Surface Integral and Finite

Element Hybrid Method for

W. D. Keat
D. M. Maybury
Department of Mechanical and
Aeronautical Engineering,
Clarkson University,
Potsdam, NY 13699-5725
B. S. Annigeri
United Technologies Research Center,
East Hartford, CT 06040
Three-Dimensional Analysis of
Arbitrarily Shaped Surface
Cracks
A three-dimensional surface integral and finite element hybrid method has been
developed for modeling arbitrarily shaped surface cracks in complex structural com-
ponents. Accurate stress intensity factors were obtained by decomposing the problem
into a surface integral model of the fracture in a domain of infinite extent and a finite
element model of the uncracked domain. Boundary conditions were enforced by
applying corrective traction to the surfaces of both constituent models. Coupling
between the two formulations was minimized by implementing the fundamental solu-
tion for a force multipole near a planar free surface. Surface cracks intersecting
nonplanar free surfaces were modeled in a piecewise linear fashion by deploying
multiple sets of these functions. The effectiveness of this approach was demonstrated
for well-documented cases, including a corner crack in a thick plate and a three-
dimensional edge crack. The results of these benchmark studies will be used to
develop a set of heuristics for assuring suitable finite element mesh densities in the
vicinity of the fracture.

Introduction hybrid formulation to model embedded cracks in finite bodies

of arbitrary shape (Keat et al., 1988). The same multipole
Alternative computational methods for three-dimensional
solutions proved ineffective at modeling surface cracks because
fracture analysis range from finite elements, boundary elements,
the strong coupling induced between the two formulations re-
and weight function techniques, to line spring models and the
quired finite element meshes so refined as to negate the advan-
surface integral method. It is fair to conclude that each method
tages of a hybrid approach. The fundamental solutions for
can be identified with its own unique set of strengths and weak-
multipoles in a half-space were thus introduced to handle the
nesses. The rationale for the hybrid method discussed here is
problem of a surface crack intersecting a planar free surface in
that each of the methods listed above has attributes, often not
a finite body (Annigeri et al, 1988). This paper further general-
duplicated by the other methods, that can contribute to an effec-
izes the hybrid method by considering surface cracks inter-
tive three-dimensional fracture analysis.
secting free surfaces of arbitrary shape.
The surface integral and finite element hybrid method (SI-
FEH) employs the principle of superposition to combine a sur-
face integral model of the fracture in an infinite (or semi-infi- Surface Integral Formulation
nite) region with a finite element model of the uncracked finite The surface integral method (Cleary, 1977) represents the
body. The two formulations are coupled through the need to fracture as a distribution of force multipoles, which through
enforce boundary conditions on the external boundaries and on Hooke's law may equivalently be expressed in terms of dis-
the fracture surface. The coupling is generally weak so that the placement discontinuities. The key feature of the method is that
two meshes can be set up independently. Hence, when conduct- only the surface of the fracture needs to be discretized.
ing a crack propagation study, only the surface integral discreti- The traction induced at a point on the fracture can be found
zation of the fracture surface needs to be remeshed; the volume by summing the effects of the multipoles acting over the crack
discretization associated with the finite element model remains surface, Sc. If expressed in terms of components of relative
fixed. One may note that for a linear elastic analysis, the use displacement 8k rather than dipole intensity, the following inte-
of boundary elements would appear to have been an equally gral equation results:
valid choice for representing the finite body. However, the finite
element method was retained here in anticipation of future mod-
eling of material nonlinearities and thermal effects.
The generality of this hybrid method depends to a large de-
-11m l)nj6kdA

where the fundamental solution has been decomposed into sin-

(1)

