Engineering Analysis with Boundary Elements 64 (2016) 176–195

Contents lists available at ScienceDirect

Engineering Analysis with Boundary Elements

journal homepage: www.elsevier.com/locate/enganabound

Dual boundary element analysis of fatigue crack growth, interaction

and linkup
E. Santana, A. Portela n
Department of Civil Engineering, Faculty of Technology, University of Brasília, Brasília-DF 70910-900, Brazil

art ic l e i nf o a b s t r a c t

Article history: Multiple-site and widespread fatigue damage have been an issue to the aircraft and construction
Received 28 October 2015 industry for a long period. Structural components develop cracks at several locations which grow with
Received in revised form crack paths that are difficult to predict. When two cracks approach one another, their stress fields
2 December 2015
influence each other leading to an enhancing or shielding effect which depends on the position and
Accepted 7 December 2015
Available online 29 December 2015
orientation of the cracks. Since there are no generalized analytical methods for predicting crack stress
fields, simulation of multiple-crack growth is an important and challenging task which is still an evolving
Keywords: area of research.
Multiple-cracked plates This paper describes a two-dimensional application of the dual boundary element method (DBEM) to
Dual boundary element method
the analysis of mixed-mode multiple-crack growth in linear elastic fracture mechanics, under fatigue
J-integral
loading. The crack-growth process is simulated with an incremental multiple-crack extension analysis
Stress intensity factors
Crack coalescence based on the maximum principal stress criterion. For each increment of the analysis, in which crack
Crack linkup extensions are modelled with new straight boundary elements, the DBEM is applied to perform a single-
Fatigue analysis region stress analysis of the cracked structure and the J-integral is used to compute the stress intensity
factors. The incremental analysis is based on a prediction–correction technique that defines, in each
increment of the analysis, the direction and the extension of the multiple interacting cracks, thus taking
into account the discreteness of the analysis and ensuring that the requirement of the path uniqueness is
satisfied. Based on the ligament yield criterion which assumes that when the plastic zones of two
adjacent cracks touch each other, the ligament between the cracks fails and the cracks coalesce, plates
with multiple-site damage can be analysed. The fatigue life and residual strength of the structure are
introduced as a post-processing procedure on the results of the multiple-crack growth. Results of this
incremental analysis are presented for several geometries with multiple-site damage, demonstrating the
accuracy and efficiency of the strategies adopted in the analysis.
& 2015 Elsevier Ltd. All rights reserved.

1. Introduction elastic fracture mechanics is that the crack behaviour is deter-

Catastrophic fracture failure of engineering structures is caused
by cracks that extend beyond a safe size. Cracks, present to some
extent in all structures, either as a result of fabrication defects or
localized damage in service, may grow. The crack growth leads to a
decrease in the structural strength. Thus, when the service loading
cannot be sustained by the current residual strength, fracture
occurs leading to the failure of the structure. Fracture, the final
catastrophic event which takes place very rapidly, is preceded by
crack growth which develops slowly during normal service con-
ditions, mainly by fatigue due to cyclic loading.
Linear elastic fracture mechanics can be used to describe the
behaviour of the cracks. The fundamental postulate of the linear
n
Corresponding author. Tel.: 55 61 8154 7262.
E-mail address: aportela@unb.br (A. Portela).
mined solely by the values of the stress intensity factors which are
a function of the applied load and the geometry of the cracked
structure and thus, play a fundamental role in linear elastic frac-
ture mechanics applications.
Crack-growth processes are simulated with an incremental
crack-extension analysis. For each increment of the crack exten-
sion, a stress analysis is carried out and the stress intensity factors
are evaluated. The paths of the multiple cracks, predicted thus on
an incremental basis, are computed through a criterion defined in
terms of the stress intensity factors. Crack-tip linkup, of adjacent
cracks, is considered in this process.
The geometry of engineering structures, which is continuously
changing with the extension of the multiple cracks, requires the
use of numerical methods to evaluate the stress intensity factors.
The boundary element method is a well-established numerical
technique in the engineering community, see Brebbia [1] and
Brebbia et al. [2]. The boundary element method has been

