an improved, fully symmetric, finite-strain phenomenological constitutive model for shape memory...
TRANSCRIPT
Finite Elements in Analysis and Design 47 (2011) 166–174
Contents lists available at ScienceDirect
Finite Elements in Analysis and Design
0168-87
doi:10.1
� Corr
E-m1 Cu
Univers
journal homepage: www.elsevier.com/locate/finel
An improved, fully symmetric, finite-strain phenomenological constitutivemodel for shape memory alloys
J. Arghavani a,�,1, F. Auricchio b,c,d, R. Naghdabadi a,e, A. Reali b,d
a Department of Mechanical Engineering, Sharif University of Technology, 11155-9567 Tehran, Iranb Dipartimento di Meccanica Strutturale, Universit �a degli Studi di Pavia, Italyc Center for Advanced Numerical Simulation (CeSNA), IUSS, Pavia, Italyd European Centre for Training and Research in Earthquake Engineering (EUCENTRE), Pavia, Italye Institute for Nano-Science and Technology, Sharif University of Technology, 11155-9567 Tehran, Iran
a r t i c l e i n f o
Article history:
Received 31 May 2010
Accepted 6 September 2010Available online 28 September 2010
Keywords:
Shape memory alloys
Finite strain
Solution algorithm
Finite element
Exponential mapping
UMAT
4X/$ - see front matter & 2010 Elsevier B.V.
016/j.finel.2010.09.001
esponding author. Tel.: +98 21 6616 5511; fa
ail address: [email protected] (J. Argh
rrently visiting Ph.D. student, Dipartiment
it�a degli Studi di Pavia, Italy.
a b s t r a c t
The ever increasing number of shape memory alloy applications has motivated the development of
appropriate constitutive models taking into account large rotations and moderate or finite strains. Up to
now proposed finite-strain constitutive models usually contain an asymmetric tensor in the definition
of the limit (yield) function. To this end, in the present work, we propose an improved alternative
constitutive model in which all quantities are symmetric. To conserve the volume during inelastic
deformation, an exponential mapping is used to arrive at the time-discrete form of the evolution
equation. Such a symmetric model simplifies the constitutive relations and as a result of less
nonlinearity in the equations to be solved, numerical efficiency increases. Implementing the proposed
constitutive model within a user-defined subroutine UMAT in the nonlinear finite element software
ABAQUS/Standard, we solve different boundary value problems. Comparing the solution CPU times for
symmetric and asymmetric cases, we show the effectiveness of the proposed constitutive model as well
as of the solution algorithm. The presented procedure can also be used for other finite-strain
constitutive models in plasticity and shape memory alloy constitutive modeling.
& 2010 Elsevier B.V. All rights reserved.
1. Introduction
Shape memory alloys (SMAs) are unique materials with theability to undergo large deformations regaining the original shapeeither during unloading (superelastic or pseudo-elastic (PE)effect) or through a thermal cycle (shape-memory effect (SME))[1,2]. Since such effects are in general not present in standardalloys, SMAs are often used in innovative applications. Forexample, nowadays pseudo-elastic Nitinol is a common andwell-known engineering material in the medical industry [3,4].
The origin of SMA features is a reversible thermo-elasticmartensitic phase transformation between a high symmetry,austenitic phase and a low symmetry, martensitic phase.Austenite is a solid phase, usually characterized by a body-centered cubic crystallographic structure, which transforms intomartensite by means of a lattice shearing mechanism. When thetransformation is driven by a temperature decrease, martensite
All rights reserved.
x: +98 21 6600 0021.
avani).
o di Meccanica Strutturale,
variants compensate each other, resulting in no macroscopicdeformation. However, when the transformation is driven by theapplication of a load, specific martensite variants, favorable to theapplied stress, are preferentially formed, exhibiting a macroscopicshape change in the direction of the applied stress. Uponunloading or heating, this shape change disappears through thereversible conversion of the martensite variants into the parentphase [1,5].
For a stress-free SMA material, four characteristic tempera-tures can be identified, defined as the starting and finishingtemperatures during forward transformation (austenite to mar-tensite), Ms and Mf, and as the starting and finishing temperaturesduring reverse transformation, As and Af. Accordingly, in a stress-free condition, at a temperature above Af, only the austeniticphase is stable, while at a temperature below Mf, only themartensitic phase is stable. As a consequence, applying a stress ata temperature above Af, SMAs exhibit a pseudo-elastic behaviorwith a full recovery of inelastic strain upon unloading, while at atemperature below Ms, the material presents the shape-memoryeffect with permanent inelastic strains upon unloading whichmay be recovered by heating.
In most applications, SMAs experience a general thermo-mechanical loading conditions more complicated than uniaxial or
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174 167
multiaxial proportional loadings, typically undergoing very largerotations and moderate strains (i.e., in the range of 10% forpolycrystals [1]). For example, with reference to biomedicalapplications, stent structures are usually designed to significantlyreduce their diameter during the insertion into a catheter;thereby, large rotations combined with moderate strains occurand the use of a finite deformation scheme is preferred.
