Finite Elements in Analysis and Design 86 (2014) 61–70

Contents lists available at ScienceDirect

Finite Elements in Analysis and Design

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

Integrating a logarithmic-strain based hyperelastic formulation into

a three-field mixed finite element formulation to deal with
incompressibility in finite-strain elastoplasticity
Dina Al Akhrass a,b,c,n, Julien Bruchon a, Sylvain Drapier a, Sébastien Fayolle b,c
a
Ecole des Mines de Saint-Etienne, Centre SMS, LGF UMR CNRS 5307, 158 cours Fauriel, 42023 Saint-Etienne Cedex 2, France
b
EDF R&D, Département AMA, 1, Avenue du Général de Gaulle, 92141 Clamart, France
c
Laboratoire de Mécanique des Structures Industrielles Durables, UMR EDF-CNRS-CEA 8193, 1, Avenue du Général de Gaulle, 92141 Clamart, France

art ic l e i nf o a b s t r a c t

Article history: This paper deals with the treatment of incompressibility in solid mechanics in finite-strain elastoplas-
Received 6 September 2013 ticity. A finite-strain model proposed by Miehe, Apel and Lambrecht, which is based on a logarithmic
Received in revised form strain measure and its work-conjugate stress tensor is chosen. Its main interest is that it allows for the
20 February 2014
adoption of standard constitutive models established in a small-strain framework. This model is
Accepted 1 April 2014
extended to take into account the plastic incompressibility constraint intrinsically. In that purpose, an
Available online 28 April 2014
extension of this model to a three-field mixed finite element formulation is proposed, involving
Keywords: displacements, a strain variable and pressure as nodal variables with respect to standard finite element.
Incompressibility Numerical examples of finite-strain problems are presented to assess the performance of the formula-
Finite-strain
tion. To conclude, an industrial case for which the classical under-integrated elements fail is considered.
Mixed finite element method
& 2014 Elsevier B.V. All rights reserved.
Logarithmic strains

1. Introduction geometric processing. In such a way, the relationship between the

large-strain and the small-strain setting is defined by purely
Over the past three decades, the development of finite element geometric transformations. When modeling finite-strain elastoplas-
formulations capable of modeling large elastoplastic strains has tic processes by the finite element method, it is important to
received considerable attention. The earliest formulations, intro- consider the nearly incompressible plastic behavior, which is typical
duced in the 1970s, were rate-type formulations based on a of metals, for instance. It is well-known that the standard
hypoelastic stress–strain relationship [1–3]. The main drawback of displacement-based finite element formulation performs poorly in
this kind of formulations is that they introduce elastic dissipation. quasi-incompressible situations, producing stiff solutions and oscil-
Despite that, they are still commonly used today. In the 1980s, Simo lations of the stress field. Over the years, and particularly in the
and co-workers [4–7] introduced a formulation based on hyper- 1990s, different strategies were proposed to reduce or avoid
elastic constitutive formulations that does not exhibit elastic volumetric locking and pressure oscillations in finite element
dissipation. Many researchers, such as Eterovic and Bathe [8] or solutions. Several methods to deal with incompressibility have
Weber and Anand [9], developed some formulations that use a been developed, such as under-integrated elements [11], Enhanced
hyperelastic constitutive model based on the logarithmic strain Assumed Strain (EAS) methods [12–14], B-bar and F-bar methods
tensor, also called the Hencky tensor. In these works, an additive [15,16], or mixed formulations [17,18]. The mixed finite element
decomposition of the elastic and plastic strains in the absence of method is robust and generic and it is a popular and efficient way to
rotations is assumed, which is a typical feature of the geometrically deal with incompressibility. Mixed elements for finite strains were
linear theory of plasticity. This provides a natural basis for a first introduced by Simo et al. [19] and have been since developed
material-independent extension of constitutive structures from by many authors [20,21]. In non-linear solid mechanics, the use of a
the geometrically linear to the nonlinear theory at finite-strain. two-field formulation is not convenient for many constitutive
More recently, Miehe et al. [10] developed a formulation for which models [20,22]. For example when the plasticity criterion depends
the kinematic setting consists of a constitutive model in the on the hydrostatic stress like in Rousselier or GTN laws [23,24],
logarithmic strain space that is preceded and followed by purely a two-field formulation based on displacement and pressure is not
straightforward and differ from the chosen law. For this kind of
laws, the use of a three-field formulation in which the unknowns
n
Corresponding author. Tel.: þ 33 6 11 98 75 89. are the displacement, the pressure and the volumetric strain fields,
E-mail address: alakhrass@emse.fr (D. Al Akhrass). allows us to have a generic method. Furthermore, in the finite-

http://dx.doi.org/10.1016/j.finel.2014.04.004
0168-874X/& 2014 Elsevier B.V. All rights reserved.
62 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70

strain framework, it is important to express these formulations in a and can be expressed as