http://dx.doi.org/10.1016/j.enganabound.2015.12.002
0955-7997/& 2015 Elsevier Ltd. All rights reserved.
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
successfully applied to linear elastic problems in domains con-
taining no cracks. For symmetric crack problems only one side of
the crack need be modelled and a single-region boundary element
analysis may be used. However, in a single-region analysis, the
solution of general crack problems cannot be achieved with the
direct application of the boundary element method, because the
coincidence of the crack boundaries causes an ill-posed problem.
Among the techniques devised to overcome this difficulty, the
most general are the sub-regions method, presented by Blandford
et al. [8] and the dual boundary-element method (DBEM), first
presented by Portela et al. [5] in elastostatics. The main drawback
of the method of subregions is that the introduction of artificial
boundaries, which connect the cracks to the boundary so that the
domain is partitioned into subregions without cracks, is not
unique and thus it cannot easily be implemented into an auto-
matic procedure to simulate the growth of multiple cracks. On the
other hand, the DBEM is the most efficient technique devised to
overcome this difficulty. It introduces two independent equations,
the displacement and traction boundary integral equations, with
the displacement equation applied for collocation on one of the
crack surfaces and the traction equation on the other. With this
strategy, general mixed-mode crack problems can be solved in a
single-region boundary element formulation, with both crack
surfaces discretized with boundary elements.
Historically, the use of dual integral equations was first repor-
ted by Bueckner [3] in crack problems, by Watson [4] in the
boundary element method and by Hong and Chen [6] who derived
integral equations of elasticity. However, it is well known in the
scientific community that the effective implementation of the dual
boundary element method for crack problems was first reported
by Portela et al. [5]. A thorough review article of dual boundary
element methods, with emphasis on hypersingular integrals and
divergent series, was presented by Chen and Hong [7].
Within the limits of linear elastic analysis, the stress field is
unbounded at the tip of a crack. This was early reported by Brahtz
[11] and later by Williams [12] who, after an investigation of the
analytical form of these singularities demonstrated that under all
possible combinations of boundary conditions, the stress becomes
infinite at the tip of a crack. From a physical point of view,
unbounded elastic fields are meaningless. Nevertheless, unboun-
ded stresses cannot be ignored as their presence indicates that
new phenomena (e.g. plasticity and fracture) may occur, leading to
localized damage in practical situations. In this paper, the term
singularity is used to denote the cases in which the elastic stress
field becomes unbounded. If r denotes the distance measured from
the crack tip, the stress field is of the order r
1=2 which becomes
singular as r tends to zero. The stress intensity factor (SIF), defined
at the crack tip, is a measure of the strength of this singularity.
The presence of the stress singularity in the numerical model
raises considerable numerical difficulties by virtue of the need of
simultaneously representing both the singular and the finite
stresses in the numerical model. The performances of the most
important approaches that have been devised to overcome this
difficulty, in the finite element method (FEM), in the extended
finite element method (XFEM), in the boundary element method
(BEM) and in meshfree or meshless methods, are briefly reviewed
in the following.
A common procedure, used in the very early fracture-
mechanics applications of the finite element method, is to ignore
the presence of the singularity and to attempt to diminish its effect
on the overall solution by using mesh refinement in the neigh-
bourhood of the crack tip. The numerical value of the calculated
stress components at the crack tip will always be finite, but it can
be made as large as one desires by increasing refinement of the
mesh. Obviously, this procedure is mesh dependent and, if it
converges, will produce a slow-convergence ratio in the entire
177
domain of the problem, as shown by Tong et al. [18]. This is
obviously a consequence of the impossibility of representing
simultaneously both the singular and the finite stresses in the
numerical model, simply with a mesh refinement procedure. In
this approach, the stress intensity factors are evaluated from a
correlation procedure, involving a comparison between the
numerical results of either the displacement or the stress fields
and the respective analytical solutions, represented in the form of
an eigenfunction expansion series around the crack tip. Typically,
the stress intensity factors obtained by application of this corre-
lation procedure at crack-face nodal points are then extrapolated
to the crack tip. Consequently, stress intensity factors cannot be
computed accurately only with the mesh refinement procedure.
This was shown, for instance, in the work of Portela et al. [5],
where values of the stress intensity factors, computed by a dis-
placement correlation procedure, are compared with those values
obtained with the J-integral technique, for several crack problems
analysed by the dual boundary element method.
The use of quarter-point isoparametric finite elements, intro-
duced by Henshell [40] and Barsoum [41], suggested the applica-
tion of quarter-point boundary elements at the crack tip, as an
alternative to the mesh refinement procedure. However, while
quarter-point finite elements both represent the r 1=2 displacement
behaviour and introduce a r
1=2 singularity in the stress field, the
use of quarter-point elements in the boundary element method, in
which displacements and tractions are approximated indepen-
dently, enables only the displacement behaviour to be properly
represented. This feature was early noticed by Cruse et al. [42]
who introduced traction-singular quarter-point boundary ele-
ments for the correct representation of the singularity in the stress
field. Stress intensity factors can be computed from quarter-point
elements by the displacement correlation procedure. The appli-
cation of this procedure over quarter-point elements, first pre-
sented by Blandford et al. [43], was called a two-point formula by
Smith [44]. The computation of stress intensity factors from
traction-singular quarter-point boundary elements was presented
by Martinez et al. [45] who have shown that the use of the crack-
tip traction nodal values of the singular element is less sensitive to
the discretization than any of the displacement correlation pro-
cedures. In general, the accuracy of stress intensity factors, com-
puted from quarter-point boundary elements by the displacement
correlation procedure, depends on the size of these elements, as
reported by Harrop [46] who studied the case of quarter-point
finite elements and concluded that it is impossible to recommend
a particular size for the quarter-point element, suitable for all
situations.
While the above methods represent the stress singularity in the
numerical model, an alternative approach, developed by Symm
[47] in potential theory, is based on the subtraction of this sin-
gularity from the numerical model. In fracture mechanics appli-
cations, the singularity subtraction technique is a procedure that
uses a singular particular solution of the crack problem to reg-
ularize the stress field and to introduce, simultaneously, the stress
intensity factors as additional primary unknowns in the problem.
This approach was first applied by Xanthis et al. [48] for anti-plane
problems and by Aliabadi et al. [14] to solve symmetrical crack
problems using the boundary element method. Analysis of sym-
metrical problems with the singularity subtraction technique is
straightforward, because the singular tractions are among the
problem unknowns, when only half of the problem is considered
with the proper boundary conditions along the symmetry line.
In the case of non-symmetrical problems, the singular tractions
are not among the boundary element unknowns and conse-
quently, there is no singularity in the numerical model to be
subtracted. The application of the sub-regions boundary element
method is an obvious way to circumvent this difficulty, as shown
178 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
by Aliabadi [49]. However, artificial boundaries introduced by this
method are not strictly necessary in the analysis of a crack pro-
blem. An alternative strategy, developed by Portela et al. [9], first
introduces the stress equations of an internal point, approaching
the crack tip, as primary unknowns in the boundary element for-
mulation. Then, the stress field, singular at this internal point, can
now be regularized with the singularity subtraction technique. The
extension of this singularity subtraction technique to pure opening
mode analysis of sharp notches was first reported by Portela et al.
[15].
Alternatively, the evaluation of stress intensity factors can be
based on contour integrals which are path-independent. The J-
integral has been used quite effectively in the dual boundary ele-
ment method, as a post-processing technique, for the evaluation of
stress intensity factors by Portela et al. [5]. A simple procedure,
based on the decomposition of the elastic field into its respective
symmetric and anti-symmetric mode components, is used to
decouple the stress intensity factors of a mixed-mode problem.
Although this technique does not perform a regularization of the
elastic field with the crack-tip singularity subtraction, it is very
accurate because it uses the elastic field computed at internal
points which is a highly accurate operation in the boundary ele-
ment method due to the use of the fundamental solutions of the
elastic field.
As an alternative to the J-integral post-processing technique,
the direct computation of the stress intensity factors, as additional
primary unknowns in the dual boundary element method, was
first presented by Portela et al. [9]. In order to avoid numerical
difficulties that arise from the presence of a singularity in the
numerical model, it is convenient to subtract this singularity from
the original problem, before it is solved by the numerical method.
This regularization considers a singular particular solution of the
problem and forces the original elastic field to be identical to this
particular solution, at the singular point. By virtue of the analytical
structure of the singular particular solution that represents the
crack-tip elastic field, the modified problem includes the stress
intensity factors as additional primary unknowns. Finally, the
numerical method can be easily applied to solve the modified
problem which is now regular and consequently can lead to highly
accurate solutions simply with coarse meshes. Despite its accu-
racy, this technique was never applied to simulate crack-growth
processes.
Following in the opposite direction of the singularity subtrac-
tion, Alatawi et al. [17] recently introduced an alternative method
to evaluate SIFs using an enrichment approach in the dual
boundary element method, for 2D problems. The enrichment
approach provides the values of SIFs directly without any need for
post-processing.
In the FEM, the hybrid crack-tip element (HCE), see Tong et al.
[18], is a very accurate method for the direct computation of stress
intensity factors and coefficients of the higher order terms of
William's expansion, as reported by Karihaloo et al. [19] and Xiao
et al. [20]. The HCE represents a crack by only one super-element
which is connected compatible with the surrounding finite ele-
ments. The HCE is formulated from a simplified variational func-
tional using truncated asymptotic crack-tip displacement and
stress expansions and interelement boundary displacements
compatible with the surrounding regular finite elements. Despite
its accuracy, the use of the HCE presents some difficulties. Effec-
tively, the exclusion of the rigid body modes in the truncated
asymptotic displacements creates jumps between these displace-
ments and element boundary displacements. If the HCE only is
used, the part of the crack inside the HCE need not conform to the
finite-element mesh. However, crack faces away from the crack tip,
outside the HCE region, must conform to the finite-element mesh.
This is a further disadvantage of the HCE in the simulation of
crack-growth processes.
Meshless or meshfree methods, see Belytschko et al. [34], have
received much attention recently, since they eliminate the need for
a discretization mesh and hence, they appear to demonstrate
significant potential to the moving boundary problem inherent in
crack-growth processes. Since these methods use only a set of
nodal points scattered in the domain with no element con-
nectivity, the remeshing required by the FEM is avoided which
dramatically simplifies the modelling process. Comprehensive
reviews of meshfree methods can be found in Li et al. [35] and Liu
et al. [36]. In these methods, the moving least squares approx-
imation is possibly the most used method to interpolate discrete
data with a good accuracy. The order of continuity of the
approximation can be set to a desired value, as reported by Sladek
et al. [31]. The treatment of crack discontinuities, the main issue of
modern meshfree methods, has been modelled in different ways;
Carpinteri et al. [32] used a virtual extension of the crack in the
direction of the tangent at the crack tip, while Wen et al. [33]
considered enriched basis functions in the moving least squares
interpolation. Enriched weight or basis functions, by incorporating
a priori knowledge of the solution that is a jump function along
the discontinuity and the asymptotic crack-tip displacement field,
have been successfully applied to fracture problems, as reported
by Fleming et al. [37], Lu et al. [38] and Gu et al. [39]. However, the
main difficulty of this strategy is that the enrichment area must be
limited, when multiple cracks are densely distributed or when
crack tips are close to the boundaries which is a drawback of this
new generation meshfree methods.
The extended finite element method (XFEM), developed by
Belytschko et al. [28], is a modern numerical modelling tool. The
XFEM enriches the local standard finite-element approximation
space to incorporate a priori knowledge of the solution, with a
displacement discontinuity function across the crack and the
asymptotic solution at nodes surrounding the crack tips, with the
use of the partition of unity method (PUM), see Melenk et al. [21].
As a result, the numerical model consists of three types of finite
elements: non-enriched elements, fully enriched elements and
partially enriched elements, the so-called blending elements. In a
blending element, some of the nodes only are enriched which
adds to the approximation parasitic terms. The error caused by
these parasitic terms, which is partly responsible for the degra-
dation of the convergence rate and limits the accuracy of the
method, was identified by Chessa et al. [29]. By virtue of the
enrichment process, the XFEM overcomes the need of using finite-
element meshes conforming with the crack discontinuity, as well
as the adaptive remeshing as the crack grows. Nevertheless, it does
not completely eliminate the need for a suitable mesh refinement
in the vicinity of the crack. The main reasons behind that feature of
the method are firstly, and just as with the FEM, the accuracy of
the solution increases with decreasing finite-element sizes; sec-
ondly, the finite-element mesh is also used to define the geometry
of the crack in an implicit way. The element sizes must therefore
be small enough to properly capture the features of the crack
geometry and consequently, the XFEM still needs a mesh that is
fine enough in the vicinity of the cracks with respect to the
characteristic lengths of these cracks. The solution accuracy of the
local fields around crack tips is a direct consequence of the choice
carried out for the enrichment functions which define a priori
knowledge of the solution. Effectively, the classical enrichment
fields are not able to provide reliable estimates of the stress
intensity factors directly, as shown by Chahine et al. [22]. The
closer these enrichment functions are to the exact asymptotic
fields, the better is the solution accuracy. Since the exact asymp-
totic fields are known only for very simple crack geometries and
loadings, this direct dependence of the solution accuracy on the
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
enrichment process is a serious drawback of the XFEM. Alternative
crack-tip enrichment techniques have been devised to simulate
failure and yet allow for direct estimation of the stress intensity
factors. In this regard, Liu et al. [23] introduced a method which is
still relying on the PUM, but with specific enrichment functions
that are Williams' series. The area in which the nodes are enriched
by the singular functions can be different from the area used to
extract the stress intensity factors that is the area where the
degrees of freedom associated with the same singular enrichment
function are constrained to be equal. The underlying partition of
the unity is inactivated by simply setting the regular degrees of
freedom to zero in the extraction zone of the stress intensity
factors. Modern formulations of the XFEM rely on truncated Wil-
liams' expansion which is dedicated to straight cracks only. In the
general case of non-straight cracks, the singular enrichment zone
must be defined onthe scale on which the crack can be considered
straight. Therefore the finite-element mesh must be fine enough to
fit with this scale. Strategies have been developed to handle these
very different scales that are required to simulate cracked bodies.
These strategies can use energy coupling methods, as presented by
Ben Dhia et al. [24], domain decomposition methods, as presented
by Guidault et al. [25], homogenization, as presented by
Belytschko et al. [26] or generalized FEM, as reported by Kim et al.
[27]. The XFEM key feature resides in its great flexibility in the
numerical analysis of the fracture process, achieved by the
enrichment process implemented in the local standard finite ele-
ment approximation space. The accuracy of the XFEM is thus
totally dependent on the choice of the enrichment functions used
in the formulation, in order to approximate the crack-tip singular
elastic field.
An evaluation of the performances of BEM-based methods and
their comparison with XFEM, in modelling cracked structures
undergoing fatigue crack-growth, was carried out by Dong and
Atluri [30]. After a thorough examination of a large set of
numerical examples of varying degrees of complexity these
researchers concluded that the BEM-based methods: (a) are far
more accurate than XFEM, for computing stress intensity factors
and thus the fatigue-crack-growth-rates; (b) require significantly
coarser meshes than in XFEM and thus result in significant savings
of computational costs and much importantly, in considerable
savings of the human-labour costs of generating meshes;
(c) require minimal effort for modelling the non-collinear propa-
gation of cracks under fatigue, without using the Level Set or Fast
Marching methods to track the crack surface; (d) can easily per-
form fracture and fatigue analysis of complex structures, such as
repaired cracked structures with composite patches and damage
in heterogeneous materials.
Multiple-site damage (MSD) is a typical problem for ageing
aircraft, where the large number of fuselage pressure cycles may
cause fatigue cracking at multiple rivet-hole locations. The
assessment of the residual strength of panels with MSD cracks
uses models to predict crack coalescence or linkup. These models
mainly differ on the linkup criteria they consider. Swift [55] first
proposed the linkup or plastic-zone touch criterion, using Irwin
formulas to determine the extent of crack tip plasticity. This cri-
terion implies that a ligament will fail if the sum of the sizes of the
plastic zones of the two crack tips equals the ligament size. This
method has also been adopted by Broek et al. [56] and De Wit et al.
[57], who determined the size of the plastic zone with the Dugdale
[58] formulas. Broek et al. [56] also applied a modified linkup
criterion by including the effect of stable tearing, which reduces
the distance between cracks. Smith et al. [59] modified the linkup
model empirically to improve the accuracy of the model fit to the
test data. Nilsson et al. [60] defined linkup as actual crack impin-
gement of the leading crack and an adjacent crack.
179
This paper is concerned with the application of the DBEM to
the analysis of multiple-crack growth in linear elastic fracture
mechanics. The dual boundary integral equations are presented,
the crack modelling strategy defined and the stress intensity fac-
tors evaluated by the J-integral technique. An incremental crack-
extension analysis is performed to determine the crack paths. For
each increment of the analysis, in which the crack extensions are
modelled with new boundary elements, the DBEM is applied for
the stress analysis and the J-integral is used for the stress intensity
factors evaluation. The incremental analysis is based on a predic-
tion–correction technique to define the direction of the multiple
cracks in each increment; the maximum principal stress crack-
growth direction criterion is applied to predict the tangent direc-
tion of each crack path and then a global correction is introduced
to determine the actual direction of each crack-extension. Crack
linkup of adjacent cracks is dealt with the ligament yield criterion.
The number of loading cycles of the current analysis increment is
computed with Paris law, for the dominant crack. With this value
and still with Paris law, the extension of the multiple cracks is also
corrected. Results of this incremental analysis of fatigue multiple-
crack growth are presented for several geometries.
The organization of the paper is as follows. After the Intro-
duction, the dual boundary integral equations are summarized in
Section 2. The computation of the stress intensity factors is pre-
sented in Section 3. Section 4 summarizes the residual strength
and fatigue life driven by the fatigue loading. The crack-tip linkup
criterion is presented in Section 5. The strategies adopted in the
incremental analysis of multiple-crack growth are presented in
Sections 6–8. Section 9 presents some numerical results, illus-
trating the effectiveness and robustness of the present analysis
procedure. Finally, Section 10 presents the concluding remarks.
2. The dual boundary element method
The DBEM performs the analysis of general crack problems in a
single-region boundary element formulation, as represented in
Fig. 1. The equations on which the DBEM is based are the dis-
placement and the traction Somigliana's boundary integral equa-
tions, as presented by Portela et al. [5]. In the absence of body
forces and assuming continuity of the displacements at a bound-
ary point P, the boundary integral representation of the displace-
ment components ui is given by
Z
cij ðPÞuj ðPÞ þ ⨍Γ T ij ðP; Q Þuj ðQ Þ dSðQ Þ ¼ U ij ðP; Q Þt j ðQ Þ dSðQ Þ; ð1Þ
Γ
where i and j denote Cartesian components; T ij ðP; Q Þ and U ij ðP; Q Þ
represent, respectively the traction and displacement Kelvin fun-
damental solutions, at a boundary point Q; the symbol ⨍Γ stands
for Cauchy principal-value integral, and the coefficient cij(P) is
given by 12δij for a smooth boundary at the point P in which δij is
the Kronecker delta. In the absence of body forces and assuming
continuity of both strains and tractions at P on a smooth boundary,
Fig. 1. Single-region analysis of the DBEM.
180 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195

