You are on page 1of 24

Int J Fract (2020) 225:129–152

https://doi.org/10.1007/s10704-020-00469-9

ORIGINAL PAPER

Validation of a 3-D adaptive stable generalized/eXtended finite


element method for mixed-mode brittle fracture propagation
Faisal M. Mukhtar1 · Phillipe D. Alves2 · C. Armando Duarte2

Received: 17 April 2020 / Accepted: 12 July 2020 / Published online: 8 August 2020
© Springer Nature B.V. 2020

Abstract
In this paper, a Stable Generalized/eXtended Finite Element Method (SGFEM) is combined with mesh adaptivity for the robust
and computationally efficient simulation of mixed-mode brittle fracture propagation. Both h-refinement around the fracture
front and p-enrichment of the analysis domain are used to control discretization errors. A Linear Elastic Fracture Mechanics
(LEFM) model based on Griffith’s criterion is adopted. LEFM scaling relations are used at each fracture propagation step to
back calculate SIFs that meet Griffith’s criterion. As a result, no iterations are necessary to find loading scaling parameters
or fracture size that meets Griffith’s criterion. The method is validated against several experimental data sets for mode I
and mode I+II fracture propagation problems. Very good agreement between SGFEM and experimental results is observed.
These include fracture path, Crack Opening Displacement (COD), and load and fracture length versus COD curves. The
computational efficiency of the method is also assessed.

Keywords Generalized FEM · EXtended FEM · Fracture propagation · Brittle fracture · Validation · Mixed-mode · Mesh
adaptivity

1 Introduction Melenk 1997; Melenk and Babuška 1996; Moës et al. 1999;
Belytschko and Black 1999) can be interpreted as a Finite
Generalized or eXtended FEMs for the solution of fracture Element Method (FEM) with an enriched test/trial space
mechanics problems (Oden et al. 1998; Oden and Duarte which is obtained by hierarchically augmenting a standard
1997; Duarte et al. 2000; Belytschko and Black 1999; Dol- finite element approximation space, S F E M , with an enrich-
bow et al. 2000) have been developed and improved for more ment space, S E N R derived from a priori knowledge about
than two decades and are available in several mainstream the solution of a problem. The enrichment space used in the
commercial software such as Abaqus, ANSYS, LS-DYNA, GFEM enables the use of finite element meshes that do not fit
OptiStruct, etc. The GFEM and XFEM are essentially crack surfaces or their fronts. This flexibility of the method is
identical methods (Belytschko et al. 2009). The acronym particularly appealing in the case of 3-D simulations of crack
GFEM is adopted in this paper. The GFEM (Babuška et al. propagation, like those considered in this paper. Furthermore,
1994; Duarte et al. 2000; Oden et al. 1998; Babuška and the convergence rate of the GFEM presented in (Sanchez-
Rivadeneira et al. 2020) is at least three times higher than
quadratic finite elements with quarter-point elements (Hen-
B Faisal M. Mukhtar
shell and Shaw 1975; Barsoum 1976) around the crack front,
faisalmu@kfupm.edu.sa
which are broadly used to solve fracture problems. However,
Phillipe D. Alves
alves2@illinois.edu the condition number of the GFEM stiffness matrix grows
much faster than in the classical FEM as the mesh is refined.
C. Armando Duarte
caduarte@illinois.edu There is a considerable amount of work on approaches to
address the conditioning issue of the method (Babuška and
1 Dept. of Civil and Env. Eng., King Fahd University of Banerjee 2012; Zhang et al. 2016, 2014; Kergrene et al.
Petroleum & Minerals, Dhahran 31261, Saudi Arabia 2016; Babuška et al. 2017; Ndeffo et al. 2017). The approach
2 Dept. of Civil and Env. Eng., University of Illinois at adopted here is the so-called Stable GFEM (SGFEM) with a
Urbana-Champaign, 205 North Mathews Av, Urbana, discontinuous interpolant enrichment modification and a p-
IL 61801, USA

123
130 F. M. Mukhtar et al.

hierarchical FEM basis—the pFEM-DSGFEM introduced in Several of the algorithmic aspects of the GFEM adopted
Sanchez-Rivadeneira et al. (2020) for 3-D fracture problems. here have been presented in Pereira et al. (2009, 2010), Gar-
This SGFEM, with the proper choice of singular enrichments, zon et al. (2014), Gupta and Duarte (2014). They include the
leads to stiffness matrices with a scaled condition number of representation and update of non-planar fracture surfaces,
the same order as in the standard FEM. A summary of the the integration of the weak form over element intersected by
pFEM-GFEM and its well-conditioned counterpart pFEM- the fracture, etc. A brief discussion on the main computa-
DSGFEM is presented in Sect. 3. Details can be found in tional geometry strategies adopted in this paper is presented
Sanchez-Rivadeneira et al. (2020). in Sect. 3.3 while the crack propagation algorithm with mesh
Available analytical information for fracture mechan- adaptivity is presented in Sect. 4. This algorithm requires at
ics problems is limited to simple fracture geometries and each crack propagation step the following information: (i) the
physics, for example, 2-D planar cracks with straight fronts. crack propagation direction; (ii) the magnitude of the crack
While this limited information can still be used to define an front advancement; (iii) a crack growth criterion to deter-
enrichment space S E N R when solving problems with more mine whether a point on the crack front propagates or not.
realistic fracture geometries, mesh refinement is required to These issues are also discussed in Sect. 4. The satisfaction
compensate for the suboptimal enrichments. In this paper, the of a brittle fracture propagation criterion in 3-D may require
pFEM-DSGFEM presented in Sanchez-Rivadeneira et al. an iterative procedure to find loading scaling factors or the
(2020) is combined with mesh adaptivity as described in crack size that meets the criterion. In this paper, LEFM scal-
Sect. 4. ing relations are used to back calculate scaling factors that
The goal of this paper is the validation of an adaptive lead to the satisfaction of the propagation criterion. These
pFEM-DSGFEM for linear elastic fracture mechanics prob- scaling factors are also used to compute other quantities of
lems. Hereafter, the acronyms GFEM and pFEM-DSGFEM interest such as reaction forces and Crack Opening Displace-
refer to the method presented in Sanchez-Rivadeneira et al. ment (COD). These are low-cost post-processing operations
(2020), unless the context indicates otherwise. Crack prop- performed on the GFEM solution of each crack propagation
agation is assumed to be governed by Griffith’s criterion step. They involve no iterations or recalculation of the GFEM
(Griffith 1921; Nuismer 1975). The governing equations solution. To our knowledge, this is a novel strategy to solve 3-
of the problem solved at each crack propagation step are D mixed-mode brittle fracture propagation problems. Details
summarized in Sect. 2. The GFEM and some of the algo- are presented in Sect. 4.1.3. Similar ideas are used by Shabir
rithms adopted here have been extensively verified (Sanchez- et al. Shabir et al. (2019) to find load-displacement curves for
Rivadeneira et al. 2020; Pereira et al. 2009, 2010; Garzon crack propagation in polycrystalline specimen and by Gupta
et al. 2014; Gupta and Duarte 2014) but not systematically and Duarte (2014) to solve a Mode I hydraulic fracture prop-
validated. While the methodology adopted here is capable of agation problem.
simulating brittle fracture propagation under general mixed- The simulations presented in Sect. 5 validate not only an
mode conditions, in this paper the focus is on the validation adaptive GFEM for brittle fracture propagation but also the
of the method under in-plane modes I and II conditions. The algorithmic aspects of the method and the adopted computa-
method is validated in Sect. 5 against several experimental tional fracture growth model for LEFM presented in Sect. 4.
data sets for mode I and mode I+II fracture propagation prob- The main conclusions of this work are summarized in Sect.
lems. The GFEM solution of a mixed mode problem is also 6.
compared with phase-field FEM solutions presented in Pham
et al. (2017) and Kumar and Lopez-Pamies (2020).
The literature on GFEMs for fracture propagation is quite
extensive but most of the works have focused on the devel-
opment of the method and its verification against analytical
or reference solutions. Examples of works dedicated to the 2 Governing equations
validation of the method for problems governed by LEFM
include the works of Grégoire et al. (2007) on 2-D dynamic Linear Elastic Fracture Mechanics (LEFM) assumptions are
crack propagation, the work of Bergara et al. (2017), Ferrie adopted. Inertial effects and gravity load are neglected. The
et al. (2006), Shi et al. (2010), Nikfam et al. (2019), Tian prescribed loads and displacements are assumed to be pro-
et al. (2019), and Varfolomeev et al. (2014) on 3-D fatigue portional. Under these assumptions, the simulation of brittle
crack propagation. We can also mention the work of Giner fracture propagation using the GFEM involves the solution of
et al. (2009), and Ventura et al. (2003) on 2-D simulations of a sequence of linear elasticity problems—one LEFM prob-
brittle crack propagation. The validation of 3-D models for lem for each fracture propagation step. The strong and weak
brittle crack propagation solved with a GFEM seems to be formulations of this problem are as follows.
not available in the literature.