gree on the library of fundamental solutions that are available gular (Ts) and nonsingular parts ( r w s ) to facilitate the use of
for use with the surface integral method. The multipole solutions specialized integration procedures, «, is the unit normal to the
corresponding to a fracture event in an infinite region are well fracture surface, and r, is the traction at a point on the fracture.
documented and have been used with the three-dimensional A technique known as subtraction of the singularity was in-
voked to render the first part integrable. In effect, a rigid body
displacement was subtracted from the singular integrand,
Contributed by the International Gas Turbine Institute and presented at the 40th
International Gas Turbine and Aeroengine Congress and Exhibition, Houston,
allowing it to be integrated in a Cauchy principal-value sense:
Texas, June 5 - 8 , 1995. Manuscript received by the International Gas Turbine
Institute March 23, 1995. Paper No. 95-GT-456. Associate Technical Editor:
C. J. Russo.
U =
Jl Tsmn0k - 6k)dA +
Jl Tfknj8kdA (2)

406 / V o l . 118, APRIL 1996 Transactions of the ASME

Copyright © 1996 by ASME
Downloaded From: http://gasturbinespower.asmedigitalcollection.asme.org/ on 11/30/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use
 where Sk is the relative displacement at the point of collocation.
The domain of the first integral was redefined to be a plane of
infinite extent, ST, consistent with the integral representation of
the rigid body displacement. The numerical problem of solving
for the unknown crack opening/shear magnitudes may thus be
formulated as a system of equations, each of the form of Eq.
(2), which enforce the known traction boundary conditions at
as many points on the crack surface as there are unknown
parameters defining the crack opening variation.
Selection of the appropriate interpolation functions for crack
opening/shear is crucial to the overall accuracy of the formula-
tion. The variation of crack opening within each element of the
fracture discretization was assumed to have the following form:
L {TC} = [S]{U}
6k = I h'8[ (3)
where h' are the interpolation functions, and 6'k are nodal values
of crack opening. In the present study, all crack opening nodes
were assumed to be situated at collocation points located in the
interior of the elements, leading to the result that opening/shear
variations between elements are discontinuous. This assumption
simplified some of the analytical integrations and was justified
by the high accuracy that has been consistently observed in
benchmark studies (Keat, 1989). The interpolation functions
for crack opening assigned to the interior and perimeter crack
elements are represented in Eqs. (4) and (5), respectively:
6k = C, + C2x + C3y + Ctxy (4)
m
6k = C5p (5)
where the C, are constants that depend on the crack opening/
shear nodal values and their locations; p is the perpendicular
distance from the crack front. Equation (4) corresponds to a
four-noded (or linear) element. A constant element is also
sometimes used in the interior and can be obtained from Eq.
(4) by truncating the last three terms. The form of Eq. (5) was
extracted from the asymptotic solution for a crack in a plane
strain, homogeneous region.
Hybrid Formulation
The basic idea behind the hybrid formulation as originally
developed by Annigeri and Cleary (1984) is to decompose the
problem of a crack in a finite body into: (1) a finite element
model of the finite body without the crack and (2) a surface
integral model of the crack in an infinite (or semi-infinite)
region. For the superposition of these models to represent a
valid solution, the boundary conditions in the original problem
must be satisfied. Toward this end, the finite element model is
used to enforce all exterior boundary conditions while simulta-
neously the surface integral model is used to enforce boundary
conditions on the crack surfaces. Enforcement of the boundary
conditions requires the calculation of two correction vectors,
here denoted as { T"} and {R c }. They are needed to cancel out
the nonzero tractions occurring in the two models where sur-