the traction components tj are given by is given by

2 t j ðPÞ þni ðPÞ

1
S ðP; Q Þuk ðQ Þ dΓ ðQ Þ ¼ ni ðPÞ⨍Γ Dijk ðP; Q Þt k ðQ Þ dΓ ðQ Þ;
Γ ijk K 2I þ K 2II
J¼ ; ð6Þ
ð2Þ E0

in which Sijk ðP; Q Þ and Dijk ðP; Q Þ contain derivatives respectively of
T ij ðP; Q Þ and U ij ðP; Q Þ; the symbol Γ stands for the Hadamard
principal-value and ni denotes the ith component of the unit
outward normal to the boundary at the point P. Eqs. (1) and (2) are
the basis of the DBEM, as presented by Portela et al. [5].
On a traction-free crack Γc, Eqs. (1) and (2) are simplified; the
displacement and the traction equations are given respectively by
cij ðPÞuj ðPÞ þ ⨍Γ c T ij ðP; Q Þuj ðQ Þ dΓ ðQ Þ ¼ 0
and
ni ðPÞ Γ Sijk ðP; Q Þuk ðQ Þ dΓ ðQ Þ ¼ 0:
c j uj;1 dS;
The DBEM transforms the boundary integral equations (1)–(4),
accordingly, into a system of linear algebraic equations, by means
of nodal collocation and integration over the boundary elements.
The unknown boundary tractions and displacements can then be
obtained by solving this system of equations which includes the
boundary conditions.
This work adopts the DBEM modelling strategy presented by
Portela et al. [5] which considers piecewise-straight cracks dis-
cretized with discontinuous quadratic boundary elements. Con-
tinuous quadratic boundary elements are used along the remain-
ing general boundaries of the problem, except at the intersection
between a crack and an edge, where semi-discontinuous boundary
elements are used on the edge. Self-point discontinuous boundary
elements are integrated analytically, while Gaussian quadrature,
performed with sub-element integration, is carried out for the
remaining integrations, see Liu et al. [10].
3. Computation of the stress intensity factors
The J-integral is used to compute the stress intensity factors.
Consider a Cartesian reference system defined at the tip of a
traction-free crack, as represented in Fig. 2. The J-integral is
defined as
Z
 
J¼ Wn1
t j uj;1 dS; ð5Þ
S
where S is an arbitrary contour surrounding the crack tip, W is the
strain energy density, tj are the traction components and n1 is the
x-component of the unit outward normal to the contour path. The
relationship between the J-integral and the stress intensity factors
where E0 is the elasticity modulus E for plane stress conditions and
E0 ¼ E=ð1
ν2 Þ for plane strain conditions.
To decouple the stress intensity factors in Eq. (6), the integral J
is represented by the sum of two integrals as J ¼ J I þ J II , where the
superscripts indicate the pertinent deformation mode. For this
representation to be possible, it is sufficient to consider a contour
that is symmetrical about the crack axis, as shown in Fig. 2 and
decompose the displacement and stress fields into their sym-
ð3Þ
metric and anti-symmetric components. Following this procedure
the J-integral components can be expressed as
Z  
ð4Þ Jm ¼ W m n1
t m m
ð7Þ
S
for m ¼I or m ¼ II. The relationship between the J-integral com-
ponents and the stress intensity factors is now given by
K 2I K 2II
JI ¼ ; J II ¼ : ð8Þ
E0 E0
The implementation of this procedure into the DBEM is
straightforward, as reported by Portela et al. [5]. A circular contour
path around the crack tip is automatically defined with a set of
internal points at symmetrical positions about the crack axis, as
shown in Fig. 3.
The integration along the contour path is performed by the
trapezoidal rule. For the sake of simplicity only circular paths,
centred at the crack tip and containing a pair of crack-nodes are
considered; each path is referred to by a path number that
increases as the radius of the contour increases, as represented in
Fig. 3. The J-integral is an effective method for the determination
of stress intensity factors, because the interior elastic field can be
accurately determined along the contour path in the dual
boundary element method, since the exact variation of the interior
elastic field is built into the fundamental solution of the problem.
4. Residual strength and fatigue life
In a mixed-mode analysis, an equivalent mode-I stress intensity
factor can be defined, see Broek [51]. For the maximum principal
stress criterion, it is given by
θt θt θt
K Ieq ¼ K I cos 3
3K II cos 2 sin : ð9Þ
2 2 2