The majority of the currently available 3D macroscopicconstitutive models for SMAs has been developed in the smalldeformation regime (see, e.g., [6–16] among others). Finitedeformation SMA constitutive models proposed in the literaturehave been mainly developed by extending small strain models.The approach in most cases is based on the multiplicativedecomposition of the deformation gradient into an elasticand an inelastic or transformation part [17–25], thoughthere are some models which have utilized an additive decom-position of the strain rate tensor into an elastic and an inelasticpart [26].
In the present work we focus on a finite-strain extension of thesmall-strain constitutive model initially proposed by Souza et al.[8] and extensively studied in Refs. [27–29]. We first develop afinite-strain constitutive model containing an asymmetric tensor,also observed in the constitutive equations of [20–24]. To thisend, we propose an improved alternative constitutive modelwhich is expressed in terms of symmetric tensors only. We thenimplement the proposed model in a user-defined subroutine(UMAT) in the nonlinear finite element software ABAQUS/Standard and compare the solution CPU times for differentboundary value problems. The results show the increasedcomputational efficiency (in terms of solution CPU time), whenthe proposed alternative symmetric form is used. This is mostlydue to the simplification in computing fourth-order tensorsappearing when a tensorial equation is linearized.
The structure of the paper is as follows. In Section 2, based on amultiplicative decomposition of the deformation gradient intoelastic and transformation parts, we present the time-continuousfinite-strain constitutive model. In Section 3, we propose analternative constitutive equation which includes only symmetrictensors. In Section 4, based on an exponential mapping, the time-discrete form and the solution algorithm are discussed. In Section5, implementing the proposed integration algorithm within thecommercial nonlinear finite element software ABAQUS/Standard,we simulate different boundary value problems. We finally drawconclusions in Section 6.
2. A 3D finite-strain SMA constitutive model: time-continuousframe
We use a multiplicative decomposition of the deformationgradient and present a thermodynamically consistent finite strainconstitutive model as done by Reese and Christ [20,21],Evangelista et al. [22] and, more recently, by Arghavani et al.[23,24]. The finite-strain constitutive model takes its origin fromthe small-strain constitutive model proposed by Souza et al. [8]and improved and discussed by Auricchio and Petrini [27–29].
2.1. Constitutive model development
Considering a deformable body, we denote with F thedeformation gradient and with J its determinant, supposed to bepositive. The tensor F can be uniquely decomposed as
F ¼ RU ¼VR ð1Þ
where U and V are the right and left stretch tensors, respectively,both positive definite and symmetric, while R is a proper
orthogonal rotation tensor. The right and left Cauchy–Greendeformation tensors are then, respectively, defined as
C ¼ FT F , b¼ FFTð2Þ
and the Green–Lagrange strain tensor, E, reads as
E¼C�1
2ð3Þ
where 1 is the second-order identity tensor. Moreover, thevelocity gradient tensor l is given as
l¼ _F F�1ð4Þ
The symmetric and anti-symmetric parts of l supply the strainrate tensor d and the vorticity tensor w, i.e.,
d¼ 12 ðlþ lT
Þ, w¼ 12ðl�lT
Þ ð5Þ
Taking the time derivative of Eq. (3) and using (4) and (5), it canbe shown that
_E ¼ FT dF ð6Þ
Following a well-established approach adopted in plasticity[30,31] and already used for SMAs [17–24], we assume a localmultiplicative decomposition of the deformation gradient into anelastic part Fe, defined with respect to an intermediate config-uration, and a transformation one Ft , defined with respect to thereference configuration. Accordingly,
F ¼ FeFtð7Þ
Since experimental evidences indicate that the transformationflow is nearly isochoric, we impose detðFt
Þ ¼ 1, which after takingthe time derivative results in
trðdtÞ ¼ 0 ð8Þ
We define Ce¼ FeT
Fe and Ct¼ FtT
Ft as the elastic and thetransformation right Cauchy–Green deformation tensors, respec-tively, and using definitions (2) and (7), we obtain
Ce¼ Ft�T
CFt�1
ð9Þ
To satisfy the principle of material objectivity, the Helmholtz freeenergy has to depend on Fe only through the elastic right Cauchy–Green deformation tensor; it is moreover assumed to be afunction of the transformation right Cauchy–Green deformationtensor and of the temperature, T, in the following form [22–24]
C¼CðCe,Ct ,TÞ ¼ceðCeÞþct
ðEt ,TÞ ð10Þ
where ceðCeÞ is a hyperelastic strain energy function and
Et¼ ðCt
�1Þ=2 is the transformation strain. We remark that inproposing decomposition (10), we have assumed the samematerial behavior for the austenite and martensite phases. Inaddition, we assume ce
ðCeÞ to be an isotropic function of Ce; it can
be therefore expressed as
ceðCeÞ ¼ce
ðICe ,IICe ,IIICe Þ ð11Þ
where ICe ,IICe ,IIICe are the invariants of Ce. We also define ct in thefollowing form [8,22,27–29]
ctðEt ,TÞ ¼ tMðTÞJEtJþ1
2hJEtJ2þI eLðJEtJÞ ð12Þ
where tMðTÞ ¼ b/T�T0S and b, T0 and h are material parameters;the MacCauley brackets calculate the positive part of theargument, i.e., /xS¼ ðxþjxjÞ=2, and the norm operator is definedas JAJ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiA : AT
p, with A : B¼ AijBij.