way that makes the implementation of the constitutive laws simple.
1
That is why the idea in this work is to adopt the model of Miehe Ep ¼ lnðGp Þ ð6Þ
2
et al. [10], and to adapt it to a three-field formulation.
This paper presents a robust non-linear mixed finite element Thus, the logarithmic strain tensor allows us to switch from
procedure for the numerical analysis of finite-strain elastoplasti- multiplicative properties for the elastoplasticity in finite strain to
city, where the use of specific elements is considered to deal with the additive structure of the small strain theory. Furthermore, the
plastic incompressibility. The Miehe's model, based on a logarith- plastic Jacobian denoted Jp is such that
mic description of the strain tensor, is exposed in Section 2. qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
A three-field mixed finite element formulation is then presented J p ≔ detðGp Þ ¼ exp½trðE p Þ ð7Þ
in Section 3. The extension of the finite-strain model to this three-
field formulation is described in Section 4. Finally, some numerical which gives
simulations are presented in Section 5. The efficiency of the trðE p Þ ¼ lnðJ p Þ ð8Þ
developed model is first shown through numerical tests. A case
of industrial interest, for which under-integration technique fails, Thus, it can be written that
is then considered in order to assert the robustness of the detðGp Þ ¼ 1 3 trðE p Þ ¼ 0 ð9Þ
presented approach.
Indeed, the multiplicative constraint on the determinant of the
plastic metric is described by the additive constraint on the trace
of the logarithmic plastic strain.
2. Description of the finite-strain model
The model is based on the logarithmic strain tensor (1) and its
work-conjugate stress tensor denoted T. In order to get the
In order to describe the finite-strain framework, a hyperelastic-
expression of T , in terms of standard tensors, the expended power
based model developed by Miehe, Apel and Lambrecht is con-
is expressed as a function of the second Piola–Kirchhoff stress
sidered [10,25]. The kinematic framework consists of a constitutive
tensor S and the right Cauchy–Green strain tensor C
model in the logarithmic strain space that is framed by purely
geometric pre-processing and post-processing steps. This numer- 1_
PðtÞ ¼ SðtÞ : C ðtÞ ð10Þ
ical approach of the material response computation can be split 2
into three steps: a geometric pre-processing step in which the
with
logarithmic strain tensor is defined, a second step to get its work-
∂C
conjugate stresses computed from the constitutive model, and a C_ ðtÞ ¼ ð11Þ
third one to get back to classical tensors using geometric post- ∂t
processing. Note that in the present work, Lagrangian tensors are and as a function of E and T
considered.
_
PðtÞ ¼ T ðtÞ : EðtÞ ð12Þ

so that T can be expressed as [10]

2.1. Geometric pre-processing
T ¼ S : P L 1 ð13Þ
This step consists in defining the logarithmic strain tensor, and
the geometric transformations necessary to get it along with its with
associated stress tensor, from standard tensors. The logarithmic ∂E
strain tensor E is defined by PL ¼ 2 ð14Þ
∂C
1 Thus, geometric relationships between the logarithmic strain
E¼ lnðCÞ ð1Þ
2 tensor E and its work-conjugate stress tensor T with respectively,
the right Cauchy–Green strain tensor, and the second Piola–
with C the right Cauchy–Green strain tensor,
Kirchhoff stress tensor, have been established.
C ¼ FT F ð2Þ

where F is the deformation gradient tensor defined as the relative 2.2. Constitutive model in the logarithmic strain space
deformation of the medium from its initial state (position X) to its
current state (position x) Let us assume a constitutive model of plasticity that is written
in the logarithmic strain space. From the logarithmic strain tensor
∂xi
F ij ¼ ð3Þ E and the logarithmic plastic strain tensor E p , calculated respec-
∂X j
tively by (1) and (6), and some hardening variables denoted α, the
Following Miehe's approach [10], an additive decomposition of the constitutive model provides the stress tensor T and the associated
logarithmic strain is considered elastoplastic tangent modulus Eep

E ¼ Ee þ Ep ð4Þ fE; E p ; αg ) Constitutive model ) fT; Eep g ð15Þ

where E e is referred as the elastic strain, and E p as the plastic strain. The tangent modulus yields the rate of the stress T with respect to
It has been shown in Refs. [10,25,26] that this additive decomposi- the rate of the logarithmic strain
tion provides results close to those obtained by assuming Lee's T_ ¼ Eep : E_ ð16Þ
multiplicative decomposition of the deformation gradient [27].
E p is assumed to be a function of the plastic metric tensors Note that, in the logarithmic strain space, the constitutive model
p
G [26] defined by has the same structure as models for small-strain plasticity. It is
hence possible to adopt simply, for finite-strain elastoplasticity,
Gp ¼ F pT F p ð5Þ the frame of constitutive models from the small-strain theory.
 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70 63

