Computers and Structures xxx (2018) xxx–xxx

Contents lists available at ScienceDirect

Computers and Structures

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

Large deformation analysis of functionally graded visco-hyperelastic

materials
João Paulo Pascon
Materials Engineering Department, Lorena School of Engineering, University of São Paulo, Pólo-Urbo Industrial, Gleba AI-6, s/n°, 12602-810 Lorena, SP, Brazil

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

Article history: In this paper, a numerical formulation for the analysis of viscoelastic functionally graded materials under
Received 1 March 2018 finite strains is presented. The general constitutive modeling is described within the context of
Accepted 1 June 2018 Lagrangian and isotropic visco-hyperelasticity. The specific models selected are the compressible neo-
Available online xxxx
Hookean hyperelastic law, the Zener rheological model and the isochoric evolution law described in
terms of the rate of the viscous right Cauchy-Green stretch tensor. The material coefficients may vary
Keywords: smooth and continuously along one direction according to the power law. The viscous update is per-
Large deformation analysis
formed via the exponential rule. The main novelty of this paper is the use of gradually variable viscoelas-
Functionally graded materials
Visco-hyperelasticity
tic coefficients in the finite strain regime.
3D finite elements Four numerical examples involving functionally graded materials and finite viscoelastic strains are
originally analyzed to assess the formulation proposed: a bar under uniaxial extension, a block under
simple shear, the Cook’s membrane and an elastomeric bridge bearing. Isoparametric solid tetrahedral
finite elements of linear, quadratic and cubic orders are employed. The influence of the material vis-
coelastic parameters on the mechanical behavior is analyzed in detail. Results confirm that mesh refine-
ment provides more accuracy and the present model can reproduce large levels of viscoelastic strains in
functionally graded materials.
Ó 2018 Elsevier Ltd. All rights reserved.

1. Introduction
Viscoelastic functionally graded materials (VFGMs) have been
widely used in engineering and industry. In these advanced com-
posites, the mechanical behavior is time-dependent and the mate-
rial properties vary gradually (smooth and continuously) over the
volume. This gradual variation avoids the occurrence of material
mismatch and stress discontinuities, which lead to delamination
problems in composite laminates, for example. Functionally
graded materials (FGMs) have many potential applications in
high-temperature environments, e.g. nuclear reactor components,
chambers of internal combustion engines, blade casing in thermal
power plants and spacecraft. Such composites usually exhibit
creep and relaxation viscoelastic behavior at high temperatures.
A representative material having such properties is the polymer
(or elastomer).
Although the damping mechanisms vary with temperature,
many elastomers are viscoelastic even at room temperature, for
instance. There are several works in the context of mechanical/
structural analysis of VFGMs under isothermal conditions. In the
E-mail address: jppascon@usp.br
work of [1], a new plate theory is developed considering the vis-
coelasticity of polymer foams. It is stated that, since the mechani-
cal properties vary over the thickness direction, the foam can be
approximately modeled as a FGM. As pointed out by [2], due to
their impact loading resistance and low weight, polymeric foams
are employed in automotives, spacecraft, submarines and airplane.
To investigate the dynamic behavior, those authors develop a
method for free vibration analysis of cylindrical panels composed
of VFGM. In [3], a dynamic analysis of multi-span VFGM nanopipes
conveying fluid is performed based on nonlocal elasticity theory,
showing that the concept of FGM also has applications in nan-
otechnology. Moreover, since the mechanical behavior is signifi-
cantly affected by the nature of spatial variation of material
properties, FGMs can be tailored for specific task performances.
Optimization design procedures involving VFGMs can be found,
for instance, in [4] and [5]. Another fact to be highlighted is that
cracks can eventually appear in FGMs under high temperatures.
Examples of fracture mechanics models devoted to VFGMs are
described in [6] and [7].
The usual modeling of FGMs involves the definition of a contin-
uous function that describes the gradual variation of the material
coefficients along one or more directions. The most common
functionally graded (FG) models are the power law (p-FGM), the

https://doi.org/10.1016/j.compstruc.2018.06.001
0045-7949/Ó 2018 Elsevier Ltd. All rights reserved.

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
2 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx
sigmoidal law (S-FGM) and the exponential law (E-FGM) (see, for
instance, the work of [8]). Other FG models can also be adopted,
such as the two- and the five-parameter exponential laws, as well
as the Weibull exponential function, usually employed for carbon
nanotube (CNT)-reinforced laminated composites [9–11]. For
model calibration of VFGMs, the reader can be referred to the work
of [12], in which the generalized Kelvin model of arbitrary order is
used.
An important limitation of the abovementioned works involv-
ing VFGMs is the restriction to small strain regime. It is well known
that elastomers, for example, are usually highly deformable and
present nonlinear time-dependent behavior. The modeling of elas-
tomers under finite viscoelastic strains remains a major challenge.
The finite elastic response of polymers is often modeled within the
context of hyperelasticity, defining a scalar strain energy function
that can reproduce large strain levels and, in general, a nonlinear
material model (see, for instance, the works of [13] and [14]. The
inclusion of the viscoelastic behavior can be performed by setting
the time-dependence of the material properties (or the damping
response). Two common frameworks appear in the context of finite
viscoelastic strains: Convolution Integral Model (CIM), based on
convolution or hereditary integrals [15–17]; and Internal Variable
Model (IVM), defined in terms of hidden (or internal) strain history
variables (see, for example, the works of [18–23]). A comparison of
both frameworks is done in [24], which have concluded that it is
very difficult to decide which model is more appropriate for
specific applications. The present formulation belongs to the
second framework and can be considered as a particular case of
the finite strain models proposed in [18] and [19], in which the
thermal effects are considered, or the damage model suggested
in [25]. The model of this paper is also restricted to isotropic
materials. An example of general anisotropic visco-hyperelastic
model is found in the phenomenological formulation proposed
by [26].
The Finite Element Method (FEM) is employed to deal with gen-
eral structural problems involving FG visco-hyperelastic materials.
The element adopted is the isoparametric tetrahedral solid of any-
order based on positional description, successfully employed, for
instance, in [27–29]. In these works, it is demonstrated that full
integration scheme together with high order polynomials is very
effective to avoid locking problems even in complex structural
problems, without using logarithmic strains, corotational rates
and mixed formulations. The present work can be considered as
an extension of the homogeneous visco-hyperelastic formulation
proposed in [29] to the case in which the material has gradually
variable properties. Some alternative numerical formulations have
been recently proposed to solve the same problems analyzed in the
present work. One example is the Consecutive-Interpolation Proce-
dure (CIP) addressed in the works of [30–33], in which the nodal
average gradients are interpolated in a second step, resulting in
continuous nodal stresses without smoothing operation and with-
out increasing the number of degrees of freedom. Another interest-
ing approach is the Isogeometric Analysis (IGA) used, for example,
in the works of [34–36]. In that formulation, the geometry is
exactly described using CAD basis functions (e.g. NURBS) and fewer
control points (when compared to traditional FEM), resulting in a
high-order continuity and a simple mesh refinement.
According to the author’s bibliographic review, there are few
works in which finite visco-hyperelasticity is employed together
with the concept of FGM. In the study of [37], for example, a for-
mulation is proposed to analyze the mechanical behavior of trans-
versely isotropic FG rubbers under finite viscoelastic strains. It is
demonstrated that the proposed constitutive model agrees with
uniaxial experimental data of polyurea. However, that work is the-
oretical and no finite element analysis is performed for general
mechanical or structural problems. The purpose of the present
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
work, motivated by the lack of studies, is to perform a finite ele-
ment analysis of general 3D structural problems involving FG
visco-hyperelastic materials under isothermal and quasi-static
conditions. The development of this original formulation is essen-
tial mainly for elastomeric structures under finite strains and com-
plex boundary conditions. Two novelties of the present study can
be cited: the use of the FG concept to describe the gradual variation
of visco-hyperelastic coefficients; and the unique finite element
analysis of general structural problems involving VFGMs under
finite strain levels.
This paper is organized as follows. The kinematic formulation is
described in Section 2. The constitutive models, the evolution
equations and the FG law adopted are provided in Section 3. The
numerical solution procedure to deal with the nonlinear expres-
sions involved is given in Section 4. The illustrative numerical
examples used to validate the present methodology are described
in Section 5. Finally, the main conclusions of the work are high-
lighted in Section 6.
2. Kinematics
The present kinematic description is the same as the one
adopted in [29] and, thus, is briefly described.
The kinematics is based on the multiplicative split of the defor-
mation gradient F, similar to the Kröner-Lee decomposition
employed in finite elastoplasticity:
@y
F¼ ¼ FeFv ð1Þ
@x
where the vector fields x and y are the initial and the final
(deformed) configurations, respectively; and the subscripts ()e and
()v denote, in this order, the elastic and the viscous parts. The mul-
tiplicative decomposition (1) is also used in [19,20,23,38], for
instance. According to [20], the multiplicative decomposition of
the deformation gradient is conceptual and is not defined
experimentally.
The use of decomposition (1) leads to the concept of an inter-
mediate configuration, described by setting F e ¼ I ) F ¼ F v .
According to [24], this configuration can be used only for quasi-
static problems, which is the case of the present study. An alterna-
tive multiplicative decomposition of the gradient is found in [26],
in which the reversed form is adopted: F ¼ F v F e . As pointed out
by those authors, the intermediate configuration F ¼ F v defined
in (1) is a stress-free (or relaxed) configuration only when internal
static equilibrium is reached or when time scales becomes infi-
nitely large (t ! 1). However, the difference between such
decompositions does not influence the present approach, since it
is not necessary to determine the individual components of the
gradients Fe and Fv (see Section 3).
The numerical approximation is based on positional description
and follows the usual FEM procedure, i.e., is based on nodal posi-
tions and Lagrange shape functions. Isoparametric solid tetrahedral
finite elements of any-order are employed. The degrees of freedom
are the current spatial positions of the nodes, instead of nodal dis-
placements. Further details on how to determine the deformation
gradient (1) based on the nodal positions can be found, for
instance, in the works of [27–29,39]. One should note that, in such
references, no strain enhancement is used and, although the anal-
ysis is geometrically nonlinear, no special treatment of element
distortions is carried out.
The strain measures adopted in this work are the symmetric
right Cauchy-Green stretch tensor and its invariants:
C ¼ F T F ¼ F Tv F Te F e F v ¼ F Tv C e F v ð2Þ
i1 ¼ trðCÞ ¼ C e : C v ð3Þ
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx 3