123
Validation of 3-D SGFEM for brittle fracture growth 131

with E () the energy space Szabo and Babuška (1991) with
norm

u ε() = B(u, u) (7)

3 A generalized Finite Element Method for


3-D fractures

Consider standard linear finite element shape functions


Nα (x), α ∈ Ih = {1, 2, . . . , n}, associated with node x α
of a mesh with n nodes. These FE shape functions form a
partition of unity (PoU), i.e.:
Fig. 1 Schematic of a 3-D domain  with a crack surface ΓC

Nα (x) = 1 ∀x ∈  (8)
Strong form: Consider a cracked elastic domain  ¯ = ∪ α∈Ih
∂ in R and a crack surface ΓC with crack front γF like the
3

one illustrated in Fig.1. A GFEM shape function φαi (x) is given by the product
The equilibrium equations are given by of a FE shape function Nα (x) and an enrichment function
E αi (x), i.e.,
∇· σ = 0 in  (1)
φαi (x) = Nα (x)E αi (x) (no summation on α) (9)
where σ denotes the Cauchy stress tensor. The boundary
conditions are given by where α ∈ Ihe ⊂ Ih is the index of a node in a FE mesh
and i ∈ I (α) = {1, 2, . . . , m α } is the index of the enrich-
u = ū on Γu (2) ment function at the node. These enrichment functions form

σ · n = t̄ on Γt (3) a basis of the local space χα (ωα ) = span{E αi }i=1 called
patch approximation space with patch ωα being the union of
where ū and t̄ are prescribed displacements and tractions, all finite elements sharing node x α .
respectively, n is the unit normal vector to Γ ≡ ∂, Γu ∪Γt = Let S F E M be the underlying finite element approximation
Γ , and Γu ∩ Γt = ∅. space given by
Hooke’s law, i.e., σ = Cε, where C is a fourth-order

tensor, is adopted. Small deformations are also assumed so SF E M = cα Nα , cα ∈ R3 (10)
that the strain tensor can be written as ε = 21 (∇u + ∇u ). α∈Ih

While the enrichment space is given by


Weak form: The variational form of the problem described
above is given by the following statement: ⎡ ⎤
 1
Find u ∈ V () such that, ∀v ∈ V0 () SE N R = 13 Nα χα , 13 = ⎣1⎦ (11)
α∈Ih 1
B(u, v) = F(v) (4)
The test/trial GFEM space SG F E M is given by
where

SG F E M = S F E M + S E N R (12)
B(u, v) = σ (u) : ε(v)d (5)

That is, SG F E M is obtained by hierarchically augment-
F(v) = t̄· vdΓ (6)
Γu ing the FEM space S F E M with the global enrichment space
SE N R .
and For linear elastic fracture problems, the enrichment space
  S E N R is typically partitioned into three subspaces
V () = v ∈ E () : v|Γu = ū
 
S E N R = S E N R ⊕ SH
p
V0 () = v ∈ E () : v|Γu = 0 E N R ⊕ SE N R
br
(13)

123
132 F. M. Mukhtar et al.

p
where S E N R represents the polynomial enrichment func-
tion space (which hierarchically increases the polynomial
order of the approximation). SH E N R contains the space of
piecewise constant and piecewise high-order polynomials,
which allows representing the discontinuity along the crack
surface. Sbr E N R contains branch (singular) enrichment func-
tions which approximate the asymptotic behavior of the
solution field in a neighborhood of the crack front.
It is noted that SH E N R is not empty only for patches ωα ,
α ∈ IhH ⊆ Ih , where IhH is the set of nodes whose support
is intersected by the crack surface, but not the crack front, Fig. 2 Crack front Cartesian and polar coordinate systems
while Sbr E N R is defined only for patches ωα , α ∈ Ih ⊆ Ih ,
br
br
where Ih is the set of nodes within a neighborhood of the
crack front. 3.1.2 Singular enrichments

3.1 pFEM-GFEM approximation for 3-D fracture The singular enrichment space Sbr E N R is built based on the
problems Oden-Duarte (OD) branch enrichment functions (Oden and
Duarte 1997; Duarte et al. 2000) which uses the first term
The pFEM-GFEM is based on a p-hierarchical FEM basis from each mode of the 2-D asymptotic expansion of the
( pFEM), constructed via the use of orthogonal polynomi- elasticity solution in the neighborhood of a planar crack in an
als and standard linear finite element shape functions (Szabo infinite plate with stress-free crack faces (Szabo and Babuška
and Babuška 1991; Bravos 2004). The first order approxi- 1991).
mation consists of the standard linear finite element shape Let (r , θ, z̄) and (x̄, ȳ, z̄) denote the local crack front
functions associated with vertex nodes in a FE mesh. High- cylindrical and Cartesian coordinate system respectively at
order hierarchical shape functions can be associated with point (x, y, z) denoted in the global Cartesian coordinate sys-
high dimensional geometrical entities of the mesh: edges, tem as shown in Fig. 2. Let κ denote the Kolosov’s constant
faces, and volume of 3-D elements. In this paper, we adopt defined as κ = 3 − 4ν for plane strain and κ = 3−ν 1+ν for
second and third order p-hierarchical basis. The polyno- plane stress, and ν the Poisson’s ratio. The scalar-valued OD
p
mial enrichment function space S E N R is defined based on enrichments adopted in this paper are given by
these p-hierarchical basis. We refer to Section 3 of Sanchez-
Rivadeneira et al. (2020) and the monographs Szabo and √
Babuška (1991); Bravos (2004) for details. L OD6-scalar
front (x) = r R̄(x)
⎡  θ
  θ ⎤
κ − 21 cos θ2 − 21 cos 3θ2  κ + 32 sin θ2 +
1 1 3θ 3 1
2 sin 3θ
2
3.1.1 Discontinuous enrichments ⎣ κ + sin − sin κ − 2 cos 2 + 1
cos 3θ ⎦
2 2 2 2 2 2
sin θ2 cos θ2
The discontinuous enrichment space SH E N R is defined using (16)
the piecewise constant Heaviside function H (x)
where R̄(x) is defined as


⎨1 if x is on the positive side
⎧⎡ ⎤
H (x) = or on the crack surface (14) ⎪
⎪ R11 R12 R13

⎩ ⎪
⎪⎢ ⎥
0 otherwise ⎪
⎪⎣ R21 R22 R23 ⎦ if z̄is parallel tox, yorzaxis





⎨ R31 R32 R33
The high-order discontinuous enrichment space SH R̄(x) = (17)
E N R is ⎪⎡ ⎤

⎪ R11 R12 0
given by ⎪


⎪⎢ ⎥

⎪⎣ R21 R22 0⎦ otherwise
  ⎪

SH
p ⎩
E N R = H (x) S F E M ⊕ S E N R (15) R31 R32 0

p
with S E N R as defined earlier. In the case of quadratic approx- Entries Ri j , i = 1, 2, 3, represent the components of the
imation adopted in this paper, this leads to the definition of base vector in the direction j of crack coordinate system with
vertex and edge Heaviside enrichment functions. respect to the global base vectors. Basis L OD6-scalar
front is able

123
Validation of 3-D SGFEM for brittle fracture growth 133

to approximate both modes I and II uncoupled with mode III


Sanchez-Rivadeneira et al. (2020).
αi (x) be the i-th column of L front
Let L br OD6-scalar (x). The sin-

gular enrichment space is given by


⎧ ⎫

⎨  m
br =2
α ⎪

Sbr
ENR = αi  L αi (x)Nα (x),
a br br
αi ∈ R
a br 3

⎩α∈I br ⎪

i=1
h

(18)
Fig. 3 Non-planar crack surface and base vectors at crack front vertices
where  stands for the entry-wise product of two vectors (Pereira et al. 2009)
operator (Hadamard product).
The quadratic pFEM-GFEM approximation of a displace-
ment field for a linear elastic fracture mechanics problem
u(x) ∈ R3 follows from the definition of space SG F E M and generalization of the modification based on finite element
is given by interpolants introduced in Babuška and Banerjee (2012).
A pFEM-GFEM approximation space defined using these
   enrichment modifications is denoted a Discontinuous Inter-
uh (x) = a α Nα (x) + c αβ Nα (x)Nβ (x)
α∈Ih α∈Ih x β ∈ωα
   polant Stable pFEM-GFEM ( pFEM-DSGFEM) space. Both
β =α pF E M edge sh fn
 enrichment modifications are space-preserving. This means
H
+ a α H (x)Nα (x) vertex sh fn that pFEM-GFEM and pFEM-DSGFEM spaces are the
  
α∈IhH GFEM same—only their bases differ (Sanchez-Rivadeneira and
 
+ cH
αβ H (x)Nα (x)Nβ (x) Duarte 2019; Sanchez-Rivadeneira et al. 2020). It is also
  
α∈IhH x β ∈ωα
pF E M−G F E M edge sh fn
noted that the computational cost of these enrichment mod-
β =α
ifications is low and their implementation not difficult.
  α =2
m br
+ αi  L αi (x)Nα (x)
a br br
(19)
  
α∈Ihbr i=1
G F E M vertex sh fn
3.3 Computational geometry issues
where a α , a H H
α , c αβ , c αβ , a αi ∈ R (for a fixed i). Here
br 3

α and β refer to vertex node indices, therefore Nα (x), α ∈ The implementation of the GFEM requires efficient and
Ih , represent vertex shape functions – the linear partition of robust computational geometry strategies. These include the
unity. The quadratic pFEM edge shape functions appear as representation of non-planar crack surfaces, the definition of
the product Nα (x)Nβ (x). Cubic approximations are defined base vectors of coordinate systems used in the computation of
in a similar manner. singular enrichment functions, and algorithms for automatic
generation of elements used for the numerical integration
3.2 Discontinuous interpolant stable pFEM-GFEM of the weak form. In this paper, crack surfaces are repre-
(pFEM-DSGFEM) sented using surface meshes like those shown in Figs. 3 and
12 . These triangulations are known as geometrical crack
The pFEM-GFEM approximation (19) leads to a stiffness surfaces. The surface mesh is independent of the 3-D mesh
matrix with a condition number that is higher than in the used to solve the problem and composed of a set of triangles
classical finite element. This issue can be addressed by mod- (facets) that represent the crack geometry and location. It is
ifying the enrichment functions introduced earlier. A brief noted that these triangulations have no degrees of freedom.
summary of the enrichment modification techniques used They serve only for computational geometry purposes. Fig-
in pFEM-DSGFEM is presented next. Detailed expositions ure 3 shows base vectors of coordinate systems defined at
can be found in Sanchez-Rivadeneira and Duarte (2019) and crack front vertices of a geometrical crack surface. Besides
Sanchez-Rivadeneira et al. (2020). the computation of singular enrichments, these base vectors
The first enrichment modification simply shifts the enrich- are also used for the extraction of stress intensity factors at
ment functions such that they are zero at the mesh nodes crack front vertices (cf. Appendix A), and for the crack front
they are assigned to. The second enrichment modification advancement as described in Sect. 4.1. We refer to Appendix
entails subtracting from an enrichment function its discon- A of Sanchez-Rivadeneira et al. (2020) and the references
tinuous piecewise linear interpolant (Sanchez-Rivadeneira therein for details on computational geometry aspects of the
and Duarte 2019; Sanchez-Rivadeneira et al. 2020). This is a implementation of the GFEM adopted in this paper.

123
134 F. M. Mukhtar et al.