faces have not been explicitly represented.
Thus the system of equations associated with the finite ele-
ment model may be stated as follows:
[K]{ U} = {R}- {Rc} (6)
where [K] is the finite element stiffness matrix, { U) is the nodal
displacement vector, {R} is the external nodal load vector. The
correction load vector {Rc} consists of nodal forces that are
statically equivalent (but opposite in sign) to the tractions oc-
curring in the surface integral model on surfaces coincident
with the external boundaries. It can therefore be expressed in
terms of the degrees of freedom associated with the surface
integral model, i.e., the crack opening displacements {6}:
[R'} = [G]{6] (7)
Journal of Engineering for Gas Turbines and Power
Downloaded From: http://gasturbinespower.asmedigitalcollection.asme.org/ on 11/30/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use
The system of equations defining the surface integral model
is generated using boundary collocation:
[C]{6] = [T] - {D (8)
where [C] is the surface integral coefficient matrix, [T] are
traction boundary conditions at the collocation point positions.
The correction traction vector {Tc} is equal (but opposite in
sign) to tractions occurring in the finite element model at posi-
tions coincident with the collocation point locations. It can
therefore be expressed in terms of the degrees of freedom asso-
ciated with the finite element model, i.e., the finite element
nodal displacements { U}:
(9)
Substituting Eqs. (7) and (9) into Eqs. (6) and (8), respec-
tively, and writing the results in partitioned matrix form leads
to:
~K G l f I/I (R)
(10)
.s cJUrW
which can be solved directly for the unknown nodal displace-
ments and crack openings. Equation (10) can be further modi-
fied to allow for the imposition of displacement boundary condi-
tions on the external boundaries (Annigeri and Cleary, 1984).
Accurate Modeling of Surface Flaws
The original version of the hybrid formulation, with its reli-
ance on the multipole solutions for an infinite region, proved
ineffective at modeling surface cracks, as successive finite ele-
ment mesh refinements failed to lead to monotonic convergence.
The coarse finite element meshes used with these early models
were unable to capture the singular Rc fields produced along
the line of intersection of the fracture with the free surface.
To reduce the coupling between the finite element and surface
integral models, fundamental solutions for multipoles in a half-
space were derived (Keat, 1989) by a process of differentiation
starting from the Papkovitch functions of Rongved (1955). Use
of these functions with the hybrid formulation eliminated the
Rc field on the free surface explicitly represented in the half-
space solution; the Rc fields were applied to the remaining faces
in the usual manner.
Because of the added cost involved in using the fundamental
solutions for the half-space, a criterion is needed to determine
whether a given crack element lies close enough to a planar
interface to warrant the use of these functions. Any valid test
must take into consideration the global dimensions of the frac-
ture. To understand this, assume that a given crack element has
a small enough nondimensional depth to classify it as being a
near-interface fracture event. The nonuniqueness, and thus the
fallibility, of this test can be observed if the element is subdi-
vided into smaller and smaller elements, until each subelement
has a large enough nondimensional depth to overturn its original
classification.
A test based on the following ratio has been proposed:
IS, T"dA
Kz) = -7ff (11)
dA
))J'
where the fundamental solution has been resolved into singular
(Vs) and nonsingular (VNS) components, which are integrated
over the surface of the fracture. For a given crack geometry
and material property set, this ratio will depend exclusively
upon the location of the field point, with the perpendicular
distance (z) of the field point from the interface having the
primary influence. Equation (11) has been used in a Newton-
APRIL 1996, Vol. 1 1 8 / 4 0 7
 H—i—i—i——i—i—i—i—M—i—m—f~i—i—i—i——'—'—i—r