i3 ¼ J 2 ¼ detðCÞ ¼ J 2e J2v ¼ ði3 Þe ði3 Þv ð4Þ the shear modulus, which corresponds to the neo-Hookean law ini-
tially proposed by [43] for incompressible elastomers.
where scalar J = det(F) is the Jacobian, which is a measure of volu- The stress tensor S and the internal dissipation inequality can
metric strain (or volume changes) also defined as the local ratio be found by means of the Clausius-Duhem inequality of thermody-
between infinitesimal volume elements in the reference and cur- namics and the Coleman-Noll procedure (see, for example, the
rent configurations (dv =dV 0 ). Because of decomposition (1), the work of [13]):
quantities Je and Jv measures the volumetric changes of the elastic
and viscous parts of deformation, respectively. As one can note, @w
S¼2 ¼ S1 þ Sq ð7Þ
the present formulation is Lagrangian, since the reference is the ini- @C
tial undeformed configuration, and objectivity is assured by the use
1
of tensor C and its invariants (see [13], for instance). dint ¼ S : C_  w_ P 0 ) dint ¼ ðC e S eq Þ : Dv P 0 ð8Þ
2
3. Constitutive modeling where S1 is the equilibrium stress tensor at the spring corre-
sponding to w1; Sq is the non-equilibrium stress tensor at the
As in [29], the constitutive modeling adopted is based on the Maxwell element; scalar dint is the internal dissipation; Seq is a
visco-hyperelastic models proposed in [19,20,23,40,41], which fictitious stress tensor; and Dv is the viscous rate of deformation
can reproduce finite levels of elastic and viscous strains. tensor. The stresses S1 and Sq are defined at the reference
configuration. The fictitious stress Seq defined at the intermediate
3.1. Stress configuration can be considered the push-forward operation on
Sq via Fv:
The rheological model of Zener is adopted, i.e., the time-

@w
dependent response is described by two elements arranged in par- S q ¼ F 1 T 1
v S eq F v ¼ F v 2 e F T ð9Þ
allel (see Fig. 1): an elastic spring with strain energy w1; and a @C e v
Maxwell element containing an elastic spring with strain energy
According to the above expression, when time scale becomes
we and a dashpot with viscosity g. One can say that w1 is the extremely large the effect of viscosity disappears (F v ¼ I) and,
stored strain energy between the reference and the final configura-
thus: S q ¼ S eq . In other words, if the reference configuration were
tions, and we is the stored strain energy between the intermediate
the intermediate configuration, the fictitious tensor S eq would be
configuration (F = Fv) and the final one. Therefore, the expressions
the stress at the Maxwell element [29].
defining the energies w1 and we can be written in terms of the
The following stress-strain relations developed in [29] are nec-
stretch tensors C and Ce, respectively, or in terms of C and the pair
essary in order to determine the stress tensor S in terms of tensors
(C,Cv). The influence of the rheological model adopted on the
C and Cv for models (5) and (6):
mechanical response is much greater in the finite strain regime,
when compared to the small strain case, as pointed out by [20]. @w1
S1 ¼ 2 ¼ K 1 lnðJÞC 1 þ l1 ðI  C 1 Þ ð10Þ
The present model is defined following the compressible and @C
isotropic neo-Hookean hyperelastic law adopted, for example,

@w
in [42]: S q ¼ F 1 2 e F T ¼ K e lnðJ e ÞC 1 þ le ðC 1
v C
1
Þ ð11Þ
v
@C e v
K1 2 l
w1 ¼ ½lnðJÞ þ 1 ði1  3  2lnJÞ ð5Þ
2 2 One can see that it is not necessary to determine the viscous
gradient Fv and the elastic Cauchy-Green stretch tensor Ce, as the
Ke 2 l stress tensor can be completely obtained from tensors C and Cv.
we ¼ ½lnðJ e Þ þ e ðie1  3  2lnJ e Þ ð6Þ
2 2 This fact is convenient for implementation since both tensors C
where K and l are the bulk and shear moduli of each spring. The and Cv are symmetric and, thus, have only six independent compo-
neo-Hookean model is usually employed to reproduce the nonlinear nents (not nine, as in the case of the gradients F and Fv). Moreover,
stress-strain behavior of elastomeric materials under finite levels of the gradients are not objective strain measures. Although the neo-
elastic strains. The expressions (5) and (6) correspond to a model Hookean model is nonlinear and can be used to reproduce finite
version in which the compressibility condition is included. The elastic strains, expressions (10) and (11) are not so complex when
near-incompressibility condition observed in polymers can be compared to other nonlinear hyperelastic models found in the sci-
enforced by choosing a value of bulk modulus much larger than entific literature.

3.2. Evolution equation

In order to obtain the stress tensor S, it remains to deter-

mine tensor Cv. Following the usual procedure, an evolution
equation in terms of viscous strain rate is employed. This
equation describes the variation of the viscous strains over time
and must satisfy inequality (8). The evolution equation adopted
in this work is similar to the formulations proposed in [20]
and [40]:
1
Dv ¼ dev ðC e S eq Þ ð12Þ
g
Applying the relation between Dv and the rate of Cv, a Lagran-
gian version of the evolution Eq. (12) can be defined:
2
Fig. 1. Rheological visco-hyperelastic model adopted. The symmetric second-order C_ v ¼ 2F Tv Dv F v ¼ F Tv dev ðC e S eq ÞF v ð13Þ
tensor S is the second Piola-Kirchhoff stress tensor. g
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
4 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx

As tensor Cv is Lagrangian (or defined in terms of the initial con- 4. Numerical algorithm
figuration), there is no need of using a corotational Eulerian strain
rate. As demonstrated in [29], the use of some properties of tensor The numerical solution strategies adopted are described in this
algebra from (13) results in: section.