Moreover, in Eq. (12) we also use the indicator function I eL
defined as
I eLðJEtJÞ ¼
0 if JEtJreL
þ1 otherwise
(ð13Þ
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174168
to satisfy the constraint on the transformation strain norm(i.e., JEtJreL ). The material parameter eL is the maximumtransformation strain norm in a uniaxial test.
Taking the time derivative of Eq. (9) and using (4), the materialtime derivative of the elastic right Cauchy–Green deformationtensor is obtained as
_Ce¼�ltT
CeþF t�T _CF t�1
�Celtð14Þ
We now use the Clausius–Duhem inequality form of the secondlaw of thermodynamics
S : 12_C�ð _CþZ _T ÞZ0 ð15Þ
where S is the second Piola–Kirchhoff stress tensor related to theCauchy stress r through
S ¼ JF�1rF�Tð16Þ
and Z is the entropy. Substituting (10) into (15), we obtain
S :1
2_C�
@ce
@Ce :_C
eþ@ct
@Et :_E
tþ@ct
@T_T
!�Z _T Z0 ð17Þ
Now, substituting (14) into (17), after some mathematicalmanipulations, we obtain
S�2Ft�1 @ce
@Ce Ft�T
� �:
1
2_Cþ2Ce @c
e
@Ce : lt�X : _E
t� Zþ @c
t
@T
!_T Z0
ð18Þ
where
X ¼ hEtþðtMðTÞþgÞN ð19Þ
and
N ¼Et
JEtJð20Þ
The positive variable g results from the indicator functionsubdifferential @I eL
ðJEtJÞ and it is such that
gZ0 if JEtJ¼ eL
g¼ 0 otherwise
(ð21Þ
In deriving (18) we have used the isotropic property of ceðCeÞ,
implying that Ce and @ce=@Ce are coaxial. Following standardarguments, we finally conclude that
S ¼ 2Ft�1@ce
@Ce Ft�T
Z¼�@ct
@T
8>>><>>>:
ð22Þ
Using a property of second-order tensor double contraction, i.e.,A : ðBCÞ ¼ B : ðACT
Þ ¼ C : ðBT AÞ, we obtain
X : _Et¼ ðF tXFtT
Þ : dtð23Þ
In deriving (23) we have used the following relation:
_Et¼ FtT
dtFtð24Þ
Substituting (22) and (23) into (18) and taking advantage of thesymmetry of Ce
ð@ce=@CeÞ, the dissipation inequality can be
written as
ðP�KÞ : dtZ0 ð25Þ
where
P ¼ 2Ce@ce
@Ce
K ¼ FtXF tT
8>>><>>>:
ð26Þ
To satisfy the second law of thermodynamics, (25), we define thefollowing evolution equation:
dt¼ _zðP�KÞD
JðP�KÞDJð27Þ
where we denote with a superscript D the deviatoric part of atensor. Moreover, we highlight that (27) satisfies the inelasticdeformation incompressibility condition (8).
To describe phase transformation, we choose the followinglimit function:
f ¼ JðP�KÞDJ�R ð28Þ
where the material parameter R is the elastic region radius.Similarly to plasticity, we also introduce the consistencyparameter and the limit function satisfying the Kuhn–Tuckerconditions
f r0, _zZ0, _zf ¼ 0 ð29Þ
2.2. Representation with respect to the reference configuration
In the previous section, using a multiplicative decomposition ofthe deformation gradient into elastic and transformation parts, wederived a finite-strain constitutive model. However, while F and Ft
are defined with respect to the reference configuration, Fe is definedwith respect to an intermediate configuration. It is then necessary torecast all equations in terms of C (control variable) and Ct (internalvariable) which are described with respect to the reference config-uration, allowing to express all equations in a Lagrangian form.
Since a hyperelastic strain energy function ce has beenintroduced in terms of Ce (i.e., with respect to the intermediateconfiguration), we first investigate it.
As introduced in (11), ce depends on Ce only through itsinvariants, which are equal to those of CCt�1
. In fact, for the firstinvariant the following identity holds
ICe ¼ trðCeÞ ¼ trðFt�T
CFt�1
Þ ¼ tr ðCFt�1
Ft�T
Þ
¼ trðCCt�1
Þ ¼ ICCt�1 ð30Þ
and similar relations hold for the second and the third invariants.Accordingly, we may write
@ce
@Ce ¼ a11þa2Ceþa3Ce2
ð31Þ
where ai ¼ aiðICCtr�1 ,IICCt�1 ,III
CCt�1 Þ. Substituting (31) into (22)1 andusing (9), we conclude that
S ¼ 2ða1Ct�1
þa2Ct�1
CCt�1
þa3Ct�1
ðCCt�1
Þ2Þ ð32Þ
which expresses the second Piola–Kirchhoff stress tensor interms of quantities computed with respect to the referenceconfiguration.
