JID: ADES

ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

Advances in Engineering Software 0 0 0 (2017) 1–14

Contents lists available at ScienceDirect

Advances in Engineering Software

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

Research paper

A computational framework for G/XFEM material nonlinear analysis

A.B. Monteiro∗, A.R.V. Wolenski, F.B. Barros, R.L.S. Pitangueira, S.S. Penna
Graduate Program in Structural Engineering from School of Engineering, Federal University of Minas Gerais, Antonio Carlos Avenue 6627, Pampulha,
31270-901 Belo Horizonte, MG, Brazil

a r t i c l e i n f o a b s t r a c t

Article history: The Generalized/eXtended Finite Element Method (G/XFEM) has been developed with the purpose of
Received 11 April 2017 overcoming some limitations inherent to the Finite Element Method (FEM). Different kinds of functions
Revised 12 July 2017
can be used to enrich the original FEM approximation, building a solution specially tailored to problem.
Accepted 6 August 2017
Certain obstacles related to the nonlinear analysis can be mitigated with the use of such strategy and
Available online xxx
the damage and plasticity fronts can be precisely represented. A FEM computational environment has
Keywords: been previously enclosed the G/XFEM formulation to linear analysis with minimum impact in the code
G/XFEM structure and with requirements for extensibility and robustness. An expansion of the G/XFEM imple-
Nonlinear analysis mentation to physically nonlinear analysis under the approach of an Unified Framework for constitutive
Constitutive models models based on elastic degradation is firstly presented here. The flexibility of the proposed framework
is illustrated by several examples with different constitutive models, enrichment functions and analysis
models.
© 2017 Elsevier Ltd. All rights reserved.

1. Introduction Investigating these capabilities, Torres and Proena [5] used

G/XFEM in 3D numerical simulations to solve structural problems
The Finite Element Method (FEM) is a quite consolidated nu- with damage and plasticity (coupled or not). For this purpose, con-
merical technique in the study of several structural engineering stitutive models of elastoplasticity (with von Mises function), of
problems, but it presents some limitations related specially to the brittle evolutive damage [6] and of elastoplasticity with damage
description of the behavior of phenomena such as crack and dam- [7] were employed.
age propagation and large deformations. The nature of these phe- In [8] G/XFEM was extended to treat progressive damage with
nomena leads to the modification of the mesh in a very costly pro- a p adaptive strategy. The authors analyzed a reinforced concrete
cess. This issue encouraged the proposition of methods such as the beam with progressive damage described by the Mazars model [6].
ones based on meshless formulations. The study of Elguedj et al. [9] evaluated crack propagation due to
As an interface between meshless methods and the FEM, meth- fatigue using plastic materials in the G/XFEM framework with the
ods based on elements and on the concept of Partition of Unity enrichment functions proposed by Elguedj et al. [10], incorporating
(PU) have been developed such as the Partition of Unity Finite El- material properties on such functions. Combescure et al. [11] de-
ement Method (PUFEM), the Generalized Finite Element Method veloped and implemented the G/XFEM framework with enrich-
(GFEM) and the eXtended Finite Element Method (XFEM). The fun- ment functions to reproduce a cohesive crack propagation [12,13].
damental concepts of the PUFEM were introduced by Melenk and Piedade Neto et al. [14] treated the nonlinear analysis of Solid Me-
Babuška [1] and Babuška and Melenk [2] and they are grounded chanics through G/XFEM to contact problems.
in the enrichment strategy based on the multiplication of the PU In the paper of Crt et al. [15], crack propagation due to initia-
by customized functions. Posteriorly, Belytschko and Black [3] pre- tion, growth and coalescence of voids in structures was evaluated
sented the XFEM with minimum refinement of the mesh due to enriching the approximation with Heaviside and singular functions.
the enrichment of the approximations of FEM with discontinuous Broumand and Khoei [16] employed the G/XFEM framework (with
functions associated with the presence of crack. The GFEM and Heaviside enrichment function) to simulate the full process of dy-
XFEM are basically identical methods [4], so hereafter the term namic ductile fracture based on a nonlocal damage viscoplasticity
G/XFEM will be used to refer to these methods. model. Freitas et al. [17] noted that the convergence rates of the
G/XFEM in the modeling of cracks are similar to the FEM ones
since a geometric enrichment with crack-type functions is used
∗
Corresponding author. around the crack tip.
E-mail address: [email protected] (A.B. Monteiro).

http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
0965-9978/© 2017 Elsevier Ltd. All rights reserved.

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

2 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

For quasi brittle materials, Unger et al. [18] used a cohesive

model for the tractions across the crack and enrichment crack tip
functions that allow the representation of the singular stress fields.
Mos and Belytschko [13] applied the G/XFEM to model the crack
growth in concrete considering a cohesive law on the crack faces.
Mariani and Perego [19] discussed a methodology to simulate the
propagation of cohesive crack in quasi brittle materials, applying
Heaviside cubic polynomial enrichment functions. Recently, Kim
and Duarte [20] applied the G/XFEM-global-local in 3D analysis of
cohesive crack propagation in concrete structures.
Fig. 1. PU functions to the cloud ωj [21].
Many of the researches previously listed were not concerned
with the issue of computational implementation. In this paper, that
continues and complements the research of Alves et al. [21], an ob-
ject oriented design of G/XFEM to physically nonlinear analysis is
firstly introduced, following the literature discussed here. An FEM
programming environment proposed by Fonseca and Pitangueira
[22] and study in [23], so called INSANE, has been expanded by
Alves et al. [21] enclosing the standard version of G/XFEM formula-
tion to linear analysis with minimum impact in the code structure.
Requirenments of extensibility and robustness were adopted, mak-
ing it possible to combine different kinds of elements and analysis Fig. 2. Local approach and cloud enrichment ωj [21].
models with the G/XFEM enrichment strategies
Furthermore, Gori et al. [24] performed, in the INSANE system,
an expansion of the theoretical framework proposed by Carol et al. The clouds are formed by sets of finite elements that share the
[25], making the constitutive models implementation independent same nodal points xj (ωj in Fig. 1(a)).
on the analysis model and on the adopted numerical method. Such The Lagrangian Finite Element functions Nj , Fig. 1(b), associated
unified framework enables the use of different constitutive models with each one of the n nodes, can be considered as a PU because
in the numerical examples performed in this paper, demonstrating for any position x:
the versatility of the G/XFEM implementation to physically nonlin-