2.3. Geometric post-processing Denoting ð:ÞD the deviatoric part of a tensor and 1 the second rank unit
tensor, the enhanced logarithmic strain tensor can be expressed as
Once the stress tensor and the tangent modulus have been
θ
obtained in the logarithmic strain space, the last step consists in E ¼ ED þ 1 ð25Þ
3
recovering the standard stress tensor, which is easier to handle,
and its associated tangent modulus, by the application of the Thus, with the choice of relation (20), θ corresponds to the spherical
transformation rules introduced above. The second Piola–Kirchh- part of the enhanced logarithmic strain tensor E [21].
off stress tensor S is obtained from relation (13) The Lagrangian multiplier p is introduced in order to enforce
the relationship between θ and J (20). It is a third variable
S ¼ T : PL ð17Þ
independent of the two others, which corresponds to the hydro-
and it can be shown that the associated tangent modulus which is static pressure. Thus, a variational form for the finite-strain
such as problem is given on the initial domain Ω by [20]
Z
1_
S_ ¼ Cep
L : C ð18Þ Π ðu; p; θÞ ¼ ½WðC Þ þ pðlnðJÞ  θÞ dΩ  W ext ð26Þ
2 Ω
may be written [10] with W being the stored energy function expressed in terms of the
     2  enhanced right Cauchy–Green tensor, and Wext the potential of
∂E T ∂E ∂ E
Cep
L ¼ 2 : Eep : 2 þT : 4 2 ð19Þ external forces.
∂C ∂C ∂C
The weak formulation of the problem is obtained by looking for
To summarize, the strain tensor C is first transformed into the the saddle-point of the Lagrangian functional Π , which is char-
logarithmic strain tensor E, for which the decomposition in elastic acterized by the Euler–Lagrange optimality conditions
and plastic parts is additive. This property allows us to adopt a
constitutive law from the small strain theory, and yields the stress δΠ ðu; p; θÞ½δu; δp; δθ ¼ 0 ð27Þ
tensor T and the tangent modulus Eep L in the logarithmic strain for all (δu; δθ; δp) sufficiently regular vanishing on the domain
space, which are, eventually, transformed into the standard boundary where u, θ and p are prescribed, denoted from here as
tensors S and Cep
L . admissible.
The second Piola–Kirchhoff stress S is derived from the poten-
tial W
3. A three-field mixed finite element formulation
∂W
S ¼2 ð28Þ
The mixed finite element method is an efficient way to deal ∂C
with incompressibility in solid or fluid mechanics. Let us recall and the variation of (26) leads to
that such mixed finite element formulation is used to ensure the Z    
finite element convergence and stability [15]. There exist different 1 δJ
δΠ ðu; p; θÞ ¼ S : δC þ p  δθ þ δpðlnðJÞ  θÞ dΩ  δW ext ¼ 0
possible formulations, such as three-field formulations, involving Ω 2 J
displacements u, a strain-like variable θ, and pressure p, or two- ð29Þ
field formulations involving only displacements and pressure The variation of the displacement gradient with respect to the
[28–30]. The choice of variable combinations may depend on the current configuration, δL, is given by
form of the constitutive law employed. As explained in introduction,
to be able to naturally deal with a plasticity criterion depending on δL ¼ δFF  1 ð30Þ
the hydrostatic stress, the three-field formulations are preferred [18].
Next, from the relation
As the aim of this work is to develop a method as generic as possible,
we decide to concentrate on a three-field formulation. ∂J
¼ JF  T ð31Þ
∂F
3.1. Weak mixed formulation which is established in [31], it can be shown that
δJ
In order to establish the weak formulation, the deformation ¼ 1 : δL ð32Þ
J
gradient F is enhanced by introducing a strain-like variable θ
which allows us to measure the volume variation. This latter is The variation of the enhanced right Cauchy–Green tensor is thus
weakly related to the Jacobian of the deformation gradient J [20]. given by
Several relations are possible, among which  2=3  
expðθÞ 2 δJ
lnðJÞ ¼ θ ð20Þ δC ¼ δC þ δθ  C ð33Þ
J 3 J
In that case, the enhanced deformation gradient F can be expressed as where
 1=3
expðθÞ δC ¼ F T ðδL þ δLT ÞF ð34Þ
F¼ F ð21Þ
J
Using relation (32), it can then be written
The enhanced right Cauchy–Green tensor C is then computed from  
T T 2
C ¼F F
T
ð22Þ δC ¼ F δLD þ δLD þ δθ1 F ð35Þ
3
Note that, using this relation, Using relations (32) and (35), the variation of the Lagrangian (29)
1 T may be expressed as
E ¼ lnðF F Þ ð23Þ Z   
2 1
can be written as
δΠ ðu; p; θÞ ¼ ðτ D þ p1Þ : δL þ δθ 1 : τ  p dΩ
Ω 3
Z
1  
E ¼ E þ ðθ  lnðJÞÞ1 ð24Þ þ ½δp ln J  θÞ dΩ  δW ext ¼ 0 ð36Þ
3 Ω
64 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70

where the enhanced Kirchhoff stress is given by 3.2.2. Tangent stiffness expression
T Due to the non-linearities, the resolution of a such problem
τ ¼F S F ð37Þ requires the use of an appropriate incremental/iterative procedure
Thus, the variation of the Lagrangian with respect to the displace- as the Newton–Raphson method. For instance, to build the tangent
ment field u, the strain variable θ and the pressure p can be matrix it is necessary to linearize the variation of the Lagrangian
respectively written, for any admissible virtual fields δΠ (36).
Z Using the fact that
δu Π ¼ ðτ D þ p1Þ : δL dΩ  δW ext ¼ 0 ð38Þ  
Ω 1
ΔF ¼ Δθ1 þ ΔLD F ð49Þ
Z 3
δp Π ¼ δpðlnðJÞ  θÞ dΩ ¼ 0 ð39Þ and after some algebraic manipulations, the linearization of (38)
Ω
may be written
 Z 
1 Z Z
δθ Π ¼ δθ 1 : τ p dΩ ¼ 0 ð40Þ
Δðδu Π Þ ¼  ðτ D þ p1Þ : ðΔLδLÞ dΩ þ
2
δL : τ D Δθ dΩ
Ω 3
Ω Ω 3
Z Z
Note that (40) represents the weak formulation of the pressure
þ τ : ½ðδLD ÞT ΔLD þ ΔLD δLD  dΩ þ Δp1 : δL dΩ
constitutive equation (p ¼ trðτ Þ=3) and (39), the one of the kine- Ω Ω
Z Z
matic relation between J and θ (20). 2
þ ΔθδL : D : 1 dΩ þ 2δL : ðD Þ : ΔL dΩ
D D D
ð50Þ
Ω3 Ω
3.2. Finite element discretization
where ΔL, Δθ and Δp denote incremental quantities and the
fourth-order tensor D is defined by
To solve these equations, the mixed finite element method is
used. The continuous spaces are approximated with discrete ∂S T T
spaces, in which the approximated solutions ðuh ; ph; θh Þ are D¼F F F F ð51Þ
∂C
searched [15]. Domain Ω is discretized by the mesh T h Ω . The
The tensors ðDD ÞD and DD are respectively defined by
three discretized fields can be written as
nbnodes dim ðDD ÞD ¼ I d DI d ; DD ¼ I d D ð52Þ
uh ¼ ∑ ∑ ei Nuk U ki ð41Þ
k¼1 i¼1 with
1
nbnodes I d ¼ I  1  1T ð53Þ
ph ¼ ∑ Npk P k ð42Þ 3
k¼1
and I the fourth rank unit tensor. Furthermore, it can be easily
nbnodes shown that the linearization of the two additional variational
θ
θh ¼ ∑ N k Θk ð43Þ equations (39) and (40) leads to
k¼1 Z
where dim is the space dimension, and ei is the ith vector of the Δðδp Π Þ ¼ δp½1 : ΔL  Δθ dΩ ð54Þ
Ω
canonical basis of the space Rdim . Nuk, Npk and N θk denote the shape
functions for respectively displacement, pressure and strain vari- Z
2
able fields, and U, P and Θ are the nodal variables. Δðδθ Π Þ ¼ δθΔθðtrðτ Þ þ1 : D : 1Þ dΩ
Ω9
The variation of the corresponding Eulerian gradient δLh is Z
2
computed from þ δθτ : ΔLD dΩ
Ω3
Z Z
nbnodes 2
δLhij ¼ ∑ δU ki Bkj ð44Þ þ δθð1 : DÞ : ΔLD dΩ  δθΔp dΩ ð55Þ
k¼1 Ω3 Ω