In order to find the Lagrangian form of the evolution equation,we substitute (27) into (24) and obtain
_Et¼ _zFtT ðP�KÞD
JðP�KÞDJFt
ð33Þ
or equivalently,
_Ct¼ 2 _zFtT ðP�KÞD
JðP�KÞDJFt
ð34Þ
Focusing on the first term of the right hand side of Eq. (34), wemay observe that
FtT
PFt¼ ðFtCeFt
Þ 2Ft�1 @ce
@Ce F t�T
� �Ct¼ CSCt
ð35Þ
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174 169
while, focusing on the second term of the right hand side ofEq. (34), we may observe that
FtT
KFt¼ FtT
FtXFtT
Ft¼ CtXCt
ð36Þ
We may then conclude that
FtT
ðP�KÞFt¼ YCt
ð37Þ
and obtain from (35)–(37)
Y ¼ CS�CtX ð38Þ
Moreover, we have
FtT
ðP�KÞDFt¼ FtT
ðP�K�13trðP�KÞ1ÞFt
¼ ðCSCt�CtXCt
�13 trðCSÞCt
þ13trðCtXÞCt
Þ ¼ YDCtð39Þ
yielding
ðP�KÞD ¼ Ft�T
YDFtT
ð40Þ
and
JðP�KÞDJ¼ JYDJ ð41Þ
Now, we substitute (40) and (41) into (28) and (33) to obtain
f ¼ JYDJ�R ð42Þ
and
_Ct¼ 2 _z
YD
JYDJCt
ð43Þ
Table 1 presents the time-continuous finite-strain constitutivemodel expressed only in terms of Lagrangian quantities, whichhas already been presented in [22,24].
3. An alternative form of the constitutive model in terms ofsymmetric quantities
We observe that due to the asymmetry of CS, the quantity Y isnot symmetric. The asymmetric tensor Y also appears in theconstitutive equations proposed in [20–24]. To this end, we
Table 1Finite-strain constitutive model in the time-continuous frame: original form
[22,24].
External variables: C ,T
Internal variable: Ct
Stress quantities:
S ¼ 2a1Ct�1þ2a2Ct�1
CCt�1
þ2a3Ct�1ðCCt�1
Þ2
Y ¼ CS�CtX
X ¼ hEtþðtMþgÞN
with
gZ0 ifJEtJ¼ eL
g¼ 0 otherwise
(
and
N ¼ Et
JEtJ
where Et¼ ðCt
�1Þ=2
Evolution equation:
_Ct¼ 2 _z YD
JYDJCt
Limit function:
f ¼ JYDJ�R
Kuhn–Tucker conditions:
f r0, _zZ0, _zf ¼ 0
Norm operator:
JYDJ¼ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiYD : YDT
p
present an alternative formulation which is in terms of symmetrictensors only. According to (32), we may write
CS ¼ 2ða1CCt�1
þa2ðCCt�1
Þ2þa3ðCCt�1
Þ3Þ ð44Þ
Eq. (44) shows that the asymmetry in CS is due to the asymmetricterm CCt�1. We now present the following identity:
CCt�1
¼UtðUt�1
CUt�1
ÞUt�1
¼UtCUt�1
ð45Þ
where
C ¼Ut�1
CUt�1
ð46Þ
Substituting (45) into (44), we obtain
CS ¼UtC SUt�1
ð47Þ
where
S ¼ 2ða11þa2Cþa3C2Þ ð48Þ
Moreover, due to the coaxiality of Ct and X, we may write
CtX ¼UtCtXUt�1
ð49Þ
Substituting (47) and (49) into (38), we obtain
Y ¼UtQUt�1
ð50Þ
where Q is defined as
Q ¼ C S�CtX ð51Þ
Using the property trðUtQUt�1
Þ ¼ trðQ Þ, we also obtain
YD¼UtQ DUt�1
ð52Þ
We now substitute (52) into (42) and (43) and obtain (53) and(54), respectively, as follows:
YD : YDT
¼Q D : Q D or f ¼ JYDJ�R¼ JQ DJ�R ð53Þ
and
_Ct¼ 2 _zUt Q D
JQ DJUt
ð54Þ
We remark that in the proposed formulation we can express S as
S ¼Ut�1
SUt�1
ð55Þ
Finally, we summarize the proposed improved time-continuousfinite-strain constitutive models in Table 2.
We may compare the original and the proposed alternativetime-continuous forms summarized in Tables 1 and 2, respec-tively. It is clearly observed that in the proposed alternativeconstitutive model, all quantities are symmetric and the normoperator argument is also a symmetric tensor.
Remark 1. We note that according to (20), the variable N is notdefined for the case of vanishing transformation strain. To thisend, we use a regularization scheme proposed by Helm and Haupt[9] and also used in [7,20,21] to overcome this problem:
JEtJ ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiJEtJ2
þdq
ð56Þ
where d is a user-defined parameter (typical value: 10�7).