idipole
Error
in Rc

I I I

0
z/D
Fig. 1 Effect of dipole depth on the error induced in the finite element representation of the R° field. Insert
shows simplified geometry used to evaluate the error analytically.

Raphson algorithm to search iteratively for the critical depth at dimensionless dipole depth in Fig. 1. Note that this result does
which the fundamental solution for the half-space should be- not take into account the level of mesh refinement in the z
come active. direction and thus does not address the accuracy of Tc.
In a hybrid analysis, the largest value of r(z) that can be
tolerated, without sacrificing accuracy, will depend to a large Computational Strategy for Cracks Intersecting Mul-
extent on the sizing of the finite element mesh. An estimate of tiple Free Surfaces
how well the Rc field can be represented using thefiniteelement
interpolation functions was derived analytically for the simpli- For this class of problem, the fracture is allowed to approach
fied geometry of Fig. 1. In thefigure,an individualfiniteelement and/or intersect more than one planar free surface. Through
surface, over which the Rc field is presumed to occur, has been cracks and corner cracks are typical examples.
idealized by a planar circular area. A tensile dipole is con- The hybrid method was adapted to this class of problem by
strained to lie on a central axis perpendicular to the circular introducing modifications that allow for the use of multiple sets
area. The dipole might represent, for example, the nearest tip of of fundamental solutions. Figure 2 shows the component models
an approaching fracture. The influence function {Vs) associated that would be used to model a fracture situated close to two
with the dipole is assumed to have the following form: planar free surfaces. Two surface integral models, each one
modeling separate halves of the fracture, have been employed
to represent the fracture. Each crack element is assigned to one
(12) of the surface integral models on the basis of which planar free
R3
surface exerts the greatest influence upon it. This influence can
where R is the distance from the dipole to a point on the circular be assessed, for example, by evaluating the ratio in Eq. (11),
c
area. One may note that the resulting R field will become or by calculating the perpendicular distances between the crack
increasingly singular as z is allowed to approach zero. Yet, due element and the free surfaces.
to the symmetry of the configuration and the low order of the All three models are coupled as a result of satisfying traction
finite element interpolation functions (linear with cross terms), and displacement boundary conditions. Each surface integral
the best that the element can do is to represent this variation as model contributes tractions in the form of an {Rc}fieldto those
a uniform stress field of intensity: surfaces of the finite element model that it does not explicitly
model. Also, since each surface integral domain only partially
11 TsdA models the fracture, it will receive corrective tractions from all
(13) of the other models including its surface integral complements.
The corrective tractions {Tc2} and {Tc%} in Fig. 2 are absorbed
which is applied to the surface of the finite element in the in the surface integral coefficient matrix and thus do not alter
form of four equal nodal forces. The error measure is therefore the form of the governing matrix equations.
defined as
Convergence Studies
11 Vs A
Ts\dA
(14)
The convergence studies were aimed at determining appro-
priate meshes first for a corner crack in a thick plate and second
for a three-dimensional edge crack in a thick plate (see Fig. 3).
where Eq. (14) indicates the relative error between the actual Component solutions used with the hybrid formulation to model
and idealized Rc fields. The error is plotted as a function of these geometries are shown in Fig. 4.

408 / Vol. 118, APRIL 1996 Transactions of the ASME

Downloaded From: http://gasturbinespower.asmedigitalcollection.asme.org/ on 11/30/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 Actual P r o b l e m Model I Model II Model III

a .. i\ ,. ..

4"

,
«\^\v •J*.
Finite Element Model of
the bounded domain (no
fracture)
Surface Integral Model of Another Surface Integral
a fracture in a semi-infinite Model of a fracture in a
domain semi-infinite domain

Fig. 2 Component geometries used with the hybrid method to model a fracture near multiple planar surfaces
in a bounded domain

The surface integral discretizations can generally be set up

Comer Crack
without reference to the demands on the finite element model. (a/t=0.8)
Previous experience with embedded fractures (Keat et al., 1988)
and semi-elliptical surface cracks (Annigeri et al., 1988) sug-
gested that the meshes depicted in Fig. 5 would be more than
adequate for representing the elastic behavior of the crack geom- Edge Crack
(a/w=0.2)
etries under consideration. The discretizations were created us-

It
t
/
Edge Crack
(a/w=0.8)

