J. L. Du.B. Yu.P.commatsci. Yildirim. H. A. Sci.010 . A. (2011). Ayhan.. S. J.L. Canonsburg PA. International Journal of Solids and Structures 47 (2010) 2178–2189. D. Zhang. Joseph. Cui. International Journal of Fracture 132 (2005) 369–395.1016/j. Guo. J. 1997. F. Campbell.. S. Sladek.O. ANSYS Basic Analysis Procedures Guide. Ritchie. Engineering Fracture Mechanics 78 (2011) 585–604. M.M. M.W. Please cite this article in press as: B. Sabuncuoglu et al. Wu. Erdogan. Release 5. [5] H.06. Comput.L. H. G. Mater.4. F. Boyce. Martin-Meizoso. Fatigue and Fracture of Engineering Materials and Structures 22 (1999) 621–631. Wu.F. [6] Ch. F. ASME Journal of Basic Engineering 85 (1963) 528–534. International Journal of Fracture 109 (2001) L17–L22. Nied. International Journal of Solids and Structures 46 (2009) 796– 810. Paris. doi:10. ASME Journal of Applied Mechanics 29 (1962) 651–654. Materials Science and Engineering A 232 (1997) 103–109. / Computational Materials Science xxx (2011) xxx–xxx [3] B. [9] [10] [11] [12] [13] [14] 7 ANSYS.O. Erdogan. Erdogan. Sabuncuoglu et al. O. B. V.2011. J. Wang. R. Gao. J. Davidson.R.F. Roder. [8] P. Dag. Ayhan.O. USA. Sladek. ANSYS Inc. Anglada. Martinez-Esnaola. L. [7] P. International Journal for Numerical Methods in Engineering 54 (2002) 899–921. International Journal of Fracture 41 (1989) 105–131. Irwin. [4] A. Acala.