Composite Structures
journal homepage: www.elsevier.com/locate/compstruct
University of Michigan, Department of Mechanical Engineering, 2350 Hayward Street, Ann Arbor, MI 48109, USA
University of Michigan, Department of Aerospace Engineering, FXB Building, 1320 Beal Avenue, Ann Arbor, MI 48109, USA
a r t i c l e
i n f o
Article history:
Available online 11 April 2012
Keywords:
Variational multiscale
VMCM
Cohesive crack propagation
Curved crack
Mixed mode
Laminated ber composite
a b s t r a c t
Predictions of crack propagation is a valuable resource for ensuring structural integrity and damage tolerance of aerospace structures. Towards that end, a variational multiscale approach to predict mixed
mode in-plane cohesive crack propagation is presented here. To demonstrate applicability and to provide
validation of the nite element based predictive methodology, a comparative study of the numerical
results with the corresponding experimental observations of crack propagation in laminated ber reinforced composite panels is presented.
2012 Elsevier Ltd. All rights reserved.
1. Introduction
The aerospace industry relies on extensive testing to ensure the
structural integrity and damage tolerance of vehicle structures, thus
leading to a substantial increase in the total design cost of many
aerospace vehicles. These costs can be reduced by developing high
delity computational models which can provide valuable information regarding the performance of a structure up to and including
failure, provided the modeling is based on material parameters that
can be measured, and is validated using laboratory tests that are
designed to be discriminatory. The nite element method is a key
computational tool in solid mechanics and has become the mainstay
of problems involving any of the broad phenomena of material
deformation elasticity, plasticity and damage. However, its utility
for problems of crack propagation has met with mixed success. The
distinguishing characteristic of crack problems, in general, is the formation and propagation of boundaries, which are not part of the original boundary value problem. These boundaries emerge from
within the domain of the problem conguration, under load, rendering a departure from a traditional continuum description for the
initially solid body, at least across the newly formed boundaries.
This is not an obstacle, if the resulting crack path is known a priori,
and the mesh is ensured to have elemental surfaces align along possible crack surfaces; but often, neither conditions are feasible. For all
but trivial crack propagation problems, the crack path is not known
Corresponding author at: University of Michigan, Department of Mechanical
Engineering, 2350 Hayward Street, Ann Arbor, MI 48109, USA. Tel.: +1 734 764
8227.
E-mail address: dcw@umich.edu (A.M. Waas).
0263-8223/$ - see front matter 2012 Elsevier Ltd. All rights reserved.
http://dx.doi.org/10.1016/j.compstruct.2012.03.035
3337
enhanced nite elements are popularly classied as element enrichment methods [2,9,12,4,17,16,10] and nodal enrichment methods
like XFEM [15,14,30]. Interested readers are referred to Oliver
et al. [18], for detailed discussion and comparison of these methods.
The primary task of this paper is presenting a new multiscale framework, referred to as the Variational Multiscale Cohesive Method
(VMCM), to embed cohesive models into continuum elements
through discontinuous enrichment of the shape functions and demonstrating its effectiveness by simulating mixed-mode failure in ber-reinforced composites.
The paper is organized as follows: Section 2 provides the motivation for a multiscale approach to capture displacement discontinuities involved in crack propagation. Then, Section 3 presents the
variational multiscale formulation for modeling cohesive crack
propagation, followed by details of the nite dimensional implementation in Section 4. Simulations and experimental validation
of the multiscale approach are provided in Section 5 and the concluding remarks follow in Section 6.
2. Crack propagation as a ne scale problem
Physically, crack propagation is a process of congurational
change by which new surfaces are created. The creation of new surfaces is governed by surface laws, different from the constitutive
laws of the continuum. Classically, this process of surface creation
is handled by affecting changes in the numerical discretization,
involving incremental grid renement and remeshing. However,
changing the grid to reect the evolving domain boundaries is computationally very expensive. Instead, an alternative view of cracks
as displacement discontinuities in the continuum domain is considered here. The concept of discontinuous displacement elds and the
resulting singular strains nds its mathematical treatment in the
work of Temam and Strang [28] on BD(X), the space of bounded
deformations for which all components of the strain are bounded
measures. This idea was used to develop a numerical framework
for the problem of strong discontinuities due to strain localization
by Simo et al. [26], Simo and Oliver [25] and Armero and Garikipati
[1]. The physical process of strain localization involves localized
changes in the continuum constitutive response and no new
boundaries and surface laws appear, but its numerical treatment
introduced the use of the distributional framework and discontinuous basis functions, which was adopted in Garikipati [8] for embedding micromechanical surface laws into a macroscopic continuum
formulation, albeit in a multiscale setting. The presentation in this
work follows and extends these multiscale arguments specically
for numerical representation and evolution of cohesive cracks.