4 Crack propagation algorithm with mesh (2014). Figure 12 shows the evolution of a crack surface
adaptivity and its update.
8. After updating the geometrical crack surface, the 3-D
This section presents the main steps in a 3-D GFEM GFEM mesh is unrefined to the initial mesh provided by
simulation of fracture propagation. Linear elastic fracture the user. The algorithm is then repeated until no crack
mechanics is adopted. The propagation algorithm is based on front vertex meet the propagation criterion or the user-
those proposed in Pereira et al. (2010), Garzon et al. (2014), prescribed number of propagation steps is reached.
Gupta and Duarte (2014) and is implemented in the 3-D adap-
tive GFEM software ISET Duarte (2016). The algorithm enables automated, efficient, and accurate
Given the representation of the crack surface at propaga- simulations of 3-D fracture propagation as demonstrated
tion step (cf. Sect. 3.3), the following is performed: in Sect. 5. It is noted that in the propagation algorithm,
the boundary conditions of the problem are unchanged
1. Refine the 3-D GFEM mesh around the crack front throughout the simulation—no iterations are necessary to
such that the element size meets user-prescribed crite- find loading parameters or fracture size that meet a propaga-
ria. Examples of these meshes are shown in Figs. 7 and tion criterion. The rational for this is presented in Sect. 4.1.3.
16 . Details on the refinement algorithm can be found in
Section 4.2 of Gupta and Duarte (2014). 4.1 A computational fracture growth model for
2. Compute the intersection between the geometrical crack linear elastic fracture mechanics
surface and the GFEM mesh. The Computational Geom-
etry Algorithms Library (CGAL) [62] implementation The propagation algorithm described in Sect. 4 is based on
of AABB trees (trees of axis-aligned bounding boxes) is linear elastic fracture mechanics assumptions and requires,
adopted for these computations. These intersections are at a crack propagation step, the following information at each
in turn used for the generation of integration sub-elements crack front vertex of the geometrical fracture surface:
as described in Pereira et al. (2009); Sanchez-Rivadeneira
et al. (2020). 1. The crack propagation direction. In this work, we adopt
3. Enrich the finite element nodes with singular, Heaviside Schöllmann’s criterion (Schöllmann et al. 2002). A sum-
and polynomial functions, as described in Sect. 3.1. mary of the equations used in this criterion is presented
4. Integrate the stiffness matrix and load vector associated in Sect. 4.1.1.
with the GFEM discretization of (4). This is performed 2. The magnitude of the crack front advancement. A Paris-
with the aid of integration sub-elements. type law is used to compute this at each front vertex based
5. Solve the linear elastic problem (4) discretized with the on the maximum advancement magnitude, Δamax , pre-
GFEM. scribed by the user and other law-specific parameters.
6. Extract mode I, II and III Stress Intensity Factors (SIFs) The law adopted in this work and the rational for its adop-
at every vertex on the crack front of the geometrical crack tion are presented in Sect. 4.1.2
surface. The extraction method adopted in this paper is 3. A crack growth criterion to determine whether a point
described in Appendix A. on the crack front propagates or not. Griffith’s criterion
The SIFs are used to determine, at each crack front vertex, (Griffith 1921; Kuna 2013) is typically adopted for brittle
if the crack will propagate and the direction and mag- materials. The satisfaction of this criterion may require
nitude of propagation as described in Sect. 4.1. These an iterative procedure to find loading parameters or the
quantities define propagation vectors at each crack front crack size that meets the criterion. In this paper, LEFM
vertex. scaling relations are used to back calculate, at each prop-
7. The propagation vectors are used to define the position of agation step, loading factors that lead to the satisfaction
the crack front vertices at the next propagation step. The of Griffith’s criterion. Details are provided in Sect. 4.1.3.
geometrical crack surface is updated by adding a new
layer of triangular facets to the current crack front or 4.1.1 Crack propagation direction: Schöllmann’s criterion
by stretching the current crack front vertices. The strat-
egy used depends on the magnitude of the crack front Crack front deflection can be represented by the kinking
advancement vector at each front vertex. angle θ0 and the twisting angle ψ0 , as shown in Fig. 4.
After several propagation steps, the crack surface trian- These quantities are computed using Schöllmann’s criterion
gulation may be remeshed to reduce the number of facets (Schöllmann et al. 2002). It assumes, like in the 2-D Max-
and speed up computational geometry operations. imum Tangential Stress (MTS) criterion (Erdogan and Sih
Details on these algorithms for geometrical crack surface 1963), that the crack growth direction is perpendicular to the
updating and remeshing are described in Garzon et al. maximum principal stress.

123
Validation of 3-D SGFEM for brittle fracture growth 135

4.1.2 Crack front advancement magnitude

Crack growth in linear elastic fracture mechanics is depen-


dent on the stress intensity factors or, equivalently, on the
energy release rate (Paris and Erdogan 1963; Kuna 2013).
Griffith’s criterion is commonly used for brittle fracture
growth (Griffith 1921; Nuismer 1975; Kuna 2013). It is given
by

0, if Gmixed ( j) < Gc
Δa( j) = (22)
> 0, if Gmixed ( j) = Gc

where Δa( j) is the magnitude of crack advancement at crack


front vertex j,
Fig. 4 Crack growth under each mode and their superposition. The Gmixed ( j) is the energy release rate in the direction of
kinking angle θ0 and the twisting angle ψ0 define the direction of crack crack propagation at front vertex j, and Gc is the critical
front propagation (Gupta and Duarte 2014) energy release rate of the material. In the case of plane strain
conditions, Gc is given by
The kinking angle θ0 is a non-linear function of all three
1 − ν2 2
SIFs and is determined by the conditions Gc = KIc (23)
E
 
∂σ1  ∂ 2 σ1 
= 0 < 0 

where K I c is the fracture toughness of the material under
∂θ θ=θ0 ∂θ 2 θ=θ0 mode I loading.
The energy release rate at each crack front vertex is used
where σ1 is the maximum principal stress. to determine whether the crack front propagates or not. The
Once the kinking angle θ0 is calculated, the twisting angle, crack propagates when the energy release rate at any crack
ψ0 , is given by the direction of σ1 and can be computed using front vertex in the direction of crack propagation, Gmixed ,
  reaches the critical energy release rate of the material (Griffith
1 2τθ z (θ0 ) 1921; Nuismer 1975). Gmixed is computed using (Nuismer
ψ0 = arctan (20)
2 σθ (θ0 ) − σz (θ0 ) 1975),

where σθ , τθ z and σz are components of the stress tensor Gmixed = G I /I I + G I I I (24)


in a cylindrical coordinate system defined at the crack front.
For further details, the reader is referred to Schöllmann et al.
where G I /I I and G I I I are given by Nuismer (1975),
(2002); Richard et al. (2005).
The kinking and twisting angles are used to prescribe the   ∗ "
1 − ν2 1 θ   ! 2
direction of crack propagation at each crack front vertex of G I /I I = cos K I 1 + cos θ ∗ − 3K I I sin θ ∗
E 2 2
the geometrical crack surface (Pereira et al. 2010). An effec-
(25)
tive crack front propagation direction θ ∗ is defined based on
θ0 and an increment accounting for the twisting angle ψ0 , and
and
thus the effect of K I I I . Details on the procedure to compute
this increment are provided in Pereira (2010). #  ∗ $2
1+ν θ
Schöllmann’s criterion degenerates to Erdogan and Sih’s GI I I = K I I I cos (26)
E 2
MTS criterion (Erdogan and Sih 1963) if K I I I = 0 which is
the case of all problems solved in this paper. In this case, the where direction θ ∗ is the effective kinking angle as defined
kinking angle θ0 is given by Kuna (2013) in Sect. 4.1.1.
⎧ ⎡  ⎤ It is noted that the criterion given by Eq. (22) does not pro-
⎪ 

⎨2 arctan ⎣ 1 K I − 1 KI 2 vide the magnitude of the crack front advancement along the
+ 8⎦ , if K I I  = 0
θ0 =

4 KI I 4 KI I (21) crack front once the condition Gmixed = Gc is met at a crack


0, if K I I = 0 front vertex. The criterion as stated in (22) is not suitable for
a computational implementation. Lazarus (2003) presented
and the twisting angle ψ0 = 0. a regularization of Irwin’s criterion – which is equivalent to

123
136 F. M. Mukhtar et al.

Griffith’s criterion—based on a Paris-type law with a high evolves continuously from the front due to the absence
exponent. This law provides the magnitude of the crack front of Mode III effects. Experimental and analytical evidence
advancement along the crack front. It scales the magnitude (Sommer 1969; Lazarus et al. 2001; Lin et al. 2010; Pham
of the advancement such that the point on the crack front and Ravi-Chandar 2014; Leblond and Frelat 2014; Pham and
with the largest mode I stress intensity factor advances the Ravi-Chandar 2017) have shown that crack growth under
most. We adopt a similar strategy which is suitable for the K I I I = 0 presents fracture facets rotated with respect to
simulation of 3-D crack propagation in brittle materials. The the crack front forward direction. These facets eventually
magnitude of crack advancement, Δa( j), at a crack front coalesce to form a continuous crack surface.
vertex j is taken as,
⎧ 4.1.3 Crack growth criterion and scaling factors
⎨0, if K I ,eq ( j) ≤ K I c
 m
Δa( j) = K I ,eq ( j)−K I c The crack propagation algorithm described so far does not
⎩Δamax , if K I ,eq ( j) > K I c
,eq −K I c
K Imax
enforce that Gmixed ( j) ≤ Gc at every crack front vertex j. In
(27) the crack propagation simulations presented in Sect. 5, this
condition is enforced using LEFM scaling relations. Crack
where m and Δamax are model parameters. A model sim- propagation is simulated with prescribed displacements or
ilar to (27) is adopted in Rungamornrat et al. (2005). The loads that are equal to or larger than the maximum dis-
equivalent mode I stress intensity factor, K I ,eq ( j), at crack placements/loads applied to the experimental specimen. The
front vertex j is computed using the expression proposed by scaling factors computed as described below are then used,
Schöllmann et al. (2002) and Richard et al. (2005), at each fracture propagation step, to back calculate SIFs such
    that Gmixed ( j) ≤ Gc . These scaling factors are also used to
1 θ0 2 θ0 3 compute other quantities of interest such as reaction forces
K I ,eq ( j) = cos K I cos − K I I sin(θ0 )
2 2 2 2 and Crack Opening Displacement (COD). As a result, no
 ⎫
#   $2 ⎬ iterations are necessary to find loading parameters or fracture
θ0 3
+ K I cos2 − K I I sin(θ0 ) + 4K I2I I size that meet Griffith’s criterion, thus saving a considerable
2 2 ⎭
amount of computational work.
(28) Let K I , K I I , and K I I I denote the SIFs extracted from
the GFEM solution at a crack propagation step. LEFM is
where K I , K I I , K I I I , and θ0 are computed at crack front assumed to be valid. Furthermore, the loads, t̄, and displace-
vertex j with θ0 computed using Schöllmann’s criterion as ments, ū, prescribed on Γt and Γu , respectively, are assumed
described in Sect. 4.1.1. to be proportional.
If K I I I = 0, then Let the scaled SIFs be defined by
 
