You are on page 1of 6

Transactions of JSCES, Paper No.

19990024

Development of an Automatic Quadrilateral Mesh Generator


for the Simulation of Curved Crack Growth
Yasumi KAWAMURA1), Yang MU1) and Yoichi SUMI1)
1) Department of Naval Architecture and Ocean Engineering, Yokohama National University
(79-5 Tokiwadai Hodogaya-ku, Yokohama 240-8501, Japan)

A step-by-step finite element approach has been developed for the simulation of curved crack growth,
where a cracked domain is successively re-meshed by an automatic mesh generator. It is well known that
the pre-processing is a cumbersome procedure in such a simulation, so that an efficient pre-processor has
been developed by using an automatic quadrilateral mesh generator based on the paving method. In order
to obtain a favorable mesh density distribution, improvements of the algorithms and implementation of a
new conversion technique for quadrilateral elements are made for the paving method. An automatic
boundary nodes generator with an embedded template-mesh around a crack tip is developed for the preprocessor of a crack analysis, which is applied to problems with single and multiple-crack tip. The
accuracy and the effectiveness of the crack analysis are confirmed by a numerical example. Since the path
of a curved crack may sometimes be influenced by the selection of the increment size of a crack, an
appropriate numerical procedure including re-analysis is proposed. The validity of the method is then
demonstrated by the simulation of wavy crack growth of thermally induced brittle cracking.
Key Words: Curved crack analysis, Crack path prediction, Pre-processor, Mesh generation, Paving
method

1.

INTRODUCTION

One of the authors, Sumi and his associates have


developed a simulation method of crack growth including the
crack path prediction (Sumi(1); Sumi et al.(2,3), Sumi(4)), where
the stress parameters such as stress intensity factors are
computed at a pre-existing crack tip. The incremental
curvilinear profiles of crack growth are determined by using
the first order perturbation method with the local symmetry
condition (KII=0). Having extended the crack tip to an
estimated point, the above mentioned process of the finite
element analysis is repeated with the re-generation of the mesh,
where one of the difficulties in this method is the preprocessing including mesh generation. In the simulation
procedure developed by Sumi et al., the modified quadtree
method has been used for the mesh generation. (2-4) Although it
may generally form regular mesh pattern, poor mesh or failure
of meshing could sometimes be experienced near the
boundary of an analysis domain. In this paper, a robust preprocessor for the simulation of curved crack growth is
developed including the quadrilateral mesh generator based on
the paving method.
In the simulation of curved crack growth, accurate crack
paths can be predicted, only if the size of an incremental crack
growth is satisfactorily small so that the local symmetry
condition is approximately satisfied along the entire trajectory
of a curved crack. It is obvious, however, that if the increment
size is too small, too many iterative finite element analyses are
required, resulting in an excessive computational time. We
shall discuss the proper selection of the size of the increments
on the curved paths of crack growth so that the crack path can
be predicted within a certain allowable limit.

2. PRE-PROCESSOR
ANALYSIS

FOR

CRACK

2.1. Prerequisites of Pre-processors for Crack Analysis

* 1999430, w199972, 
1999722, 1999 G
Manuscript received, April 30, 1999; final revision, July 2, 1999;
published, July 22, 1999. Copyright 1999 by the Japan Society for
Computational Engineering and Science.