/*
Fig. 5 Surface integral discretizations

ing an automatic mesh generator specially designed for use

with the surface integral models under mode I conditions. Mesh
a»@5@
density was controlled by specifying the number of crack ele-
H H Fraaum wrfaco
V ESS Fradure a u t a *
ments along each edge. Tip elements, defined by Eq. (5), were
EH] MocW iyitmrty p i n * E D Model «ynw«iy plan* employed along the crack fronts; linear and constant elements
were used to model the interiors of the corner and edge cracks,
Fig. 3 Corner crack and edge crack geometries used to assess effec- respectively. Also, each crack element was associated, based
tiveness of hybrid strategy for surface cracks. Plate dimensions are de- on proximity, with one of the half-space solutions depicted in
fined such that: T/W = 0.2.
Fig. 4.
A succession of finite element meshes (see Fig. 6) of increas-
F.B. MODEL
ing density was used to analyze both crack geometries. Uniform
(WITHOUT CRACK) element spacing was used in all three coordinate directions
F.E. MODEL
(WITHOUT CRACK)

IT*= * = * *= ^

\
\
\

c -A
2nd HALF-SPACE
SOLUTION
} (
J
W l 3rd HALF-S
SOLUTION
^
Comer Crack (a/t=0.2)
r « - Corner Crack (a/t=0.8)
! I '
K - Edge Crack (a/w=0.2)

Fig. 4 Cross-sectional view of component geometries used to model Fig. 6 Finite element discretizations corresponding to 5 x 5 x 3 and 25
corner crack (left) and edge crack (right) x 25 x 3 mesh refinements

Journal of Engineering for Gas Turbines and Power APRIL 1996, Vol. 1 1 8 / 4 0 9

Downloaded From: http://gasturbinespower.asmedigitalcollection.asme.org/ on 11/30/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use