4. A 3D finite-strain SMA constitutive model: time-discreteframe
In this section we investigate the numerical solution of theconstitutive model derived in Section 3 and summarized inTable 2, with the final goal of using it within a finite elementprogram. The main task is to apply an appropriate numericaltime-integration scheme to the evolution equation of the internalvariable, considering that, in general, implicit schemes are
Table 2Finite-strain constitutive model in the time-continuous frame: proposed alter-
native form.
External variables: C,T
Internal variable: Ct
Stress quantities:
S ¼Ut�1
SUt�1
S ¼ 2ða11þa2C þa3C2Þ
Q ¼ C S�CtX
X ¼ hEtþðtMþgÞN
with
gZ0 if JEtJ¼ eL
g¼ 0 otherwise
(
and
N ¼ Et
JEtJ,C ¼Ut�1
CUt�1
where Et¼ ðCt
�1Þ=2
Evolution equation:
_Ct¼ 2 _zUt Q D
JQ DJUt
Limit function:
f ¼ JQ DJ�R
Kuhn–Tucker conditions:
f r0, _zZ0, _zf ¼ 0
Norm operator:
JQ DJ¼
ffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiQ D : Q D
q
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174170
preferred because of their stability at larger time step sizes.Moreover, the present section provides some details about thestress update and the computation of the consistent tangentmatrix, which are the two points where the material model isdirectly connected to the finite element solution procedure.
We treat the nonlinear problem described in Section 3 as animplicit time-discrete deformation-driven problem. Accordingly,we subdivide the time interval of interest [0, t] in sub-incrementsand we solve the evolution problem over the generic interval[tn,tn +1] with tnþ14tn. To simplify the notation, we indicate withthe subscript n a quantity evaluated at time tn, and with nosubscript a quantity evaluated at time tn +1.Assuming to know thesolution and the deformation gradient Fn at time tn as well as thedeformation gradient F at time tn +1, the stress and the internalvariable have to be updated from the deformation history.
4.1. Time integration
Exponential-based integration schemes are frequently appliedto problems in plasticity and isotropic inelasticity [32,33], sincethe use of the exponential mapping allows to exactly conserve thevolume during an inelastic deformation. Thus, larger time stepsizes than any other first-order accurate integration scheme maybe used.
We now consider an evolution equation in the following form:
_Ct¼ _zA¼ _zA1Ct
ð57Þ
where A1 ¼ACt�1
with A being a symmetric tensor, while A1 is notsymmetric in general.
Applying the exponential mapping scheme to the evolutionequation (57), we obtain
Ct¼ expðDzA1ÞC
tn ð58Þ
In the equation above, we need to compute the exponential of anasymmetric tensor, which is a problematic computation due tothe impossibility of using a spectral decomposition method.2
2 The exponent of an asymmetric tensor is computed with a series expansion.
However, the exponent of a symmetric tensor can also be computed in closed form
by means of a spectral decomposition.
Accordingly, using the strategy initially proposed in [20–22,34]and also exploited in [35,24], we can find an alternativeexpression where the argument of the exponential operator is asymmetric tensor which can be computed through spectraldecomposition. Following [20], we can write
expðDzA1Þ ¼ 1þDzA1þDz2
2!A2
1þ � � �
¼ 1þDzACt�1
þDz2
2!ðACt�1
Þ2þ � � �
¼Ut 1þDzUt�1
AUt�1
þDz2
2!ðUt�1
AUt�1
Þ2þ � � �
!Ut�1
¼UtexpðDzUt�1
AUt�1
ÞUt�1
ð59Þ
We now right- and left-multiply (58) by Ct�1
n and Ct�1
, respec-tively, to obtain
Ct�1
n ¼ Ct�1
expðDzA1Þ ð60Þ
Then, substituting (59) into (60), we can write the integrationformula as
�Ct�1
n þUt�1
expðDzUt�1AUt�1
ÞUt�1
¼ 0 ð61Þ
which is the time-discrete form associated to (57) presented so farin the literature [20–22,34,35,24].
Comparing (54) and (57), we conclude that
A¼ 2Ut Q D
JQ DJUt
ð62Þ
We now substitute (62) into (61) and after right- and left-multiplying by Ut , we obtain the time-discrete form of theproposed evolution Eq. (54) as
�UtCt�1
n Utþexpð2DzWÞ ¼ 0 ð63Þ
where
W ¼Q D
JQ DJð64Þ
4.2. Solution algorithm
As usual in computational inelasticity problems, to solve thetime-discrete constitutive model we use an elastic predictor-inelastic corrector procedure. The algorithm consists of evaluatingan elastic trial state, in which the internal variable remainsconstant, and of verifying the admissibility of the trial function. Ifthe trial state is admissible, the step is elastic; otherwise, the stepis inelastic and the transformation internal variable has to beupdated through integration of the evolution equation.