One of the key issues for the step-by-step crack analyses is


the generation of appropriate finite element mesh for general
geometry, which may be tedious and sometimes error prone.
Obviously, accuracy and computation time directly depend on
the size, the shape and the number of elements in the mesh.
For the development of the proposed pre-processor, we shall
consider the following aspects;
i) Regular mesh pattern in the vicinity of a crack tip
In order to avoid excessive numerical error, regular mesh
patterns are embedded by using certain template-mesh
around a crack tip.
ii) Smooth transition from high to low density mesh
Unnecessary mesh-refinement is avoided for the
computational efficiency by implementing a gentle gradient
of mesh density from a crack tip to the boundary of a domain,
which confirms to minimize the expense of solution without
loss of the accuracy.
iii) Automatic mesh generation by quadrilateral elements
It is well known that in the analyses with low order
elements, triangular elements are generally less accurate and
less versatile than quadrilateral elements so that an automatic
mesh generated by the latter elements is preferable from the
view point of accuracy.
iv) Minimum input effort
Since analyses are repeatedly carried out for the simulation
of crack growth, the amount of input data should be reduced
as little as possible.
2.2. Development of the Automatic Mesh Generator
(1) Outline of the Paving Method (5)
Several methods of quadrilateral mesh generation have
recently been proposed such as the conversion method from
triangle mesh (Johnston et al.(6); Shimada and Itoh(7); Potyondy
et al.(8)), and the Triquamesh method (Sluiter and Hansen (9)).
In these methods difficulties are observed to generate wellshaped quadrilaterals near the boundary, where high solution
accuracies are normally required in structural analyses. The
paving method, which is proposed by Blacker and
Stephenson(5), generates quadrilateral elements from the
boundary so that well-shaped elements can naturally be
allocated near the boundary, which is the main reason why we
apply and further develop the method for the simulation of
crack growth in the present study.

Figure 1 shows the process of the paving mesh generation,


where the input data are only those corresponding to a set of
boundary nodes shown in Fig.1 (a). In this case, the element 1
to element 13 are generated first, and they form the first row
of the elements (Fig. 1(b)). Then, the elements from 14 to 25
form the second row of the elements (Fig.1(c)). Repeating this
procedure, boundary (which is called as the paving boundary)
progresses towards the inside of the domain, resulting in the
one filled with quadrilateral elements (Fig. 1(d)). In order to
generate a high quality mesh, several algorithms such as
smoothing, seaming, intersection handling, row adjustment,
and clean-up, are involved in the procedure.

(b) Improvement of the Smoothing Process


In the original version of the paving method, the boundary
node smoothing is applied after the element row generation. In
Fig.1(b), for example, it is applied to the nodes denoted by
open circle after the element row generation (element 1 to 13).
Considerably distorted elements, however, could be formed, if
the original elements (before smoothing) are distorted, or if
the element size is extremely different from the adjacent
elements. In order to avoid this inexpedience, the procedure of
the boundary node smoothing has been improved in the
present study.
Let vector Vi be the position vector of a target node on the
paving boundary (see Fig. 3(a)). The shifting vector for Vi due
to boundary node smoothing is defined as,
=

(a)






           

(b)




A = Vi Vi

 





(d)

Outline of the paving method

(2) Improvements of the Paving Method


The paving method based on the algorithm by Blacker and
Stephenson(5) is reliable to fill the domain with quadrilaterals
of almost the same size. However, if we directly apply the
method to crack problems, it sometimes generates ill-shaped
elements or it even does not work because of the transition of
the size of the elements in a cracked domain. The following
improvements including a new technique are implemented in
the present meshing algorithm.
(a) Termination of the Element Row Generation
In the case where stress concentration could be expected,
the boundary node data may have different intervals in order
to generate mesh with high resolution near the stress
concentrated region. If the original version of the paving
algorithm is directly applied to such problems, the row
generation of large elements sometimes progresses deep into
the inside of the paving boundary (see Fig. 2). In this example
the larger elements are generated from the bottom, while
smaller elements are formed from the top, and they intersect as
illustrated in the figure. In such cases, ill-shaped elements are
generated even if the intersection and the smoothing
procedures are applied after the generation of the element row.
To avoid this inexpedience, the algorithm is improved in such
a way that under these circumstances the element row
generation is terminated and the smoothing, intersection, and
seaming processes follow immediately.

upper paving
boundary
lower paving
boundary

Fig. 2

(2)

n
1
Vi = Vmj + Vml + Vmk
n m

           