2
C_ v ¼ dev ðCS q ÞC v ð14Þ 4.1. Equilibrium
g
The deviatoric character of the evolution equations adopted First, the static equilibrium is expressed at the current position
leads to an isochoric viscous deformation, i.e.: J v ¼ detF v ¼ 1 or via the nonlinear version of the Principle of Minimum Total Poten-
tial Energy (see [13], among many others):
J 2v ¼ detC v ¼ 1. It means that the volumetric changes correspond
Z Z
 
only to the elastic part of the deformation occurred. Evolution @w
dP ¼ ðdwÞdV 0  dðf ext  ydf Þ ¼ dV 0  f ext  dydf ¼ 0 ()
equations with the same type as (14) are similar to the incom- X X0 @ydf
Z 0

pressible flow rule adopted in finite elastoplasticity. The inclusion @w
f int ¼ dV 0 ¼ f ext
of viscous volumetric changes can be performed by modifying the X0 @ydf
evolution Eqs. (12) or (14). ð16Þ
where the symbol d denotes a virtual variation (similarly to the
3.3. Functionally graded variation Principle of Virtual Work); scalar P is the total potential energy;
dV0 is an infinitesimal volume element of the initial configuration
Following the usual procedure in the modeling of FGMs, an X0; ydf is the vector that contains the degrees of freedom, which
expression describing the gradual variation of some material coef- are the three current positions of the nodes; and the vector fields fint
ficient, called FG law, must be defined. In this study, the material and fext are the internal and external (applied) force vectors, respec-
parameters vary smooth and continuously along one direction tively. Vector fext corresponds to the mechanical point loads on the
according to the power FG law (see, for instance, the work of [8], structure and vector fint must be determined for the final configura-
for further details). If a general coefficient (a) varies along direction tion described by ydf at which equilibrium (16) is achieved. As the
x (which can be x1, x2 or x3) from a2 at x = 0 to a1 at x = xmax, the analysis is geometrically nonlinear, a general expression to deter-
resultant FG power law is: mine explicitly the unknown configuration ydf from the known vec-

p 
p 
p tor fext is not available. Therefore, the Newton-Raphson iterative
x x x technique is employed, i.e., for a given trial configuration, the resid-
a ¼ a1 þ a2 1  ¼ a2 þ ða1  a2 Þ ð15Þ
xmax xmax xmax ual force vector and the Hessian matrix are determined until the
equilibrium (16) is sufficiently satisfied within a given tolerance:
where p is the FG power coefficient, employed to define the gradual Z

@w
variation of parameter a. If the power coefficient p approaches zero, r ¼ f int  f ext ¼ dV 0  f ext ð17Þ
X0 @y df
the material parameter a is approximately constant and equal to a1.
On the other hand, if p approaches infinity, a is equal to a2. Thus, the Z !
@r @f int @2w
cases p ! 0 and p ! 1 are considered the two homogeneous limit- H¼ ¼ ¼ dV 0 ð18Þ
ing cases. The value of the graded parameter a can be determined @ydf @ydf X0 @ydf @ydf
from (15) at any point based on its initial position. Examples of
gradual variation are depicted in Fig. 2. It should be pointed out that Dydf ¼ H 1  r ð19Þ
expression (15) is not restricted to prismatic bodies and can be
One can note that the applied forces considered in the present
applied to more complex geometries, including radial variation in
work are conservative, as their values do not depend on the final
axisymmetric problems, for example.
configuration. The prescribed loads (or prescribed displacements)
Expression (15) can be employed for any of the five material
are divided into a large number of steps in order to improve con-
coefficients of the present model (K1, Ke, l1, le or g). If all these
vergence of the analysis. The integrands above can be replaced by:
coefficients are gradually variable according to (15), the stresses
and the rate of Cv will change along direction x even in the case @w @w @C 1 @C
¼ : ¼ S: ð20Þ
of homogeneous deformation. @ydf @C @ydf 2 @ydf
!
@2w 1 @ 2 w @C @w @C @C
¼ : þ : :
@ydf @ydf 2 @C@C @ydf @C @ydf @ydf


1 @S @C @C @C
¼ : þS : : ð21Þ
2 @C @ydf @ydf @ydf
The derivatives of tensor C in respect to the vector of degrees of
freedom ydf can be obtained, for example, in [39]. The partial
derivatives @S=@C represent a fourth-order tensor called consistent
tangent operator in terms of tensor C and the expressions to deter-
mine it for the present model can be found in [29]. In expressions
(20) and (21), one can note the separation between the finite ele-
ment approach and the constitutive modeling. The determination
of stretch tensor C, as well as its derivatives with respect to ydf,
is performed based on the kinematical approximation employed.
The determination of the stress tensor S and the consistent tangent
Fig. 2. Gradual variation of a generic material coefficient a according to the power operator, in turn, depends on the material model and on the previ-
FG law (15) for the case in which a1 > a2. ously calculated strain field.

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx
The integrals that appear in expressions (17) and (18) are
numerically evaluated via Gaussian quadrature formulas, whose
coordinates and weights can be found in the work of [44]. As in
the works of [27–29], the elements are fully integrated, avoiding
the use of reduced or selective reduced integration schemes.
4.2. Viscous update
The determination of the consistent tangent operator @S=@C,
which appears in expression (21), involves the calculation of the
derivative @C v =@C, as the stress S depends on both stretch tensors
C and Cv:
@S @S @S @C v
¼ þ : ð22Þ
@C @C @C v @C
This means that the stretch tensors above depend on each
other. Actually, tensor Cv evolves as the tensor C is changed. Due
to the evolution equation adopted, a general formula to determine
@C v =@C is not available. This problem can be overcome via a dual-
step method: elastic predictor followed by a viscous update. In the
first phase, the deformation is assumed to be purely elastic, i.e., the
viscous deformation remains constant. In the second phase, the
viscous tensor Cv is update, which is done in the present study
via the following exponential scheme applied to the evolution Eq.
(14):
ðC v ÞNþ1 ¼ expðDtRv ÞNþ1 ðC v ÞN ð23Þ
2 tolerance adopted for the error regarding the norm of the residual
C_ v ¼ Rv C v ) Rv ¼ dev ðCS q Þ ð24Þ
g
where N and N + 1 denote, in this order, the beginning and the end
of step N; Dt is the time increment adopted; and Rv is the tensor
that defines the ‘‘direction” of the rate of C v . The total strain C is
kept constant along the second phase. Alternative common integra-
tion schemes include backward Euler method, midpoint rule and
Runge-Kutta procedures.
Fig. 3. Flowchart of the developed computer code.
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
5
As in the case of the equilibrium, expressions (23) and (24) are
nonlinear regarding the components of Cv and, thus, a Newton-
Raphson iterative technique is employed to perform the viscous
update:
M v ¼ ðC v ÞNþ1  expðDtRv ÞNþ1 ðC v ÞN ð25Þ