n
ear analysis. N j ( x ) = 1. (1)
In this context, the main purpose of this paper is to demon- j=1
strate the G/XFEM enrichment capacity to nonlinear analysis,
which flexibilizes the change of the approach along the domain A set of enrichment functions, so-called local approximation
according to the characteristics of the analyzed problem, provid- functions, Fig. 2(a) is composed by qj linearly independent func-
ing different ways of investigating the problem and performing tions defined to each node xj with support on the cloud ωj :
relevant numerical simulations. The intrinsic ability of G/XFEM to
develop customized approximations for specific applications is ex- I j = {L j1 (x ), L j2 (x ), ..., L jq (x )} = {L ji (x )}qi=1 , with L ji (x ) = 1.
tremely useful to nonlinear problems and if classical FEM is used, (2)
the meshes can not properly model the singularities present.
At the end of the process, the shape functions φ ji (x) of G/XFEM,
This paper is organized as follows: a brief explanation of
shown in Fig. 2(b), associated with the node xj are built through
G/XFEM are presented in Section 2. The INSANE class organization
the enrichment of the PU functions by the components of the set
and its expansion to include G/XFEM method to nonlinear analysis
I j.
are discussed in Section 3. In order to validate and justify the rele-
Thus, according to Eq. (3), φ ji (x) can be obtained by the product
vance of the studied procedure, nonlinear numerical examples are
between the basic functions that form the PU (Fig. 1(b)) and the
discussed in Section 4. The final is devoted to concluding remarks
enrichment functions (Fig. 2(a)).
and discussion.
{φ ji }qi=1 = N j (x ) × {L ji }qi=1 (no summation on j ). (3)
2. G/XFEM, a summary explanation A generic approximation u˜ is obtained by the following linear
combination of the shape functions:
The G/XFEM can be considered a variation of the conven-  
tional Finite Element Method (FEM). According to Duarte et al. [26], 
N 
q

G/XFEM was independently proposed by: u˜ (x ) = N j (x ) u j + L ji (x )b ji , (4)

j=1 i=2
• Babuška et al., initially named Special Finite Element Method
where uj and bji are nodal parameters associated with standard
[27], and later as the Partition of Unity Method [1].
(N j ) and G/XFEM (N j × L ji (x )) shape functions, respectively.
• Duarte and Oden, as a meshless formulation in the hp cloud
The function u˜ (x ) presents the approximative characteristics of
method [28] and [29], and later as a new cloud-based hp finite
the local approximation function while inherits the compact sup-
element method [30].
port of the PU.
• Belytschko [3], [31], for discontinuous solutions, named eX-
tended Finite Element Method.
3. INSANE system, object oriented design
The strategy used in G/XFEM consists of PU functions to de-
fine the shape functions. The conventional functions of FEM (such The INSANE (INteractive Structural ANalysis Environment) sys-
as Lagrangian functions) facilitates the application of the G/XFEM tem [21–23] is an open source software for computational me-
and, differently from the meshless methods, directly verifies the chanics. It is based on the Java language, and relies on the object-
boundary conditions [8]. oriented paradigm. There is a constant concern in conceiving the

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14 3

Fig. 3. Numerical core. Fig. 4. UML to obtain the incremental stiffness matrix.

implementations at the highest level of generalization, permit- increment, δ Uij , is calculated from it:
ting future applications with minimum changes in the pre-existing
δ Uij = δλij δ UPj + δ UQj ,
i i

code. (6)
In order to introduce the INSANE, a summary of several mod- that is decomposed in a portion of the incremental displacements
ules of the numerical core application are presented, corresponding
due to the reference load, δ UPj , obtained by Eq. (7), and other por-
i

to the standard FEM and G/XFEM approach, and also to show the i
generalization performed here to enclose the new implementation tion associated to the residual load, δ UQj , calculated by Eq. (8):
on the G/XFEM to nonlinear analysis.
Kij−1 δ UPj = P;
i
(7)

3.1. General view

Kij−1 δ UQj = Qij−1 .
i
(8)
The numerical core is composed by the abstract classes
The convergence of the incremental-iterative process is verified
Solution and Model and the interfaces Assembler and at the end of each iteration, comparing a given tolerance value
Persistence, according to the UML diagram in Fig. 3. against the magnitude of the residual force vector (Eq. (9)), and/or
The task of Persistence is to treat the input data and to pro-
the magnitude of the incremental displacements vector (Eq. (10)):
vide the output data to the other applications. Assembler inter-
face is responsible for mounting the matrix system of the model,

returning the stiffness matrix.  Qij = Qij · Qij ; (9)
The abstract class Solution provides a set of methods de-