(c)

(1)

where A and B are the modification vectors which improve


the shape of adjacent elements and modify the direction from
node Vj to Vi, respectively. In the original algorithm, A is
defined by



Fig.1

A + B
2

The case where the element row generation must be


Stopped

(3)

where n is a number of elements connected to node Vi. The


vectors Vmj , Vml and Vmk represent the positions of the node of
the m-th element. Using this smoothing procedure, all of the
elements connected to the node Vi become closer to
parallelograms in the average, and it works very well for the
elements of almost the same size without much distortion. On
the contrary, if gradation of element size exists, small
elements become ill-shaped after the application of the above
operations. In order to avoid this situation, equation (3) has
been modified in the following form;

Vi =

m =1

( Vmj + Vml + Vmk )


(4)

1
m =1 l m

lm =

Vml Vi + V j Vi

(5)

In this equation, by using the inverse of the characteristic


length (lm) of each element as a weight, the modification of the
shape of the larger elements is restrained so that the smaller
elements are prevented from excessive distortion.
Vi' A
Vi

PB

PA

V2 l
2

PA1

Vj = V1 j = V2j

V1k

(a)

Fig.3

V2k

V2 l

Vi

V1l

V1l

lQ

V1k
Vj
(b)

PC

PA2
lD
V2k

Boundary node smoothing

The vector B in equation (1) is defined in the case where


the node Vi is connected to only two elements. As illustrated
in Fig. 3(b), the vector PA is defined as the vector oriented to
bisect the angle between vector PA1 (= V1l Vj) and vector PA2
(= V2l Vj). The vector PB is oriented to bisect the angle
between PA and PC (=Vi Vj). The modification vector B is
defined by
P
(6)
B = L B P C
PB

(lQ + l D ) / 2
L=
lD

if (lQ > l D )
otherwise

(7)

where lD is the length of the vector PC and lQ is the distance


between Vj and the line that links V1l and V2l (see Fig.3(b)). In
equation (7), the projection length, L, is determined in such a
way that the paving boundary becomes smooth when it forms
a concave shape. On the other hand when its shape is convex,
B modifies the direction of the projection to avoid the
generation of narrow elements.
(c) Intersection of Line Segments with Different Length
A paving boundary may intersect the pre-existing paving
boundary, as it progresses towards the inside of a domain. In
this case, the intersection procedure is applied, where the two
intersected line segments are selected and connected.(5)
However, if the lengths of the two selected segments are
considerably different as shown in Fig. 4, the connection
sometimes causes ill-shaped elements. In this case the small
elements form ill-shaped mesh if the smaller segment is
moved to and connected to the larger segment (see Fig. 4(a)).
In the present paving method, the smaller segment is selected
as the connecting position of the two segments so that the
generation of ill-shaped mesh is avoided (Fig. 4(b)). This
improvement is also applied to the seaming procedure.

(a)
Before connection

(b)
After connection

Fig.4 Connection of two segments with different size


(d) A Conversion Technique for Quadrilaterals
Figure 5(a) shows the results of the mesh generated by
using the aforementioned algorithm for the problem with the
boundary nodes having considerably different intervals. The
intersection procedure or the seaming procedure is applied, as
the smaller and the larger elements are generated and
intersected. Many distorted elements are generated in the
transition area of the element size as shown in Fig. 5(a)
because of the connection of different size elements. In the
present study we propose a new conversion technique for the
improvement of the shape of quadrilateral elements, which can
be applied after the mesh generation.