1
@M v
DC v ¼  : Mv ð26Þ
@C v
where Mv is the residual matrix of the viscous update scheme. It
should be remembered that this procedure is done at each Gaussian
point. As the present finite element is a solid tetrahedron of any-
order, the processing time may become extremely large and, thus,
a high computational effort is required for complex analyzes. How-
ever, this can be overcome by using parallel processing techniques,
for example. In this case, the accuracy of numerical results is
assured if the number of integration points is sufficiently large.
The algorithm proposed has been implemented in a computer
code and the corresponding flowchart is provided in Fig. 3.
5. Numerical examples
Four structural problems involving FMGs under finite viscoelas-
tic strains have been analyzed in order to assess the performance of
the proposed numerical formulation in predicting their mechanical
behavior. The analyzes have been performed via computer simula-
tions in a parallel programming cluster with eight processors. The
vector (17) is 106. The focus of the present study is the gradual
variation of viscoelastic parameters and, thus, the effect of the FG
power coefficient on the mechanical behavior is analyzed in detail
regarding displacements, strains and stresses. Five values for coef-
ficient p have been selected for all the numerical examples: p ? 0,
p = 0.25, p = 1.0, p = 2.5, and p ? 1. The first and the last values
correspond to the two homogeneous limiting cases, and the other
three values denote the FG intermediate cases. This procedure is
6 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx
usually done in the scientific literature for modeling FGMs (see, for
instance, [8], [45] and [46]).
Since the integrals in (17) and (18) are numerically evaluated,
the value of the material coefficients must be determined at each
integration point in the case of a FGM. The number of integration
points per element in this case should be larger than the corre-
sponding homogeneous case, which increases the processing time.
Although homogenization techniques and selective/reduced inte-
gration schemes can reduce considerably the processing time, full
integration is adopted in this study, as the present numerical ana-
lyzes are performed with the abovementioned cluster. It has been
demonstrated in the works of [27], [29] and [46] that full integra-
tion provides accuracy of numerical results in the context of non-
linear analyzes involving high-order solid tetrahedral finite
elements. Three polynomial degrees are used in the present study:
linear (four nodes), quadratic (10 nodes) and cubic (20 nodes per
element).
Finite element discretizations (or meshes) with different orders
and amounts of elements can be employed for each structural
problem to analyze the influence of the mesh refinement on the
mechanical behavior. The generation of such discretizations is per-
formed here from the same base mesh, composed of hexahedra of
linear order. Each hexahedron of the base mesh is divided into a
certain number of hexahedra according to the element order and
the number of divisions selected. The resultant hexahedra are
divided into six tetrahedral elements each. For example, if a base
mesh with four hexahedra has two divisions along the three spatial
directions, the resultant number of tetrahedral elements is: 4  (2
 2  2)  6 = 192. All the nodal coordinates of the resultant tetra-
hedral mesh are automatically determined from the coordinates of
the base mesh and the element order selected.
5.1. Uniaxial stress
The first problem is the prismatic specimen showed in Fig. 4.
The purpose is to analyze the influence of the variable viscoelastic
parameters on the response of a bar under uniaxial stress. The
boundary conditions depicted in Fig. 4 are employed to enforce this
stress state along x1. The material parameters adopted at x2 = 0
have been extracted from the numerical analysis performed by
[23] and correspond to a carbon-black filled natural rubber, which
is a usual elastomer employed, for example, in tires and bushings.
The values of the coefficients at x2 = 1.0 are 10% higher, except for
Fig. 4. Prismatic bar under uniaxial stress state (geometry, mesh, boundary conditions and material parameters).
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
the viscosity, which is 10 times higher. Two situations are analyzed
with the same specimen in order to show the capabilities of the
present formulation in reproducing the basic time-dependent phe-
nomena in large-strain viscoelasticity: creep under constant force
and stress relaxation under constant strain. The time increment
adopted is Dt = 0.1 s (per step). The mesh division employed in
both cases is 4-10-2, whose values denote the number of divisions
along the x1, x2 and x3-directions, respectively. The reason for this
mesh division is to enrich the discretization along the direction
with gradual variation (x2) and, at the same time, to avoid very
elongated elements. Only the linear order is used, as the present
example is very simple, which results in 165 nodes and 480 tetra-
hedral elements. The number of integration points per element is
one if the material is homogeneous and four if the material has
gradual variation.
The following linearized measures of small strain theory of
Solid Mechanics can be determined in uniaxial problems:
pffiffiffiffiffiffi
eeng
1 ¼ k1  1 ¼ C1  1 ð27Þ
pffiffiffiffiffiffi
rnom
1 ¼ k 1 S1 ¼ C 1 S1 ð28Þ
where eeng
1 is the longitudinal engineering (linearized) strain; rnom
1 is
the longitudinal nominal stress; and k1 is the longitudinal stretch.
The reason of employing these measures is that they have a direct
physical meaning, unlike stretch C1 and stress S1.
According to expression (15), increasing the FG power coeffi-
cient p leads to a higher volumetric ratio of the material defined
at x2 = 0. For the present example, this material is slightly less rigid
and has a much lower viscosity when compared to the material
defined at x2 = 1.0. The results obtained from the creep analysis
provided in Fig. 5 show that, as the power coefficient p increases,
the instantaneous elastic response is higher and the time spent
to stabilize the response is lower. This fact is expected because,
as demonstrated in [29] for the homogeneous case, an increase
in the viscosity parameter g leads to a lower instantaneous
response and a higher stabilization time. The creep curves of
Fig. 5 are similar to the numerical results obtained, for instance,
by [41] for a carbon-black filled chloroprene rubber. The constitu-
tive law adopted by those authors is the Bergström-Boyce model,
which is an extension of the elastic Arruda-Boyce model and has
the Zener rheological representation with a nonlinear viscous rate
evolution law (see, for example, [47]). The boundary conditions
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx 7

Fig. 5. Results for the prismatic bar under creep test: (a) prescribed nominal stress (28); (b) engineering strain (27).

employed in [41] are different from the present example, as one of
the faces normal to the longitudinal axis is completely clamped
and, thus, the stress state near this face is not uniaxial. Considering
the entire length of the bar, the maximum longitudinal mean strain
achieved in that work is 70%, and the maximum strain of the pre-
sent example varies from 740% to 820%, which represent extremely
large values of deformation. The initial undeformed configuration
is recovered in all cases when time becomes extremely large
(t ! 1). Moreover, the analysis of the constant-force intervals
reveals that the differences along the first interval (1 < t < 11 s)
are much higher when compared to the second one (12 < t < 50
s). This is probably due to the level of force achieved at each inter-
val. The stabilization time becomes higher by increasing the stress
level because, as pointed out by [23], the long molecular chains of
the polymer demands some time to deform and may entangle.
The influence of the constitutive law on the distribution of Cv11
can be seen in Fig. 6. The tendency mentioned in the last paragraph
is also present. Comparing the five distributions in each graph, an
increase in the power coefficient leads to an increase in the viscous
stretch Cv11, as the viscosity becomes smaller. Besides, for the FG
cases, the values of Cv11 decrease as the initial coordinate x2
increases. This is expected because, in such cases, the viscosity
parameter becomes higher (see Figs. 2 and 4).
The numerical results obtained from the relaxation test shows
that, as the power coefficient p increases, the instantaneous elastic
response and the stabilization time decreases (see Fig. 7). As in the
creep test, an increase in the value of p leads to a higher ratio of the
material with lower viscosity and, as demonstrated in the homoge-
neous study of [29], when the viscosity parameter adopted is
reduced, both instantaneous elastic response and stabilization
time decreases. The homogeneous material with the properties
defined at x2 = 1.0 is the most rigid material and, thus, has the lar-
gest value of instantaneous stress response. The reason for the
smaller stabilization time can be seen in the evolution Eq. (19),

in which one can note that, as the viscosity decreases, the rate C
v
becomes larger. In relaxation uniaxial tests, the longitudinal non-
equilibrium stress Sq1 vanish as the time evolves (see, for instance,
[29]) and the rate at which this happens depends on the rate of
stretch tensor Cv. In other words, as the coefficient p increases,
the peak stress achieved at t = 1 s and also at t = 12 s becomes
smaller. For all cases, however, the applied force stabilizes at t =
11 s, becomes a compressive load after t = 11.9 s and returns to
zero when t ! 1. This stress inversion is probably due to the dis-
sipative nature of the polymer deformation, as the entangled
molecular chains require a compressive stress to recover and pre-
serve their initial configuration. One can say that the maximum
strain level of 800% achieved along the interval 1 < t < 11 s corre-
sponds to a very high uniaxial deformation. Therefore, the present
model can reproduce both creep and relaxation tests of VFGMs in
the large-strain regime.
It is demonstrated that, even in the present simple example, the
effect of the constitutive model on the mechanical behavior of
VFGMs at finite strain levels is considerably great. The nature of

Fig. 6. Distributions of the longitudinal viscous stretch along the line segment x1 = x3 = 0 for the creep test: (a) t = 10 s; (b) t = 11 s.