θ0 θ0 θ0
K I ,eq ( j) = cos2 K I cos − 3K I I sin (29) K ic := αc K i i = I , I I , I I I (32)
2 2 2
 
1 θ0
= cos [K I (1 + cos θ0 ) − 3K I I sin θ0 ] with the scaling factor αc a parameter such that
2 2
(30)
% Gmixed (K Ic , K Ic I , K Ic I I ) = Gc (33)
E
= G I /I I (θ0 ) (31)
1 − ν2
These last two equations and (24) give
Parameter Δamax represents the maximum displacement
magnitude among all front vertices. It is applied to the crack   ∗ "
1 − ν2
1 θ   ! 2
αc2 cos K I 1 + cos θ ∗ − 3K I I sin θ ∗
front vertex with the highest equivalent stress intensity fac- E 2 2
#  ∗ $2
,eq . In this work, parameter Δamax is selected by the
tor K Imax 1+ν θ
user. The effect of this parameter on the simulation results is + αc2 K I I I cos = αc2 Gmixed (K I , K I I , K I I I ) = Gc
E 2
investigated in Sect. 5. Parameter m is taken as unity in all
problems solved in this paper. It is noted that a crack front Therefore, the scaling factor at a crack propagation step
advancement is prescribed at a front vertex only if the crack can be computed using
growth criterion (22) is satisfied.

Remark 1 The foregoing discussions (Sects. 4.1.1 and 4.1.2) Gc
αc = (34)
and simulations presented in Sect. 5 assume that the crack Gmixed (K I , K I I , K I I I )

123
Validation of 3-D SGFEM for brittle fracture growth 137

A similar expression can also be derived using the equiv- et al. (2017). The third problem is the fracture test conducted
alent mode I stress intensity factor K I ,eq . by Ingraffea and Grigoriu (1990) on deep beams with a series
In a 3-D problem, the energy release rate Gmixed is not of holes. This problem is used to further verify the capability
constant along the crack front. Thus, one must select where of the GFEM in predicting crack paths influenced by sev-
to evaluate (34). One option is to evaluate it at the crack front eral holes that make the crack propagation sensitive to initial
vertex that has the highest value of Gmixed . This guarantees crack geometry and location relative to the holes. In all simu-
that Griffith’s criterion is satisfied at all crack front vertices. lations, Stress Intensity Factors (SIFs) are extracted using the
In the case of problems solved in Sect. 5, Gmixed is nearly Displacement Correlation Method (DCM). A brief review of
constant along the crack front. Thus, in this paper, the scaling the DCM formulation adopted in this work is presented in
factor αc is computed using the average of Gmixed values Appendix A.
extracted at all crack front vertices. We found that this average
value provides slightly more accurate results than taking the 5.1 Mode I fracture propagation: compact tension
maximum value of Gmixed . Specimen CT24 of Pham et al. (2017)
Based on previous assumptions, the following relations
hold The systematically designed experiments conducted by
Pham et al. (2017) consists of two groups of specimens.
ūc t̄ c Kc The first one includes six compact tension (CT) specimens
= = I = αc
ū t̄ KI while the second one adopts three specimens and a modi-
fied CT specimen geometry. Figure 5 shows the geometry
where ū and t̄ are displacements and tractions prescribed on and dimensions of the specimens in the first group (mode
Γu and Γt and ūc and t̄ c are these quantities scaled such that I conditions). The material of the specimen is a thermo-
the GFEM solution satisfies Griffith’s criterion at a propa- plastic polymer, polymethylmethacrylate (PMMA), whose
gation step. Thus, the scaling factor αc computed at a given elastic material properties and fracture parameters, as deter-
crack propagation step using (34) can be used to scale the mined in Pham et al. (2017), are: Modulus of elasticity
prescribed displacements and tractions such that the GFEM E = 2.98 GPa; Poisson’s ratio ν = 0.35; and Fracture
solution satisfies Griffith’s criterion at that step. It is noted energy G c = 0.285 kJ/m2 that resulted in an estimated frac-
that the GFEM solution does not need to be re-computed ture toughness K I c = 0.98 MPa m1/2 .
using ūc and t̄ c as boundary conditions. Instead, quantities The initial sharp crack is generated in the specimen ahead
of interest computed from the available GFEM solution are of the U-notch, using a wedge impact along with a blade.
scaled using αc . For example, the reaction forces, r cu , at Γu The top and bottom holes are loaded monotonically, with dis-
and the Crack Opening Displacement, COD, can be com- placements u+ −
y and u y , respectively, as illustrated in Fig. 5.
puted using Figure 6 shows the setup used in the experiments of
Pham et al. (2017). It consists of an Instron testing machine,
r cu = αc ru CODc = αc COD (35) cameras, and loading grips. The test proceeds using a
displacement-controlled loading whose magnitude, and the
This procedure is adopted in the simulations presented in displacement of the load-points are monitored. At the same
Sect. 5. Linear elastic fracture mechanics scaling relations are time, a set of two cameras are placed on opposite sides of
also used by Shabir et al. (2019) to find load–displacement the specimen to capture the images used for calculating the
curves for crack propagation in polycrystalline specimen. crack opening displacement (COD) at the load line, using
DIC analysis, and tracking the crack front location. The COD
corresponding to each load step is determined using time cor-
5 Validation with experimental results relation between the image and loading time sequences. More
details about the experiment and a summary about the image
In this section, the GFEM and algorithms presented ear- analysis can be found in Pham et al. (2017).
lier are validated against three experimental data sets. The
experiments involve mode I and mode I+II fracture propaga-
tion problems with increasing complexity in crack path. The Boundary conditions at specimen holes: The displacements
first two problems were proposed by Pham et al. (2017) and imposed by loading pins on the top and bottom holes of the
adopt a standard and a modified compact tension (CT) speci- specimen are modeled as displacements prescribed to nodes
men. The latter is a mixed-mode problem while the former is located along the load lines shown in Fig. 5. These nodes are
mode I. Load versus Crack Opening Displacement (COD), equally spaced in the thickness direction of the specimen.
crack length versus COD, and crack path predicted by the Both vertical and lateral movements of the so defined nodes
GFEM are compared with the experimental data of Pham in the bottom hole are prevented in the GFEM model. Vertical

123
138 F. M. Mukhtar et al.

reasonably coarse mesh is used at the specimen holes, the


results are not sensitive with respect to discretization param-
eters adopted in this paper, such as the polynomial order of
the GFEM approximation or element size at the holes.

Experimental results of Specimen CT24 of Pham et al.


(2017) is used in this section, unless stated otherwise, for the
validation of the adaptive GFEM and the fracture propagation
algorithm presented in Sects. 3 and 4 , respectively. This
specimen was tested under mode I condition. Two sets of
meshes and GFEM approximations of degree p = 2 and
p = 3 are adopted in the studies presented next. The effect
of crack advancement magnitude, Δamax , used in Equation
(27) as well as the effect of the initial crack front geometry
are investigated next.
Fig. 5 Dimensions (in mm) of the CT specimen used in the mode I
fracture experiment
5.1.1 Mesh refinement and enrichment study

Figure 7a, c show the two user-provided meshes used in the


convergence studies presented in this section. Hereafter, the
sequence of meshes based on them and used for the simu-
lation of crack growth are denoted by “coarse” and “fine”
meshes. The meshes ignore the initial crack ahead of the
specimen notch. The crack is instead discretized using the
GFEM enrichment functions presented in Sects. 3.1.1 and
3.1.2. The nearly uniform meshes provided by the user are
automatically refined around the crack front as shown in
Fig. 7b and d such that the maximum edge length of ele-
ments intersected by the crack front, h max , is close to 0.1 mm.
Thus, h max /a ≈ 3.65 × 10−2 , where a = 2.74 mm is the
length of the initial sharp crack. This ratio follows the rec-
ommendations by Gupta et al. (2017). This same h max is
used in all simulations presented in this section. The problem
size using the coarse meshes and GFEM approximations of
degree p = 2 ranges between 41,751 and 121,347 equations.
In the case of simulations with fines meshes and p = 2, the
Fig. 6 Experimental setup showing the CT specimen (1), cameras (2), number of equations ranges between 194,661 and 227,475.
and loading grips (3), (Pham et al. 2017)
Figure 8 shows plots of load versus COD and crack length
versus COD. The crack length measurements shown in these
upward displacement of magnitude 1.5 mm is applied to the plots starts from the left edge of the specimen shown in
nodes of the top hole while its lateral movement is restricted. Fig. 5. The load is the reaction force at the nodes with pre-
To prevent rigid body motion, one node in each of the top scribed displacement as explained earlier. These quantities
and bottom loaded lines is restricted from movement in all are computed at each crack propagation step using the scal-
directions. A discussion on the influence of these modeling ing relations defined in Sect. 4.1.3. GFEM results for the
assumptions is presented in Sect. 5.2.5. sequence of coarse and fine meshes with p = 2 basis func-
tions and Δamax = 0.1 mm are shown in the figure, together
Remark 2 It is well known that point displacement bound- with the experimental results from Pham et al. (2017). The
ary condition is inconsistent in elasticity problems, except GFEM curves match well the experimental data. In addition,
to prevent rigid body motion (Szabo and Babuška 1991). the two GFEM curves in each plot are nearly identical. This
Numerical experiments not presented here were performed to is the case since the same element size is adopted near the
check the sensitivity of computed reaction forces at the nodes crack front for both sequences of meshes. The mesh refine-
with prescribed displacements. They show that, as long as a ment away from the crack front has no impact on the results

123
Validation of 3-D SGFEM for brittle fracture growth 139

Fig. 7 Discretization using coarse and fine meshes (a and c) and the
corresponding typical automatic mesh refinement around the crack front
during the fracture propagation (b and d) for the CT specimen used for
mode I fracture

for this problem. Based on this, hereafter only the coarse


mesh is used in the simulations.