(a)

(b)

= +
+ =
+ =
+ + = 2
6

Fig. 6 Conversion of quadrilateral elements


( from two elements (a) to three elements (b) )

A
5
D
4

Fig. 7

C
3

B
4

The case where the conversion procedure should not


be applied

First, all inner nodes which are connected more than five
elements are sorted. The following conversion technique is
then applied to these nodes in the descent order.
(Case 1) A node connected to five elements
As shown in Fig. 6 (a), two neighboring elements which are
connected to the node (node A) are tested whether the pair is
appropriate for the candidate of the conversion. If the pair
satisfies the following two conditions; i.e. the connectivity
condition and the anti-distortion condition, they are subjected
to the conversion to three elements as illustrated in Fig. 6(b).
[Condition 1] The connectivity condition is given by
n A 4 + n B 4 + nC 4 > nA 4 + nB 4 + nC 4

(8)

where nA, nB and nC are the numbers of connected elements


to nodes A, B and C, respectively before the conversion,
and nA , nB and nC are those after the conversion. This
condition is intended to make the number of connecting
elements to nodes A, B and C closer to 4. It should be noted
that even when the pair of elements satisfies this condition,
the pair is not subjected to the conversion in the case as
illustrated in Fig. 7, where the numbers of elements
connecting to the adjacent nodes B and C are 4 and 3,
respectively. Furthermore, the number of elements
connected to node D is 4. This requirement is essential to
avoid the successive propagation of the conversion.
[Condition 2] The anti-distortion condition is given by
e0 > e1

(9)

where

1
e0 =

+ 2 + 3 + 4
2
2
2
2
4

(10)

and

1 + 2
e1 =

(a)
(b)
Fig. 5 Mesh generation for the problem with gradation of
element size ( (a) generated mesh without the conversion
procedure; (b) generated mesh with the conversion
procedure)

+ 3 + 4 +
3
2
3 2
. (11)
8

In equation (10), 1, 2, 3, and 4 are the angles at nodes A,


B, C in Fig.6(a), respectively, so that e0 defines the
averaged deviation of these angles from rectangle. Also, e1
can be regarded as the mean deviation from rectangle of the
eight angles defined in Fig.6(b). Obviously, this condition is
expected to reduce the distortion of the elements.
From all candidates, which satisfy the above two
conditions, the pair which has the smallest value of e1 is
selected for the conversion.

(Case 2) A node connected to more than five elements


The pair of the elements is selected from the candidate
which satisfies the [Condition 2].
The conversion procedure is repeated until no node
satisfies the above conversion conditions. Figure 5 (b) shows
the mesh generated after applying the proposed conversion
technique. In comparing Figs. 5(a) and (b), the improvement
of the shape of the quadrilateral elements is confirmed.
2.3. Pre-Processing for the Simulation of Crack Growth
In order to apply the mesh generator to the simulation of
crack growth, we developed the following pre-processing
programs.
(1) Automatic Generation of Boundary Nodes
A set of boundary nodes must be prepared as the input data
of the paving method, where appropriate adjustment of their
intervals are essential for the favorable mesh density
distribution.
An expert of finite element analysis generally discretizes
the parts of a domain with relatively small elements, where
high stress gradient could be expected from his intuitive
knowledge based on his experience. In this paper, the special
parts of the boundary, 1,2,3,...,n, are selected as shown in
Fig.8(a) based on the geometrical features with certain
characteristic lengths such as the length of a segment, li, or a
radius of curvature, ri, of the boundary. Based on experts'
knowledge, the required density of the generated boundary
nodes corresponding to each part is defined in Table 1.
Fig.8(b) shows the generated boundary nodes, where the
curvature of the crack line is the dominant characteristic
length of the geometric features of the domain. It should be
noted that the interactive effects of adjacent special parts of
boundaries are also taken into account for the determination of
proper mesh size.
(2) Template-Mesh around a Crack Tip
In order to avoid an excessive irregular numerical error
around a crack tip, proper mesh pattern is arranged in the
vicinity of a crack tip by introducing the templates of mesh
shown in Fig.9, in which the four or 16 square elements are
embedded in this special region. Figure 10 shows the meshes
generated for a crack growth simulation, where highly wavy