Fig. 2. Coordinate reference system and contour path for the J-integral: (a) general contour path and (b) circular contour path.
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
Fig. 3. Contour path numbers for the J-integral in the DBEM.
The fracture condition then follows from K Ieq ¼ K Ic , in which KIc is
the fracture toughness.
A residual-strength diagram shows the variation in the max-
imum load that the cracked structure can sustain, the load that
causes fracture instability, as the crack length varies. For a given
crack length, the relationship between the residual strength σc and
the applied reference stress σr is defined by
K Ic
σc ¼ σr :
K Ieq
At each step of the incremental analysis the residual strength is
represented conveniently in a normalized form as
σ c K Ieqn
¼ ð11Þ
σ cn K Ieq
where σ cn and KIeqn represent the residual strength and the cor-
responding mode-I equivalent stress intensity factor respectively,
computed at the initial crack length.
In general, fatigue crack-growth is driven by variable amplitude
loading. In the simplest case, the loading cycles have a constant
amplitude and may be described by a constant amplitude load
with a constant amplitude stress ratio.
A fatigue-life diagram shows the variation, in the number of
loading cycles, required to extend the crack, as a function of the
crack length. It is computed from the empirical Paris model
defined as
da
¼ CðΔK eff Þm ;
dN
where a is the crack length, N is the number of load cycles, C and m
are material dependent constants and ΔK eff is the range of the
effective stress intensity factor. The model of Tanaka was applied
in the present work with ΔK eff defined by
ΔK 2eff ¼ ΔK 2I þ 2ΔK 2II :
The stress intensity factor range of each individual mode is given
by ΔK ¼ K max
K min ¼ K max ð1
RÞ in which R ¼ K min =K max ¼ σ min =
σ max is the stress amplitude ratio of the loading cycle. Integration
of Eq. (12), with the trapezoidal rule, is used in this work to
compute either the number of loading cycles required to extend
the crack a given increment or the crack-extension increment that
results from a given number of loading cycles.
181
5. Crack-tip linkup
Multiple site damage (MSD) is a typical problem for ageing
aircraft, starting when the fuselage pressure cycling fatigue loads
lead to crack initiation and propagation at multiple rivet locations,
as schematically represented in Fig. 4. These cracks interact and,
when a critical situation is reached, sudden crack coalescence or
linkup may occur reducing the structural residual strength. The
failure scenario is governed by a long lead crack that is formed by
successive linkups of in-line short cracks which reduces the overall
structural integrity.
The coalescence or linkup of multiple cracks is considered in
this paper. For the sake of simplicity, the most extensively applied
crack-tip linkup criterion that is the ligament-yield criterion, also
known as the plastic zone touch model of Swift [55], is considered
in this paper. The ligament yield criterion is based on the
assumption that the ligament between two adjacent cracks fails
and the cracks coalesce, when their plastic zones touch each other,
as schematically represented in Fig. 5. The crack-tip plastic zone
sizes can be estimated using Irwin or Dugdale plastic zone models.
As the stress intensity factors increase, the plastic zone sizes will
increase until they touch each other and the cracks linkup. In this
paper the plastic zone sizes are calculated using the Dugdale [58]
model as
 
π KI 2
ry ¼ ; ð14Þ
8 σy
where σy is the yield-stress value, ry is the value of the plastic-zone
ð10Þ
radius, represented in Fig. 5 and the values of the stress intensity
factor are determined by the DBEM analysis in each computational
cycle of the growth of cracks.
The computed size of the yield-ligament can be smaller or
longer than the size of the current extension increments of the
cracks linking up. Hence, to ensure efficiency of the numerical
analysis, the size of the effective linkup distance that is considered
is taken as the maximum value between the computed size of the
yield-ligament and the size of the crack-extension increments; it is
discretized with one single new boundary element on each crack
face. Note that the length of the dominant-crack extension incre-
ment can always be defined as the result of a compromise
between accuracy and computational cost that is, the smaller the
size of the crack-extension increment the more accurate and
expensive is the analysis.
6. Direction of the multiple-crack growth increments
Among several criteria that have been proposed to describe the
ð12Þ local direction of mixed-mode crack growth, one of the most
commonly used is based on the maximum principal stress at the
crack tip, see Erdogan and Sih [50]. The maximum principal stress
criterion postulates that the growth of a crack will occur in a
direction perpendicular to the maximum principal stress. Thus, the
ð13Þ
Fig. 4. Multiple-site fatigue damage at rivet locations of a fuselage.
182 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
Fig. 5. Plastic zone touch or ligament yield criterion for crack linkup.
Fig. 6. Incremental crack-extension direction.
local crack-growth direction θt is determined by the condition that
the local shear stress is zero, that is
K I sin θt þ K II ð3 cos θt
1Þ ¼ 0;
where θt is the angular coordinate of the tangent to the crack path,
centred at the crack tip and measured from the crack axis ahead of
the crack tip. As a continuous criterion, the maximum principal
stress does not take account of the discreteness of the crack-
extension procedure. Therefore, in an incremental analysis the
trajectories of the multiple-crack paths, predicted by Eq. (15), must
be corrected to give the direction of the actual crack-extension
increment, such that the requirement of the path uniqueness be
satisfied.
The procedure applied to define the direction of the n-th crack-
extension increment of an active crack introduces a correction
angle β to the tangent direction θtðnÞ predicted by the maximum
principal stress criterion, as shown in Fig. 6. This correction angle
is given by β ¼ θtðn þ 1Þ =2, in which θtðn þ 1Þ is the direction of the
next crack-extension increment, also computed with the max-
imum principal stress criterion. This predictor–corrector proce-
dure is applied iteratively, for all the active cracks simultaneously
to take into account their interaction, while each correction is
smaller than the previous one. For the current n-th crack-
extension increment of the active cracks, the i-th iteration can
be summarized as follows:
 Before the first iteration, for each active crack evaluate the
crack-path tangent direction θtðnÞ with the maximum principal
i
stress criterion, Eq. (15) and extend the crack one increment da
to pi.
 Start an iteration loop to predict and correct the direction of the
extension of the active cracks, simultaneously.
 Perform a stress analysis to evaluate the stress intensity factors
for the current crack-extension increments.
 Loop on the active cracks and with the new stress intensity
factors and the maximum principal stress criterion, Eq. (15),
evaluate the next crack-path direction θtðn þ 1Þ .
i
 Define the correction angle β ¼ θtðn þ 1Þ =2, measured from the
i i
crack-extension increment in the previous iteration.
 Correct the crack-extension increment da to its new direction,
iþ1
given by θtðnÞ ¼ θtðnÞ þ β , moving the crack-tip to pi þ 1 .
i i
 Finish the loop running on the active cracks and go back to the
iþ1
second step to update the iteration loop, while j β j o j β j or,
i
equivalently j K II ðP i þ 1 Þj o j K II ðP i Þj .
7. Extension of the multiple-crack growth increments
When the analysis regards the growth of a single crack, the
length of the crack-extension increment da, can be defined arbi-
trarily. However, some restrictions on this size are introduced to
ensure efficiency of the numerical analysis. This increment of the
crack extension can be discretized with a fixed number of new
boundary elements; for the sake of simplicity, two pairs of new
identical boundary elements are used in this work. In order to
avoid numerical difficulties, concerned with the relative size of
neighbouring elements, the length of the crack-extension incre-
ment is kept between convenient limiting bounds defined in terms
of the length of the neighbouring element, computed in the pre-
vious cycle of the incremental analysis. These limiting bounds of
the crack-extension increment are usually defined as 1 and 4 times
respectively, the length of the neighbouring element. Apart from
this constraint, the length of the crack extension increment may
be defined as the result of a compromise between accuracy and
computational cost that is, the smaller the size of the crack-
extension increment the more accurate and expensive is the
ð15Þ analysis.
The incremental analysis of fatigue growth of a single crack is
based on the extension of the crack with a specified increment
length which leads to the computation of the corresponding
increment number of loading cycles required to extend the crack
that length, through integration of Eq. (12).
In the analysis of multiple-crack growth, this procedure is used
only for dominant cracks, defined within a small neighbourhood of
the maximum value of KIeq, see Eq. (9), which are computed in
each cycle of the incremental analysis. Hence, the analysis is based
on an increment number of loading cycles dN, computed from
dominant cracks that extend a specified increment length. This
increment dN, assigned to all the cracks in the current computa-
tional cycle, is then used to compute the individual crack-
extension increments dai of non-dominant cracks, through inte-
gration of Eq. (12). The basic procedure used to define the exten-
sion of the multiple-crack growth increments, in each computa-
tional cycle of the incremental analysis, can be summarized as
follows:
 Define dominant cracks within a small neighbourhood of the
maximum value of the mode-I equivalent stress intensity factor
KIeq, computed from all the cracks.
 The increment number of loading cycles dN, used to extend
dominant cracks the specified increment length da, is computed
through integration of Eq. (12).
 For each crack evaluate the length of the extension increment
dai , through integration of Eq. (12), with the value of dN
obtained from dominant cracks.
 Turn off the growing of a crack whenever its computed dai , does
not comply with the usual length ratio of neighbouring ele-
ments that is smaller than half the length of the previous crack-
extension increment, or is smaller than a given percentage of
the extension increment of dominant cracks; in this case, store
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195 183

the computed small value dai , of the crack that was not
extended, to perform a possible restart in a later cycle of the
analysis.
 Enforce the restarting of each crack that did not grow, in the
previous cycle of the analysis, whenever its stored extension is
greater than the given percentage (in the previous item) of the
extension increment of the dominant cracks.

The computed extension increment of each non-dominant

crack, always smaller than the extension increment of dominant
cracks, is also discretized with two new boundary elements.
Hence, to ensure efficiency of the numerical analysis, the same
restrictions on the length of the crack-extension increments,
defined for dominant cracks, are considered here. Over-sized
increments do not occur, since the given extension increment of
dominant cracks is the limiting bound. On the other hand, under-
sized increments can delay their growing to another cycle of the
incremental analysis, as already explained.