Remark 3 The small “bumps” observed in the experimental


curves shown in Fig. 8 is an indication of the several instances
where the crack stopped growing, momentarily, while the
COD and the load continue to increase until at a later instant
when the crack restarts its growth (Pham et al. 2017). This
was attributed to the local non-homogeneity in the material b
of the specimen.
Fig. 8 Mesh convergence in the GFEM simulation of Specimen CT24
Figure 9 shows GFEM results for the sequence of coarse using Δamax = 0.1 mm and p = 2 a Load variation, b Crack length
meshes with approximations of degree p = 2 and p = 3, versus COD
and Δamax = 0.1 mm. The GFEM curves, again, match well
the experimental data and show that a quadratic basis suffices
for the sequence of coarse meshes adopted for this problem. 5.1.3 Effect of initial crack front geometry

5.1.2 Effect of crack advancement magnitude 1amax In all simulations presented earlier, the initial sharp crack cre-
ated from the specimen notch was assumed to be planar, with
The effect of crack advancement magnitude Δamax is con- a straight crack front, and of length a = 2.74 mm. However,
sidered in this section. Figure 10 shows plots of load versus the initial crack front generated by the razor blade impact
COD and crack length versus COD for the coarse mesh with for Specimen CT24 was not straight (Pham et al. 2017). In
p = 2 GFEM basis functions, and Δamax = 0.3, 0.1 and this section, we investigate the effects of the geometry of the
0.035 mm. Good agreement with the experimental data is initial crack front on the results presented so far.
observed in all cases. Hereafter, Δamax = 0.1 mm is adopted, A zoomed view of the micrograph of the initial crack sur-
unless stated otherwise. face of Specimen CT24 of Pham et al. (2017) is shown in

123
140 F. M. Mukhtar et al.

a a

b b
Fig. 9 p-adaptivity in the GFEM simulation of Specimen CT24 using Fig. 10 GFEM simulation of Specimen CT24 versus experimental
coarse meshes and Δamax = 0.1 mm a Load variation, b Crack length results for different crack increments using coarse meshes and p = 2 a
versus COD Load variation b Crack length versus COD

Fig. 11. The geometry of the curved crack front shown in the
figure was digitally extracted and used to define the geomet-
rical crack surface shown at the top of Fig. 12. The GFEM initial crack geometry is a drop in the threshold load value
simulation with the sequence of coarse meshes, p = 2, required to propagate the crack. The compliance of the model
and Δamax = 0.1 mm were repeated using this initial crack with a curved initial crack front is larger than the one with a
geometry. The results are compared in Fig. 13 with the pre- straight crack front. However, once the propagation starts, the
vious ones (based on a straight initial crack front) and the two GFEM curves become nearly identical. Figure 12 shows
experimental data for Specimen CT24. The length of the snapshots of the geometrical crack surface at propagation
crack with a curved front is measured at the side of the steps 40 and 120. Each layer of triangular facets (added to
specimen with a shorter initial crack, which approximately the initial surface) corresponds to a propagation step. The
coincides with the length of the initial crack with a straight snapshots show that the shorter portion of the crack surface
front. The only observed effect of the fabrication error in the catches up with the longer one after several propagation steps.

123
Validation of 3-D SGFEM for brittle fracture growth 141

Fig. 11 Micrograph of the fracture surface for Specimen CT24 (Pham


et al. 2017)

Fig. 12 Geometrical crack surface for the initial crack with a curved
front (top). The geometrical crack surface at propagation steps 40 and
120 are shown in the middle and bottom of the figure

This section illustrates the flexibility that the GFEM with


mesh adaptivity provides to the user. The method can handle
a broad range of initial crack geometries and changes to that
do not require any user change to the input 3-D mesh used b
in the simulations. Fig. 13 Influence of the initial crack front geometry in the GFEM sim-
ulation of Specimen CT24 using coarse mesh, Δamax = 0.1 mm, and
p = 2 a Load variation, b Crack length versus COD

5.1.4 Comparison with other specimens and phase-field


FEM
5.1.5 Computational performance
Figure 14 compares the GFEM simulation of Specimen CT24
(using coarse meshes, Δamax = 0.1 mm, and p = 2) with While the GFEM and phase-field results are close, as shown
experimental results from specimens CT21, CT22, and CT24 in Fig. 14, the computational cost of phase-field simulations
of Pham et al. (2017). The observed scatter in experimental is higher than for the GFEM with mesh adaptivity. The phase-
data is due to variations in initial sharp crack geometry, dif- field FEM model for Specimen CT24 adopts an unstructured
ferences in initial crack bluntness, etc. (Pham et al. 2017). mesh with 52,000 hexahedron elements. The element size
Results of a phase-field FEM simulation of Specimen CT24 in the region that contains the expected crack path is 25μm
reported in Pham et al. (2017) are also included in the figure. (Pham et al. 2017). The average maximum (minimum) edge
The GFEM and phase-field curves agree very well with the length of elements intersected by the crack front in the coarse
experimental results of Specimen CT24. GFEM meshes, like the one shown in Fig. 7(b), is h̄ max =

123
142 F. M. Mukhtar et al.

Fig. 15 Dimensions (in mm) of the modified CT specimen of Pham


a et al. (2017)

simulation. The superior computational performance of the


GFEM is due to: (i) smaller problem size since in the GFEM
refinement is performed along a line (crack front) while in
the phase-field FEM the mesh needs to be refined along the
entire crack surface; (ii) At each crack propagation step, a lin-
ear problem is solved by the GFEM and a highly non-linear
one is solved by the phase-field FEM.

5.2 Mode I+II fracture propagation: Modified CT


Specimen CT31 of Pham et al. (2017)

This section presents the validation of the adaptive GFEM


and the fracture propagation algorithm against data sets from
a mode I+II fracture propagation problem. The experimental
data are from Specimen CT31 of Pham et al. (2017) unless
stated otherwise. Figure 15 shows the geometry and dimen-
sions of the specimens of Pham et al. (2017) tested under
b mixed-mode conditions. Each specimen is created by drilling
Fig. 14 GFEM simulation of Specimen CT24 (with coarse meshes, a third hole in a CT specimen as shown in the figure. This
Δamax = 0.1 mm, and p = 2) compared with data from three speci- hole influences the crack path as shown later on, leading to
mens and phase-field FEM simulations reported in Pham et al. (2017). mode I+II fracture propagation.
a Load variation b Crack length versus COD
The material properties of the specimen and the prescribed
boundary conditions are the same as in Sect. 5.1. The bound-
ary conditions are modeled as displacements prescribed to
96μm (h̄ min = 22μmm). The average number of tetrahedron nodes of the GFEM mesh as described in Sect. 5.1.
elements in the sequence of coarse meshes is 12,658. Mesh refinement and enrichment studies are presented
It is reported in Pham et al. (2017) that the phase-field sim- next. The effect of crack advancement magnitude, Δamax ,
ulation of crack propagation in Specimen CT24 took about is also investigated. The actual geometry of the initial crack
two hours to complete on a machine with 80 processors. The front in Specimen CT31 is not available. Thus, its effect on
GFEM simulation of Specimen CT24 on the sequence of the GFEM results is not considered—the initial crack front is
coarse meshes using Δamax = 0.1 mm and p = 2 took a assumed to be straight in all simulations presented in this sec-
little less than 3 hours on a 2019 MacBook Pro with 8 cores. tion. The initial crack surface, however, is slightly curved and
The average time spent to solve one propagation step was 28 the sharp crack does not start exactly from the U-notch tip.
sec. and 365 propagation steps were solved during the entire This happens due to fabrication imperfections, and it results

123
Validation of 3-D SGFEM for brittle fracture growth 143

in both the initial crack tail and front having some offset
away from the intended nominal initial crack surface loca-
tion. Hence, the mixed-mode simulations presented in this
section model the initial crack surface as non-planar with
a vertical off-set estimated from the experimental results of
Pham et al. (2017). The adopted crack front is given by the
segment (x = 13.221, y = 0.374, −1.5 ≤ z ≤ 1.5) mm,
with the origin of the coordinate system at U-notch “tip”
(located at the middle plane of the specimen and at a dis-
tance of 40.64 mm from the right face of the specimen).