where B is the shape function gradient matrix defined on each Thus, the system of linear equations to be solved for the (iþ 1)th
element Ωe by equilibrium iteration of the (n þ 1)th time step is
0 uu 1ðn þ 1;iÞ 0 1 0 u 1ðn þ 1;iÞ
∂N iujΩe K K up K uθ ΔU ðn þ 1;i þ 1Þ R
Bij jΩe ¼ ð45Þ B pu C B C B C
∂xj @K K pp K pθ A @ ΔP A ¼ @ Rp A ð56Þ
K θu K θp K θθ ΔΘ Rθ

where the tangent terms are expressed as

3.2.1. Residual expression Z  
2
The discretization of (38)–(40) allows us to express the resi- K uu
inlm ¼  Bnj Bmk ðτ D
ik þ pδik Þδlj þ trðτ Þδlk δij dΩ
duals which are sums over elements. The (i; k)th component of the Ω 9
Z  
2 2
displacement residual, the kth component of the pressure and þ Bnj Bmk τ lj δki þ τ jk δli  τ lk δij  τ ij δlk dΩ
strain variable residuals can be written, respectively, on each Ω 3 3
Z
element Ωe þ 2 Bnj ðDijlk Þ Bmk dΩ
D D
ð57Þ
Z Ω
Ruik ¼ ðτ D
ij þ P m N m δij ÞBkj dΩe
p
ð46Þ Z
Ωe 2 θ
K uinrθ ¼ Nr ðBnj τ D
ij þ Bnj Dijkl δkl Þ dΩ
D
ð58Þ
Z Ω3
θ
Rpk ¼ N pk ðlnðJÞ  Θm Nm Þ dΩe ð47Þ Z
Ωe
K up
ins ¼ Nps Bnj δij dΩ ð59Þ
Z   Ω
1
Rθk ¼ N θk trðτ Þ  P m N pm dΩe ð48Þ Z  
Ωe 3 2 θ D
K θrinu ¼ N r ðτ ij þ δkl DD
klij ÞBnj dΩ ð60Þ
with δij being the Kronecker delta symbol. Ω 3
 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70 65

Z
θp ðn;iÞ
K rs ¼ N θr N ps dΩ ð61Þ deformation gradient F can be easily calculated from Eq. (21)
Ω
ðn;iÞ
!1=3
Z ðn;iÞ expðθ Þ
F ðn;iÞ ¼ 1 þ ∇x ðuðn;iÞ Þ; F ¼ F ðn;iÞ ð70Þ
K psrθ ¼  N θr N ps dΩ ð62Þ J ðn;iÞ
Ω

Z with J ðn;iÞ being the Jacobian of the transformation at iteration i.

K pu ¼ δ dΩ
N ps Bnj ij ð63Þ Furthermore, from the displacement increment Δuðn;iÞ , the
sin
Ω deformation gradient increment ΔF ðn;iÞ can be calculated, and
Z ðn;iÞ
the enhanced deformation gradient increment ΔF can be
K θθ
rt ¼ N θr N θt ðtrðτ Þ þ δij Dijkl δkl Þ dΩ ð64Þ deduced
Ω
 
ðn;iÞ 1 D ðn;iÞ
ΔF ðn;iÞ ¼ ∇x ðΔuðn;iÞ Þ; ΔF ¼ Δθðn;iÞ 1 þ ΔLðn;iÞ F ð71Þ
K pp ¼ 0 ð65Þ 3