8. Multiple-crack growth incremental analysis Fig. 7. Rectangular plate with an internal kinked crack ðH=W ¼ 2; a=W ¼ 0:1Þ and
applied remote stress σ.
The incremental crack-extension analysis assumes a piece-wise
linear discretization of the unknown crack paths. For each incre-
ment of the crack extension, the DBEM is applied to carry out a Table 1
pffiffiffiffiffi
stress analysis of the cracked structure and the J-integral is used Stress intensity factor K I =ðσ πcÞ for the kinked crack, tip A.
for the evaluation of the stress intensity factors. The steps of this pffiffiffiffiffi
b/a K I =ðσ πcÞ
basic computational cycle, repeatedly executed for any number of
cycles of the incremental analysis, are summarized as follows: J-integral Murakami [54]

 As an initialization, input the upper bound value for the crack-
extension increments; perform a DBEM stress analysis of the
cracked structure and evaluate the stress intensity factors;
assume that all cracks can grow and specify an initial crack-
extension da, that is the minimum value between the given
upper bound value and three times the length of the crack-tip
element, for dominant cracks and half of that value for the
remaining cracks;
 Compute the direction of the tangent to the crack-paths, with
the maximum principal stress criterion and define dominant
cracks, within a small neighbourhood of the maximum value of
the mode-I equivalent stress intensity factor KIeq, see Eq. (9),
computed from all the cracks;
 Use two pairs of identical new boundary elements to mesh the
crack-extension increments, in the tangent direction, of active
cracks;
 sented in Fig. 7. One of the segments of the crack is horizontal with
Correct the direction of the crack-extension increments, as
explained in Section 6;
 horizontal and has a length b;pffiffiffithe horizontal projection of the total
Perform a DBEM stress analysis of the cracked structure and
evaluate the stress intensity factors, as explained in Section 3;
 Correct the length of the crack extension increments, as
explained in Section 7;
 Assess crack linkup, as explained in Section 5;
 Repeat all the above steps sequentially until a specified number
of cycles of the incremental analysis is reached.
The results obtained from an incremental crack-extension
analysis are a diagram of the crack paths, diagrams of SIF varia-
tion along each crack path, fatigue-life (FL) and residual-strength
(RS) diagrams along each crack path, as explained in Section 4.
9. Numerical results
The DBEM modelling strategy, presented in Section 2, is used to
obtain numerical results. The stress intensity factors are computed
0.2 0.996 0.995
0.4 0.991 0.990
0.6 0.988 0.986
from the J-integral, as referred in Section 3, in general using the
contour path number 5 with 32 points. Paris equation, presented
in Section 4, is used to compute the crack-growth rate and the
number of fatigue loading cycles required for a crack-extension
increment, with the parameters C ¼ 4:624  10
12 and m ¼3.3
and with the stress-amplitude ratio of the loading cycle R ¼2/3,
unless otherwise is referred.
9.1. Plate with an internal kinked crack
Consider now a plate with an internal kinked crack, as repre-
length a while the other segment makes an angle of 45° with the
crack is given by 2c ¼ a þb 2=2. The kink of the crack is at the
centre of the plate which has a height equal to twice the width and
is loaded at the ends with a uniform traction σ. Three cases were
considered, b=a ¼ 0:2, 0.4, and 0.6 with a=W ¼ 0:1. A mesh of 48
quadratic boundary elements was set up, in which the horizontal
and the inclined crack segments were discretized with 5 and
4 discontinuous elements on each crack face, respectively. The
stress intensity factors were obtained for both crack tips A and B.
Accurate results for comparison are published by Murakami [54].
pffiffiffiffiffiffi pffiffiffiffiffiffi
Tables 1–4 present the values of K I =ðσ π cÞ and K II =ðσ π cÞ, as well
as the appropriate values of the reference [54], as a function of b=a,
respectively for crack tips A and B. Even with the present coarse
mesh, the results obtained for this case show an excellent accu-
racy, matching those reference values within two decimal places.
The kinked-crack dimensions were changed for the values b=a
¼ 0:6 with a=W ¼ 0:3 to perform a fatigue analysis and better
enhance graphic results. Paris equation was used with the
184 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195

Table 2 9.2. Plate with three edge cracks

pffiffiffiffiffi
Stress intensity factor K II =ðσ πcÞ for the kinked crack, tip A.

pffiffiffiffiffi Consider now a rectangular plate with three edge cracks, as

b/a K II =ðσ πcÞ
represented in Fig. 10. The plate, loaded at the ends with a uniform
J-integral Murakami [54] traction σ, is symmetric about a horizontal axis containing the
crack number 3 which is on the right edge. A fatigue analysis was
0.2 0.030 0.028 carried out using Paris equation with the parameters C ¼ 4:624 
0.4 0.035 0.033
0.6 0.032 0.030
10
12 and m ¼3.3 and with the stress-amplitude ratio of the
loading cycle R¼2/3.
A mesh of 54 quadratic boundary elements was set up, with
Table 3 5 elements on each of the horizontal sides, 10 elements on each of
pffiffiffiffiffi
Stress intensity factor K I =ðσ πcÞ for the kinked crack, tip B. the vertical edges and 4 discontinuous elements on each crack face
pffiffiffiffiffi graded towards the crack tip, respectively with the ratios 0.4, 0.3,
b/a K I =ðσ πcÞ
0.2, and 0.1. A crack-extension increment, equal to 2 times the
J-integral Murakami [54] length of the smallest crack-tip element, was specified for domi-
nant cracks, in a total number of 14 crack-growth increments. The
0.2 0.604 0.598 cracks have grown to the final shape represented in Fig. 11, where
0.4 0.576 0.574
0.6 0.570 0.568
it can be seen that the crack-path uniqueness criterion is satisfied
for all cracks, independently of the size of the crack-extension
increment considered.
SIF and RS diagrams of crack-tips number 1 and number 3, as a
Table 4
pffiffiffiffiffi function of the crack growth, are represented in Fig. 12; they show
Stress intensity factor K II =ðσ πcÞ for the kinked crack, tip B.
that along the path of the crack number 1, the mode I stress
pffiffiffiffiffi intensity factor increases slowly while the mode II stress intensity
b/a K II =ðσ πcÞ
factor tends to zero very rapidly. Regarding the path of the crack
J-integral Murakami [54] number 3, note that although the monotonically increasing
behaviour of the mode I SIF along the crack path is a typical feature
0.2
0.555
0.557
0.4
0.602
0.607 of this formulation, it is not a general rule. Indeed, cases of
0.6
0.623
0.627 decreasing mode I stress intensity factor along the crack path are
quite common, leading to an increase of the respective residual
strength. This effect is called arrest of the crack growth or simply
crack arrest, see Broek [52], as represented in Fig. 13. As it can be
seen in Figs. 12 and 13, the crack number 3 that eventually had its
growth arrested by the simultaneous opening of the other cracks
Fig. 8. The final shape of the kinked crack, after a fatigue crack-growth of 1; 012;
does not influence the residual strength of the structure during the
419  109 loading cycles. whole crack-growth process.

9.3. Plate with a branching crack

parameters C ¼ 4:624  10
12 and m¼ 3.3 and with the stress-
amplitude ratio of the loading cycle R ¼ 2=3. A maximum value of
the crack-extension increment equal to 3 times the length of the
smallest crack-tip element was specified for the dominant crack, in
a total number of 10 crack-growth increments. The kinked crack
has grown to the final shape represented in Fig. 8, where it can be
seen that crack paths tend to be perpendicular to the load path
which roughly follows the stress trajectories that are in the
direction of the maximum principal stress, at each point of the
crack paths.
The diagrams of the stress intensity factors and residual
strength of both crack-tips, as a function of the crack growth, are
represented in Fig. 9; they show that along the crack paths, the
mode I stress intensity factor increases slowly while the mode II
stress intensity factor tends to zero very rapidly. This result is in
agreement with the generalized concept that crack extension
usually takes place in mode I or close to it, see Broek [52]. Fur-
thermore, the residual strength decreases monotonically along the
crack paths, as a direct consequence of the stress intensity factor
variation.
As it can be seen in Fig. 9, the crack B was the dominant one
during the first 3 crack-extension increments which corresponds
to 6:523960  108 loading cycles. In the next 2 increments, both
cracks played the role of dominant cracks while the crack A was
the dominant crack from the 6th increment until the end of the
growth analysis.
Consider now a plate with an internal branching crack, as
represented in Fig. 14, with b/a¼ 1, H/W ¼1, a/W ¼0.025 and
θ ¼ 45°; the
pffiffiffi horizontal projection of the total crack is given by
2c ¼ a þ b 2=2. The plate is loaded by a remote uniform traction σ,
applied to the upper and lower edges.
A mesh of 44 quadratic boundary elements was set up, in
which the horizontal and the inclined crack segments were dis-
cretized with 4 discontinuous elements on each crack face, graded
towards the respective tip with the ratios 0.4, 0.3, 0.2 and 0.1. The
stress intensity factors were obtained for the crack tips A and B
with the J-integral contour path number 5 with 32 points. Accu-
rate results for comparison are published by Chen and Hasebe
[53].
Even with the present extremely coarse mesh, the results
obtained for this case given in Table 5, show an excellent accuracy,
matching those reference values within two decimal places.
The branching-crack dimensions were changed for the values
ða=W ¼ 0:05Þ to perform a fatigue analysis and better enhance
graphic results. Paris equation was used with the parameters C ¼
6:9  10
12 and m ¼3 and with the stress-amplitude ratio of the
loading cycle R ¼ 2=3. A maximum value of the crack-extension
increment equal to 2 times the length of the smallest crack-tip
element was specified for dominant cracks, in a total number of 20
crack-growth increments. The branching crack has grown to the
final shape represented in Fig. 15, where it can be seen again that
crack paths tend to be perpendicular to the load path which
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195 185

Fig. 9. Stress intensity factors and residual strength of the kinked crack, along a fatigue crack-growth of 1; 012; 419  109 loading cycles: (a) SIF – crack-tip A; (b) SIF – crack-
tip B; (c) RS – crack-tip A; and (d) RS – crack-tip B.

roughly follows the stress trajectories that are in the direction of

the maximum principal stress, at each point of the crack paths.
Fig. 16 represents the diagrams of the stress intensity factors of
both crack-tips A and B, as a function of the crack growth, as well
as the respective residual-strength diagrams of the plate. As it can
be seen in Fig. 16, the crack A was the dominant one during all the
growth analysis and consequently has a direct influence in the
residual strength of the plate.