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
8 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx

Fig. 7. Results for the prismatic bar under relaxation test: (a) prescribed engineering strain (27); (b) nominal stress (28).

the spatial variation of material coefficients will clearly affect the suspension system or the rolling resistance of tires. The material
overall and the local behavior in terms of creep, relaxation, instan- coefficients defined at x3 = 1.0 are the same as those reported
taneous elastic response and stabilization time. Because of the by [23]. The parameters at x3 = 0 have the same values, except
highly nonlinear character of the present material model, analyti- for Ke and le, which are eight times larger. The mesh division
cal solutions for both homogeneous and FG cases are practically employed for this case is 2-2-10 (along directions x1, x2 and x3,
impossible to be obtained. Then, the use of numerical methods respectively). As in the first example, only linear order is
during the design of a structure made of VFGMs is imperative even used, resulting in 240 tetrahedra and 99 nodes. The numerical
for simple cases. integration schemes adopted are the same as those of the first
problem.
5.2. Simple shear The measures from Solid Mechanics employed in this case are
the amount of shear and the mean nominal shear stress:
The second numerical example is a cubic block under cyclic

u1
simple shear condition (see Fig. 8). All the displacements along c12 ¼ arctan ð29Þ
h0
the x2-direction are restricted and the normal strain along the x3
are allowed. The objective in the present analysis is to evaluate
F 12
the influence of the graded material parameters (Ke and le) snom
12 ¼ ð30Þ
L 0 b0
on the usual hysteretic behavior of polymers under simple shear.
The prediction of the hysteresis phenomenon is important, for where L0, h0 and b0 are the initial dimensions along x1, x2 and
example, in the analysis of rubber springs used in a vehicle x3-directions, respectively.

Fig. 8. Block under simple shear (geometry, mesh, boundary conditions and material parameters).

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx 9

A triangular pulse with increasing amplitude is employed for
the prescribed displacement at the upper face. Two values for the
time increment have been adopted: 0.1 s and 0.01 s. These values
correspond to a prescribed displacement rate u_ 1 of 0.50 and 5.0,
respectively. The comparison of results provided in Figs. 9 and 10
hysteresis loop. According to Fig. 10d–f, it can be observed that
the hysteresis is also higher when the shearing limits of the
cycle becomes larger, as the stabilization time increases with
the stress level achieved. The finite levels of shearing achieved
(cmin
12 ¼ 1:25 and c12 ¼ 1:19) indicate that the present formula-
max

indicates that, as the displacement rate increases, the variation of
the applied forces and, thus, the hysteresis loop become larger.
This fact confirms the viscoelastic rate dependent stiffening behav-
ior usually observed in elastomers and reported, for instance, in
[41] for highly saturated nitrile HNBR50 rubber. Moreover, the
results of Fig. 10 show that the hysteresis loop becomes larger
when the FG power coefficient p increases, which results in a larger
ratio of the material with lower values of Ke and le. As pointed out
by [48], the hysteresis is caused by internal friction, which is
directly related to damping. When the moduli Ke and le decrease,
the relative effect of the dashpot of the Maxwell element in Fig. 1
increase, leading to a higher internal friction and, thus, a higher
tion can reproduce the hysteretic behavior of VFGMs under simple
shear and large-strain regime.
5.3. Cook’s membrane
The third problem is the Cook’s membrane, showed in Fig. 11.
This benchmark example is often used to assess the finite element
performance in a large deformation problem with combined bend-
ing and shear (see the works of [49–51], among many others).
Although it seems a plane structural problem, the singularity at
the upper left corner leads to a complex 3D behavior around this
point. The Cook’s membrane is usually analyzed under the

Fig. 9. Results for the block under simple shear with Dt = 0.1 s: (a) evolution of the prescribed displacement; graphs stress (30) versus strain (29) for: (b) homogeneous case 1
(p ! 0); (c) FG case p = 0.25; (d) FG case p = 1.0; (e) FG case p = 2.50; (f) homogeneous case 2 (p ! 1).

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
10 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx

Fig. 10. Results for the block under simple shear with Dt = 0.01 s: (a) evolution of the prescribed displacement; graphs stress (30) versus strain (29) for: (b) homogeneous
case 1 (p ! 0); (c) FG case p = 0.25; (d) FG case p = 1.0; (e) FG case p = 2.50; (f) homogeneous case 2 (p ! 1).

assumption of plane strain conditions, in which the planes x3 = 0.0
and x3 = 0.5 do not move along the x3-direction. In this paper, how-
ever, the stress-free conditions at these out-of-plane surfaces are
employed, following the work of [50], which have adopted the
nonlinear Hartmann-Neff hyperelastic model [52] without viscous
effects. The shear load is uniformly distributed along the right face
(x1 = 48.0) and is always vertical (not tangential to the deformed
face). One can note that the refinement of the base mesh is concen-
trated around the upper left corner, in which both stress and strain
fields are expected to be more complex. The time increment per
step adopted is Dt = 0.1 s. All five material parameters vary along
the thickness direction (0 6 x3 6 0:5). The reason for adopting this
kind of gradation is that there are many structural components
composed of thin FG panels and plates in which the material coef-
ficients vary across the thickness, such as the thermal barriers
employed in space structures and turbine rotors. The adopted val-
ues of the bulk moduli are much higher than the shear moduli and,
thus, the near-incompressibility condition is assumed, i.e., the vol-
umetric changes are supposed to be small. This condition usually
leads to numerical instabilities in the analysis of the Cook’s mem-
brane. Therefore, the objective in analyzing the present example is
to verify if the finite element formulation proposed can reproduce
the correct mechanical behavior in a complex problem. There is no
study in which the Cook’s membrane is analyzed within the con-
text of visco-hyperelasticity, according to the author’s biblio-
graphic review.
A convergence analysis regarding the most flexible homoge-
neous case (K1 = 1000, l1 = 10, Ke = 500, le = 8 and g = 200) has
been performed in order to assess the influence of the mesh refine-
ment on the membrane behavior. The plane displacements of point
A (48.0, 60.0, 0.0) at the end of phases 1 and 2 converge with mesh
refinement, according to Table 1. The influence of the power coef-
ficient p on the membrane behavior has been analyzed for the most
refined mesh: 3325 nodes and 576 cubic elements. The graph

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx 11

Fig. 11. Cook’s membrane: geometry, boundary conditions and material data (the dashed lines correspond to the base mesh).

Table 1
Convergence analysis for the most flexible homogeneous Cook’s membrane.

ORD NN NE NIP u1 (100) u2 (100) u1 (200) u2 (200) PT (s) TNI

1 60 144 1 9.28 14.19 1.63 3.21 11 979
189 576 12.56 15.89 1.99 3.55 41 900
390 1296 14.75 17.02 2.24 3.79 130 900
663 2304 16.48 17.92 2.48 4.04 306 883
1008 3600 17.81 18.60 2.71 4.29 751 828
1425 5184 18.83 19.10 2.90 4.51 3039 800
2 315 144 11 19.95 20.14 3.15 4.91 196 974
1105 576 –22.61 20.83 3.86 5.56 1322 900
2375 1296 –22.96 20.93 3.95 5.65 15,494 800
4125 2304 –23.05 20.98 3.98 5.67 48,085 800
3 910 144 15 –22.97 20.90 3.97 5.66 1014 900
3325 576 –23.09 21.01 3.99 5.68 39,808 800

ORD = element order. NN = number of nodes. NE = number of elements. NIP = number of integration points (per element). u1 (X) and u2 (X) = displacements of point A at the
end of step X.
PT (s) = processing time (in seconds). TNI = total number of iterations.

force-displacement depicted in Fig. 12 shows that the membrane
becomes more rigid when the power coefficient increases. As in
the first numerical examples, this trend is expected because the
ratio of the most rigid material (defined at x3 = 0.0) becomes larger
as the coefficient p increases. The maximum displacements pre-
sented in Fig. 12 shows that the membrane achieves large defor-
mation levels for all five cases, reinforcing the need of a
geometrically nonlinear analysis.
The results in terms of stresses are provided in Figs. 13–17 con-
sidering the most refined mesh (3325 nodes and 576 cubic order
elements). Due to the singularity point, individual stress compo-
nents may not converge with mesh refinement (see, for instance,
the work of [50]) and, thus, the present results are given in terms
of the equivalent Cauchy stress:
rffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
3
req ¼ dev r : dev r
2 be able to resolve possible singularities. Some finite element
where dev r is the deviatoric part of the Cauchy stress tensor. The
noticeable concentration of stresses at the upper left corner in all
five cases may give rise to crack formation or structural instability,
for example. The comparison of Figs. 13–17 indicates that the
equivalent stresses are higher for the homogeneous cases. The trend
between the maximum equivalent stress and the power coefficient
is not clear. Except for the homogeneous case 1 (p = 0.0), as the
power coefficient increases the maximum equivalent stress
becomes slightly larger. Comparing the zoom views, the stress
(31) seems to concentrate at the corner x3 = 0.5 for the homoge-
neous cases, and a little more at the corner closer to the plane x3
= 0 for the FG cases. The tendency in each of the heterogeneous
cases is expected because, as the x3-coordinate increases, the mate-
rial becomes less rigid.
As expected, the upper left corner is a critical point for all five
ð31Þ cases and, thus, the finite element formulation employed must

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
12 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx

Fig. 12. Graph force–displacement for the Cook’s membrane.