voted to the solution of the matrix system. The abstract class
 δ Uij = δ Uij · δ Uij ; (10)
Model provides the information to represent a generic discrete
model. It is composed by several lists of objects, each one rep-
In the numerical examples of this paper, the convergence of the
resenting a peculiar component of a discrete model, like nodes,
incremental-iterative process is verified by the relative tolerance
elements, type of problem, type of analysis model and materials,
(Tol) in relation to the value of the incremental displacements vec-
for example. It provides to Assembler information to assemble
tor (Eq. (10)), according to Eq. (11):
the final matrix system. Both Model and Solution communi-
cate with the interface Persistence, which is responsible for  δ Uij 
the processing of input and output data, persisted as XML (eXten- T ol < . (11)
 Uij 
sible Markup Language) files.
A new iteration is necessary until convergence is attained, in
which case the value of the new load increment is calculated
3.2. Physically nonlinear analysis from the restriction equation (depending on the control method
adopted). The update of the variables follows:
As mentioned earlier, the abstract class Solution is dedicated
to the solution of both linear and nonlinear systems of equations λij = λij−1 + δλij ; (12)
generated by each of the numerical methods embedded in the
software. In the nonlinear analysis of structures it is necessary to
Uij = Uij−1 + δ Uij . (13)
solve a (n + 1 ) × (n + 1 ) system of equations, with n equilibrium
equations plus a restriction equation, to obtain (n + 1 ) unknowns Finally, the residual load vector is evaluated by:
consisting of the increase in the load factor plus n incremental dis-
Qij = λij P − Fij , (14)
placements.
An incremental-iterative process, in which a system of incre- where Fij is the equivalent load vector corresponding to the inter-
mental equilibrium equations is established to the loading step i nal stress (defined ahead – Eq. (16)). In the first iteration, i.e. j = 1,
in the jth iteration, is used to solve the nonlinear problem: the residual load vector is null.
In the case of G/XFEM, the solution of the linear equation sys-
Kij−1 δ Uij = δλij P + Qij−1 , (5)
tem results in nodal parameters from which the nodal displace-
where Kij−1 is the tangent stiffness matrix, which is a function ments are calculated by Eq. (4). In G/XFEM, as in FEM, the matrix
and vectors of the linear equations systems (7) and (8) are assem-
of the displacement field Uij−1 , δ Uij is the incremental displace-
bled from the contribution of each finite element e. This process is
ments vector, δλij is the load factor increment, P is the refer- based on two fundamental operations:
ence load vector and Qij−1 is the residual load vector. The abstract 
class Solution (Fig. 4) is responsible for solving the algorithm ex- Kij−1 = BT E B dVe (15)
Ve
pressed by Eq. (5).
Depending on the control parameter adopted, a corresponding where Kij−1 is the tangent stiffness matrix, B is the strain–
increment in the load factor, δλij , is established and a displacement displacement matrix, E is the constitutive matrix, and σ are the

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

4 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

Fig. 6. UML diagram of the class ProblemDriver.

obtain the stiffness matrix of each element. The stiffness matrix is

sent to Assembler after the evaluation of the material degrada-
tion process. The class ProblemDriver informs the Assembler
Fig. 5. UML diagram of the abstract class Model. all the data for assembling the final system of the model. Such
class has the necessary attributes to particularize the problem to
the nonlinear analysis, including the class Element, responsible
stresses, in which all are integrated on the volume Ve of each ele- for the information about of each finite element.
ment e. The procedure allowing to mount the tangent stiffness ma-
The equivalent load vector is calculated by Eq. (16): trix and the vector of internal forces, used in Eq. (5), is shown
 in Fig. 7. The entire process is triggered by the class Solution
Fij = BT σ ij dVe (16) that initiates a loop over the steps of the incremental pro-
Ve
cess, while the method execute() triggers the iterative pro-
In the G/XFEM approach, when the PU is enriched to build the cess for each step. For each iteration n and each step k, the class
shape functions (Eq. (3)), the final stiffness matrix Kij−1 may be Assembler is requested to provide the tangent stiffness matrix
positive semi-definite. This happens, for example, when both the [K t ]kn−1 (getIncrementalCuu).
PU and the functions Lji (x) of Eq. (2) are of polynomial type. In fact, The method getIncrementalCuu contains the main opera-
the solutions of linear equation systems (3) and (4) exist although tions that allow to assemble the global stiffness matrix, making use
they are not unique. In [32] an interactive procedure is proposed to of contributions of the elements composing the discrete model (Eq.
extract such solution and the corresponding algorithm was imple- (15)). Each element is then required to provide its tangent stiffness
mented in INSANE by Alves et al. [21]. This strategy is one of sev- matrix. Such object is evaluated by the numerical integration of Eq.
eral other methods of the library of linear equation system solvers (15). The abstract class ProblemDriver (called by the method
currently implemented in INSANE. Those methods are accessed by getIncrementalC(Element e)) manages the integration pro-
StaticEquilibriumPath class, derived from Solution class, cedure, that may vary depending on the peculiar problem ana-
and that implements the incremental iterative process, assembled lyzed.
from FEM or G/XFEM, Fig. 4.
The StaticEquilibriumPath is also supported by the 4. Numerical simulations
Step interface that implements the standard version of the
Newton–Raphson method, by class StandardNewtonRaphson The main idea of this paper is to demonstrate the G/XFEM en-
that implements the execute() method. To perform the richment capacity to nonlinear analysis and not exactly present the
Newton–Raphson process, Step requires the GFemAssembler, performance of the method. Such capacity flexibilizes the change
derived from FemAssembler, for mounting all the matrices and of the approach along the domain according to the characteris-
vectors of Eq. (5). tics of the analyzed problem, providing different ways of investi-
The FemAssembler class, in turn, has the FemModel object gating the problem and performing the relevant numerical simu-
able to define which discrete model will be assembled (Fig. 4), lations, with the choice of the enrichment functions, the nodes to
whether coming from the FEM or G/XFEM. Aiming this, the class be enriched, combination of different functions in the same prob-
FemAssembler uses its object to ask Element to integrate ma- lem, among other possibilities. The intrinsic ability of G/XFEM to
trix Kij−1 and vector Fij of Eqs. (15) and (16), respectively. develop customized approximations for specific applications is ex-
Thus, the stiffness matrix is required by the FemAssembler tremely useful to nonlinear problems and if classical FEM is used,
through the methods getIncrementalCuu(), getTotalCuu() or getCuu(), the meshes can not properly model the singularities present.
respectively related to the tangent, Eq. (15), secant or elastic ap- In this section, some numerical simulations are presented aim-
proximations to the constitutive tensor, shown in Fig. 5. ing to illustrate the implementation presented here. The solu-
The abstract class Model, shown in Figs. 4 and 5, represents the tion procedure for the nonlinear problems described here is the
discrete model of the problem. The class GFemModel, whose im- Newton–Raphson Algorithm, it is used the aforementioned Uni-
plementation is detailed in [21], derives from FemModel and in- fied and Theoretical and Computational Framework to Constitutive
corporates all its instances and attributes. Models [24] and other characteristics of each simulation are de-
This class is composed by several lists of objects associated tailed in the respective item.
to specific features of the model such as: nodes (Node), ele- In Sections 4.1 and 4.2, the geometry and boundary conditions
ments (Element), types of shape functions (Shape), peculiar are very simple, since the goal is not to demonstrate the G/XFEM
problem analyzed (ProblemDriver), types of analysis models capabilities neither the computational efficiency of the implemen-
(AnalysisModel), materials (Material), types of constitutive tation, but the general approach proposed to enclose the G/XFEM
models (ConstitutiveModel), among others. nonlinear formulation in the INSANE environment. Indeed, the fo-
The expansion to the computational system to include new ca- cus is to use several kinds of elements, analysis models and con-
pabilities into the G/XFEM to nonlinear analysis is related to the stitutive models combined with different strategies of enrichment
implementation of the class GFemPhysicallyNonLinear that of the PU.
extends the class GFemParametric, according to Fig. 6. Comparisons with results presented in the literature are made
The classes getC and getF were modified, while in Sections 4.3, 4.4 and 4.5 in order to demonstrated that G/XFEM
getIncrementalC and getTotalC classes were created to is able to reproduce the results of FEM and even improve the so-