2 l1

Table 1 Special parts and their required node intervals


Part Characteristic length
Node interval
1
Curvature r1
r1/12

Length of line segment l1

Concave curvature r2

Length of caved in line segment l2

Initial crack curvature r3

Crack path curvature r4

7
8

l1

r /30
2

l2/4

r /30
r /30
r /30
3
4

Minimum curvature of adjacent


increment r5
The shortest distance from crack tip
to boundary Rmin

Rmin/4

X2

X2

X1

X1

(a) Four elements


(b) 16 elements
Fig.9 Template for the crack tip

(a) Mesh for the initial geometry

(b)Mesh at a certain step

r4

r3 6
5
7
r
5
4
l2
8

r2

crack tip

3
r1
Rmin

c ra c k

(a) Definition of special parts


(c) Close-up of the mesh around the crack tip in Fig.10(b)
Fig.10 Examples of the mesh for a crack growth simulation

(b) Generated boundary nodes and the template-mesh


Fig.8 Special parts of the domain

crack path is observed. As shown in Fig.10(b), the mesh


around the wavy crack is dense compared with the mesh for
the initial straight crack in Fig.10(a). This example illustrates
the validity of the developed mesh generator and preprocessor for a complicated profile of a crack path.
(3) Paving Method for Multi-domain
In order to apply the present method to predict the
formation of growth pattern of interacting cracks in a
heterogeneous material, the whole domain is first subdivided
into M sub-domains, so that each sub-domain has only one
crack tip. (3) For different sub-domains, the material properties
may be defined separately to model heterogeneous materials.

In this paper, the pre-processor of the paving method is


developed to implement automatic mesh generation for each
sub-domain with identical boundary nodes, where boundary
nodes are doubly defined on the each interface of subdomains.

3.
NUMERICAL
DISCUSSIONS

ANALYSIS

AND

Accuracy of the Results for Problems with Multiple


Crack Tips
Figure 11 shows the result of the mesh generation for a
slightly grooved plate with a center crack. This problem has
two crack tips so that the analyzed model consists of two subdomains, 1 and 2, where the two crack tips 1 and 2 are
respectively defined in the figure. Table 2 shows the computed
stress intensity factors and the so-called T-stress for the plate
under uniform tension by the method of superposition of
analytical and finite element solutions. (10) The present results
agree well with the results given by Isida. (11) The developed
pre-processor and the mesh generator seem to work
satisfactorily for a crack analysis.

Domain 1

1.25W

3.1.

the following manner. If KII/KI exceeds the allowable limit


after a certain step, a is reduced to C a (c.f. C=2/3), and
the re-analysis of the step has been carried out.
Figure 12 shows a problem of a crack in a plate under
biaxial tension. The crack growth simulation is carried out by
the present method for the cases with the allowable limits of
KII/KI being 3%, 5%, 10% and 20%, respectively. Figure 13
shows the simulated results. In the case where the allowable
limits are 3% or 5%, the increment size of the crac k
propagation is small enough so that the crack path becomes
smooth as illustrated in the figure. On the other hand, when
the allowable limit becomes large, so is the corresponding
incremental crack extension, which results in the alternate
changes of direction of the crack path during the growth. As
shown in this example, it is possible to carry out the
simulation within a certain limit of error, whose allowable
level could be 3% to 5% in the present case.

t : Thickness

15mm

X
500mm

6mm
85mm

0.4W

1.5W
W

500mm

Fig. 12

An example problem for crack propagation analysis


0

Domain 2
1.5W
2W

Table 2 Computed results for the center cracked plate

Present results
Isidas solution (11)