Fig. 13. Equivalent Cauchy stress field for the homogeneous case 1 of Cook’s Fig. 15. Equivalent Cauchy stress field for the FG case p = 1.0 of Cook’s membrane at
membrane at t = 10 s. t = 10 s.

Fig. 14. Equivalent Cauchy stress field for the FG case p = 0.25 of Cook’s membrane Fig. 16. Equivalent Cauchy stress field for the FG case p = 2.50 of Cook’s membrane
at t = 10 s. at t = 10 s.

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx
Fig. 17. Equivalent Cauchy stress field for the homogeneous case 2 of Cook’s
membrane at t = 10 s.
formulations fail in analyzing this singularity problem due to the
inability to reproduce the correct membrane flexibility, presenting
severe locking behavior. One can also note, in the perspective
views of Figs. 13–17, the displacements along the thickness
Fig. 18. Transverse Cauchy-Green stretch (C33) around the upper left corner of the Cook’s membrane at t = 10 s: (a) homogeneous case 1; (b) FG case p = 0.5; (c) FG case p =
1.0; (d) FG case p = 2.50; (e) homogeneous case 2.
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
13
direction (x3) around the upper left corner, which leads to a finite
3D strain field especially in this region. The transverse Cauchy-
Green stretch (C33) is also concentrated at the upper left corner
(see Fig. 18), reaching high levels. This may explain the reason
why some plane strain formulations can present numerical
instabilities even in linear-elastic analyzes. As in the case of
displacements, the values of C33 become smaller by increasing
the power coefficient (the membrane becomes more rigid).
Therefore, the present numerical formulation can reproduce the
large-deformation viscoelastic behavior of a Cook’s membrane
composed of FGM, capturing the singularity and the variation of
stresses and strains across the thickness direction.
5.4. Elastomeric bridge bearing
The last example is the rubber bearing depicted in Fig. 19. The
geometry is extracted from [53], which have analyzed a bridge
bearing composed of four rubber layers reinforced by three rigid
steel plates, including frictional contact between the pad and the
rigid blocks on top and bottom. The constitutive laws adopted in
that work are the nonlinear hyperelastic Yeoh model [54] for the
rubber and a bilinear elastoplastic model for the steel. In the pre-
sent study, only one elastomeric layer between two rigid faces is
analyzed. The prescribed displacement at the upper face is applied
along the five phases showed in Fig. 20. One can note that the ver-
tical normal strain achieves 10% at the end of phase 1 and is held
14 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx

Fig. 19. Elastomeric bridge bearing: geometry, boundary conditions and material data.

Fig. 20. Phases of the prescribed displacement imposed at the upper face of the elastomeric bearing.

constant along phase 2. The idea is to simulate the effect of the
bridge weight on the support that contains the rubber bearing
before the passage of the vehicles. The normal strain reached in
[54] along the vertical loading, considering the material data and
the average vertical stress applied, is approximately 7.2%, which
explains the reason of selecting 10% of compression in the present
study. After the end of phase 2, the face is moved simultaneously
along the x1 and x3 directions, until a maximum horizontal dis-
placement equal to the rubber thickness and 20% of compression
along the vertical direction. Then the face is brought back to the
same position reached at the beginning of phase 3 and do not move
along the last phase. The objective of analyzing the bearing with
these last three phases is to simulate the effect of the traffic load-
ing, which causes both vertical and horizontal displacements at the
pad. One should highlight that possible rotations of the upper face
are neglected in the present study, considering that they are pre-
vented by the rigid faces. In the work of [53], these rotations occur
but are negligible even after the horizontal traffic loading.
The present numerical example is analyzed considering the
symmetry regarding the plane x3 = 228.6 mm. The base mesh
employed is divided into three parts along x1 and x2-directions,
and two parts along x3-direction (see Fig. 19). The material coeffi-
cients adopted for the upper face (x2 = 11.113 mm) are extracted
from [23] and the parameters at the lower face (x2 = 0) are defined
as in the first example (K1, l1, Ke and le increased by 10% and vis-
cosity g 10 times higher). The FG material distribution through the
thickness direction (x2) can be justified by possible temperature
cycling which, according to the study of [55] on asphaltic pave-
ments, leads to variation in temperature with depth. In this case,
since the viscoelastic response of an elastomer depends on the
temperature level, the material coefficients will vary across the
thickness.
As in the third example, a convergence analysis of the vertical
reaction forces at the top has been carried out for the most flexible
homogeneous case. As expected, the reaction forces (at time t = 11
s) converge with mesh refinement, according to table 2. The values
of the forces computed are the sum of all the vertical forces at the
upper face necessary to reproduce the prescribed displacement.
The evolution of the vertical normal stress can be seen in
Figs. 21–25 for the most refined mesh. The maximum compressive

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx 15

Table 2
Convergence analysis for the elastomeric bridge bearing.

ORD NN NE NIP RF1 RF2 PT (s) TNI

1 64 162 1 0.1202 1.1976 2 273
343 1296 0.1199 1.1181 29 256
1000 4374 0.1198 1.0905 412 245
2197 10,368 0.1197 1.0771 5628 260
2 343 162 11 0.1197 1.0794 25 259
2197 1296 0.1191 1.0570 3949 261
3 1000 162 15 0.1191 1.0557 235 269
6859 1296 0.1173 1.0410 60,136 278

ORD = element order. NN = number of nodes. NE = number of elements. NIP = number of integration points (per element). RF1 and RF2 = reaction forces at t = 11 s. PT (s) =
processing time (in seconds). TNI = total number of iterations.

Fig. 21. Distribution of the normal Cauchy stress r22 for the homogeneous case 1 of the bridge bearing at: (a) t = 1 s; (b) t = 10 s; (c) t = 11 s; (d) t = 20 s.

Fig. 22. Distribution of the normal Cauchy stress r22 for the FG case p = 0.25 of the bridge bearing at: (a) t = 1 s; (b) t = 10 s; (c) t = 11 s; (d) t = 20 s.

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
16 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx

Fig. 23. Distribution of the normal Cauchy stress r22 for the FG case p = 1.0 of the bridge bearing at: (a) t = 1 s; (b) t = 10 s; (c) t = 11 s; (d) t = 20 s.