In order to solve the inelastic step, we use another predictor-corrector scheme, that is, we assume g¼ 0 (i.e., we predict anunsaturated transformation strain case with JEtJreL) and wesolve the following system of nonlinear equations (we refer to thissystem as PT 1 system):
Rt¼�UtCt�1
n Utþexpð2DzWÞ ¼ 0
Rz ¼ JQ DJ�R¼ 0
(ð65Þ
If the solution is not admissible (i.e., JEtJPT14eL), we assume gZ0(i.e., we consider a saturated transformation case with JEtJ¼ eL)and we solve the following system of nonlinear equations (werefer to this system as PT2 system):
Rt¼�UtCt�1
n Utþexpð2DzWÞ ¼ 0
Rz ¼ JQ DJ�R¼ 0
Rg ¼ JEtJ�eL ¼ 0
8>><>>: ð66Þ
Table 3Solution algorithm for the proposed improved finite-strain constitutive model.
1. compute C ¼ FT F2. set UtTR
¼Utn to compute fTR and trial solution
3. if ðf TR o0Þ then
Accept trial solution, i.e., Ut¼UtTR
elseSolve PT1 system.
if ðJEtJPT1 oeLÞ then
Accept PT1 solution, i.e., Ut¼ ðUt
ÞPT1
else
Solve PT2 system and set Ut¼ ðUt
ÞPT2
end ifend if
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174 171
The solution of Eqs. (65) and (66) is, in general, approachedthrough a straightforward Newton-Raphson method.
Remark 2. The corresponding PT 1 and PT 2 systems for theoriginal form of the constitutive model are, respectively, asfollows [22,24]:
Rt¼�Ct�1
n þUt�1expðDzUt�1
AUt�1
ÞUt�1
¼ 0
Rz ¼ JYDJ�R¼ 0
(ð67Þ
Rt¼�Ct�1
n þUt�1
expðDzUt�1
AUt�1
ÞUt�1
¼ 0
Rz ¼ JYDJ�R¼ 0
Rg ¼ JEtJ�eL ¼ 0
8>><>>: ð68Þ
where
A¼ 2YD
JYDJCt
4.3. Consistent tangent matrix
We now linearize the nonlinear equations as it is required forthe iterative Newton-Raphson method. For brevity, we report theconstruction of the tangent matrix only for the case of thesaturated phase transformation, corresponding to (66). Lineariz-ing (66), we obtain
RtþRt
,Ut : dUtþRt
,Dz dDzþRt,g dg¼ 0
RzþRz,Ut : dUt
þRz,Dz dDzþRz
,g dg¼ 0
RgþRg,Ut : dUt
þRg,Dz dDzþRg
,g dg¼ 0
8>>><>>>:
ð69Þ
where subscripts following a comma indicate differentiation withrespect to that quantity. The derivatives appearing in the aboveequations are detailed in [36].
Utilizing the linearized form (69), after converting it to matrixform, a system of eight nonlinear scalar equations is obtained thatis solved for dUt , dDz and dg.
We now address the construction of the tangent tensorconsistent with the time-discrete constitutive model. The use ofa consistent tensor preserves the quadratic convergence of theNewton–Raphson method for the incremental solution of theglobal time-discrete problem, as it is done in the framework of afinite element scheme.
The consistent tangent is computed by linearizing the secondPiola–Kirchhof tensor, i.e.,
dS ¼D : dE¼D : 12dC ð70Þ
Recalling that S is a function of C and Ct , we can write
dS ¼@S
@C: dCþ
@S
@Ut : dUtð71Þ
Recasting (71) into a matrix form, we obtain
½dS� ¼@S
@C
� �½dC�þ
@S
@Ut
� �½dUt� ð72Þ
where we have used ½�� to denote the matrix form of the tensorialargument.
We now consider Eq. (66) as a function of C,Ut ,Dz and g, andthen the corresponding linearization gives
Rt,C : dCþRt
,Ut : dUtþRt
,Dz dDzþRt,g dg¼ 0
Rz,C : dCþRz
,Ut : dUtþRz
,Dz dDzþRz,g dg¼ 0
Rg,C : dCþRg
,Ut : dUtþRg
,Dz dDzþRg,g dg¼ 0
8>>><>>>:
ð73Þ
Recasting (73) into a matrix form, we obtain
dUt
dzdg
264
375¼�
Rt,Ut Rt
,Dz Rt,g
Rz,Ut Rz
,Dz Rz,g
Rg,Ut Rg
,Dz Rg,g
26664
37775�1
Rt,C
Rz,C
Rg,C
26664
37775½dC� ð74Þ
Now, using (74), we can compute the matrix ½B� such that
½dUt� ¼ ½B�½dC� ð75Þ
We then substitute (75) into (72) to obtain the consistent tangentmatrix as
½D� ¼ 2@S
@C
� �þ2
@S
@Ut
� �½B� ð76Þ
Finally, Table 3 presents the solution algorithm for the proposedimproved, fully symmetric, finite-strain constitutive model.
Remark 3. We highlight that in [22,24] and similarly in [20], Ct ,Dz and g are considered as unknowns in (68), while in this workwe assume Ut , Dz and g as unknowns in (66) as it has alreadybeen done in a similar works [21,34].