5.2.1 Mesh refinement and enrichment study

Figure 16a, c show the two user-provided meshes used in the


convergence studies presented in this section. Hereafter, the
sequence of meshes based on them and used for the simu-
lation of crack growth are denoted by “coarse” and “fine”
meshes. The meshes ignore the initial crack ahead of the
specimen notch as in the previous section. The nearly uni-
form meshes provided by the user are automatically refined
around the propagating crack front as shown in Fig. 16b, d
such that the maximum edge length of elements intersected
by the crack front, h max , is close to 0.1 mm. This same h max is
used in all simulations presented in this section. The problem
Fig. 16 Discretization using coarse and fine meshes (a and c) and the
size using the coarse meshes and GFEM approximations of corresponding typical automatic mesh refinement around the crack front
degree p = 2 ranges between 36,597 and 87,381 equations. during the fracture propagation (b and d) for the modified CT specimen
In the case of simulations with fines meshes and p = 2, the used for mixed-mode fracture
problem size ranges between 262,572 and 309,360.
Figure 17 shows plots of load versus COD and the pre-
dicted crack path. The load is the reaction force at the nodes
with prescribed displacement as explained earlier. These sonably well the experimental data and show that a quadratic
quantities are computed at each crack propagation step using basis suffices for the sequence of coarse meshes adopted for
the scaling relations defined in Sect. 4.1.3. GFEM results this problem. Figure 19 shows contour plots of von Mises
for the sequence of coarse and fine meshes with p = 2 stress field on the deformed configuration (with magnifi-
basis functions and Δamax = 0.1 mm are shown in the cation) at crack propagation steps 20 and 120. Figure 20
figure, together with the experimental results from Pham compares the crack path predicted by the GFEM with the
et al. (2017). The stiffness of the specimen before the crack experiment of Pham et al. (2017). The GFEM simulation
propagates is matched very well by the GFEM model. The adopts coarse meshes, Δamax = 0.1 mm, and p = 2. It is
predicted critical load for crack propagation is slightly higher clear that the GFEM path agrees well with the experimental
than the measured one but the GFEM curves match reason- one.
ably well the experimental data. The observed difference in
Fig. 17a between the GFEM curves and experimental data for
COD greater than ≈ 0.32 mm is related to the model adopted
for the boundary conditions at the holes of the specimen. This
is further discussed in Sect. 5.2.5. The two GFEM curves in 5.2.2 Effect of crack advancement magnitude 1amax
each plot shown in Fig. 17 are nearly identical. This is the
case since, as in the previous problem, the same element size The effect of crack advancement magnitude Δamax is consid-
is adopted near the crack front for both sequences of meshes. ered in this section. Figure 21 shows plots of load versus COD
Based on this, hereafter only the coarse mesh is used in the and the crack path for the coarse mesh with p = 2 GFEM
simulations. basis functions, and Δamax = 0.3, 0.1 and 0.035 mm. Good
Figure 18 shows GFEM results for the sequence of coarse agreement with the experimental data is observed in all cases.
meshes with approximations of degree p = 2 and p = 3, This justifies the adopted Δamax = 0.1 mm in the simulations
and Δamax = 0.1 mm. The GFEM curves, again, match rea- presented earlier.

123
144 F. M. Mukhtar et al.

a a

b
b
Fig. 18 p-adaptivity in the GFEM simulation of Specimen CT31 using
Fig. 17 Mesh convergence in the GFEM simulation of Specimen CT31 coarse meshes and Δamax = 0.1 mm a Load variation, b Crack path
using Δamax = 0.1 mm, and p = 2 a Load variation, b Crack path

of Specimen CT31 by Pham et al. (2017) and by Kumar and


5.2.3 Comparison with other specimens and phase-field Lopez-Pamies (2020). Phase-field results from Pham et al.
FEM (2017) are for two choices of the ratio l0 /h where l0 is an
intrinsic size scale for the damage model in the phase-field
Figure 22 compares the GFEM simulation of Specimen CT31 simulation and h is the element size in the region that contains
(using coarse meshes, Δamax = 0.1 mm, and p = 2) with the expected crack path. Unstructured meshes with small ele-
experimental results from specimens CT31, CT32, and CT33 ments (h = 25μm and h = 6.25μm ) are used (Pham et al.
of Pham et al. (2017). It can be observed that the load versus 2017). The GFEM and phase-field load versus COD curves
COD curve for Specimen CT33 deviates from the curves for agree very well with the experimental results of Specimen
specimens CT31 and CT32. The crack path from all experi- CT31. The crack paths predicted by GFEM is close to the
ments are, however, close for x > 17 mm. The GFEM results phase-field results of Pham et al. (2017) and nearly identical
are compared in Fig. 23 with phase-field FEM simulations to the phase-field results of Kumar and Lopez-Pamies (2020).

123
Validation of 3-D SGFEM for brittle fracture growth 145

Fig. 20 Final crack path for Specimen CT31 a GFEM simulation with
coarse meshes, Δamax = 0.1 mm, and p = 2, b Experiment of Pham
et al. (2017)

12,581. The average time spent to solve one propagation step


was 13 sec. and 127 propagation steps were solved during
the entire simulation. The reduced cost of a propagation step
compared to Specimen CT24 is due to the smaller problem
size at each propagation step. Fewer propagation steps are
required for the simulation of Specimen CT31 than Specimen
CT24 since the initial (final) crack of Specimen CT31 is
longer (shorter) than for Specimen CT24.

5.2.5 Effects of model adopted for boundary conditions at


CT specimen holes

The observed difference in, for example, Fig. 17a between


the GFEM load versus COD curves and experimental data is
discussed in this section.
The displacements imposed by the loading pins on the top
Fig. 19 Contour plots of von Mises stress field on the deformed con-
and bottom holes of the CT and modified CT specimens are
figuration (with magnification) for two crack propagation steps modeled as displacements prescribed to nodes located along
the load lines, as described in Sect. 5.1. The actual loading
is, however, a contact problem where one of the contacting
5.2.4 Computational performance surfaces (the specimen’s circular surfaces of the hole faces)
rotates relative to the vertically moving cylindrical loading
The GFEM simulation of Specimen CT31 on the sequence pin. As such, the location of the effective loaded zone changes
of coarse meshes using Δamax = 0.1 mm and p = 2 took during the experiment. The adopted model, on the other hand,
about 27.5 minutes on a 2019 MacBook Pro with 8 cores. The assumes that the loaded/contacting surfaces remain the same
average number of tetrahedron elements in these meshes is throughout the test. The loading pins do not prevent rotation

123
146 F. M. Mukhtar et al.

a
a

b b
Fig. 21 GFEM simulation of Specimen CT31 versus experimental Fig. 22 GFEM simulation of Specimen CT31 (with coarse meshes,
results for different crack increments using coarse meshes and p = 2 a Δamax = 0.1 mm, and p = 2) compared with data from three speci-
Load variation, b Crack path mens of the modified CT test of Pham et al. Pham et al. (2017). a Load
variation b Crack path

of the specimen. This motivated the adoption of point con-


straints in our model. If, however, displacements were instead imen larger than in the Mode I problem. This may explain the
prescribed on element faces, the relative rolling movement better matching between the GFEM load versus COD curves
between the loading pins and the holes would be prevented, and experimental data shown in Fig. 8a (Mode I problem)
leading to an excessively constrained model. than in Fig. 17a (Mode I+II problem).
It is noted that the rotation of the CT specimen analyzed The relative motion between the loading pins and the mod-
in Sect. 5.1 is smaller than in the case of the specimen ana- ified CT specimen holes is illustrated in Fig. 24. The locations
lyzed in this section (Pham et al. 2017). In the first case the of the contact on the hole surfaces change from A to A and
crack propagates along a plane of symmetry and the speci- A at the top and bottom holes, respectively, as the speci-
men rotation is small (Pham et al. 2017). The turning of the men deforms. This movement is prevented by the adopted
crack in the Mode I+II problem causes a rotation of the spec- boundary conditions.

123
Validation of 3-D SGFEM for brittle fracture growth 147

[Pham et al.(2017)]
[Pham et al.(2017)]

a
Fig. 24 Illustration of the deformed (black dashed line) and unde-
formed (solid blue line) configurations of the modified CT specimen
[Pham et al. (2017)] showing the rotations of points along the loaded holes
[Pham et al. (2017)]

[Kumar and Lopez-Pamies (2020)]

Fig. 25 Geometry (with dimensions in inches) and boundary conditions


of the beam with holes under 3-point loading (Bittencourt et al. 1996;
Ingraffea and Grigoriu 1990)

b
crack. Two case are considered: Case (i) adopts an initial
Fig. 23 GFEM simulation of Specimen CT31 (with coarse meshes, crack size a = 1 in (25.4 mm) and its distance to the beam
Δamax = 0.1 mm, and p = 2) compared with phase-field FEM simu-
lations by Pham et al. (2017) and by Kumar and Lopez-Pamies (2020).
centerline b = 6 in (152.4 mm). These parameters are illus-
a Load variation, b Crack path trated in Fig. 25; Case (ii) adopts a = 2.5 in (63.5 mm) and
b = 6 in (152.4 mm). The specimens were prepared out of a
plexiglass sheet with Modulus of elasticity E = 450, 000 psi
5.3 Mode I+II fracture propagation: Cornell fracture (3.102 GPa), Poisson’s ratio ν = 0.35 and Fracture energy
group beam with holes G c = 1.73 lbf/in (0.304 N/m).
This problem, or some slight variation of it, has been
This section presents the validation of the adaptive GFEM used to validate several methods and fracture propagation
using experiments conducted by the Cornell University frac- algorithms. Among them, we can mention quarter-point
ture group (Ingraffea and Grigoriu 1990; Bittencourt et al. finite elements with remeshing (Bittencourt et al. 1996), the
1996). The tests adopt deep beams with a series of holes Element-Free Galerkin Method (Ventura et al. 2002), phase-
as illustrated in Fig. 25. The goal is to validate the capabil- field FEMs (Mesgarnejad et al. 2015; Wu et al. 2019), and
ity of the GFEM in predicting crack paths that are sensitive eXtended FEM (Loehnert 2014). We comment of the com-
to the size and location, relative to the holes, of the initial putational cost of the GFEM later in this section.

123
148 F. M. Mukhtar et al.

Fig. 27 Comparison between GFEM crack path and experimental


results (Ingraffea and Grigoriu 1990; Bittencourt et al. 1996) for Case
1 (a = 2.5 in and b = 6 in)

Fig. 26 GFEM simulation of fracture propagation in a beam with holes.


a User-provided mesh, b Predicted crack path for Case 1, and c Predicted
crack path for Case 2

Figure 26a shows the user-provided mesh adopted in the


simulations presented in this section. The same mesh is used
for both initial fracture cases described above. The input Fig. 28 Comparison between GFEM crack path and experimental
results (Ingraffea and Grigoriu 1990; Bittencourt et al. 1996) for Case
mesh is nearly uniform and has 98,644 tetrahedral elements. 2 (a = 2.5 in and b = 6 in)
Elements intersected by the crack front were refined as in
previous problems. The average number of tetrahedron ele-
ments used in the sequence of meshes to solve cases (i) and agreement is very good for both cases, in spite of using a rela-
(ii) was 111,150 and 111,005, respectively. A polynomial tively small number of 3-D elements—thanks for the adopted
order of p = 2 and a value of Δamax = 1 mm were adopted singular enrichments and mesh adaptivity. The average time
in the simulations. Simulations with Δamax = 0.1 mm were spent to solve one propagation step was 35.43 s. and 41.35 s.
also performed and produced the same crack paths as those for cases (i) and (ii), respectively. A total of 103 and 119 prop-
with Δamax = 1 mm. agation steps were solved in the crack propagation simulation
Figure 26b, c show predicted crack paths for cases 1 and of cases (i) and (ii), respectively. The 2-D phase-field FEM
2 described earlier. The sensitivity of the crack path with model of Mesgarnejad et al. (2015) uses a mesh with approxi-
respect to initial crack size and distance between the crack mately 1.35 million linear triangular elements, which is about
and holes is clear. Case 1 leads to the arrest of the crack at one order of magnitude more elements than in the adopted
the middle hole while in Case 2 the crack arrests at the top 3-D GFEM meshes. This problem, with different choices for
hole. parameters a and b, was solved by Loehnert (2014) using a
The GFEM crack paths for cases 1 and 2 are compared 2-D XFEM and a mesh of 272,800 bi-linear elements, which
with experimental results (Ingraffea and Grigoriu 1990; Bit- is more than double the number of elements in the adopted
tencourt et al. 1996) in Figs. 27 and 28 , respectively. The 3-D GFEM meshes.