As shown in Fig. 1, a crack opening can be represented by a discontinuous displacement eld over an uncracked body. The use of
zero volume elements (interface elements, standard cohesive zone
elements, etc.), to capture such a discontinuous displacement eld,
rw : r dV
w f dV
w T dS
Ch
where f is the body force, g and T are the prescribed boundary displacement and surface traction, respectively. r is the (Cauchy)
stress tensor given by r = C:sym(ru), where C is the fourth-order
elasticity tensor.
Now, scale decompositions of u and w are introduced. The
decompositions are qualied by requiring that the ne scales, u0
and w0 , vanish outside the neighborhood of the crack path, which
is contained in X0 (Fig. 2), referred to as the microstructural or
ne-scale subdomain
u
|{z}
|{z}
u0
2a
w
|{z}
|{z}
w
2b
coarse scale
coarse scale
fine scale
0
fine scale
2 S fv j
u
2c
2 V fv j
w
v g on Cg g
v 0 on Cg g
u0 2 S 0 fv j v 0 on X n intX0 g
w0 2 V 0 fv j v 0 on X n intX0 g
2d
2e
2f
: r dV
rw
X0
f dV
w
rw0 : r dV
X0
T dS
w
3a
Ch
w0 f dV
w0 T dS
C0h
3b
Z
Cc
w0 r n dS
Z
Cc
w0 Tc dS
W0
Fig. 2. The microstructural domain, X0 , and the crack surface, Cc. Shown in the inset
are the crack orientation vectors and the crack surface traction.
3338
(a)
(b)
Fig. 4. Possible constructions of the discontinuous multiscale shape function in 2D. n is the normal to the crack path in the direction of the desired jump in displacement.
Fig. 5. The crack path is chosen along the direction of minimum average shear traction along a nite length ahead of the crack tip. Stress, strain and/or strain energy metrics
are needed in the neighborhood of the crack tip for mesh insensitive path prediction.
where Tc is the external traction on the crack faces. In the subsequent sections, Equations W and (W0 ) are referred to as the coarse
scale and ne scale weak forms, respectively.
3.1. Fine-scale eld and micromechanics embedding
Equation (W0 ) allows us to embed any traction based cohesive
surface-law, Tc, into the continuum formulation. Writing the trac-
Tc T cn n T cm m
The ne scale eld, u0 , for crack problems is composed of a displacement discontinuity, sut, which can be expressed in terms of
the components sunt and s umt along n and m respectively,
3339
(a)
(b)
(c)
Fig. 6. Mesh objectivity for straight crack propagation. (A) Symmetrically loaded SETB specimen, (B) displacement magnitude contours for different mesh densities, and (C)
corresponding loaddisplacement response. The P and D values have been normalized with xed reference values. Mesh objectivity for curved crack propagation in
eccentrically loaded SETB specimen has been shown in Rudraraju et al. [23].
shear
sunt and sumt are referred to as the crack face opening displacement
and crack face shear displacement, respectively. Similarly, the crack
face opening mode is referred to as Mode-I and crack face shear
mode is referred to as Mode-II. Now consider a simple micromechanical surface traction relation given by:
T cn T cn0 Hn sun t
7a
T cm T cm0 Hm sum t
7b
where T cn0 and Hn are the Mode-I critical opening traction and
Mode-I softening modulus, and T cm0 and Hm are the Mode-II critical
shear traction and Mode-II softening modulus. Using Eqs. (5) and
(7), u0 (characterized by sut) can be eliminated from Equation
. Once u
is obtained it can be
W, which can then be solved for u
used to recover u0 , thereby determining the complete displacement
he n; g
u
n
node
X
NA n; gde
8a
A1
c
u0h
e n; g M C n; gsute
8b
A
3340
(a)
(b)
(c)
Fig. 7. Mesh objectivity for straight crack propagation. (A) CTS specimen, (B) displacement magnitude contours for different mesh densities, (C) corresponding load
displacement response. The P and D values have been normalized with xed reference values.
Table 1
Lamina and laminate properties of carbon ber/epoxy [45/0/+45/90]6s laminated
ber reinforced composite.