Remark 4. Up to now, all relations have been derived in acompletely general manner without specifying the form of theHelmholtz free energy ce, apart from the fact that it is an isotropicfunction of Ce. Despite the hyperelastic strain energy function ce
can take any well-known form in finite elasticity, for thenumerical examples discussed in Section 5 we use the commonlyused Saint–Venant Kirchhoff strain energy function:
ce¼
l2ðtr Ee
Þ2þm tr Ee2
ð77Þ
which yields
a1 ¼l4ðC : 1�3Þ�
1
2m, a2 ¼
1
2m, a3 ¼ 0 ð78Þ
where l and m are the Lam�e constants.
5. Boundary value problems
In this section, we solve four boundary value problems tovalidate the adopted fully symmetric model as well as theproposed integration algorithm and the solution procedure. Ahelical spring and a medical stent are simulated at two differenttemperatures to show the model capability of capturing bothpseudo-elasticity and shape-memory effect. We have recentlyused these boundary value problems [24] as benchmarks toinvestigate the efficiency and robustness of different integrationschemes. Although the results presented here are the samepresented in [24], in this work we are interested in comparing theCPU times for both the original asymmetric and the proposedsymmetric constitutive models. In all boundary value examples
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174172
we use the following material properties:
E¼ 51 700 MPa, n¼ 0:3, h¼ 750 MPa,
eL ¼ 7:5%, b¼ 5:6 MPa 3C�1
T0 ¼�25 3C, R¼ 140 MPa, Af ¼ 0 3C, Mf ¼�25 3C
The temperature in the pseudo-elastic simulations is set to 37 1C,while in the shape-memory effect simulations a temperature of�25 1C is adopted.
For all simulations, we use the commercial nonlinear finiteelement software ABAQUS/Standard, implementing the describedalgorithm within a user-defined subroutine UMAT. In order todemonstrate the 1D SMA behavior, we first simulate a cube withan edge length of 1 mm under an applied displacement on oneface while the opposite face is fixed. The applied displacement isincreased from zero to a maximum value of 0.1 mm andsubsequently decreased to zero and increased in the oppositedirection to a value of 0.1 mm and finally decreased back to zero.Fig. 1 shows the SMA behavior at two different temperatures.
0
200
400
600
800
1000
1200
1400
1600
Forc
e [N
]T = 37 °C
5.1. Helical spring: pseudo-elastic test
A helical spring (with a wire diameter of 4 mm, a springexternal diameter of 24 mm, a pitch size of 12 mm and with twocoils and an initial length of 28 mm) is simulated using 9453quadratic tetrahedron (C3D10) elements and 15 764 nodes. Anaxial force of 1525 N is applied to one end while the other end isfully clamped. The force is increased from zero to its maximumvalue and unloaded back to zero. Fig. 2 shows the spring initialgeometry, the adopted mesh and the deformed shape under themaximum force. After unloading, the spring recovers the originalshape as it is expected in the pseudo-elastic regime. Fig. 3 (top)shows the force-displacement diagram. It is observed that thespring shape is fully recovered after load removal.
−11−10 −8 −6 −4 −2 0 2 4 6 8 10−2000
−1000
0
1000
2000
3000
True (Logarithmic) Strain [%]
Cau
chy
Stre
ss [
MPa
] T = 37° CT = −25° C
Fig. 1. 1D SMA behavior: pseudo-elasticity at T¼37 1C and shape-memory effect
at T¼�25 1C.
Fig. 2. Pseudo-elastic spring: comparison of ini
5.2. Helical spring: shape-memory effect test
We also simulate the same spring of Section 5.1 in the case ofshape-memory effect. An axial force of 427 N is applied atT¼�25 1C (Fig. 4, top) and, after unloading, the spring does notrecover its initial shape (Fig. 4, bottom). After heating, up to atemperature of 10 1 C, the spring recovers its original shape. Fig. 3(right) shows the force-displacement-temperature behavior.According to Fig. 3 (bottom), heating the spring leads to fullrecovery of the original shape at T¼Af (0 1C) and subsequentheating does not change its shape anymore.
5.3. Crimping of a medical stent: pseudo-elastic test
In this example, the crimping of a pseudo-elastic medical stentis simulated. To this end a medical stent with 0.216 mm thicknessand an initial outer diameter of 6.3 mm (Fig. 5, top-left) is crimpedto an outer diameter of 1.5 mm. Utilizing the ABAQUS/Standardcontact module, the contact between the catheter and the stent isconsidered in the simulation. A radial displacement is applied to
tial geometry and deformed configuration.
0 10 20 30 40 50 60 70 80
Displacement [mm]
020
4060
80−25−20
−15−10−50510
0
100
200
300
400
500
Temperature [°C]
Displacement [mm]
Forc
e [N
]
Fig. 3. Force-displacement diagram for the SMA spring: pseudo-elasticity (top)
and shape-memory effect (bottom).
Fig. 4. Shape-memory effect in the simulated spring: deformed shape under maximum load (top) and after unloading (bottom).
Fig. 5. Stent crimping: initial geometry (top-left), crimped shape in PE case (top-right), crimped shape in SME case (bottom-left) and deformed shape in SME case after
uncrimping (bottom-right).
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174 173
the catheter and is then released to reach the initial diameter. Inthis process, the stent recovers its original shape after unloading.Fig. 5 (top-right) shows the stent deformed shape when crimped.