123
Validation of 3-D SGFEM for brittle fracture growth 149

6 Conclusions Further validation of the GFEM presented here will be


pursued. This includes, for example, problems with mode III
An adaptive Generalized FEM for the simulation of mixed- effects and problems with multiple spatial scales of interest.
mode brittle fracture propagation is presented and validated
against several experimental data sets for mode I and mode Acknowledgements F. M. Mukhtar gratefully acknowledges the partial
support from King Fahd University of Petroleum & Minerals. Ph. D.
I+II fracture propagation problems. Very good agreement Alves and C. A. Duarte gratefully acknowledge the partial support from
between the GFEM and experimental results is observed. the University of Illinois at Urbana-Champaign.
These include fracture path, Crack Opening Displacement
(COD), and load and fracture length versus COD curves. The
computational efficiency of the method is also assessed. The Compliance with ethical standards
simulation cost of the GFEM is shown to be significantly
Conflicts of interest The authors declare that they have no conflict of
smaller than state-of-the-art phase-field FEMs. This is due interest.
to (i) smaller problem size since in the GFEM refinement is
required only along a line (crack front) while in the phase-
field FEM the mesh needs to be refined along the entire crack A The displacement correlation method
surface; (ii) At each crack propagation step, a linear problem
is solved by the GFEM and a highly non-linear one is solved This section presents the expression for the extraction of SIFs
by the phase-field FEM. based on the Displacement Correlation Method (DCM)
Three benchmark problems with available experimental (Chan et al. 1970; Banks-Sills and Sherman 1986)
data are used for the validation of the GFEM formulation and and the averaging scheme adopted in this paper. A more
crack propagation algorithms. The first problem is a Compact detailed derivation can be found in Gupta et al. (2017).
Tension (CT) specimen under mode I loading. Mesh refine- The approximation of K I can be computed as follows
ment and enrichment studies are presented and show that a
polynomial order p = 2, combined with mesh refinement %
2π G
around the crack front, suffices to deliver accurate results. K I∗ (r ) = u y (r ), (36)
r κ +1
The effect of crack advancement magnitude and initial crack
front geometry are also investigated. They demonstrate the where
flexibility of the method in adopting different starting crack
geometries while adopting the same user-defined mesh. u y (r ) = u y (r , π ) − u y (r , −π ) (37)
The second benchmark problem is a mode I+II fracture
propagation simulation in a modified CT specimen. Dis- is the displacement jump in the ȳ-direction of the crack
cretization parametric studies are also performed for this front coordinate system (cf. Fig. 2).
problem with the same conclusions as for the mode I problem. The error of this approximation is O(r ). It can be improved
The effects of the model adopted for boundary conditions at using Richardson extrapolation Heath (1997) which is given
CT specimen holes are discussed in details. They show that by
the contact between the loading pins and the specimen needs
to be modeled in order to remove observed inconsistencies
K I∗ (ra ) = K I + C̃ra
of the adopted model. We plan to investigate this by combin-
ing, in a non-intrusive manner (Gupta et al. 2012), the GFEM K I∗ (rb ) = K I + C̃rb , (38)
code used in the simulations presented in this paper with a
standard FEM code like Abaqus [1]. This will leverage and where ra and rb define the location of sampling points
complement the capabilities of the methods and algorithms with ra < rb .
implemented in these codes. The above equations can be solved for K I
The third benchmark problem aimed at validating the  
capability of the GFEM in predicting mode I+II crack paths rb ra ∗
KI = K I∗ (ra ) − K I (rb ) . (39)
in deep beams with a series of holes. These paths are sensitive rb − ra rb
to the size and location of initial cracks relative to the holes.
The crack paths predicted by the GFEM once again matched This approximation of K I has an error of O(r 2 ) instead
very well experimental results (Ingraffea and Grigoriu 1990; of O(r ). Analogous expressions can be derived to calculate
Bittencourt et al. 1996). The number of 3-D finite elements approximations for K I I and K I I I .
used by the GFEM is shown to be significantly smaller the The averaging scheme is adopted in this work. In this
number of 2-D elements used by a phase-field FEM. strategy, several pairs of sampling points at ra and rb are
used to compute approximations of K I with Eq. (39). The

123
150 F. M. Mukhtar et al.

final approximation of K I is taken as the average of those Bergara A, Dorado J, Martin-Meizoso A, Martínez-Esnaola J (2017)
approximations. Fatigue crack propagation in complex stress fields: experiments
and numerical simulations using the extended finite element
Let K I∗ (rk ) be an approximation of K I evaluated at rk
method (XFEM). Int J Fatigue 103:112–121. https://doi.org/10.
using Eq. (36), k = 1, ..., Nsampl , with rk < rk+1 . Nsampl 1016/j.ijfatigue.2017.05.026
denotes the total number of sampling points used in the aver- Bittencourt T, Wawezynek P, Ingraffea A, Sousa J (1996) Quasi-
aging scheme. automatic simulation of crack propagation for 2D LEFM problems.
Eng Fract Mech 55(2):321–334. https://doi.org/10.1016/0013-
Let K I ,k , be an approximation of K I computed using 7944(95)00247-2
Eq. (39) with ra = rk and rb = rk+1 = rk + Δk , i.e., Bravos CMAA (2004) Um sistema de refinamento h-p adaptativo uti-
  lizando elementos finitos hierarquicos multidimensionais. Ph.D.
rk+1 rk thesis, Faculdade de Engenharia Mecanica, Universidade Estad-
K I ,k = K I∗ (rk ) − K I∗ (rk+1 ) . (40) ual de Campinas
Δk rk+1
CGAL: Computational Geometry Algorithms Library (2012). http://
www.cgal.org
A simple average of K I ,k , k = 1, . . . , Nsampl − 1, gives Chan S, Tuba I, Wilson W (1970) On the finite element method in linear
the final approximation for K I fracture mechanics. Eng Fract Mech 2:1–17
Dolbow J, Moës N, Belytschko T (2000) Discontinuous enrichment in
Nsampl −1 finite elements with a partition of unity method. Finite Elem Analy
1 
Des 36:235–260
KI = K I ,k . (41)
Nsampl − 1 Duarte C (2016) ISET-An Adaptive Generalized Finite Element Solver,
k=1 Reference Manual. ISET Developers, http://gfem.cee.illinois.edu
Duarte C, Babuška I, Oden J (2000) Generalized finite element
Analogous averaging procedures are used for K I I and methods for three dimensional structural mechanics prob-
KI I I . lems. Comput Struct 77:215–232. https://doi.org/10.1016/S0045-
In this paper, the same distance between two consecu- 7949(99)00211-4
Erdogan F, Sih G (1963) On the crack extension in plates under plane
tive sampling points is adopted, i.e., Δk = Δ = 0.1h min loading and transverse shear. J Basic Eng 85:519–527
where h min is the minimum edge size of the elements cut by Ferrie E, Buffiere JY, Ludwig W, Gravouil A, Edwards L (2006) Fatigue
the crack front. The location of the first sampling point is crack propagation: In situ visualization using X-ray microtomog-
defined by ρ̄ = r1 /h min = 0.4. The total number of sam- raphy and 3D similation using the extended finite element method.
Acta Materialia 54:1111–1122
pling points used for the averaging scheme is Nsampl = 10. Garzon J, O’Hara P, Duarte C, Buttlar W (2014) Improvements of
These parameters are selected based on recommendations explicit crack surface representation and update within the gener-
from Gupta et al. (2017). alized finite element method with application to three-dimensional
crack coalescence. Int J Num Methods Eng 97(4):231–273. https://
doi.org/10.1002/nme.4573
Giner E, Sukumar N, Tarancón JE, Fuenmayor FJ (2009)
References An Abaqus implementation of the extended finite element
method. Eng Fract Mech 76(3):347–368. https://doi.org/10.1016/
Abaqus: Version 6.14 Documentation. Dassault Systemes Simulia Cor- j.engfracmech.2008.10.015
poration, Providence, RI, USA (2014) Grégoire D, Maigre H, Réthoré J, Combescure A (2007) Dynamic crack
Babuška I, Banerjee U (2012) Stable generalized finite element method propagation under mixed-mode loading—comparison between
(SGFEM). Comput Methods Appl Mech Eng 201–204:91–111. experiments and X-FEM simulations. Int J Solids Struct
https://doi.org/10.1016/j.cma.2011.09.012 44(20):6517–6534. https://doi.org/10.1016/j.ijsolstr.2007.02.044
Babuška I, Melenk J (1997) The partition of unity method. Int J Num Griffith AA (1921) The phenomena of rupture and flow in solids. Philo-
Methods Eng 40:727–758 sophical Transactions of the Royal Society of London. Series
Babuška I, Caloz G, Osborn J (1994) Special finite element methods for A, Containing Papers of a Mathematical or Physical Character
a class of second order elliptic problems with rough coefficients. 221:163–198
SIAM J Num Analy 31(4):945–981 Gupta P, Duarte C (2014) Simulation of non-planar three-dimensional
Babuška I, Banerjee U, Kergrene K (2017) Strongly stable generalized hydraulic fracture propagation. Int J Num Analyt Methods
finite element method: application to interface problems. Com- Geomech 38:1397–1430. https://doi.org/10.1002/nag.2305
put Methods Appl Mech Eng 327:58–92. https://doi.org/10.1016/ Gupta P, Pereira J, Kim DJ, Duarte C, Eason T (2012) Analysis of
j.cma.2017.08.008 three-dimensional fracture mechanics problems: a non-intrusive
Banks-Sills L, Sherman D (1986) Comparison of methods for calculat- approach using a generalized finite element method. Eng Fract
ing stress intensity factors with quarter-point elements. Int J Fract Mech 90:41–64. https://doi.org/10.1016/j.engfracmech.2012.04.
32:127–140 014
Barsoum R (1976) On the use of isoparametric finite elements in linear Gupta P, Duarte C, Dhankhar A (2017) Accuracy and robustness
fracture mechanics. Int J Num Methods Eng 10(1):25–37. https:// of stress intensity factor extraction methods for the general-
doi.org/10.1002/nme.1620100103 ized/eXtended Finite Element Method. Eng Fract Mech 179:120–
Belytschko T, Black T (1999) Elastic crack growth in finite elements 153. https://doi.org/10.1016/j.engfracmech.2017.03.035
with minimal remeshing. Int J Num Methods Eng 45:601–620 Heath M (1997) Scientific computing: an introductory survey. McGraw-
Belytschko T, Gracie R, Ventura G (2009) A review of Hill series in computer science, McGraw-Hill
extended/generalized finite element methods for material model- Henshell R, Shaw K (1975) Crack tip finite elements are unnecessary.
ing. Model Simul Mater Sci Eng 17:043001. https://doi.org/10. Int J Num Methods Eng 9(3):495–507. https://doi.org/10.1002/
1088/0965-0393/17/4/043001 nme.1620090302