9.4. Plate with two skew-parallel embedded cracks

Consider now a square plate with two skew-parallel embedded

cracks, as represented in Fig. 17, with h=b ¼ 1, 2a=d ¼ 0:2 and
e=f ¼ 3:0; the horizontal and vertical projections of the cracks are
centred. The plate is loaded by a remote uniform traction σ,
applied to the upper and lower edges.
A mesh of 62 quadratic boundary elements was set up, in
which the crack segments were discretized with 8 discontinuous
elements on each crack face, graded towards the respective tips
with the ratios 0.4, 0.3, 0.2 and 0.1. The stress intensity factors
were obtained for the crack tip A with the J-integral contour path
number 5 with 32 points. Accurate results for comparison are
published by Murakami [54]. Fig. 10. Rectangular plate with three edge cracks ðH=W ¼ 2; a=W ¼ 0:1Þ and
The results obtained for this case, presented in Table 6, show an remote stress σ.
excellent accuracy, matching those reference values within the
error of 0.003%. amplitude ratio of the loading cycle R¼2/3. A maximum value of
Paris equation was used for a fatigue analysis with the para- the crack-extension increment equal to 4 times the length of the
meters C ¼ 4:624  10
12 and m ¼3.3 and with the stress- smallest crack-tip element was specified for dominant cracks, in a
186 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195

Fig. 11. The final shape of the three edge cracks, after 14 crack-growth increments, respectively equal to 2 and 4 times the length of the smallest crack-tip element: (a) crack
paths and (b) uniqueness of the crack paths.

Fig. 12. Stress intensity factors and residual strength of the three edge cracks, along 14 crack-growth increments: (a) SIF – crack-tip 1; (b) SIF – crack-tip 3; (c) RS – crack-tip
1; and (d) RS – crack-tip 3.
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195 187

Fig. 13. Deformed configurations of the plate with three edge cracks; note that the opening of the left-side cracks forces the right-side crack to gradually close down, after
the increment number 4: (a) Increment – 0; (b) Increment – 2; (c) Increment – 4; (d) Increment – 8; (e) Increment – 10; and (f) Increment – 14.

total number of 20 crack-growth increments. The cracks have
grown to the final shape represented in Fig. 18, where it can be
seen that the growth direction of the inner crack tips, initially
almost straight and perpendicular to the load path, changed so
that they approached each other in a typical shape that results
from the interaction of the stress fields around the tips of these
cracks; the local tensile stresses are no longer parallel to the
applied traction and the cracks bend towards each other by tilting.
On the other hand, the direction of growth of the outer crack tips
was almost straight and perpendicular to the loading path.
The diagrams of the stress intensity factors and residual
strength of the inner and outer crack-tips, as a function of the
crack growth, are represented in Fig. 19; they show that along the
crack paths, the mode I stress intensity factor increases slowly
while the mode II stress intensity factor tends to zero very rapidly.
Furthermore, the residual strength decreases slowly along the
crack paths, as a direct consequence of the stress intensity factor
variation.
As it can be seen in Fig. 19, all the crack tips were simulta-
neously dominant during the first 3 crack-extension increments
188 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195

which corresponds to 1:52199  103 loading cycles. In the next predict the growth, interaction, and coalescence of the various
9 increments which corresponds to 2:28401  103 loading cycles, cracks in the panel. The analysis provides details for the manner in
the inner crack tips played the role of dominant cracks while the which MSD cracks grow and coalesce into lead cracks that can lead
outer crack tips were the dominant ones from the 13th increment to final failure. The ligament-yield criterion for crack linkup
until the end of the growth analysis. accounts for the significant plastic zone ahead of the crack tips
which limit the ability to carry additional load. Therefore, a critical
condition of the panel is reached when the plastic zone at the tip
9.5. Plate with an array of cracked fastner holes
of a growing crack touches the plastic zone of the nearest
This application is concerned with the fatigue analysis of approaching crack tip from an adjacent hole. When this condition
occurs, a new lead crack is generated and the failure of the panel
multiple site damage (MSD) at a row of holes in a finite width
can be predicted as the new lead crack can unzip through the
panel. The influence of two MSD scenarios in the cracked panel is
remaining array of holes.
studied with the DBEM incremental analysis of crack growth to
Consider a rectangular panel of width w and height h, with an
array of three cracked fastener holes of diameter d, with an arbi-
trary arrangement of small cracks with the reference lengths a1, a2,
a3, a4 and a5, under a uniform traction σ applied at the ends of the
plate, in the direction perpendicular to the crack axis as schema-
tically represented in Fig. 20. The plate is assumed with the elastic
constants E¼ 73.1 GPa, ν ¼ 0.33 and yield stress σ y ¼ 0:345 GPa.
The holes are assumed to contain various combinations of initial
radial through-thickness small cracks which are referred to as
multiple-site damage (MSD), while the term lead crack is used to
describe a large crack connecting two or more holes. Paris equa-
tion is used for fatigue analysis with the parameters C ¼ 4:624 
10
12 and m ¼3.3 and with the stress-amplitude ratio of the
loading cycle R ¼2/3. All the cracks are modelled with 4 dis-
continuous quadratic elements on each crack face graded towards
the respective tips with the ratios 0.4, 0.3, 0.2 and 0.1. The stress
intensity factors were obtained with the J-integral contour path
number 2 with 32 points. A maximum value of the crack-extension
increment equal to 4 times the length of the smallest crack-tip
element was specified for dominant cracks, in a total number of 20
Fig. 14. Rectangular plate with an internal branching crack (H/W¼ 1, a/W ¼0.025
and 18 crack-growth increments that respectively correspond to
and b/a¼ 1) and remote uniform traction σ. two different analyses. For the sake of the simplicity, only two
different MSD scenarios of the starting crack configurations are
studied, aiming to determine how the assumed MSD scenarios can
Table 5 affect the life of the panel, as well as the trends in the crack growth
Normalized stress intensity factors for the branching crack. and coalescence.
pffiffiffiffiffi
Mode – Tip K=ðσ πcÞ
9.5.1. First MSD scenario
J-integral Chen and Hasebe [53] The initial crack geometry of this scenario, defining the size of
the initial cracks and their location relative to the holes of the
KI – A 1.044 1.046 panel, schematically represented in Fig. 20, was considered with
KI – B 0.495 0.495
KII – B 0.506 0.508
the lengths a1 =w ¼ 0:113, a2 =w ¼ 0:045, a3 =w ¼ 0:08, a4 =w ¼ 0:103
and a5 =w ¼ 0:025 as shown in Fig. 21.

Fig. 15. The final shape of the branching crack, after a fatigue crack-growth of 4; 780; 649  106 loading cycles: (a) deformed cracked plate and (b) crack paths.
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
Fig. 16. Stress intensity factors, residual strength and life of the branching crack, along a fatigue crack-growth of 4; 780; 649  106 loading cycles: (a) SIF – crack-tip A; (b) SIF
– crack-tip B; (c) RS – crack-tip A; and (d) RS – crack-tip B.
Fig. 17. Rectangular plate with two skew-parallel embedded cracks ðh=b ¼ 1; 2a=
d ¼ 0:2; e=f ¼ 3:0Þ and applied remote stress σ.
The path and the respective fatigue life of the MSD cracks of the
panel, along 20 crack-growth increments, are represented in
Fig. 22, where it can be seen that cracks start growing in a slow
stable manner with dominant cracks growing respectively from
the right side of the second hole (crack number 3) and left side of
the third hole (crack number 4). These cracks perform the first
189
Table 6
Normalized stress intensity factors for the two skew-parallel embedded cracks.
pffiffiffiffiffiffi
Mode – Tip K=ðσ πaÞ
J-integral Murakami [54]
KI – A 1.013 1.009
linkup of the panel after 63,068,071 loading cycles, in 7 crack-
growth increments, resulting in a lead crack with tips at crack
number 2 and crack number 5, as shown in Fig. 23. This linkup was
performed with a Dugdale yield ligament, normalized by the plate
width w, equal to 0.011278 that is about 10 times bigger than the
crack-tip ligament. This process continues with the lead crack
presenting a highly accelerated growth. Now, the crack on the
right side of the first hole (crack number 1) also grows quickly in
the presence of the lead crack tip to the left of the second hole
(crack number 2). These cracks perform the second linkup of the
panel after 68,095,920 loading cycles which correspond to 16
crack-growth increments, resulting in a much longer lead crack
rooted in the first hole with its tip on the right side of the third
hole (crack number 5), as shown in Fig. 24. This linkup was per-
formed now with a Dugdale yield ligament, normalized by the
plate width w, equal to 0.011172 which is about twice the crack-tip
ligament.
The stress intensity factors and the residual strength of the
cracks growing on the right side of the first and third holes,
190 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195

Fig. 18. The final shape of the skew-parallel embedded cracks, after a fatigue crack-growth of 2:40974  103 loading cycles, considering da=dN ¼ 4:624  10
12 K 3:3 :
(a) deformed cracked plate and (b) crack paths.