P
tW

tip 1
tip 2
0.967
0.966
0.975

K II /

P
tW

tip 1
0.030

a
tip 2
-0.28

P
tW
tip 1
tip 2
-1.341 -1.339
T/

3.2. Simulation of Crack Growth


In the present study, we shall apply the pre-processing
method to the simulation of curved crack growth, in which we
have to carry out a step-by-step analysis accompanied by the
incremental extension of a crack. The prediction of a crack
path and the selection of the increment of a crack extension in
each step are crucial for the accurate simulation. In the present
study, the stress field parameters ahead of a crack tip are first
determined by the method of superposition of analytical and
finite element solutions (1~4) for the initial crack geometry, then
crack path is predicted by the local symmetry condition
(KII=0) using the first order perturbation solution formulated
by Sumi, Nemat-Nasser and Keer. (12) This process is repeated
for the final stage of a crack growth. In this method, the
increment size (a) of the crack extension is assumed
satisfactorily small compared with Rmin, which is the shortest
distance between the crack tip and the nearest boundary of the
domain (see Fig. 8). In practical calculation, a/Rmin obviously
takes a small but finite value so that numerical error could be
involved in the predicted crack paths. The local symmetry
condition is, therefore, slightly violated along the curved
trajectory.
In order to keep the accuracy of the simulation within an
allowable level, we shall introduce a re-analysis procedure in

y-coordinate(mm)

3%

Fig.11 Mesh pattern for the center cracked plate in tension

KI /

Initial Crack

-50

5%
-100

10%
20%

-150

-200

-250

50

100

150

200

250

x-coordinate(mm)

Fig. 13 Simulated crack paths for various allowable limits of


KII/KI
The method is applied to the problem of thermally induced
brittle cracks in a glass plate, whose computational model is
illustrated in Fig. 14. In this problem, crack growth is induced
in a brittle half-space by the formation of a thermally cooled
layer at the free surface with a prescribed temperature profile,
in which the thickness of the thermal layer ( ) is used as a
measure of the load parameter. The one-dimensional
temperature profile is assumed as
TS
T ( x) =
T0

x ,
x > ,

(12)

where T0 is the initial temperature, and TS < T0 is the


temperature of the uniformly cooled surface layer. The
corresponding thermal stress in an intact body is calculated as
E (T0 TS )
t =
0

x ,
x > ,

(13)

which induces the driving force to extend thermal cracks. In


equation (13), and E are the linear thermal expansion
coefficient and Youngs modulus, respectively. In the
following numerical simulation, we assume the crack spacing,
B=25mm, and the initial crack length, H0=10mm. The material
properties are selected for a typical glass plate as
= 8.5 10 6 / $ C , E = 6.9 10 4 MPa, and the fracture
toughness, K C = 15.8 Nmm 3 / 2 . Taking the temperature
difference T = T0 TS =25 $ C , the thermal crack extends
with increasing the thermal penetration depth.
The simulated crack growth shows an interesting wavy
path as illustrated in Fig.15, which is known as the Hiratas
experiment in Ref. [13]. A slightly asymmetric loading
condition may induce this wavy behavior, whose physical
aspects will be discussed in elsewhere.(14) As can been seen in
Fig.16, the critical condition, K I = K C , holds along the
curved trajectory, while the local symmetry condition is
approximately satisfied by applying the re-analysis procedure.
The total number of steps in the simulation in Fig.15 is 103,
and re-analyses of once to three times are carried out at certain
steps including step 24 as shown in Fig.16.
Y

: Thickness of Thermal Layer


t: Thermally Induced Load
H0: Initial Crack Length

H0

Fig.14 Computational model of thermally induced crack

H0

Fig.15 Predicted crack path for thermally induced crack


problem
KII/KI

KI/Kc

KI /Kc

1.1

KII/KI

0.8
0.6

1.0

0.4
0.9

0.2

0.8

0.0
Y
1.0
0.0 step 20
-1.0

0.7
0.6

-0.2

35

30
25

30.0

-0.4
X
45.0

35.0
40.0
Crack path

-0.6

0.5

-0.8
20

21

22

23

24

25

26

27

28

29

30

31

32

33

34

35

Step

Fig.16 Variation of KI /KC and KII /KI from step 20 to 35


(re-analysis is carried out at step 24)