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14 5

Fig. 7. Tangent stiffness matrix and internal forces vector assembly sequence diagram.

lutions just applying enrichment functions in some nodes of the • P1 (Linear Enrichment):
meshes, allowing varied analysis in which the refinement of the
mesh (that can lead to localization numerically induced) is not
⎡     ⎤
necessary to achieve better solutions. x−x j y−y j
1 0 0 0
φ Tj (x ) = N j (x )⎣    ⎦
The approximation functions, with monomials expressed in co- hj hj

ordinate x, used to the analysis are defined by: x−x j y−y j

0 1 0 hj
0 hj
• P0 (No Enrichment):
(21)
• P2 (Quadratic Enrichment):
φ Tj (x ) = [N j (x )]; (17)
⎡      2  2 ⎤
x−x j y−y j x−x j y−y j
⎢1 0 hj
0 hj
0 hj
0 hj
0 ⎥
φ Tj (x ) = N j (x )⎣      2  2 ⎦ (22)
x−x j y−y j x−x j y−y j
0 1 0 hj
0 hj
0 hj
0 hj

4.1. Smeared crack model of Carreira and Chu (1985) [33]

• P1 (Linear Enrichment):
A beam, represented in Fig. 8, is subjected to a shear force P on
  its right end. The following analysis are performed (the G/XFEM
x − xj
φ Tj (x ) = N j (x ) N j (x ) ; (18) and FEM approximations are able to reproduce any third degree
hj
polynomial):
• P2 (Quadratic Enrichment):
• G/XFEM (L2-P2-1): 1 line element L2 (2 nodes) enriched by
P2 function (60 degrees of freedom – DOFs).
  2
• G/XFEM (L2-P2-2): 2 line elements L2 (3 nodes) enriched by
x − xj x − xj P2 function (90 DOFs).
φ (x ) = N j (x )
T
j N j (x ) N j (x ) .
hj hj • G/XFEM (L2-P2-3): 3 line elements L2 (4 nodes) enriched by
P2 function (120 DOFs).
(19) • G/XFEM (L3-P1-1): 1 line element L3 (3 nodes) enriched by
The approximation functions, with monomials expressed in co- P1 function (54 DOFs).
ordinates x and y, used to the analysis are defined by: • G/XFEM (L3-P1-2): 2 line elements L3 (5 nodes) enriched by
P1 function; (90 DOFs).
• P0 (No Enrichment): • G/XFEM (L3-P1-3): 3 line elements L3 (7 nodes) enriched by
 P1 function; (126 DOFs).
N j (x ) 0 • FEM (L4-P0-1): 1 line element L4 (4 nodes) without enrich-
φ Tj (x ) = (20)
0 N j (x ) ment function (P0) (24 DOFs).

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

6 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

Fig. 8. Geometry, loading and boundary conditions of the beam. Note that the bending moment M = P × 100 and the shear force P on the left end of the beam are support
reactions of Section 4.1 problem and they are applied as stresses in Section 4.2 problem.

Fig. 9. Equilibrium paths obtained with the smeared crack model.

Table 1 L4-P0-2 and L4-P0-3, respectively. The observation of these fig-

Parameters adopted.
ures allows to infer that the kind of enrichment function employed
Young’s modulus: E = 25, 0 0 0 MPa Poisson ratio: ν = 0.20 does not affect significantly the convergence of the nonlinear pro-
Parameters of the constitutive model of [33] cedure, but the number of iteration increases with increasing the
Fc = 20.0 MPa Ft = 2.0 MPa et = 0.0 0 016 ec = 0.0016 number of elements. As expected, more iterations were necessary
around the load peak, where the equilibrium path changes direc-
tion.
• FEM (L4-P0-2): 2 line elements L4 (7 nodes) without enrich-
ment function (P0) (42 DOFs). 4.2. Scalar damage model of Mazars (1984) [6]
• FEM (L4-P0-3): 3 line elements L4 (10 nodes) without enrich-
ment function (P0) (60 DOFs). The same example shown in Fig. 8 is analyzed here considering
a plane stress state and using the constitutive model presented by
The nonlinear behavior of the problem is defined by the Mazars [6]. Fig. 13 shows the meshes with quadrilateral (with 1, 2
smeared crack model of Carreira and Chu [33]. The material prop- and 4 elements) and triangular elements (with 2 elements) used
erties adopted are depicted in Table 1. here.
Such parameters are defined as: Fc is the compressive strength, The numerical simulations here are defined by the combina-
Ft is the tensile strength, ec is the strain related to the maximum tion of different types of elements, levels of mesh refining and
compressive stress and et is the strain related to the maximum degree of the enrichment functions. For the meshes with quadri-
tensile stress. lateral elements, the conventional Q12 cubic element is used only
In the numerical simulations, it was adopted the Timoshenko for the first mesh, (a). The result from the FEM analysis is com-
beam analysis model, 10 mm of thickness and generalized displace- pared with other ones obtained by G/XFEM for the three levels of
ment control method, with load factor of 0.009, tolerance to con- h-refiniment, (a), (b) and (c). On the other hand, the conventional
vergence of 1 × 10−4 (×100% ) = 0.010% in relation to the norm of T10 cubic element is used in the FEM analysis for the one mesh of
the incremental displacements vector and reference load of P = triangular elements (d). The following numerical simulations are
20.0 MN. performed, all of them capable of reproducing any three degree
The equilibrium paths for the simulations are plotted in Fig. 9. polynomial:
A simple h-refinement has been perfomed and from the data pre-
sented it is noted that the maximum load factor decreased as the • FEM (Q12-P0-1): 1 quadrilateral element Q12 (12 nodes)
meshes were refined for all combinations. The analysis perfomed without enrichment function (P0) (24 DOFs).
with 2 and 3 elements present better post-peak results than the • G/XFEM (Q8-P1-1): 1 quadrilateral element Q8 (8 nodes) en-
one with 1 element, capturing the concrete softening behavior. Be- riched by P1 function (48 DOFs).
sides, similar responses are reproduced by the G/XFEM and FEM • G/XFEM (Q8-P1-2): 2 quadrilateral elements Q8 (13 nodes)
simulations and this seems to be related to the same polynomial enriched by P1 function (78 DOFs).
space described by the nine analysis. • G/XFEM (Q8-P1-4): 4 quadrilateral elements Q8 (21 nodes)
Figs. 10, 11 and 12 show the number of iterations to achieve enriched by P1 function (126 DOFs).
equilibrium for each load step of the analyzed models L2-P2-1, • G/XFEM (Q4-P2-1): 1 quadrilateral element Q4 (4 nodes) en-
L2-P2-2, L2-P2-3, L3-P1-1, L3-P1-2, L3-P1-3, L4-P0-1, riched by P2 function (40 DOFs).