Fig. 19. Stress intensity factors, residual strength and fatigue life of the skew-parallel embedded cracks, along a fatigue crack-growth of 2:40974  103 loading cycles,
considering da=dN ¼ 4:624  10
12 K 3:3 : (a) SIF – crack-tip A; (b) SIF – outer crack-tip; (c) RS – crack-tip A; and (d) RS – outer crack-tip.

respectively cracks number 1 and number 5, are presented in
Figs. 25 and 26, where it can be clearly seen the strong impact of
the linkups on the cracks growing from adjacent holes; these
cracks were strongly influenced by the presence of the lead crack,
in both linkups, through the discontinuities shown in Figs. 25 and
26. Before the initial cracks have been linking, the crack on the
right side of the 3rd hole (crack number 5) has grown
comparatively slower, as can be seen in Fig. 22. However, after the
2nd linkup this crack grows very fast, as Fig. 22 shows, and can
lead to the failure of the panel.
This MSD scenario presented the unzip of a lead crack from the
third hole into the first hole. The patterns described in this MSD
scenario suggest that crack growth and coalescence will tend to
result in a much larger crack which will control the failure of the
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195 191

Fig. 20. Array of cracked fastner holes ðh=w ¼ 0:5; d=w ¼ 0:05Þ with remote uni-
form traction σ and crack reference numbers.

Fig. 21. Initial geometry of the first MSD scenario.

Fig. 23. Deformed configurations of the panel – 1st linkup, in the first MSD sce-
nario: (a) after 62,179,451 loading cycles and (b) 1st linkup, after 63,068,071
loading cycles.

Fig. 22. Paths and fatigue life of the first MSD scenario, along 20 crack-growth
increments.

panel as it extends through other holes. The presence of a lead

crack has a strong effect on the cracks located immediately adja-
cent to it and therefore this is the most damaging crack
configuration.

9.5.2. Second MSD scenario

The initial crack geometry of this scenario, defining the size of
the initial MSD cracks and their location relative to the holes of the
panel, schematically represented in Fig. 20, was considered with
the normalized lengths a1 =w ¼ 0:118, a2 =w ¼ 0:065, a3 =w ¼ 0:05,
a4 =w ¼ 0:098 and a5 =w ¼ 0:018 as shown in Fig. 27.
The path and the respective fatigue life of the MSD cracks of the
panel, along 18 crack-growth increments that is 86,334,740 load-
ing cycles, are represented in Fig. 28, where it can be seen that
cracks start growing in a slow stable manner with dominant
cracks now growing respectively from the right side of the first
hole (crack number 1) and left side of the second hole (crack
number 2). These two cracks perform the first linkup of the panel
after 76,869,823 loading cycles, in 6 crack-growth increments,
resulting in a lead crack with the tip at crack number 3, as shown
in Figs. 29. This process continues with the lead crack presenting a
highly accelerated growth. Now, the crack on the left side of the
third hole (crack number 4) also grows quickly in the presence of
the lead crack tip to the right of the second hole (crack number 3). Fig. 24. Deformed configurations of the panel – 2nd linkup, in the first MSD sce-
These cracks perform the second linkup of the panel after nario: (a) after 67,959,671 loading cycles and (b) 2nd linkup, after 68,095,920
85,749,344 loading cycles which correspond to 15 crack-growth loading cycles.
192 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195

Fig. 25. SIF of the cracks growing on the right side of the first and third holes, in Fig. 26. RS of the cracks growing on the right side of the first and third holes, in the
the first MSD scenario: (a) SIF – crack of the 1st hole and (b) SIF – crack of the 3rd first MSD scenario: (a) RS – crack of the 1st hole and (b) RS – crack of the 3rd hole,
hole, right side. right side.