with B being the shape function gradient matrix previously with ΔLðn;iÞ ¼ ΔFF  1 .
defined in (45). From the enhanced deformation gradient, an enhanced loga-
In the two-field formulation framework, the variation of the ðn;iÞ
rithmic strain tensor E is calculated as in Eq. (1), which
Lagrangian (36) with respect to the displacement field u and the
corresponds to the geometric pre-processing step (see Section 2.1)
pressure p is reduced to
Z ðn;iÞ 1 ðn;iÞ
E ¼ lnðC Þ ð72Þ
δu Π ¼ ðτ D þp1Þ : δL dΩ  δW ext ¼ 0 ð66Þ 2
Ω
ðn;iÞ
Z with C being the enhanced right Cauchy–Green strain tensor.
p ðn;iÞ
δp Π ¼ δp lnðJÞ  dΩ ¼ 0 ð67Þ In the same manner, its increment ΔE can be calculated from
Ω κ
ΔF ðn;iÞ .
with κ the bulk modulus and with The work-conjugated stress tensor T
ðn;iÞ
is defined as in (13)
e p
lnðJÞ ¼ trðEÞ ¼ trðE Þ þ trðE Þ ð68Þ ðn;iÞ ðn;iÞ 1
T ¼S : ðP L Þðn;iÞ ð73Þ
Eq. (67) is true only if, on one hand, Hooke's elastic relation is
with
assumed and if, on the other hand, lnðJÞ corresponds to the elastic
part of the strain ðn;iÞ
1 ∂E
ðP L Þðn;iÞ ¼ 2 ðn;iÞ
ð74Þ
lnðJÞ ¼ trðE e Þ 3 trðE p Þ ¼ 0 ð69Þ ∂C
1
which means that the plasticity behavior is volume-preserving. For details about the calculation of P L , it can be referred to [10].
Let us notice that the three-field formulation presented here The constitutive model with a small-strain structure is then
differs slightly from the one developed by Taylor [18,20]. The called to obtain, as in (15), the enhanced stress tensor at t n þ 1 ,
ðn þ 1;iÞ ep
difference is the choice of the relation enforced between the T , and the associated elastoplastic tangent moduli ðE Þðn þ 1;iÞ
Jacobian and the strain-like variable (20). Indeed Taylor chose to which is such as
impose J ¼ θ. It can also be reminded that the present work is
ðn þ 1;iÞ ðn þ 1;iÞ
T_ ¼ ðE Þðn þ 1;iÞ : E_
ep
developed in an industrial framework, and the main applications ð75Þ
are simulations of problems encountered in the power plants
ðn;iÞ
(repair of pipes, rolling expansion, etc.). The applications consid- For this, the enhanced logarithmic strain tensor E , its increment
ered are very sensitive and it is necessary to have accurate results. ðn;iÞ
ΔE , its plastic part, some hardening variables, and the
That is why this aspect takes priority over the computation time. ðn;iÞ
Thus, we need to consider elements with an interpolation for the enhanced stress tensor T are required. This corresponds to
pressure field which is at least linear. Furthermore, a quadratic the second step of the model described in Section 2.2.
interpolation for the displacement field is chosen in order to get a Using the geometric transformations described in Section 2.3,
ðn þ 1;iÞ
better convergence. the enhanced second Piola–Kirchhoff stress tensor S and the
ep ðn þ 1;iÞ
associated tangent modulus ðC L Þ which is such as
ðn þ 1;iÞ ep 1 ðn þ 1;iÞ
S ¼ ðC L Þðn þ 1;iÞ : C ð76Þ
4. Integration of the finite-strain model into the three-field 2
mixed finite element formulation are obtained at t n þ 1 . These variables can then be respectively
injected in (37) and (51), so that the tangent matrix and the
The goal of this work is to extend the finite-strain model residuals (46), (48), (47) can then be calculated at the ðiþ 1Þth equili-
described in the first section, to the three-field mixed finite brium iteration of the ðn þ 1Þth time step. The tangent system (56) can
element formulation, which is generic. The main interest of using
now be solved in order to get ðΔU; ΔP; ΔΘÞðn þ 1;i þ 1Þ .
this model is its capability to adopt the same framework as in
According to the Newton–Raphson method, the variables are
small strain theory, concerning the constitutive law implemen-
then updated
tation.
We consider a finite increment of time ½t n ; t n þ 1 , and assume ðU; P; ΘÞðn þ 1;i þ 1Þ ¼ ðU; P; ΘÞðn þ 1;iÞ þðΔU; ΔP; ΔΘÞTðn þ 1;i þ 1Þ ð77Þ
that all the variables are known at tn. For the ith equilibrium u p θ Tðn þ 1;iÞ
The convergence is then checked; if the residuals ðR ; R ; R Þ
iteration of the Newton–Raphson algorithm, the fields uðn;iÞ , pðn;iÞ
ðn;iÞ ðn;iÞ are small enough, ðU; P; ΘÞðn þ 1;i þ 1Þ is set to be ðU; P; ΘÞðn þ 1Þ . Else,
and θ and its increments Δu ðn;iÞ
, Δp ðn;iÞ
and Δθ are known new iterations are necessary until the convergence criterion is
ðn;iÞ
and both the deformation gradient F and the enhanced checked. A brief summary of the algorithm is presented in Table 1.
66 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70

5. Numerical applications problem is considered, as presented in [34]. Its extension to

finite-strain plasticity was proposed by Simo and Armero [35]
The formulation presented in previous sections is illustrated and has then been studied by other authors. A tapered panel is
below in some benchmark problems. Performance of the method clamped on one side and subjected to a uniform shear load of f ¼
in plastic incompressibility is tested considering 3D tetrahedral 0:3125 N=mm2 on the opposite side. The geometry, loading and
meshes. Firstly, numerical examples are presented in order to boundary conditions are given in Fig. 1(a). Plane strain conditions
compare the results of the developed elements with those of are assumed. We consider a J2-finite-strain model as described
displacement-based elements. Moreover, we use constitutive laws in [34]. The elastic part of the material constitutive law is governed
which allow us to compare results from the three-field formula- by a modified Neo-Hookean model, and its plastic part by non-
tion with those from the two-field formulation obtained by linear isotropic hardening rule with an associative flow rule based
eliminating the strain-like variable. An industrial application is on the Von Mises yield criterion, defined in terms of the yield
then presented, to evaluate the robustness of the method. Com- stress in uniaxial tension by
putations are performed with the industrial finite element code rffiffiffi
2
Code_Aster which is developed by EDF R&D [32]. sy ¼ 0:715  0:265 expð 16:93ep Þ þ 0:12924ep ð78Þ
3
5.1. Cook's plane strain problem with ep being the equivalent plastic strain.
The material parameters are those used in the study by Simo
This problem has been used by many authors [16,13,33] to test and Armero, that is a bulk modulus of κ ¼ 164:21 MPa, and a shear
finite element formulations under combined bending and shear. modulus of μ ¼ 60:1938 MPa. The quantity of interest is the
A nonlinear extension of the well-known Cook's plane strain
7
Table 1

Top right corner displacement uy(A) (mm)