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14 7

Fig. 10. Number of iterations per load step – G/XFEM (L2-P2).

Fig. 11. Number of iterations per load step – G/XFEM (L3-P1).

• G/XFEM (Q4-P2-2): 2 quadrilateral elements Q4 (6 nodes) en- In the numerical simulations, it was adopted generalized dis-
riched by P2 function (60 DOFs). placement control method, with initial load factor of 0.002, tol-
• G/XFEM (Q4-P2-4): 4 quadrilateral elements Q4 (9 nodes) en- erance to convergence of 1 × 10−4 (×100% ) = 0.010% in relation to
riched by P2 function (90 DOFs). the norm of the incremental displacements vector and reference
• FEM (T10-P0-2): 2 triangular elements T10 (16 nodes) with- load of P = 20.0 MN and M = 120.0 MN mm. Table 2 presents the
out enrichment function (P0) (32 DOFs). material parameters adopted according to Mazars [6].
• G/XFEM (T3-P2-2): 2 triangular elements T3 (4 nodes) en- The evolution law parameters At, c and Bt, c are used to modu-
riched by P2 function (40 DOFs). late the shape of the post-peak curve and their influence can be
• G/XFEM (T6-P1-2): 2 triangular elements T6 (9 nodes) en- observed from a parametric sensibility analysis. κ 0 is the equiva-
riched by P1 function (54 DOFs). lent strain from which the damage process starts.

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

8 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

Fig. 12. Number of iterations per load step – FEM (L4-P0).

Fig. 13. Quadrilateral and triangular meshes.

Table 2 4.3. Elastoplastic damage constitutive model of Lemaitre (1985)

Parameters adopted.
[7,34]
Young’s modulus: E = 29, 200 MPa Poisson ratio: ν = 0.20

Parameters of the constitutive model of Mazars [6] A simulation of the tensile test, extensively used in both exper-
At = 0.995 Bt = 80 0 0 κ0 = 0.0 0 0 07 Ac = 0.655 Bc = 1050 imental and numerical analysis of ductile fracture, is carried out
here to demonstrate the versatility of G/XFEM nonlinear imple-
mentation combined with elastoplastic damage constitutive model
of Lemaitre [7,34].
The specimen, shown in Fig. 16, is analyzed here under plane
The results are plotted in two graphs: Fig. 14 (quadrilateral ele- stress conditions. The results are compared with the three-
ments) and Fig. 15 (triangular elements). dimensional numerical analysis performed by Mashayekhi et al.
Fig. 14 presents the equilibrium paths to the quadrilateral ele- [35] (that also applied the elastoplastic damage model of Lemaitre,
ments: 1985, [7,34]) because the authors, despite performing the two-
From the data in Fig. 14, it is apparent that the more refined dimensional analysis too, just presented the three-dimensional re-
the meshes, the less rigid are the equilibrium paths. Q12-P0 and sults. The geometry of this problem induces crack initiation at the
Q4-P2 have the same approximation space whereas Q8-P1 has center and its propagation toward the outer edge.
one more term (x2 y2 ) which makes its behavior different from The following simulations are performed:
other meshes. In general for all equilibrium paths, there is a ten-
• G/XFEM (Q4-P1): P1 enriching the nodes 125, 126, 269 and 270
dency of convergence in relation to the limit load and to the
(region where damage is concentrated), with 506 nodes and
behavior of the ascending branch. With respect to the inelastic
1028 degrees of freedom;
branches, such observation can not be stated because there is no
• FEM (Q8-P0): no enrichment is applied, with 1461 nodes and
indication of convergence or localization numerically induced.
2922 degrees of freedom.
Fig. 15 shows the equilibrium paths to the triangular element:
From the data in Fig. 15, it is apparent that the behavior of the A finite element mesh, with 450 elements, is used to dis-
T10-P0-2, T6-P1-2 and T3-P2-2 equilibrium paths are equiv- cretize quarter of the domain and appropriate boundary are im-
alent, once the approximation space is the same. posed at the edges, as presented in Fig. 17. For the analysis are

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14 9

Fig. 14. Equilibrium paths to quadrilateral elements.

Fig. 15. Equilibrium paths to triangular element.

Table 3
Material data and Lemaitre elastoplastic damage constitutive model
parameters for tensile test ([37] apud [36]).