increments, resulting in a much longer lead crack rooted in the

first hole with its tip on the right side of the third hole (crack
number 5), as shown in Fig. 30.
The stress intensity factors and the residual strength of the
cracks growing on the right side of the first and third holes,
respectively cracks number 1 and number 5, are presented in
Figs. 31 and 32, where it can be clearly seen the strong impact of
the linkups on the cracks growing from adjacent holes; these
cracks were strongly influenced by the presence of the lead crack,
Fig. 27. Initial geometry of the second MSD scenario.
in both linkups, through the discontinuities shown in Figs. 31 and
32. Before the initial cracks have been linking, the crack on the
immediately adjacent to it and therefore this is the most damaging
right side of the 2nd hole (crack number 3) has grown compara-
crack configuration.
tively slower, as can be seen in Fig. 28. However, between the 1st The patterns described in both MSD scenarios suggest that
and the 2nd linkup this crack grows very fast, as Fig. 28 shows, regardless of the initial crack configuration, crack growth and
until the occurrence of the 2nd linkup. coalescence will tend to result in a much larger crack, which will
This MSD scenario presented the unzip of a lead crack from the control the expected failure of the panel as it extends through
first hole into the third hole and enforces the conclusions drawn in other holes. Additionally, the number of cycles required to the
the first MSD scenario; the patterns described in this MSD scenario expected failure of the panel, after the larger crack has reached
suggest that crack growth and coalescence will tend to result in a some critical length, is a small portion of the total panel life, and in
much larger crack which will control the failure of the panel as it real situations is certainly affected by the initiation time of the
extends through other holes. Once the unzipping process starts, initially uncracked holes. Based on this pattern the life of the panel
MSD cracks grow very quickly and can cause the panel failure. The is dependent on how long it takes for a particular MSD crack
presence of a lead crack has a strong effect on the cracks located configuration to generate the larger critical crack.
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
Fig. 28. Paths and fatigue life of the second MSD scenario, along 18 crack-growth
increments that is 86,334,740 loading cycles.
Fig. 29. Deformed configurations of the panel – 1st linkup, in the second MSD
scenario: (a) after 73,615,010 loading cycles and (b) 1st linkup, after 76,869,823
loading cycles.
10. Conclusions
The DBEM is applied to the fatigue analysis of multiple-cracked
plates for the simple case of constant amplitude loading cycles. For
each increment of the crack extension, a stress analysis is carried
out using DBEM and the stress intensity factors are evaluated by
means of the J-integral technique. This basic computational step is
repeated for a specified number of crack-extension increments.
General mixed-mode crack problems can be solved in a single-
region analysis with the DBEM. This feature constitutes a practical
advantage of the dual boundary element method over domain-
type methods and multi-region boundary element methods,
because remeshing, extensively used in these methods, is no
longer required when crack-extension increments are modelled
with new elements. The new boundary elements, introduced for
the discretization of each crack extension increment, simply
193
Fig. 30. Deformed configurations of the panel – 2nd linkup, in the second MSD
scenario: (a) after 85,554,032 loading cycles and (b) 2nd linkup, after 85,749,344
loading cycles.
generate new algebraic equations and introduce new unknowns in
the system of equations already existing.
The reliability of the whole incremental analysis process
depends heavily on the accuracy of SIF evaluation, as well as on
the accuracy of the prediction of the direction and extension of the
crack-growth increments. The J-integral technique, applied with a
circular contour around each crack tip, is used for the evaluation of
the stress intensity factors. Since the exact variation of the interior
elastic field is built into the fundamental solution of the boundary
element method, this technique is highly accurate, as shown by
the numerical results obtained in this paper. A simple prediction–
correction procedure was adopted for the definition of the actual
direction of the multiple interacting cracks in each extension
increment; it predicts the local tangent direction of the continuous
crack path using the maximum principal stress criterion and then
introduces a global correction to this direction to account for
discreteness of the crack path, with information derived from one
step ahead of the current crack-extension increment. It is of fun-
damental importance to realize that in a mixed-mode problem,
uniqueness of the crack paths cannot be obtained if the direction
of each crack-extension increment does not take into account the
discreteness of the crack path. The extension of cracks at each tip,
which are also corrected incrementally, is modelled conveniently
with new boundary elements. The real curved path of crack
growth is simulated by the piece-wise linear crack increments.
Crack linkup of adjacent cracks is dealt with the ligament yield
criterion. The number of loading cycles of the current analysis
increment is computed with Paris law, for dominant cracks. With
this value and still with Paris law, the extension of the multiple
cracks is also corrected. This incremental analysis allows for crack
coalescence, sleeping and restarting.
194 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
Fig. 31. SIF of the cracks growing on the right side of the second and third holes, in
the second MSD scenario: (a) SIF – crack of the 2nd hole, right side and (b) SIF –
crack of the 3rd hole, right side.
Results of this incremental analysis of fatigue multiple-crack
growth were presented for several geometries, demonstrating the
excellent performance of this model in simulating mixed-mode
multiple-crack growth. The sensitivity of the results to the size of
the crack-extension, based on the initial mesh, was investigated.
The tested cases produced well correlated crack paths, with some
minor differences due to the discretization error in the
coarser model.
The linkup analysis of two MSD scenarios provided details for
the manner in which MSD fatigue cracks, regardless of the initial
crack configuration, grow and coalesce into lead cracks that con-
trol the expected failure of the panel as it extends through other
holes. Additionally, the number of cycles required to the expected
failure of the panel, after the larger crack has reached some critical
length, is a small portion of the total panel life. Based on this
analysis the life of the panel is dependent on how long it takes for
a particular MSD crack configuration to generate the larger
critical crack.
As a final conclusion, the results obtained in this paper
demonstrate the accuracy and efficiency of the strategies adopted
in the analysis.
Fig. 32. RS of the cracks growing on the right side of the second and third holes, in
the second MSD scenario: (a) RS – crack of the 2nd hole, right side and (b) RS –
crack of the 3rd hole, right side.
Acknowledgements
The first author acknowledges the program PECC – Pós-
Graduação em Estruturas e Construção Civil, Department of Civil
Engineering, Faculty of Technology, University of Brasília. He also
acknowledges CAPES – Coordenação de Aperfeiçoamento de Pessoal
de Nível Superior for his Master scholarship.
References
[1] Brebbia CA. The boundary element method for engineers. London: Pentech
Press; 1978.
[2] Brebbia CA, Telles JC, Wrobel LC. Boundary element techniques. Berlin:
Springer; 1984.
[3] Bueckner HF. Field singularities and related integral representations. In: Sih
GC, editor. Mechanics of fracture, vol. 1. Leyden, The Netherlands: Nordhoff;
1973.
[4] Watson JO. Hermitian cubic and singular elements for plane strain. In: Bane-
rjee PK, Watson JO, editors. Developments in boundary element methods, vol.
4. Barking, UK: Elsevier; 1986.
[5] Portela A, Aliabadi MH, Rooke DP. The dual boundary element method:
effective implementation for crack problems. Int J Numer Methods Eng
1992;33:1269–87.
[6] Hong H-K, Chen JT. Derivation of integral equations in elasticity. J Eng Mech
(ASCE) 1988;114(6):1028–44.
 E. Santana, A. Portela / Engineering Analysis with Boundary Elements 64 (2016) 176–195
[7] Chen JT, Hong H-K. Review of dual boundary element methods with emphasis
on hypersingular integrals and divergent series. Appl Mech Rev (ASME)
1999;52(1):17–33.
[8] Blandorf GE, Ingraffea AR, Ligget JA. Two-dimensional stress intensity factor
computations using the boundary element method. Int J Numer Methods Eng
1981;17:387–404.
[9] Portela A, Aliabadi MH, Rooke DP. Dual boundary element analysis of cracked
plates: singularity subtraction technique. Int J Fract 1992;55:17–28.
[10] Liu J, Beer G, Meek JL. Efficient evaluation of integrals of order 1/r, 1=r 2 and
1=r 3 using Gauss quadrature. Eng Anal 1985;2(3):118–23.
[11] Brahtz JHA. Stress distribution in a Reentrant corner. Trans Am Soc Mech Eng
1933;55:31–7.
[12] Williams ML. Stress singularities resulting from various boundary conditions
in angular corners of plates in extension. J Appl Mech 1952:526–8.
[14] Aliabadi MH, Rooke DP, Cartwright DJ. An improved boundary element for-
mulation for calculating stress intensity factors: application to aerospace
structures. J Strain Anal 1987;22(4):1–5.
[15] Portela A, Aliabadi MH, Rooke DP. Efficient boundary element analysis of sharp
notched plates. Int J Numer Methods Eng 1991;32:445–70.
[17] Alatawi IA, Trevelyan J. A direct evaluation of stress intensity factors using the
extended dual boundary element method. Eng Anal Bound Elem 2015;52:56–
63.
[18] Tong P, Pian THH, Lasry SJ. A hybrid element approach to crack problems in
plane elasticity. Int J Numer Methods Eng 1973;7:297–308.
[19] Karihaloo BL, Xiao QZ. Accurate determination of the coefficients of elastic
crack tip asymptotic field by a hybrid crack element with p-adaptivity. Eng
Fract Mech 2001;68:1609–30.
[20] Xiao QZ, Karihaloo BL, Liu XY. Direct determination of SIF and higher order
terms of mixed-mode cracks by a hybrid crack element. Int J Fract
2004;125:207–25.
[21] Melenk JM, Babuska I. The partition of unity finite element method: basic
theory and applications. Comput Methods Appl Mech Eng 1996;139:289–314.
[22] Chahine E, Laborde P, Renard Y. Crack-tip enrichment in the XFEM method
using a cut-off function. Int J Numer Methods Eng 2008;75(6):629–46.
[23] Liu XY, Xiao QZ, Karihaloo BL. Xfem for direct evaluation of mixed mode stress
intensity factors in homogeneous and bi-materials. Int J Numer Methods Eng
2004;59:1103–18.
[24] Ben Dhia H, Rateau G. The Arlequin method as a flexible engineering design
tool. Int J Numer Methods Eng 2005;62:1442–62.
[25] Guidault PA, Allix O, Champaney L, Cornuault C. A multiscale extended finite
element method for crack propagation. Comput Methods Appl Mech Eng
2008;197(5):381–99.
[26] Belytschko T, Song JH. Coarse-graining of multiscale crack propagation. Int J
Numer Methods Eng 2009;81(5):537–63.
[27] Kim DJ, Pereira JP, Duarte CA. Analysis of three-dimensional fracture
mechanics problems: a two-scale approach using coarse-generalized fem
meshes. Int J Numer Methods Eng 2010;81:335–65.
[28] Belytschko T, Black T. Elastic crack growth in finite elements with minimal
remeshing. Int J Numer Methods Eng 1999;45:601–20.
[29] Chessa J, Wang HW, Belytschko T. On the construction of blending elements
for local partition of unity enriched finite elements. Int J Numer Methods Eng
2003;57:1015–38.
[30] Dong L, Atluri SN. Fracture and fatigue analyses: SGBEM-FEM or XFEM? Part 1:
2D structures. Comput Model Eng Sci 2013;90(2):91–146.
[31] Sladek J, Sladek V, Wunsche M, Zhang C. Interface crack problems in aniso-
tropic solids analyzed by the MLPG. Comput Model Eng Sci 2009;54:223–52.
[32] Carpinteri A, Ferro G, Ventura G. The partition of unity quadrature in element-
free crack modeling. Comput Struct 2003;81:1783–94.
[33] Wen PH, Aliabadi MH. Applications of meshless method to fracture mechanics
with enriched radial basis functions. Durab Struct Health Monit 2007;3:107–
19.
[34] Belytschko T, Krongauz Y, Organ D. Meshless methods: an overview and
recent developments. Comput Methods Appl Mech Eng 1996;139(14):347.
[35] Li S, Liu WK. Meshfree and particle methods and their applications. Appl Mech
Rev 2002;54:1–34.
195
[36] Liu GR, Gu YT. An introduction to meshfree methods and their programming.
Berlin: Springer Press; 2005.
[37] Fleming M, Chu Y, Moran B, Belytschko T. Enriched element-free Galerkin
methods for crack tip fields. Int J Numer Methods Eng 1997;40:1483–504.
[38] Lu YY, Belytschko T, Tabbara M. Element-free Galerkin method for wave
propagation and dynamic fracture. Comput Methods Appl Mech Eng
1995;126:131–53.
[39] Gu YT, Zhang LC. Coupling of the meshfree and finite element methods for
determination of the crack tip fields. Eng Fract Mech 2008;75(5):986–1004.
[40] Henshell RD, Shaw KG. Crack tip elements are unnecessary. Int J Numer
Methods Eng 1975;9:495–509.
[41] Barsoum RS. On the use of isoparametric finite elements in linear fracture
mechanics. Int J Numer Methods Eng 1976;10(1):25–37.
[42] Cruse TA, Wilson RB. Boundary integral equation method for elastic fracture
mechanics analysis. AFOSR-TR-780355, Pratt and Whitney Aircraft Group;
1977.
[43] Blandford GE, Ingraffea AR, Liggett JA. Two-dimensional stress intensity factor
computations using the boundary element method. Int J Numer Methods Eng
1981;17:387–404.
[44] Smith RNL, Mason JC. A boundary element method for curved crack problems
in two-dimensions. In: Brebbia CA, editor. Proceedings of fourth international
seminar on BEM. Berlin: Springer-Verlag; 1987.
[45] Martinez J, Dominguez J. On the use of quarter-point boundary elements for
stress intensity factor computations. Int J Numer Methods Eng 1984;20:1941–
50.
[46] Harrop LP. The optimum size of quarter-point crack tip elements. Int J Numer
Methods Eng 1982;18:1101–3.
[47] Symm GT. Integral equation methods in potential theory, II. Proc R Soc 1963;
A275:33–46.
[48] Xanthis LS, Bernal MJM, Atkinson C. The treatment of the singularities in the
calculation of stress intensity factors using the integral equation method.
Comput Methods Appl Mech Eng 1981;26:285–304.
[49] Aliabadi MH. An enhanced boundary element method for determining frac-
ture parameters. In: Proceedings of 4th international conference on numerical
methods in fracture mechanics. San Antonio, Texas: Pineridge Press; 1987. p.
27–39.
[50] Erdogan F, Sih GC. On the crack extension in plates under plane loading and
transverse shear. J Basic Eng 1963;85:519–27.
[51] Broek D. Elementary engineering fracture mechanics. Dordrecht: Martinus
Nijhoff; 1986.
[52] Broek D. The practical use of fracture mechanics. London: Kluwer Academic
Publishers; 1988.
[53] Chen YZ, Hasebe N. New integration scheme for the branch crack problem.
Eng Fract Mech 1995;52(5):791–801.
[54] Murakami Y, editor. Stress intensity factors handbook, 1st edition, vol. 2.
Pergamon Press; 1987.
[55] Swift T. Widespread fatigue damage monitoringissues and concerns. In: Pro-
ceedings of 5th international conference on structural airworthiness of new
and ageing aircraft, Hamburg. 1993. p. 113–50.
[56] Broek D, Jeong DY, Thomson D. Testing and analysis of flat and curved panels
with multiple cracks. In: Proceedings of FAA/NASA international symposium
on advanced structural integrity methods for airframe durability and damage
tolerance. NASA Conference Publication 3274, 1994. p. 85–98.
[57] De Wit R, Fields RJ, Low III SR, Harne DE, Foecke T. Fracture testing of large
scale thin-sheet aluminum alloy. National Institute of Standards and Tech-
nology, NISTIR 5661, Gaithersburg, MD, USA, 1995.
[58] Dugdale DS. Yielding of steel plates containing slits. J Mech Phys Solids
1960;8:100–8.
[59] Smith BL, Saville PA, Mouak A, Myose RY. Strength of 2024-T3 aluminum
panels with multiple site damage. J Aircr 2000;37(2):325–31.
[60] Nilsson KF. Elasto-plastic models for interaction between a major crack and
multiple small cracks. In: Proceedings of FAA/NASA symposium on continued
airworthiness of aircraft structures, Atlanta, 1996.