Laminate
Lamina
"
sut
OM Cc
ne-scale displacement discontinuity. N is the usual linear shape
function for triangle elements, and MCc is a multiscale shape function given by,
M Cc N HCc
10
sutx
suty
2
3
2
3
nix 0
nx 0
16
7
6
7
i 4 0 niy 5 dCc 4 0 ny 5
h
i
i
ny nx
ny nx
|{z}
|{z}
G
e Bd G dCc H sut
r C : Bd G sut
11a
11b
where B is the standard matrix form of the shape function gradients. Substituting the above expression for stress in Eqs. (3a) and
(4), the respective nite dimensional coarse scale and ne scale
weak forms are given by,
K u0 u H T CB
T
K u0 u0 H CG Hn n n Hm m m
3341
15c
15d
Table 2
Scaling observed in SETB specimen experiments. The P, D, and GIc values were
normalized with xed reference values.
Size
Geometry
scaling (Fig. 9)
Peak
load P/P
Load point
displacement D/D
Fracture
toughness
GIc =GIc
1
2
3
4
5
1
1.5
2
3
4
0.27
0.4
0.6
0.81
1.0
0.1
0.15
0.2
0.28
0.37
1.08
1.23
1.84
2.46
2.58
BT C : Bd G sut dV
Nf dV
N T dS
12a
Ch
H T C : Bd G sut T c
12b
c
r
BT C : Bd G sut dV
Z
X
r0 H T C : Bd G sut T c
Nf dV
N T dS
13a
Ch
13b
K u u
K u u0
K u0 u
K u0 u0
dd
dsut
r
r 0
14
where
K u u
K u u0
Z
ZX
BT CB dV
15a
BT CG dV
15b
1
We choose to represent cracks as displacement discontinuities, which implies
u R C0. This results in the strain being a singular distribution which has a bounded
measure, since u 2 BD(X). However the stress should not be a singular distribution as
required by the classical jump condition on the traction (sr nt = 0). This requirement
on the stress eld is enforced by the material constitutive response which mollies
the singular strains to yield regular stresses [26].
3342
Fig. 9. SETB specimen congurations used in the crack propagation experiments. Based on the load point location, the congurations are classied as symmetric or eccentric.
For symmetric specimens, size-1 has the dimensions shown in the gure and other sizes are scaled versions of this base size. For the eccentric specimens, size-1 has the
dimensions shown in the gure and size-2 is scaled up by a factor of two. All specimens have a nominal thickness of 6.35 mm.
Size 1
Size 2
Size 3
Size 4
Size 5
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
Fig. 10. Experimental loaddisplacement curves obtained for various sizes of SETB specimens subjected to symmetric loading conditions. Multiple specimens of each size
were tested to capture the envelope of the failure response. The P and D values have been normalized with xed reference values.
0.3
0.4
0.2
0.2
0.1
0
0.05
0.1
0.4 0.45
0.6
0.8
0.4
0.6
0.1
0.2
0.3
0.4
0.5
0.4
0.2
0
0.2
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
1
0.5
0
0.2
0.4
0.6
0.8
Fig. 11. Loaddisplacement response obtained from simulations of symmetrically loaded Size 1-5 SETB specimens. For a particular specimen size, GL and GH are the least and
highest values of fracture toughness obtained from the multiple experimental loaddisplacement curves. GL corresponds to the curve exhibiting least toughness and GH
corresponds to the curve exhibiting the highest toughness. The P and D values have been normalized with xed reference values.
3343
3
2.5
2
1.5
1
0.5
0
6
5
4
3
2
1
0
10
Fig. 12. Loaddisplacement response obtained from simulations of eccentrically loaded Size 1,2 SETB specimens. For a particular specimen size, GL and GH are the least and
highest values of fracture toughness obtained from the multiple experimental loaddisplacement curves. GL corresponds to the curve exhibiting least toughness and GH
corresponds to the curve exhibiting the highest toughness. The P and D values have been normalized with xed reference values.
(a)
(b)
Fig. 13. Eccentric compact tension specimens used in the crack propagation experiments. (a) Eccentricity: 10 mm and (b) Eccentricity: 20 mm. Both specimens have a
nominal thickness of 6.35 mm.
displacement response and the crack path, the two most important
metrics from a structural viewpoint, on the mesh density.