Solution algorithm.
6
Time step n þ 1:
Initialize iteration i ¼ 0-ðU; P; ΘÞðn þ 1;0Þ ¼ ðU; P; ΘÞðnÞ 5
For the ith iteration:
- Geometric pre-processing:
ðn;iÞ ðn;iÞ ðn;iÞ ðn;iÞ
4
Compute F and ΔF and deduce E and ΔE
- Call the constitutive model with a small strain structure to get: P2P2P1
ðn þ 1;iÞ ep 3 P2P1P1
T and ðE Þðn þ 1;iÞ P2P1
ðn þ 1;iÞ ep P2
- Geometric post-processing: Recover S and ðC L Þðn þ 1;iÞ
P1
- Compute the tangent matrix and the residuals and solve the system: 2
0 uu 1ðn þ 1;iÞ 0 1ðn þ 1;i þ 1Þ 0 u 1ðn þ 1;iÞ
K K up K uθ ΔU R
B pu pθ C B C B C
@K K pp
K A @ ΔP A ¼ @ Rp A 1
K θu K θp K θθ ΔΘ Rθ
Update variables: 0
ðU; P; ΘÞðn þ 1;i þ 1Þ ¼ ðU; P; ΘÞðn þ 1;iÞ þ ðΔU; ΔP; ΔΘÞðn þ 1;i þ 1Þ 0 5 10 15 20 25 30
Check convergence Number of elements per side
If converged set: ðU; P; ΘÞðn þ 1Þ ¼ ðU; P; ΘÞðn þ 1;i þ 1Þ
If not converged: new iteration i’iþ 1 Fig. 2. Vertical displacement of top right corner versus number of elements
per edge.

Fig. 1. (a) Geometry, loading and boundary conditions for the plane strain Cook's membrane. (b) Structured mesh of 30 elements per side.
 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70 67

vertical displacement of the top right corner of the plate (point A) and strain variable fields, referred to as P2/P1/P1, and with
as presented in Fig. 1(a). Its convergence as a function of the quadratic interpolation for displacements and pressure fields and
number of elements per side, for structured mesh (Fig. 1(b)), is linear for the strain variable fields, referred to as P2/P2/P1. Were
usually used as a criterion in investigating performance. The also considered the two-field formulation, with quadratic ele-
results are shown in Fig. 2, for several finite element formulations. ments for displacements and linear for pressure, referred to as P2/
Were considered the proposed formulation with quadratic inter- P1, and linear and quadratic displacement-based elements.
polation for the displacement field and linear for both pressure We observe, as expected, that P1 element suffers from severe
locking, even for fine meshes. The other elements seem to
converge and to have the same behavior concerning the displace-
ment for this example.

5.2. Compression of a cube: discussion on the choice

of interpolation orders

The benchmark proposed here consists of a cube specimen as

represented in Fig. 3. The loading condition consists of a pre-
scribed normal stress of 1000 Pa in the vertical direction on the
plane fz ¼ 10g. A no-slip condition is considered on plane fz ¼ 0g.
On the plane y ¼ 0, dy ¼0 is imposed and on the plane x ¼0, dx¼0
is imposed. Young's modulus is E ¼200 GPa and the Poisson's ratio
is ν ¼ 0:3. A perfectly plastic model is assumed, considering a yield
stress sY ¼ 150 MPa. The distribution of the trace of the stress is
compared at Gauss points for P2/P1/P1, P2/P2/P1 elements, and for
linear and quadratic displacement-based elements in Fig. 4.
As expected, the displacement-based formulation [15], both
Fig. 3. Geometry of the cube. with linear (Fig. 4(a)) and quadratic elements (Fig. 4(b)), is not

Fig. 4. Distribution of the trace of the stress at Gauss points (MPa) with tetrahedral. (a) P1 elements, (b) P2 elements, (c) P2/P1/P1 elements, (d) P2/P2/P1 elements.

Fig. 5. Distribution of the Lagrangian multiplier p (MPa) at the nodes with (a) P2/P1/P1, (b) P2/P2/P1.
68 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70

stable and present spurious oscillations of the stress trace. A three- element does not fulfill the Ladyjenskaia–Brezzi–Babuska (LBB)
field approach makes these oscillations disappear, as shown in stability condition. Consequently, the P2/P2/P1 element which is
Fig. 4(c) and (d), for both P2/P1/P1 or P2/P2/P1 elements. based on it cannot be stable, as shown in Fig. 5(b). In the following,
However, further analysis, as observed in Fig. 5, shows that the this element will no longer be considered.
P2/P2/P1 formulation is not stable regarding the Lagrange multi-
plier field p. This instability can be explained by considering the 5.3. Plastic sphere under internal pressure
mixed finite element corresponding to the two-field formulation
obtained by condensing the strain variable in the three-field A sphere with inner radius r i ¼ 0:2 mm and outer radius
formulation. Indeed, this element is quadratic both in displace- r o ¼ 1 mm is subjected to an internal normal stress of 1 MPa
ment and pressure (P2/P2 element). It is well known [15] that this (Fig. 6). In order to reduce the computation cost, only a quarter
of the sphere has been modeled in an axisymmetrical frame. The
ux ( y) = 0 material considered is taken as elastoplastic perfectly plastic. The
plastic flow is associated with the Von Mises criterion, with a yield
stress sY ¼ 150 MPa. Young's modulus is taken as E ¼200 GPa and
the Poisson's ratio ν ¼ 0:3.
In the case where the entire sphere has reached a plastic state,
an analytical expression of the hydrostatic stress can be obtained.
Indeed, when the plastic zone reaches the outer side, the outer
radius of the deformed sphere can be written
  
sY 1 2
r od ¼ r o exp þ ð79Þ
3 2μ 3κ
with κ being the bulk modulus and μ the shear modulus.
It can be shown that in that case, the volume change may be
expressed as
u y (x) = 0 sffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi
 ffi
2 sY sY r
f ðrÞ ¼ þ 1  1  4 ln ð80Þ
3κ κ r od

with r being the distance from the center in the deformed

Fig. 6. Geometry and boundary conditions of the sphere.
configuration.
The trace of the Cauchy stress tensor r ¼ τ =J is then
trðrÞðrÞ ¼ 3κ f ðrÞ ð81Þ

10
L2-Error of hydrostatic stress

0.1 P2
P2P1
P2P1P1

0.01

0.001

100 1000 10000

Number of elements
Fig. 9. Distribution of the hydrostatic pressure at Gauss points with under-
Fig. 7. L2-error of hydrostatic pressure vs. number of elements. integrated elements.