123
Validation of 3-D SGFEM for brittle fracture growth 151

Ingraffea AR, Grigoriu M (1990) Probabilistic fracture mechanics: A Pereira J, Duarte C, Jiao X (2010) Three-dimensional crack growth with
validation of predictive capability. Report 90-8 Department of hp-generalized finite element and face offsetting methods. Comput
Structural Engineering Cornell University Mech 46(3):431–453. https://doi.org/10.1007/s00466-010-0491-
Kergrene K, Babuška I, Banerjee U (2016) Stable generalized finite 3
element method and associated iterative schemes; application to Pham K, Ravi-Chandar K (2014) Further examination of the criterion
interface problems. Comput Methods Appl Mech Eng 305:1–36. for crack initiation under mixed-mode i+ iii loading. Int J Fract
https://doi.org/10.1016/j.cma.2016.02.030 189(2):121–138
Kumar A, Lopez-Pamies O (2020) Personal Communication Pham K, Ravi-Chandar K (2017) The formation and growth of echelon
Kuna M (2013) Finite elements in fracture mechanics: theory “numer- cracks in brittle materials. Int J Fract 206(2):229–244
ics” applications. Springer, Dordrecht Pham K, Ravi-Chandar K, Landis C (2017) Experimental validation of
Lazarus V (2003) Brittle fracture and fatigue propagation paths a phase-field model for fracture. Int J Fract 205(1):83–101. https://
of 3D plane cracks under uniform remote tensile loading. doi.org/10.1007/s10704-017-0185-3
Int J Fract 122(1–2):23–46. https://doi.org/10.1023/B:FRAC. Richard HA, Fulland M, Sander M (2005) Theoretical crack path pre-
0000005373.73286.5d diction. Fatigue Fract Eng Mater Struct 28(1–2):3–12. https://doi.
Lazarus V, Leblond JB, Mouchrif SE (2001) Crack front rotation and org/10.1111/j.1460-2695.2004.00855.x
segmentation in mixed mode i+ iii or i+ ii+ iii. Part ii: Comparison Rungamornrat J, Wheeler M, Mear M (2005) A numerical technique
with experiments. J Mech Phys Solids 49(7):1421–1443 for simulating nonplanar evolution of hydraulic fractures. In: Pro-
Leblond JB, Frelat J (2014) Development of fracture facets from a ceedings of the SPE Annual Technical Conference and Exhibition,
crack loaded in mode i+ iii: solution and application of a model p. 9. Dallas, TX, USA SPE-96968-MS
2d problem. J Mech Phys Solids 64:133–153 Sanchez-Rivadeneira A, Duarte C (2019) A stable generalized/extended
Lin B, Mear M, Ravi-Chandar K (2010) Criterion for initiation of cracks FEM with discontinuous interpolants for fracture mechanics.
under mixed-mode i+ iii loading. Int J Fract 165(2):175–188 Comput Methods Appl Mech Eng 345:876–918. https://doi.org/
Loehnert S (2014) A stabilization technique for the regularization of 10.1016/j.cma.2018.11.018
nearly singular extended finite elements. Comput Mech 54(2):523– Sanchez-Rivadeneira A, Shauer N, Mazurowski B, Duarte C (2020)
533. https://doi.org/10.1007/s00466-014-1003-7 A stable generalized/extended p-hierarchical FEM for three-
Melenk J, Babuška I (1996) The partition of unity finite element method: dimensional linear elastic fracture mechanics. Comput Methods
Basic theory and applications. Comput Methods Appl Mech Eng Appl Mech Eng 364:112970. https://doi.org/10.1016/j.cma.2020.
139:289–314 112970
Mesgarnejad A, Bourdin B, Khonsari M (2015) Validation simulations Schöllmann M, Richard H, Kullmer G, Fulland M (2002) A new crite-
for the variational approach to fracture. Comput Methods Appl rion for the prediction of crack development in multiaxially loaded
Mech Eng 290:420–437 structures. Int J Fract 117(2):129–141. https://doi.org/10.1023/A:
Moës N, Dolbow J, Belytschko T (1999) A finite element method for 1020980311611
crack growth without remeshing. Int J Num Methods Eng 46:131– Shabir Z, Van der Giessen E, Duarte C, Simone A (2019) On the appli-
150 cability of linear elastic fracture mechanics scaling relations in the
Ndeffo M, Massin P, Moës N, Martin A, Gopalakrishnan S (2017) On analysis of intergranular fracture of brittle polycrystals. Int J Fract
the construction of approximation space to model discontinuities 220(2):205–219. https://doi.org/10.1007/s10704-019-00381-x
and cracks with linear and quadratic extended finite elements. Adv Shi J, Chopp D, Lua J, Sukumar N, Belytschko T (2010) Abaqus imple-
Model Simul Eng Sci 4:1. https://doi.org/10.1186/s40323-017- mentation of extended finite element method using a level set
0090-3 representation for three-dimensional fatigue crack growth and life
Nikfam M, Zeinoddini M, Aghebati F, Arghaei A (2019) Experimental predictions. Eng Fract Mech 77(14):2840–2863. https://doi.org/
and XFEM modelling of high cycle fatigue crack growth in steel 10.1016/j.engfracmech.2010.06.009
welded T-joints. Int J Mech Sci 153–154:178–193. https://doi.org/ Sommer E (1969) Formation of fracture ‘lances’ in glass. Engineering
10.1016/j.ijmecsci.2019.01.040 Fracture Mechanics 1(3):539–546
Nuismer R (1975) An energy release rate criterion for mixed mode Szabo B, Babuška I (1991) Finite Elem Analy. Wiley, New York
fracture. Int J Fracture 11:245–250. https://doi.org/10.1007/ Tian R, Wen L, Wang L (2019) Three-dimensional improved XFEM
BF00038891 (IXFEM) for static crack problems. Comput Methods Appl Mech
Oden J, Duarte C, Zienkiewicz O (1998) A new cloud-based hp finite Eng 343:339–367. https://doi.org/10.1016/j.cma.2018.08.029
element method. Comput Methods Appl Mech Eng 153:117–126. Varfolomeev I, Burdack M, Moroz S, Siegele D, Kadau K (2014)
https://doi.org/10.1016/S0045-7825(97)00039-X Fatigue crack growth rates and paths in two planar specimens
Oden J, Duarte C (1997) Chapter: Clouds, Cracks and FEMs. In: under mixed mode loading. Int J Fatigue 58:12–19. https://doi.
B. Reddy (ed.) Recent Developments in Computational and org/10.1016/j.ijfatigue.2013.04.013
Applied Mechanics, pp. 302–321. International Center for Numer- Ventura G, Xu J, Belytschko T (2002) A vector level set method and
ical Methods in Engineering, CIMNE, Barcelona, Spain http:// new discontinuity approximations for crack growth by efg. Int J
gfem.cee.illinois.edu/papers/jMartin_color.pdf Num Methods Eng 56:923–944
Paris A, Erdogan F (1963) A critical analysis of crack propagation laws. Ventura G, Budyn E, Belytschko T (2003) Vector level sets for descrip-
J Basic Eng 85:528–534 tion of propagating cracks in finite elements. Int J Num Methods
Pereira J (2010) Generalized finite element methods for three- Eng 58(10):1571–1592. https://doi.org/10.1002/nme.829
dimensional crack growth simulations. PhD Dissertation, Univer- Wu J, Nguyen V, Nguyen C, Sutula D, Bordas S, Sinaie S (2019) Phase
sity of Illinois at Urbana-Champaign Urbana. IL, USA field modelling of fracture. Adv Appl Mech. https://doi.org/10.
Pereira J, Duarte C, Guoy D, Jiao X (2009) Hp-Generalized FEM and 1016/j.cma.2019.112614
crack surface representation for non-planar 3-D cracks. Int J Num Zhang Q, Banerjee U, Babuška I (2014) Higher order stable general-
Methods Eng 77(5):601–633. https://doi.org/10.1002/nme.2419 ized finite element method. Numerische Mathematik 128(1):1–29.
Pereira J, Duarte C, Jiao X, Guoy D (2009) Generalized finite element https://doi.org/10.1007/s00211-014-0609-1
method enrichment functions for curved singularities in 3D frac-
ture mechanics problems. Comput Mech 44(1):73–92. https://doi.
org/10.1007/s00466-008-0356-1

123
152 F. M. Mukhtar et al.

Zhang Q, Babuška I, Banerjee U (2016) Robustness in stable gen- Publisher’s Note Springer Nature remains neutral with regard to juris-
eralized finite element methods (SGFEM) applied to Poisson dictional claims in published maps and institutional affiliations.
problems with crack singularities. Comput Methods Appl Mech
Eng 311:476–502. https://doi.org/10.1016/j.cma.2016.08.019

123

You might also like