4. CONCLUSIONS
In this study, a pre-processor for the crack propagation
analysis is developed. The automatic quadrilateral mesh
generator based on the paving method is developed with the

improvements of the original algorithm and with the


implementation of a new conversion technique. The automatic
boundary nodes generator with the use of templates around a
crack tip, is also implemented in order to generate mesh with
favorable density. The method is applied to the simulation of a
curved crack growth, where the increment size of the crack
extension is crucial for the solution accuracy. In order to keep
the accuracy within a specified allowable limit of error, we
have introduced the procedure of re-analyses by decreasing
the increment size. The proposed method is verified by several
illustrative examples.

ACKNOWLEDGMENT
This research was partially supported by the Ministry of
Education, Science, Sports and Culture, Grant-in-Aid for
Scientific Research A-08305037 and B-10555338.

REFERENCES
1) Sumi, Y., Computational crack path prediction. Theor.
Appl. Fract. Mech., 4, 1985, pp.149-156.
2) Sumi, Y., Chen, Y. and Hayashi, S., Morphological aspects
of fatigue crack propagation, Part I - computational
procedure, Int. J. Fract., 82-3, 1996, pp.205-220.
3) Sumi, Y. and Wang, Z. N., A finite-element simulation
method for a system of growing cracks in a heterogeneous
material, Mechanics of Materials, 28, 1998, pp.197-206.
4) Sumi, Y., Fatigue crack propagation and computational
remaining life assessment of ship structures, Journal of
Marine Science and Technology, 3-2, 1998, pp.102-112.
5) Blacker, T. D. and Stephenson, M. B., Paving: A new
approach to automated quadrilateral mesh generation, Int. J.
Num. Meth. Eng., 32, 1991, pp.811-847.
6) Johnston, B. P., Sullivan, Jr. J. M. and Kwasnik, A.,
Automatic conversion of triangular finite element meshes
to quadrilateral elements, Int. J. Num. Meth. Eng., 31, 1991,
pp.350-355.
7) Shimada, K. and Itoh, T., Automated conversion of 2d
triangular mesh into quadrilateral mesh, in Computational
Mechanics '95, Vol.1, ed. Atluri, S. N., Yagawa, G. and
Cruse, T. A. , Springer, 1995, pp.350-355.
8) Potyondy, D. O., Wawrzynek, P. A. and Ingraffea, A. R.,
An algorithm to generate quadrilateral or triangular
element surface meshes in arbitrary domains with
applications to crack propagation, Int. J. Num. Meth. Eng.,
38, 1995, pp.2677-2701.
9) Sluiter, M. L. and Hansen, D. L., A general purpose
automatic mesh generator for shell and solid finite
elements, in Computers in Engineering, New York, ASME
1982, pp.29-34.
10)Yamamoto, Y. and Tokuda, N., Determination of stress
intensity factors in cracked plates by the finite element
method, Int. J. Num. Meth. Eng., 6, 1973, pp.427-439.
11)Isida, M., Arbitrary loading problems of doubly symmetric
regions containing a central crack, Eng. Fract. Mech., 7,
1975, pp.505-514.
12)Sumi, Y., Nemat-Nasser, S. and Keer, L. M., On crack
branching and curving in a finite body, Int. J. Fract., 21,
1983, pp.67-79.; Erratum: Int. J. Fract., 24, 1984, p.159.
13)Hirata, M., Experimental studies on form and growth of
cracks in glass plate, Scientific Papers of Institute of
Physical and Chemical Research, Vol.16, 1931, pp.172195.
14)Sumi, Y., Mu, Y. and Kawamura, Y., Morphological
aspects of thermally induced brittle cracks, Proceedings of
the Conference on Computational Engineering and Science,
Vol.4, 1999, pp.843-846.

You might also like