Young’s modulus/Poisson ratio E = 210, 0 0 0 MPa/ν = 0.20

Yield stress σ0 = 620 MPa
Hardening curve σ (κ ) = 620 + 3300[1 − e−0.40κ ]
Hardening/softening function H (κ ) = 1320 e−0.40κ
Damage exponent S = 1.0
Fig. 16. Dimensions of the specimen. Damage resistance r = 3.50 MPa

cremental displacements vector and reference load as shown in

Fig. 17.
Table 3 presents the parameters of the AISI 1010 low carbon
steel in a rolled state adopted ([37] apud [36]).
Mashayekhi et al. [35] defined a limit load to the yield stress as
PL = 15, 500 N. The normalized load P/PL x horizontal displacement
at the node 9 (Fig. 17(b)) to the simulations G/XFEM (Q4-P1) and
FEM (Q8-P0) are illustrated in Fig. 18, together with the numerical
results obtained by Mashayekhi et al. [35], where a model based
on the work of de Souza Neto et al. [36] and later by Andrade Pires
Fig. 17. Direct tensile test: (a) a quarter of geometry and (b) finite element mesh. et al. [38] is formulated and used: a single-equation return map-
ping algorithm. On the other hand, in this paper is applied the two
equation return mapping that has been proposed by Vaz and Oden
adopted 5.0 mm of thickness and generalized displacement con- [39].
trol method, with load factor of 0.50, tolerance to convergence Comparing the numerical result presented by Mashayekhi et al.
of 1 × 10−4 (×100% ) = 0.010% in relation to the norm of the in- [35] with the ones performed in this paper, it is verified that the

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

10 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

Fig. 18. Equilibrium paths obtained with elastoplastic damage model.

Fig. 19. Stress (σ xx ) x strain (ε xx ).

shape of the curves are similar. When the damage at any Gauss rium path changes direction. The total number of iterations is also
point in the structure for the first time reached 0.90, the simula- informed for each model. This number does not vary significantly
tion was terminated by Mashayekhi et al. [35]. According to them, between models and in in most part of the load steps all models
although initially the damage rate is slightly smaller than the rate needed between five and ten iterations to converge.
of effective plastic strain, as the displacement increases, the stress
state at the center of the specimen causes the damage to increase
exponentially whereas the equivalent plastic strain rate remains 4.4. Microplane constitutive model of Leukart and Ramm (2006) [40]
approximately constant.
Then the softening resulting from the damage propagation de- Petersson [41] experimentally studied concrete beams subjected
creases the load capacity of the structure. The simulations G/XFEM to the three-point bending. The experimental results obtained by
(Q4-P1) and FEM (Q8-P0) showed similar responses, but the first those authors were used here to compare with the numerical sim-
one was able to describe the equilibrium path until the collapse of ulations performed with FEM and G/XFEM approaches and the Mi-
the element due to damage, using a smaller number of degrees of croplane Constitutive Model [40].
freedom. The material parameters based on the experimental results ob-
From of the results shown in Figs. 19 and 20, it is possible to tained [41] and adopted to the numerical simulations with the ex-
show, respectively, the behavior of stress component σ xx and the ponential damage law are shown in Table 4. α mic is the maximum
damage evolution D in relation to strain component ε xx , of the material degradation, β mic is the parameter governing the shape of
node 125 (Fig. 17(b), where the worst degradation is observed). the post-peak branch and κ0mic is the limit value to initiate damage
Fig. 21 presents the evolution of damage distribution to G/XFEM to exponential law. They are dimensionless numerical parameters
(Q4-P1) analysis. In the beginning of the loading, damage pre- of the constitutive model.
sented low values, but it gradually increases and concentrates, as Fig. 23 depicts the geometric data of the beam and the mesh of
expected, at the central region of the structure. 115 four-noded quadrilateral elements. The following simulations
Fig. 22 shows the number of iterations to achieve equilibrium are performed:
for each load step of the analyzed models. The observation of
Fig. 22 allows to infer that the enrichment employed does not af- • G/XFEM (Q4-P1+P2): P1 and P2 enrichment functions high-
fect the convergence of the nonlinear procedure. As expected, more lighted at the nodes of the figure, with 384 degrees of freedom;
iterations were necessary around the load peak, where the equilib- • FEM (Q4-P0): no enrichment is applied, with 260 degrees of
freedom.

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14 11

Fig. 20. Damage (D) x strain (ε xx ).

Table 4
Parameters adopted by Petersson [41] and Leukart and Ramm [40].

Young’s modulus/Poisson ratio E = 30, 0 0 0 MPa/ν = 0.20

Uniaxial yield stress σt = 3.30 MPa
Fracture energy G f = 0.130 N/mm
Parameters of the exponential damage law [40]
Exponential α mic = 0.960 β mic = 500 κ0mic = 0.0 0 02

Fig. 23. Three-point bending: geometry and mesh.

The numerical simulations are performed with the conventional

FEM (with no enrichment Q4-P0) and with G/XFEM (Q4-P1+P2)
applying enrichment functions highlighted at the nodes of the fig-
ure, for both constitutive models. The nonlinear analysis have been
performed under plane stress conditions and with the adoption
Fig. 21. Evolution of damage to G/XFEM Q4-P1. of the generalized displacement control method, with load fac-

Fig. 22. Number of iterations per load step.

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

12 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

Fig. 24. Equilibrium paths obtained with the microplane constitutive model and an exponential damage law.

Fig. 25. Equilibrium paths obtained with the volumetric constitutive model and a polynomial damage law.

tor equal to 0.02, tolerance to convergence of 1 × 10−4 (×100% ) = Table 5

Parameters adopted by Petersson [41] and Penna [43].
0.010% in relation to the norm of the incremental displacements
vector and reference load of P = 800.0 N. Young’s modulus/Poisson ratio E = 30, 0 0 0 MPa / ν = 0.20
Fig. 24 shows the numerical results of the vertical displacement Uniaxial yield stress σt = 3.30 MPa
Fracture energy G f = 0.130 N/mm
of the node 10 (the maximum displacement point highlighted in
Fig. 23) to the exponential damage law, together with the experi- Parameters of the polynomial damage law: Eeq = 16666.666 MPa
Tension evolution f e = 1.95 MPa κ0 = 0.0 0 024
mental results obtained by Petersson [41]. Compression evolution f e = 11.1 MPa κ0 = 0.0020
The standard FEM analysis presented a worse description of the
equilibrium path due to the poorer discretization adopted (260 de-
grees of freedom to FEM and 384 degrees of freedom to G/XFEM).
A better result could be obtained if a p or h-refinement is per- 4.5. Volumetric damage model of Penna (2011) [43]
formed, significantly increasing the number of degrees of freedom.
Another strategy could be using finer mesh around the notch, but The experimental results obtained by Petersson [41] related to
special attention would be required to avoid problems related to the concrete beams subjected to the three-point bending (same ex-
the transition from elements of different size. Another procedure ample presented in Section 4.4) were used here to compare with
could be using higher order elements on that same region. In this the numerical simulations performed with FEM and G/XFEM ap-
case, elements of different orders would share a same interface, proaches and the Volumetric Damage Model [43].
with the presence of irregular nodes, as it is described in [42]. The material parameters based on the experimental results ob-
The necessity of enforcing the continuity across this interface con- tained by Petersson [41] and adopted to the numerical simula-
strains the approximation and penalizes its quality. tions with polynomial damage law are shown in Table 5. κ 0 is the
On the other hand, G/XFEM Q4-P1+P2 analysis showed a bet- limit value to initiate damage (dimensionless numerical parame-
ter concordance with the experimental results obtained by Peters- ter of the constitutive model), fe is the equivalent stress related to
son [41], thanks to the enrichment strategy that improves the solu- the material strength limit and Eeq is the linear elastic modulus.
tion around the region where the nonlinear phenomenon happens. The discretization and solution parameters are the same applied to
This enrichment strategy doesn’t require modification in the neigh- solve the nonlinear problem presented in Section 4.4.
boring elements of that region, as it would be the case of FEM, if Fig. 25 shows the numerical results of the vertical displacement
selective h or p refinments are applied. of the node 10 (the maximum displacement point highlighted in
Fig. 23) to the polynomial damage law, together with the exper-