 under the desired assumption that the finite element and surface
integral meshes can be set up independently. Furthermore, the
number of finite elements through the thickness of the plate
was held fixed at three, as being typical of what a finite element
modeler would employ with a plate structure of comparable
dimensions. Thus five different finite element meshes were em-
ployed:5 × 5 × 3, 10 × 10 × 3, 15 × 15 × 3, 20 × 20 × 3,
25 × 25 × 3. Expected major source of error is the inability
of the finite element mesh to represent exactly the R c fields.
The results for the convergence study of the corner crack
geometry are displayed in Fig. 7. Crack radius to thickness ratios
( a / t ) of 0.2 and 0.8 were examined. Accuracy was assessed by
comparison with the finite element results of Newman and Raju
( 1981 ) using the error measures defined below:
n t( 1i H - K~E) }
/ KSlFE
Avg. percent dev. = n × 100
i=1
Kli - K~E
Max. percent dev. = max --ff × 100
i=l,n gli
where KS~FEHand K f e are the hybrid and finite element results,
respectively, for the mode I stress intensity factor at the ith tip
element; Wi is a nondimensional weighting factor, which is
proportional to the length of crack front associated with the ith
tip element. Inspection of Fig. 7 indicates that convergence does
occur with the average errors flattening out at about 3 percent
for both a / t ratios. The persistent error at convergence is not
surprising in view of the fact that the crack mesh and the number
of finite elements through the thickness of the plate were both
held fixed throughout the studies. Although the number of finite
elements at convergence may appear excessive, in fact the hy-
brid method is performing well when one considers that the
smaller Corner crack ( a / t = 0.2) is fully contained within a
single element of the 25 × 25 × 3 mesh (refer to Fig. 6). The
same accuracy could have been achieved using many fewer
degrees of freedom by refining the finite element mesh in the
vicinity of the fracture. Thus one may conclude that reference
should be made to the location and proportions of the crack
when setting up the finite element model.
Results for a short three-dimensional edge crack ( a / w = 0.2)
are presented in Fig. 8. Observed rates of convergence are compa-
rable to those for the comer crack. However, when the same
studies were conducted for a long edge crack ( a / w = 0.8), conver-
gence could not be attained because of the severity of the R c field.
°
0 ~1 [ =1 0 20 0 2500
Number of Finite Element Nodes
I ~ Average (a,'t=,.2) ~ Maximum(a/t=.2) .-in- Average (a/t=.8) ~ Maximum(a/t.,.8) I
Fig. 7 Percentage deviation of hybrid results from finite element results
of Raju and Newman (1981) for corner crack geometry
410 / Vol. 118, APRIL 1996
Downloaded From: http://gasturbinespower.asmedigitalcollection.asme.org/ on 11/30/2017 Terms of Use: http://www.asme.org/about-asme/terms-of-use
5
o¸
-s ¸
._=
t~ .10
t~
.~ -is
-20 •
-25 , : . . . . , . . . . , ,
0 500 1000 1500 2oo0 2sou
Number of FE Nodes
[~ Average(a/t=.2) -o.- Maximum(a/t=.2) I
Fig. 8 Percentage deviation of hybrid results from results of Brown and
Srawley (1966) for edge crack geometry
This may suggest a need to introduce fundamental solutions that
explicitly represent both faces of an infinite plate.
(15) Summary
The use of superposition provides SIFEH with the flexibility
needed to model surface cracks of complex shape effectively.
Though specialized influence functions had to be introduced to
reduce the degree of coupling between the surface integral and
finite element models, it was sufficient to employ just the half-
space solution as it could be used with superposition to model a
nonplanar boundary in a piecewise linear sense. Results indicate
that though this strategy significantly weakens the coupling be-
tween the two discretizations, the finite element mesh may have
to be refined in the vicinity of the fracture. Additional savings in
CPU time can be obtained by using a method of substructufing
described by Annigeri et al. (1988) to restrict the surface integral
domain to a localized region containing the fracture.
The computational strategy described here is just a partial
solution to the problem of modeling a surface crack of arbitrary
shape. It should be carefully noted that each of the geometries
considered here had two geometrical characteristics in common:
( 1 ) The boundaries intersected by the surface cracks could eas-
ily be modeled as piecewise planar, and (2) all interior angles
of the crossection were less than 180 deg. It would seem on
first glance that both of these conditions must be met in the
vicinity of the fracture for the method to hold up. However,
preliminary results (e.g., for a surface crack at a hole) indicate
that there is the potential for this strategy to work effectively
for a much broader range of problems.
References
Annigeri, B. S., and Cleary, M. P., 1984, "Surface Integral Finite Element
Hybrid Method for Fracture Mechanics," International Journal for Numerical
Methods in Engineering, Vol. 20, pp. 869-885.
Annigeri, B. S., Keat, W. D., and Cleary, M. P., 1988, "Fracture Mdchanics
Research Using the Surface Integral and Finite Element Hybrid Method," Pro-
ceedings of First Joint Japan/U.S. Symposium on Boundary Element Methods,
Pergamon Press, University of Tokyo, pp. 191-202.
Brown, W. F., and Srawley, 1966, "Plane Strain Crack Toughness Testing of
High Strength Metallic Materials," ASTM STP 410, p. 12.
Cleary, M. P., 1977, "Fundamental Solutions for a Fluid-Saturated Porous
Solid," International Journal of Solids and Structures, Vol. 13, pp. 785-806.
Keat, W. D., Annigeri, B. S., and Cleary, M. P., 1988, "Surface Integral and
Finite Element Hybrid Method for Two and Three Dimensional Fracture Mechan-
ics Analysis," International Journal of Fracture, Vol. 36, pp. 35-53.
Keat, W. D., 1989, "Surface Integral and Finite Element Hybrid Method for
the Analysis of Three Dimensional Fractures," Ph.D. Thesis, M.I.T., Cambridge,
MA.
Newman, J. C., and Raju, J. S., 1981, "Stress Intensity Factor Equations for
Cracks in Three-Dimensional Finite Bodies," NASA Technical Memorandum
83200, pp. 1-49.
Rongved, 1955, "Force Interior to One of Two Joined Semi-infinite Solids,"
Proceedings of Second Midwestern Conf. on Solid Mech., pp. 1-13.
Transactions of the ASME