Consider the problem of symmetrically loaded Single Edge
Notch Three-Point Bend (SETB) specimen as shown in Fig. 6. Shown
are the problem schematic, resulting crack paths for meshes whose
density varies over an order of magnitude, and the corresponding
global loaddisplacement response. Similarly, Fig. 7 shows the
problem of a standard Compact Tension Specimen (CTS). It should
be sufciently clear from these results that the traditional mesh
dependence is absent for the case of straight crack path. However,
at rst glance, the small variation in the loaddisplacement re-
3344
1
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0
0.1
0.2
0.3
0.4
0.5
0.6
0.7
0.8
0.9
0.7
0.8
0.9
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
0.1
0.2
0.3
0.4
0.5
0.6
Fig. 14. Loaddisplacement response obtained from simulations of eccentrically loaded CTS specimens. The P and D values have been normalized with xed reference values.
ture toughness obtained from CTS experiments could be used directly in the eccentric CTS simulations, but for the symmetric
and eccentric SETB simulations, the fracture toughness was computed by normalizing the area under their respective experimental
loaddisplacement curves by the total crack area, as summarized
in Table 2. For more information on the DENT, CTS experiments
and the determination of Mode-I fracture toughness, readers are
referred to Rudraraju et al. [22]. In each of the simulations below,
the meshes contain about ten to twenty thousand elements. However, unlike the interface methods like standard cohesive zone
models, the crack is free to traverse an element along any direction,
thus removing any mesh based restriction on the crack path.
Consider the rst case-study of the symmetrically loaded SETB
specimens whose conguration is shown in Fig. 9. Five specimen
sizes with geometrically scaled planar geometry and xed thickness were considered. Multiple specimens of each size were tested
to signicantly capture the failure response envelope. The load
load point displacement responses of these specimens are shown
in Fig. 10. Table 2 summarizes the observed scaling in the loaddisplacement response, and hence in the value of fracture toughness,
calculated for each size by normalizing the area under their respective experimental loaddisplacement curves by the total crack
cross section area. P, D and GIc are xed reference values. Using
the fracture toughness values listed in Table 2, we obtain the
numerical loaddisplacement curves shown in Fig. 11. Across the
ve sizes, the numerical simulations faithfully reproduce the
experimental loaddisplacement response. Usually, the crack initiates before the peak load, and at the peak load the full bridging
zone will be formed. Further crack growth leads to a drop in the
load bearing ability of the panels due to the failure of the bers
3345
Fig. 15. Comparison of experimental and numerical crack paths for eccentrically loaded CTS specimens.
6. Conclusions
In this paper, a multiscale approach to crack propagation in
ber reinforced panels, homogenized through the panel thickness,
was introduced which treated the discontinuous displacement
eld in crack propagation as a ne scale problem. The necessary
weak formulation, nite element framework, system of equations
and code implementation details were presented. The objectivity
of the resulting computational implementation was demonstrated
through some benchmark simulations and its applicability was
shown through the comparison of numerical and experimental
loaddisplacement responses and crack paths of laminated ber
reinforced composite specimens.
Acknowledgments
This work was supported by a NASA NRA Grant under the
ARMD IVHM Project. The interest and constant encouragement of
Dr. Steven M. Arnold and Dr. Brett Bednarcyk of the MACE Center
at NASA Glenn Research Center, and Craig Collier and Phil Yarrington of Collier Research Corporation is gratefully acknowledged.
3346
References
[1] Armero F, Garikipati K. An analysis of strong discontinuities in
multiplicative nite strain plasticity and their relation with the numerical
simulation of strain localization in solids. Int J Solids Struct 1996;
33(2022):286385.
[2] Armero F, Garikipati K. An analysis of strong discontinuities in multiplicative
nite strain plasticity and their relation with the numerical simulation of
strain localization in solids. Int J Solids Struct 1996;33:286385.
[3] Bazant ZP. Mechanics of distributed cracking. Appl Mech Rev
1986;39:675705.
[4] Borja RL, Regueiro RA. A nite element model for strain localization analysis of
strongly discontinuous elds based on standard galerkin approximation.
Comput Methods Appl Mech Eng 2000;190:152949.
[5] Camacho GT, Ortiz M. Computational modeling of impact damage in brittle
materials. Int J Solids Struct 1996;33:2899938.
[6] Criseld MA, Wills J. Solution strategies and softening materials. Comput
Methods Appl Mech Eng 1988;66(3):26789.
[7] Demmel JW, Eisenstat SC, Gilbert JR, Li XS, Liu JWH. A supernodal
approach to sparse partial pivoting. SIAM J Matrix Anal Appl 1999;20(3):
72055.
[8] Garikipati K. A variational multiscale method to embed micromechanical
surface laws in the macromechanical continuum formulation. Comput Model
Eng Sci 2002;3(2):17584.