Fig. 24. Distribution of the normal Cauchy stress r22 for the FG case p = 2.50 of the bridge bearing at: (a) t = 1 s; (b) t = 10 s; (c) t = 11 s; (d) t = 20 s.

stress occurs near the symmetry plane (x3 = 0.2286 m) and
becomes smaller when the power coefficient p increases, as the
material becomes more flexible. The stress relaxation phenomenon
can be identified in all five cases comparing the stress values of t =
1 s and t = 10 s, although the differences are relatively small. This
means that the stresses (and also the strains and displacements)
in the bearing demand some time to stabilize even under a con-
stant loading, which can influence the overall structural behavior
of the bridge. The distributions at t = 10 s and t = 20 s are approxi-
mately the same, indicating that the bearing recovers the stresses
when time becomes extremely large (t ! 1) considering the same
levels of prescribed displacement. One can note the effect of the
traffic loading by comparing the stress distributions at t = 10 s
and t = 11 s. The maximum vertical normal stress becomes
approximately 2.4 times higher at t = 11 s, in which there is a shear
deformation together with a compressive strain two times larger
when compared to the time of 10 s. Moreover, the minimum stress
values are slightly more spread over the edge x1 = 228.6 mm at t =
11 s comparing to the other times, showing the influence of the
shearing on the distribution of the vertical stress r22 on the upper
face. Considering the displacement fields, one can see considerable
levels of displacement along the x1 and x3-directions in Figs. 21–25,
showing the barreling of the bearing, also noted in [53].
Some distributions of the vertical stress r22 along two line seg-
ments are provided in Figs. 26 and 27. Again, the stress values
become smaller by increasing the power coefficient p and are
higher at t = 11 s. The effect of the shearing on the curve stress ver-
sus coordinate x1 is more remarkable near the edge x1 = 0 (see
Fig. 26). This effect is different from the numerical results obtained
by [53], in which the stress curve only bend slightly to the shearing

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx 17

Fig. 25. Distribution of the normal Cauchy stress r22 for the homogeneous case 2 of the bridge bearing at: (a) t = 1 s; (b) t = 10 s; (c) t = 11 s; (d) t = 20 s.

Fig. 26. Vertical stress distribution for the elastomeric bridge bearing along the segment x2 = 11.113 mm and x3 = 228.6 mm at: (a) t = 10 s; (b) t = 11 s.

Fig. 27. Vertical stress distribution for the elastomeric bridge bearing along the segment x1 = 114.3 mm and x3 = 228.6 mm at: (a) t = 10 s; (b) t = 11 s.

Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
18 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx
direction, preserving its shape. The difference is probably related to
the constitutive models adopted in each study. The graphs in
Fig. 27 indicate that the stress becomes higher with the shearing,
but the shape of the curves is preserved. This oscillatory shape
indicates the complexity of stresses and strains inside the bearing
even with no shearing, which means that the finite element
formulation employed should be able to reproduce such fields.
It is worthwhile to mention that homogenization techniques
usually employed for FGMs may not be able to capture unexpected
oscillatory behavior of stresses and strains. Although the
processing time and the computational effort can be reduced
with the homogenization process, only the global structural
behavior can be accurately reproduced depending on the problem
analyzed and on the mesh employed. In all five cases, the
stresses at the upper face are higher than the stresses at the lower
face. This is expected for the FG cases as the material is more
rigid at the upper face. Another interesting aspect is that the FG
bearings present an intermediate behavior between the two
homogeneous limiting cases regarding the vertical normal stress
(see Figs. 26 and 27).
6. Conclusions
Large deformation analysis of VFGMs under finite strain levels
is originally presented in this paper. The material modeling at
the finite viscoelastic regime is based on a Lagrangian isotropic
visco-hyperelasticity framework. The material non-homogeneity
is described by a continuous FG power law applied to the five con-
stitutive parameters along one direction.
It is demonstrated that the present formulation can be
employed to predict the basic viscoelastic phenomena associated
with FG elastomers under finite strains: creep, stress relaxation
and hysteresis. It has also been observed that the present approach
can reproduce the mechanical behavior of complex problems even
in the near-incompressible finite strain regime, in which volumet-
ric locking problems often appear. Fully integrated solid tetrahe-
dral elements of linear, quadratic and cubic orders are used in
the present study, reinforcing that accuracy of results is guaran-
teed for sufficiently refined meshes.
The novelty of the paper is the use of gradual variation of the
visco-hyperelastic coefficients, as well as the investigation of the
material gradation on the structural behavior. Five material laws
are employed in all cases, following the usual procedure: two
homogeneous limiting cases and three intermediate FG cases.
The effect of the gradual variation on the structural behavior is
studied in terms of the power coefficient employed in each case.
As this paper is original regarding VFGMs, no reference data are
available for comparisons and, thus, the present numerical results
can be used as a basis for analyzing the performance of future
numerical formulations.
The present formulation has been successfully implemented
and, therefore, further implementations of other constitutive mod-
els can be performed, including different hyperelastic laws, rheo-
logical models and evolution laws, as well as viscoplasticity and
thermal effects, which can be observed in polymers and FGMs.
The formulation is also promising for other types of studies as,
for example, material model calibration and dynamic and impact
analyzes.
Acknowledgements
The author appreciates all the essential support given by the
following two departments, both from the University of São Paulo
(USP): the Materials Engineering Department of the Lorena School
of Engineering (DEMAR/EEL) for providing the necessary infras-
tructure to carry out the work; and the Structural Engineering
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
Department of the São Carlos School of Engineering (SET/EESC)
for allowing the remote access to their cluster.
Appendix A. Supplementary material
Supplementary data associated with this article can be found, in
the online version, at https://doi.org/10.1016/j.compstruc.2018.06.
001.
References
[1] Altenbach H, Eremeyev VA. Analysis of the viscoelastic behavior of plates made
of functionally graded materials. Z Angew Math Mech 2008;88(5):332–41.
[2] Hosseini-Hashemi Sh, Abaei AR, Ilkhani MR. Free vibrations of functionally
graded viscoelastic cylindrical panel under various boundary conditions.
Compos Struct 2015;126:1–15.
[3] Deng J, Liu Y, Zhang Z, Liu W. Size-dependent vibration and stability of multi-
span viscoelastic functionally graded material nanopipes conveying fluid using
a hybrid method. Compos Struct 2017;179:590–600.
[4] Hilton HH. Optimum linear and nonlinear viscoelastic designer functionally
graded materials – characterization and analysis. Compos A Appl Sci Manuf
2005;36:1329–34.
[5] Hilton HH, Lee DH, El Fouly ARA. Generalized viscoelastic designer functionally
graded auxetic materials engineered/tailored for specific task performances.
Mech Time-Dependent Mater 2008;12:151–78.
[6] Wang ZH, Zhang L, Guo LC. Fracture analysis of a viscoelastic functionally
graded materials strip with a central crack. J Phys: Conf Ser 2013;419:012015.
[7] Wang ZH, Zhang L, Guo LC. A viscoelastic fracture mechanics model for a
functionally graded materials strip with general mechanical properties. Eur J
Mech A/Solids 2014;44:75–81.
[8] Chi SH, Chung YL. Mechanical behavior of functionally graded material plates
under transverse load – Part I: Analysis. Int J Solids Struct 2006;43:3657–74.
[9] Banic D, Bacciocchi M, Tornabene F, Ferreira AJM. Influence of Winkler-
Pasternak foundation on the vibrational behavior of plates and shells
reinforced by agglomerated carbon nanotubes. Appl Sci 2017;7(1228):1–55.
[10] Tornabene F, Fantuzzi N, Bacciocchi M, Viola E, Reddy JN. A numerical
investigation on the natural frequencies of FGM sandwich shells with variable
thickness by the local generalized differential quadrature method. Appl Sci
2017;7(131):1–39.
[11] Fantuzzi N, Tornabene F, Bacciocchi M, Dimitri R. Free vibration analysis of
arbitrarily shaped functionally graded carbon nanotube-reinforced plates.
Composites Part B 2017;115:384–408.
[12] Amadori S, Catania G. Robust identification of the mechanical properties of
viscoelastic non-standard materials by means of frequency domain
experimental measurements. Compos Struct 2017;169:79–89.
[13] Holzapfel GA. Nonlinear solid mechanics – a continuum approach for
engineering. Chichester, England: John Wiley & Sons; 2004.
[14] Pascon JP. Constitutive Models for Hyperelastic Materials: study and computer
implementation. Master’s Dissertation (in Portuguese). Structural Engineering
Department, São Carlos School of Engineering. São Carlos, São Paulo, Brazil;
2008.
[15] Bernstein B, Kearsley EA, Zapas LJ. A study of stress relaxation with finite
strain. Trans Soc Rheol 1963;7:391–410.
[16] Holzapfel GA. On large strain viscoelasticity: continuum formulation and finite
element applications to elastomeric structures. Int J Numer Methods Eng
1996;39:3903–26.
[17] Haupt P, Lion A. On finite linear viscoelasticity of incompressible isotropic
materials. Acta Mech 2002;159:87–124.
[18] Holzapfel GA, Simo JC. A new viscoelastic constitutive model for continuous
media at finite thermomechanical changes. Int J Solids Struct
1996;33:3019–34.
[19] Reese S, Govindjee S. Theoretical and numerical aspects in the thermo-
viscoelastic material behaviour of rubber-like polymers. Mech Time-Depend
Mater 1998;1:357–96.
[20] Huber N, Tsakmakis C. Finite deformation viscoelasticity laws. Mech Mater
2000;32:1–18.
[21] Rauchs C. Finite element implementation including sensitivity analysis of a
simple finite strain viscoelastic constitutive law. Comput Struct
2010;88:825–36.
[22] Lejeunes S, Boukamel A, Méo S. Finite element implementation of nearly-
incompressible rheological models based on multiplicative decompositions.
Comput Struct 2011;89:411–21.
[23] Naghdabadi R, Baghani M, Arghavani J. A viscoelastic constitutive model for
compressible polymers based on logarithmic strain and its finite element
implementation. Finite Elem Anal Des 2012;62:18–27.
[24] Petiteau JC, Verron E, Othman R, Le Sourne H, Sigrist JF, Barras G. Large strain
rate-dependent response of elastomers at different strain rates: convolution
integral vs. internal variable formulations. Mech Time-Depend Mater
2013;17:349–67.
[25] Simo JC. On a fully three-dimensional finite-strain viscoelastic damage model:
formulation and computational aspects. Comput Methods Appl Mech Eng
1987;60:153–73.
 J.P. Pascon / Computers and Structures xxx (2018) xxx–xxx
[26] Latorre M, Montáns FJ. Fully anisotropic finite strain viscoelasticity based on a
reverse multiplicative decomposition and logarithmic strains. Comput Struct
2016;163:56–70.
[27] Pascon JP, Coda HB. Analysis of elastic functionally graded materials under
large displacements via high-order tetrahedral elements. Finite Elem Anal Des
2012;50:33–47.
[28] Pascon JP, Coda HB. High-order tetrahedral finite elements applied to large
deformation analysis of functionally graded rubber-like materials. Appl Math
Model 2013;37:8757–75.
[29] Pascon JP, Coda HB. Finite deformation analysis of visco-hyperelastic materials
via solid tetrahedral finite elements. Finite Elem Anal Des 2017;133:25–41.
[30] Bui TQ, Vo DQ, Zhang C, Nguyen DD. A consecutive-interpolation quadrilateral
element (CQ4): formulation and applications. Finite Elem Anal Des
2014;84:14–31.
[31] Bui TQ, Nguyen DD, Zhang X, Hirose S, Batra RC. Analysis of 2-dimensional
transient problems for linear elastic and piezoelectric structures using the
consecutive-interpolation quadrilateral element (CQ4). Eur J Mech A/Solids
2016;58:112–30.
[32] Nguyen MN, Bui TQ, Truong TT, Trinh NA, Singh IV, Yu T, et al. Enhanced nodal
gradient 3D consecutive-interpolation tetrahedral element (CTH4) for heat
transfer analysis. Int J Heat Mass Transf 2016;103:14–27.
[33] Nguyen MN, Bui TQ, Truong TT, Tanaka S, Hirose S. Numerical analysis of 3-D
solids and composite structures by an enhanced 8-node hexahedral element.
Finite Elem Anal Des 2017;131:1–16.
[34] Yu TT, Yin S, Bui TQ, Hirose S. A simple FSDT-based isogeometric analysis for
geometrically nonlinear analysis of functionally graded plates. Finite Elem
Anal Des 2015;96:1–10.
[35] Bui TQ, Hirose S, Zhang C, Rabczuk T, Wu CT, Saitoh T, et al. Extended
isogeometric analysis for dynamic fracture in multiphase piezoelectric/
piezomagnetic composites. Mech Mater 2016;97:135–63.
[36] Lai W, Yu T, Bui TQ, Wang Z, Curiel-Sosa JL, Das R, et al. 3-D elasto-plastic large
deformations: IGA simulation by Bézier extraction of NURBS. Adv Eng Softw
2017;108:68–82.
[37] Anani Y, Rahimi G. Modeling of visco-hyperelastic behavior of transversely
isotropic functionally graded rubbers. Polym Eng Sci 2016;56(3):342–7.
[38] Sidoroff F. Un modèle viscoélastique non linéaire avec configuration
intermédiaire. Journal de Mécanique 1974;13:679–713.
[39] Pascon JP. On nonlinear constitutive models for functionally graded materials
exhibiting large strains: numerical implementation in geometrically nonlinear
formulation. PhD Thesis (in Portuguese). Structural Engineering Department,
São Carlos School of Engineering. São Carlos, São Paulo, Brazil; 2012.
Please cite this article in press as: Pascon JP. Large deformation analysis of functionally graded visco-hyperelastic materials. Comput Struct (2018), https://
doi.org/10.1016/j.compstruc.2018.06.001
19
[40] Hasanpour K, Ziaei-Rad S, Mahzoon M. A large deformation framework for
compressible viscoelastic materials: constitutive equations and finite element
implementation. Int J Plast 2009;25:1154–76.
[41] Areias P, Matous K. Finite element formulation for modeling nonlinear
viscoelastic elastomers. Comput Methods Appl Mech Eng 2008;197:
4702–17.
[42] Sze KY, Zheng SJ, Lo SH. A stabilized eighteen-node solid element for
hyperelastic analysis of shells. Finite Elem Anal Des 2004;40:319–40.
[43] Rivlin RS. Large elastic deformations of isotropic materials. II. Some
uniqueness theorems for pure, homogeneous deformation. Philos Trans R
Soc A 1948;240:491–508.
[44] Keast P. Moderate-degree tetrahedral quadrature formulas. Comput Methods
Appl Mech Eng 1986;55:339–48.
[45] Arciniega RA, Reddy JN. Large deformation analysis of functionally graded
shells. Int J Solids Struct 2007;44:2036–52.
[46] Pascon JP, Coda HB. Large deformation analysis of functionally graded
elastoplastic materials via solid tetrahedral finite elements. Comput Struct
2015;146:59–75.
[47] Bergström JS, Boyce MC. Constitutive modelling of the time-dependent and
cyclic loading of the elastomers and application to soft biological tissues. Mech
Mater 2001;33:523–30.
[48] Bauman JT. Fatigue, stress, and strain of rubber components: a guide for design
engineers. Ed. Hanser; 2008. ISBN 3446416811, 9783446416819.
[49] César de Sá JMA, Areias PMA, Jorge MN. Quadrilateral elements for the solution
of elasto-plastic finite strain problems. Int J Numer Meth Eng
2001;51:883–917.
[50] Düster A, Hartmann S, Rank E. P-FEM applied to finite isotropic hyperelastic
bodies. Comput Methods Appl Mech Eng 2003;192:5147–66.
[51] Kadapa C, Dettmer WG, Péric D. Subdivision based mixed methods for
isogeometric analysis of linear and nonlinear nearly incompressible materials.
Comput Methods Appl Mech Eng 2016;305:241–70.
[52] Hartmann S, Neff P. Polyconvexity of generalized polynomial-type
hyperelastic strain energy functions for near-incompressibility. Int J Solids
Struct 2003;40:2767–91.
[53] Nguyen HH, Tassoulas JL. Directional effects of shear combined with
compression on bridge elastomeric bearings. J Bridge Eng 2010;15(1):73–80.
[54] Yeoh OH. Some forms of the strain energy function for rubber. Rubber Chem
Technol 1993;66(5):754–71.
[55] Dave EV, Paulino GH, Buttlar WG. Viscoelastic functionally graded finite
element method with recursive time integration and applications to flexible
pavements. Int J Numer Anal Meth Geomech 2012;36:1194–219.