Fig. 8. Geometry and mesh of the valve.

 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70 69

Fig. 10. Distribution with P2/P1/P1 elements of (a) the hydrostatic pressure at Gauss points and of (b) the Lagrangian multiplier p at the nodes.

Note that the magnitude of plastic deformations reaches 11%. The Lagrangian multiplier p (Fig. 10(b)) are very close, which guaran-
L2-error of hydrostatic pressure is computed for several structured tees the quality of the element; remind that in the three-field
triangle meshes with several number of elements. The coarser formulation presented above, the equality between these two
mesh has five elements per side. This mesh has been refined by quantities is weakly prescribed (40).
doubling the number of elements per side to get a mesh of ten
elements per side (50 elements) and so on until getting a mesh of
12,800 elements. 6. Conclusions
The L2-(absolute) error on the hydrostatic pressure is plotted in
Fig. 7 with respect to the number of elements, in order to evaluate A robust and efficient non-linear mixed finite element proce-
the convergence of the P2, P2/P1 and P2/P1/P1 elements. It can be dure for the numerical analysis of finite-strain elastoplasticity has
observed that the P2/P1/P1 element and the P2/P1 element give been presented. A finite-strain model based on the logarithmic
the same results, and when using these elements, the error is the strain tensor and its work-conjugate stress tensor has been
smallest. With the P2 element, the error is significant even with extended to a three-field formulation, which allows a generic
fine meshes. formulation to be obtained. The presented finite-strain model and
finite element formulation have the advantages of allowing an
5.4. Industrial case: assessment of constraints in the area of scour easy extension of constitutive laws from small-strain theory and
of an inlet valve are relatively easy to implement. It results in an extended finite-
strain model, which is robust, and able to take into account the
The goal of the present study is to estimate the lifetime in creep plastic incompressibility constraint intrinsically. The importance of
of an inlet valve of a thermal power station made of 15CDV-9-10 the choice of the interpolation order of the different fields has
steel submitted to a constant internal pressure and a temperature been shown through an example, and it was shown that the
ranging from 475 to 565 1C. This valve has a crack. In order to P2/P1/P1 element (quadratic interpolation for the displacement
prevent any propagation, the crack is removed by scouring a small field, linear interpolation for both pressure and strain variable
region around, leaving a smooth trench in the place of the crack. fields) is stable contrary to the P2/P2/P1 element (with a quadratic
The material is thermo-elasto-visco-plastic with isotropic hard- interpolation for the pressure field). Furthermore, the examples
ening and follows Lemaitre's law. Denoting E v the viscous portion presented above have illustrated the performance and the accu-
of the strain, it can be written racy of the proposed element, even in a case where the widely
3T D used under-integrated elements failed.
E_ v ¼ gðT eq ; λÞ ð82Þ
2 T eq
with T eq being the equivalent plastic stress, λ the cumulated References
viscous strain and T D the deviatoric part of the stress tensor.
In the case of Lemaitre's law, the plasticity criterion g may be [1] J. Argyris, M. Kleiber, Incremental formulation in nonlinear mechanics and
large strain elasto-plasticity natural approach. Part 1, Comput. Methods Appl.
expressed as
Mech. Eng. 11 (1977) 215–247.
 
1 seq n 1 1 [2] H. Hibbitt, P. Marcal, J. Rice, A finite element formulation for problems of large
gðreq ; λÞ ¼ ; with Z 0; Z 0; n 4 0 ð83Þ strain and large displacement, Int. J. Solids Struct. 6 (1970) 1069–1086.
K λ1=m K m [3] J.C. Nagtegaal, J.E. De Jong, Some computational aspects of elastic-plastic large
strain analysis, Int. J. Numer. Methods Eng. 17 (1981) 15–41.
In this application, these values have been chosen such as [4] J. Simo, On the computational significance of the intermediate configuration
n ¼5.9835 and 1=m ¼ 0. 1=K is expressed as a function of tem- and hyperelastic stress relations in finite deformation elastoplasticity, Mech.
perature, going from a value of 1.399  10  10 for a temperature of Mater. 4 (1985) 439–451.
[5] J. Simo, M. Ortiz, A unified approach to finite deformation plasticity based on
450 1C to 4.334  10  10 for a temperature of 570 1C. the use of hyperelastic constitutive equations, Comput. Methods Appl. Mech.
For further details about this law, we refer to [36]. Magnitude of Eng. 49 (1985) 241–245.
the plastic deformations reaches 20%. Fig. 8 represents the [6] J. Simo, A framework for finite strain elastoplasticity based on maximum
plastic dissipation and the multiplicative decomposition. Part I: continuum
geometry of the valve and the unstructured mesh of tetrahedra formulation, Comput. Methods Appl. Mech. Eng. 66 (1988) 199–219.
used in the computations (229,503 nodes for 151,528 elements), [7] J.C. Simo, A framework for finite strain elastoplasticity based on maximum
with the scour area surrounded. plastic dissipation and the multiplicative decomposition. Part II: computa-
tional aspects, Comput. Methods Appl. Mech. Eng. 68 (1988) 1–31.
At the scour area, there are some oscillations on the pressure
[8] A.L. Eterovic, K.-J. Bathe, A hyperelastic-based large strain elasto-plastic
values, due to incompressibility, even with under-integrated ele- constitutive formulation with combined isotropic-kinematic hardening using
ments [11], as shown in Fig. 9. Fig. 10 shows the results for the the logarithmic stress and strain measures, Int. J. Numer. Methods Eng. 30
P2/P1/P1 element. It can be observed that there is no oscillations (1990) 1099–1114.
[9] G. Weber, L. Anand, Finite deformation constitutive equations and a time
on the pressure values, and the distribution is smooth. Further- integrated procedure for isotropic hyperelastic viscoplastic solids, Comput.
more, the values of the hydrostatic pressure (Fig. 10(a)) and the Methods Appl. Mech. Eng. 79 (1990) 173–202.
70 D. Al Akhrass et al. / Finite Elements in Analysis and Design 86 (2014) 61–70