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14 13

imental results obtained by Petersson [41]. The results G/XFEM [2] Babuška I, Melenk JM. The partition of unity method. Int J Numer Methods Eng
(Q4-P1+P2) agree with the experimental ones obtained by Pe- 1997;40:727–58.
[3] Belytschko T, Black T. Elastic crack growth in finite elements with minimal
tersson [41], providing smoothness to the equilibrium path in its remeshing. Int J Numer Methods Eng 1999;45:601–20.
inelastic regime than FEM (Q4-P0). The polynomial enrichment [4] Belytschko T, Gracie R, Ventura G. A review of extended/generalized fi-
strategy provided by G/XFEM allows to improve the quality of the nite element methods for material modeling. Model Simul Mater Sci Eng
2009;17:1–24.
approximation around the notch in a very simple and straightfor- [5] Torres IFR, Proença SPB. Desenvolvimento e aplicação do método dos elemen-
ward way. tos finitos generalizados em análise tridimensional nǎo-linear de sólidos. In:
As in Section 4.4, it is apparent that the G/XFEM approximation Proceedings of actas do congresso de métodos computacionais em engenharia.
Lisboa, Portugal; 2004.
result was concordant with the experimental results mostly on the
[6] Mazars J. A description of micro and macro scale damage of concrete strucures.
post-critical regime and present better post-peak results than the Eng Fract Mech 1986;25:729–37.
FEM responses. This improvement in the results was due only to [7] Lemaitre J. A continuous damage mechanics model for ductile fracture. J Eng
Mater Technol 1985a;107:83–9.
the enrichment of the region of high stress concentration, without
[8] Barros FB, Proença SPB, de Barcellos CS. On error estimator and p-adap-
the need for refinement of the mesh. tivity in the generalized finite element method. Int J Numer Methods Eng
2004;60:2373–98.
[9] Elguedj T, Gravouil A, Combescure A. A mixed augmented Lagrangian-extended
5. Conclusions finite element method for modelling elastic-plastic fatigue crack growth with
unilateral contact. Int J Numer Methods Eng 2007;71:1569–97.
In this paper it is presented a computational framework to ex- [10] Elguedj T, Gravouil A, Combescure A. Appropriate extended functions for XFEM
simulation of plastic fracture mechanics. Comput Methods Appl Mech Eng
pand a G/XFEM implementation to nonlinear analysis. Firstly it is 2006;195:501–15.
summarized the state of art of G/XFEM, its application to physically [11] Combescure A, Coret M, Elguedj T, Cazes F, Haboussa D. Cohesive laws XFEM
nonlinear analysis of structures and recent works that obtained association for simulation of damage fracture transition and tensile shear
switch in dynamic crack propagation. Procedia IUTAM 2012;3:274–91.
good results to quasi-brittle materials, such as concrete. Secondly [12] Ortiz M, Pandolfi A. Finite-deformation irreversible cohesive elements for
an overview of the G/XFEM formulation is presented. Nevertheless, three-dimensional crack-propagation analysis. Int J Numer Methods Eng
the primary focus here is the programming environment and its 1999;44(9):1267–82.
[13] Moës N, Belytschko T. Extended finite element method for cohesive crack
extension to enclose a class to solve physically nonlinear problems. growth. Eng Fract Mech 2002;69:813–33.
This environment was conceived under a segmented conception in [14] Piedade Neto D, Ferreira MDC, Proença SPB. An object-oriented class design
[22] for FEM analysis and allows a code expansion whose complex- for the generalized finite element method programming. Lat Am J Solids Struct
2013;10:1267–91.
ity may be increased in a gradual manner. In [21] such characteris-
[15] Crété JP, Longère P, Cadou JM. Numerical modelling of crack propagation in
tic is exploited to implement a generic enrichment strategy for any ductile materials combining the GTN model and XFEM. Comput Methods Appl
PU given by the FEM formulation. Mech Eng 2014;275:204–33.
Additionally it is also demonstrated the enrichment capacity of [16] Broumand P, Khoei AR. XFEM modeling of dynamic ductile fracture prob-
lems with a nonlocal damage-viscoplasticity model. Finite Elem Anal Des
the G/XFEM to nonlinear analysis. Such capacity flexibilizes the 2015;99:49–67.
change of the approximation along the domain according to the [17] Freitas A, Torres DAF, Mendonça P, Barcellos C. Comparative analysis of ck and
characteristics of the analyzed problem. As a consequence, differ- c0 GFEM applied to two-dimensional problems of confined plasticity. Lat Am J
Solids Struct 2015;12:861–82.
ent investigating strategies can be provided with the choice of the [18] Unger JF, Eckardt S, Könke C. Modelling of cohesive crack growth in concrete
enrichment functions, the nodes to be enriched, combining differ- structures with the extended finite element method. Comput Methods Appl
ent functions in the same problem, among other possibilities. The Mech Eng 2007;196:4087–100.
[19] Mariani S, Perego U. Extended finite element method for quasi-brittle fracture.
intrinsic ability of G/XFEM to develop customized approximations Int J Numer Methods Eng 2003;58:103–26.
for specific applications is extremely useful to nonlinear problems [20] Kim DJ, Duarte CA. A new generalized finite element method for two-scale
and if standard FEM is used, the meshes cannot properly model simulations of propagating cohesive fractures in 3-d. Int J Numer Methods Eng
2015;197:487–504.
the localized phenomenon present, for example. [21] Alves PD, Barros FB, Pitangueira RLS. An object-oriented approach to the gen-
Some aspects of the design of classes are considered aiming eralized finite element method. Adv Eng Softw 2013;59:1–18.
to describe the implementation that permits the combining di- [22] Fonseca FT, Pitangueira RLS. An object oriented class organization for dynamic
geometrically non-linear FEM analysis. In: Proceedings of Iberian Latin Amer-
verse analysis approaches. The numerical examples demonstrate
ican congress on computational methods in engineering, XXVIII CILAMCE;
such flexibility. Different types of enrichments are combined with 2007.
constitutive models and elements. [23] INSANE, project. ‘http://www.insane.dees.ufmg.br/’ (Accessed in July 12). The
New investigations must be performed in future researches development code is available at the Git repository URL https://git.insane.dees.
ufmg.br/insane/insane.
aiming to verify the numerical stability of G/XFEM to the nonlinear [24] Gori L, Penna SS, Pitangueira RLS. A computational framework for constitutive
analysis applying enrichments of higher order or a large number modelling. Comput Struct 2017;187:1–23.
of finite elements, for example. Additionally, it is possible to ap- [25] Carol I, Rizzi E, Willam K. A unified theory of elastic degradation and damage
based on a loading surface. Int J Solids Struct 1994;31:2835–65.
ply other constitutive models and also to verify the G/XFEM global- [26] Duarte CA, Babuška I, Oden JT. Generalized finite element methods
local implementation to nonlinear analysis, which is under devel- for three-dimensional structural mechanics problems. Comput Struct
opment. 20 0 0;77(2):215–32.
[27] Babuška I, Caloz G, Osborn JE. Special finite element methods for a class of
second order elliptic problems with rough coefficients. SIAM J Numer Anal
Acknowledgments 1994;31:945–81.
[28] Duarte CA, Oden JT. Hp clouds – a meshless method to solve boundary-value
problem (technical report), TICAM - Texas Institute for Computational and Ap-
The authors gratefully acknowledge the important support of plied Mathematics; 1995. p. 1–66.
the Brazilian research agencies FAPEMIG (in Portuguese “Fundação [29] Duarte, Duarte CA, Oden JT. An h-p adaptive method using clouds. Comput
Methods Appl Mech Eng 1996;139:237–62.
de Amparo à Pesquisa de Minas Gerais” – Grant PPM-00669-
[30] Oden JT, Duarte CA, Zienkiewicz OC. A new cloud-based hp finite element
15) and of CNPq (in Portuguese “Conselho Nacional de Desen- method. Comput Methods Appl Mech Eng 1998;153:117–26.
volvimento Científico e Tecnológico” – Grants 308785/2014-2 and [31] Moës N, Dolbow J, Belytschko T. A finite element method for crack growth
without remeshing. Int J Numer Methods Eng 1999;46:131–50.
308932/2016-1).
[32] Strouboulis T, Babuška I, Copps K. The design and analysis of the
generalized finite element method. Comput Methods Appl Mech Eng
References 20 0 0;181(1–3):43–69.
[33] Carreira DJ, Chu K. Stress–strain relationship for plain concrete in compression.
[1] Melenk JM, Babuška I. The partition of unity finite element method: basic the- ACI J 1985;82:797–804.
ory and applications. Comput Methods Appl Mech Eng 1996;139:289–314.