[9] Garikipati K, Hughes TJR. A study of strain-localization in a multiple scale
framework. The one dimensional problem. Comput Methods Appl Mech Eng
1998;159:193222.
[10] Gasser TC, Holzapfel GA. Geometrically non-linear and consistently linearized
embedded strong discontinuity models for 3d problems with an application to
the dissection analysis of soft biological tissues. Comput Methods Appl Mech
Eng 2003;192:505998.
[11] Hughes TJR. Multiscale phenomena: Greens functions, the dirichlet-toneumann formulation, subgrid scale models, bubbles and the origins of
stabilized methods. Comput Methods Appl Mech Eng 1995;127(1
4):387401.
[12] Jirasek M. Comparative study on nite elements with embedded
discontinuities. Comput Methods Appl Mech Eng 2000;188:30730.
[13] Larsson R, Runesson K, Ottosen NS. Discontinuous displacement
approximation for capturing plastic localization. Int J Numer Methods Eng
1993;36:2087105.
[14] Moes N, Belytschko T. Extended nite element method for cohesive crack
growth. Eng Fract Mech 2002;69(7):81333.
[15] Moes N, Dolbow J, Belytschko T. A nite element method for crack growth
without remeshing. Int J Numer Methods Eng 1999;46:13150.
[16] Mosler J, Meschke G. Embedded crack vs. smeared crack models: a comparison
of elementwise discontinuous crack path approaches with emphasis on mesh
bias. Comput Methods Appl Mech Eng 2004;193:335175.
[17] Oliver J, Huespe AE. Continuum approach to material failure in strong
discontinuity settings. Comput Methods Appl Mech Eng 2004;193:3195220.
[18] Oliver J, Huespe AE, Sanchez PJ. A comparative study on nite elements for
capturing strong discontinuities: E-fem vs x-fem. Comput Methods Appl Mech
Eng 2006;195:473252.
[19] Pietruszczak ST, Mroz Z. Finite element analysis of deformation of strain
softening materials. Int J Numer Methods Eng 1981;17:32734.
[20] Ramakrishnan N, Okada H, Atluri SN. On shear band formation: II. Simulation
using nite element method. Int J Plas 1994;10(5):52134.
[21] Rudraraju SS. On the Theory and numerical simulation of cohesive crack
propagation with application to ber-reinforced composites. Ph.D. Thesis,
University of Michigan Ann Arbor; 2011.
[22] Rudraraju SS, Salvi A, Garikipati K, Waas AM. In-plane fracture of laminated
ber reinforced composites with varying fracture resistance: experimental
observations and numerical crack propagation simulations. Int J Solids Struct
2010;47(78):90111.
[23] Rudraraju SS, Salvi A, Garikipati K, Waas AM. Experimental observations and
numerical simulations of curved crack propagation in laminated ber
composites. J Compos Sci Technol 2012;72:106474.
[24] Schellekens JCJ, DeBorst R. On the numerical integration of interface elements.
Int J Numer Methods Eng 1993;36:4366.
[25] Simo JC, Oliver J. A new approach to the analysis and simulation of strain
softening in solids. Fract Damage Quasibrittle Struct 1994.
[26] Simo JC, Oliver J, Armero F. An analysis of strong discontinuities induced by
strain-softening in rate-independent inelastic solids. Comput Mech
1993;12:27796. http://dx.doi.org/10.1007/BF00372173.
[27] Song S, Waas AM. A nonlinear elastic foundation model for interlaminar
fracture of laminated composites. Compos Eng 1993;3(10):94559.
[28] Temam R, Strang G. Functions of bounded deformation. Arch Ration Mech Anal
1980;75:721. http://dx.doi.org/10.1007/BF00284617.
[29] Ungsuwarungsri T, Knauss WG. The role of damage-softened material behavior
in the fracture of composites and adhesives. Int J Fract 1987;35:22141.
[30] Wells GN, Sluys LJ. A new method for modelling cohesive cracks using nite
elements. Int J Numer Methods Eng 2001;50:266782.
[31] Xie D, Salvi A, Sun C, Waas AM, Caliskan A. Discrete cohesive zone model to
simulate static fracture in 2-d triaxially braided carbon ber composites. J
Compos Mater 2006;40:122.
[32] Xie D, Waas AM. Discrete cohesive zone model for mixed-mode fracture using
nite element analysis. Eng Fract Mech 2006;73:178396.
[33] Xu XP, Needleman A. Numerical simulation of fast crack growth in brittle
solids. J Mech Phys Solids 1994;42:1397434.