[10] C. Miehe, N. Apel, M. Lambrecht, Anisotropic additive plasticity in the [23] E. Lorentz, J. Besson, V. Cano, Numerical simulation of ductile fracture with the
logarithmic strain space: modular kinematic formulation and implementation rousselier constitutive law, Comput. Methods Appl. Mech. Eng. 197 (2008)
based on incremental minimization principles for standard materials, Comput. 1965–1982.
Methods Appl. Mech. Eng. 191 (2002) 5383–5425. [24] V. Tvergaard, A. Needleman, Analysis of the cup-cone fracture in a round
[11] D.S. Malkus, T.J. Hughes, Mixed finite element methods reduced and selective tensile bar, Acta Metall. 32 (1984) 157–169.
integration techniques: a unification of concepts, Comput. Methods Appl. [25] C. Miehe, M. Lambrecht, Algorithms for computation of stresses and elasticity
Mech. Eng. 15 (1978) 63–81. moduli in terms of Sethhill's family of generalized strain tensors, Commun.
[12] J.C. Simo, M.S. Rifai, A class of mixed assumed strain methods and the method Numer. Methods Eng. 17 (2001) 337–353.
of incompatible modes, Int. J. Numer. Methods Eng. 29 (1990) 1595–1638. [26] C. Miehe, A constitutive frame of elastoplasticity at large strains based on the
[13] E.P. Kasper, R.L. Taylor, A mixed-enhanced strain method. part I: geometrically notion of a plastic metric, Int. J. Solids Struct. 35 (1998) 3859–3897.
linear problems, Comput. Struct. 75 (2000) 237–250. [27] E. Lee, Some comments on elastic-plastic analysis, Int. J. Solids Struct. 17
[14] E.P. Kasper, R.L. Taylor, A mixed-enhanced strain method. Part II: geometrically (1981) 859–872.
nonlinear problems, Comput. Struct. 75 (2000) 251–260. [28] T. Sussman, K.-J. Bathe, A finite element formulation for nonlinear incom-
[15] F. Brezzi, M. Fortin, Mixed and Hybrid Finite Elements Methods, Springer
pressible elastic and inelastic analysis, Comput. Struct. 26 (1987) 357–409.
Series in Computational Mathematics, Springer-Verlag, New York, 1991.
[29] J. Bonet, R. Wood, Nonlinear Continuum Mechanics for Finite Element
[16] T. Elguedj, Y. Bazilevs, V. Calo, T. Hughes, B-bar and F-bar projection methods
Analysis, Cambridge University Press, Cambridge, 1997.
for nearly incompressible linear and non-linear elasticity and plasticity using
[30] R. Cisloiu, M. Lovell, J. Wang, A stabilized mixed formulation for finite strain
higher-order nurbs elements, Comput. Methods Appl. Mech. Eng. 197 (2008)
deformation for low-order tetrahedral solid elements, Finite Elem. Anal. Des.
2732–2762.
44 (2008) 472–482.
[17] T. Hughes, The Finite Element Method: Linear Static and Dynamic Finite
[31] A. Fortin, A. Garon, Les éléments finis: de la théorie à la pratique, cours de
Element Analysis, Prentice Hall International ed., Prentice-Hall, Englewood
Cliffs, New Jersey, 1987. l'universite de laval´ edition, 2010.
[18] O. Zienkiewicz, R. Taylor, The Finite Element Method for Solid and Structural [32] Code_Aster Open Source (GPL)—General FEA software 〈www.code-aster.org〉,
Mechanics, Elsevier, Butterworth-Heinemann, 2005. EDF, 1991–2013.
[19] J. Simo, R. Taylor, K. Pister, Variational and projection methods for the volume [33] B. Ramesh, A.M. Maniatty, Stabilized finite element formulation for elastic–
constraint in finite deformation elasto-plasticity, Comput. Methods Appl. plastic finite deformations, Comput. Methods Appl. Mech. Eng. 194 (2005)
Mech. Eng. 51 (1985) 177–208. 775–800.
[20] R.L. Taylor, A mixed-enhanced formulation tetrahedral finite elements, Int. J. [34] M.K. Magne, O.K. Morten, K. Trond, R.S. Bente, Isogeometric analysis of finite
Numer. Methods Eng. 47 (2000) 205–227. deformation nearly incompressible solids, J. Struct. Mech. 44 (2011) 260–278.
[21] M. Brunig, Formulation and numerical treatment of incompressibility con- [35] J.C. Simo, F. Armero, Geometrically non-linear enhanced strain mixed methods
straints in large strain elastic–plastic analysis, Int. J. Numer. Methods Eng. 45 and the method of incompatible modes, Int. J. Numer. Methods Eng. 33 (1992)
(1999) 1047–1068. 1413–1449.
[22] O.C. Zienkiewicz, R.L. Taylor, J.Z. Zhu, The Finite Element Method: Its Basis and [36] J. Lemaitre, J.-L. Chaboche, Mechanics of Solid Materials, Cambridge University
Fundamentals, Sixth ed., Butterworth-Heinemann, Oxford, 2005. Press, Cambridge, 1994.