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002
JID: ADES
ARTICLE IN PRESS [m5G;August 11, 2017;14:6]

14 A.B. Monteiro et al. / Advances in Engineering Software 000 (2017) 1–14

[34] Lemaitre J. Coupled elasto-plasticity and damage constitutive equations. Comp [40] Leukart M, Ramm E. Identification and interpretation of microplane material
Methods Appl Mech Eng 1985b;51:31–49. laws. J Eng Mech 2006;132(3):195–305.
[35] Mashayekhi M, Ziaei-Rad S, Parvizian J, Nikbin K, Hadavinia H. Numerical anal- [41] Petersson PE. Crack growth and development of fracture zones in plain con-
ysis of damage evolution in ductile solids. Tech Sci Press 2005;1:65–82. crete and similar materials. Tech report tvbm-1006. Lund, Sweden: Lund Insti-
[36] de Souza Neto EA, Perić D, Owen DRJ. Computational methods for plasticity: tute of Technology – Division of Building Materials; 1981.
theory and application. Wiley; 2008. [42] Demkowicz L, Oden JT, Rachowicz W, Hardy O. Toward a universal h-p adapta-
[37] Benallal A, Billardon R, Doghri I, Moret-Bailly L. Crack initiation and propaga- tive finite element strategy, part 1. Constrined approximations and data struc-
tion analysis taking into account initial strain hardening and damage fields. ture. Comput Methods Appl Mech Eng 1989;77:79–112.
In: Proceedings of the fourth international conference of numerical methods [43] Penna SS. Formulação multipotencial para modelos de degradação elástica:
in fracture mechanics; 1987. unificação teórica, proposta de novo modelo, implementação computacional e
[38] Andrade Pires FM, César de Sá JMA, Costa Sousa L, Natal Jorge RM. Numeri- modelagem de estruturas de concreto, Ph.D. thesis. Belo Horizonte, MG, Brazil:
cal modelling of ductile plastic damage in bulk metal forming. Int J Mech Sci Federal University of Minas Gerais; 2011.
2013;45:273–94.
[39] Vaz MJr, Oden DRJ. Aspects of ductile fracture and adaptive mesh refine-
ment in damaged elasto-plastic materials points. Int J Numer Methods Eng
2001;50:29–54.

Please cite this article as: A.B. Monteiro et al., A computational framework for G/XFEM material nonlinear analysis, Advances in Engi-
neering Software (2017), http://dx.doi.org/10.1016/j.advengsoft.2017.08.002