5.4. Crimping of a medical stent: shape-memory effect test
In this example, the same stent is crimped at a temperature of�25 1C as it is shown in Fig. 5 (bottom-left). Due to the lowtemperature, while the catheter is expanded, the stent remains ina deformed state as shown in Fig. 5 (bottom-right). The initialshape is, however, recovered after heating.
5.5. Comparison of CPU times
We now compare the CPU times for three cases, i.e., originalunsymmetric model in which Ct is considered as unknown
[22,24], the original unsymmetric model in which Ut is con-sidered as unknown and the proposed symmetric model in whichUt is considered as unknown. We denote them as original-C,original-U and improved, respectively, in the following.
We remark that we normalize all CPU times with respect to thosefor the case original-C [22,24]. Moreover, the models have beenprogrammed in a consistent manner, so that reliable comparisonsconcerning the speed of computation could be obtained.
The simulation CPU times are reported in Table 4. From suchresults we may conclude that using Ut instead of Ct , decreases theCPU time of approximately one third compared with that adoptedin [22,24].
Moreover, comparing original-U and improved CPU times, weobserve a slightly increased efficiency for the proposed symmetricmodel.
We finally, conclude that using the proposed symmetricconstitutive model and solution algorithm can improve the
Table 4Normalized CPU time comparison.
Spring Spring Stent Stent
(PE) (SME) (PE) (SME)
original-C 1.00 1.00 1.00 1.00
original-U 0.68 0.72 0.65 0.72
improved 0.64 0.69 0.63 0.73
J. Arghavani et al. / Finite Elements in Analysis and Design 47 (2011) 166–174174
numerical efficiency in the order of one third when comparedwith the previously adopted ones in [22,24].
6. Conclusions
In this paper, we investigate a 3D finite-strain constitutivemodel which is fully expressed in terms of symmetric tensors. Thederivation is based on a multiplicative decomposition of thedeformation gradient into elastic and transformation parts andsatisfying the second law of thermodynamics in Clausius–Duheminequality form. Based on an exponential mapping, we propose atime-discrete form of the evolution equation which conserves thevolume during inelastic deformations. Implementing the pro-posed model within a user-defined subroutine (UMAT) in thecommercial nonlinear finite element software ABAQUS, we solvedifferent boundary value problems. Comparing the CPU times, weshow the efficiency of the proposed constitutive model as well asthe solution algorithm, implying a solution time reduction ofaround one third (33%) with respect to the original asymmetricmodel. Moreover, we show that the improved efficiency is mostlydue to the simplification in computing fourth-order tensorsappearing in the linearization procedure.
Acknowledgements
J. Arghavani has been partially supported by Iranian Ministryof Science, Research and Technology, and partially by Diparti-mento di Meccanica Strutturale, Universit�a degli Studi di Pavia.
F. Auricchio and A. Reali have been partially supported bythe Cariplo Foundation through the Project no. 2009.2822; by theMinistero dell’Istruzione, dell’Universit�a e della Ricerca throughthe Project no. 2008MRKXLX; and by the European Research Councilthrough the Starting Independent Research Grant ‘‘BioSMA: Mathe-matics for Shape Memory Technologies in Biomechanics’’.
References
[1] K. Otsuka, C.M. Wayman, Shape Memory Materials, Cambridge UniversityPress, Cambridge, 1998.
[2] T.W. Duerig, K.N. Melton, D. Stoekel, C.M. Wayman, Engineering Aspects ofShape Memory Alloys, Butterworth-Heinemann, London, 1990.
[3] T. Duerig, A. Pelton, D. Stckel, An overview of nitinol medical applications,Materials Science and Engineering A 273–275 (1999) 149–160.
[4] K. Kuribayashi, K. Tsuchiya, Z. You, D. Tomus, M. Umemoto, T. Ito, M. Sasaki, Self-deployable origami stent grafts as a biomedical application of Ni-rich TiNi shapememory alloy foil, Materials Science and Engineering A 419 (2006) 131–137.
[5] H. Funakubo, Shape Memory Alloys, Gordon and Breach Science Publishers,New York, 1987.
[6] B. Raniecki, C. Lexcellent, Thermodynamics of isotropic pseudoelasticity in shapememory alloys, European Journal of Mechanics—A/Solids 17 (1998) 185–205.
[7] M. Panico, L. Brinson, A three-dimensional phenomenological model formartensite reorientation in shape memory alloys, Journal of the Mechanicsand Physics of Solids 55 (2007) 2491–2511.
[8] A.C. Souza, E.N. Mamiya, N. Zouain, Three-dimensional model for solidsundergoing stress-induced phase transformations, European Journal ofMechanics—A/Solids 17 (1998) 789–806.
[9] D. Helm, P. Haupt, Shape memory behaviour: modelling within continuumthermomechanics, International Journal of Solids and Structures 40 (2003)827–849.
[10] C. Bouvet, S. Calloch, C. Lexcellent, A phenomenological model for pseudoelasticityof shape memory alloys under multiaxial proportional and nonproportionalloadings, European Journal of Mechanics—A/Solids 23 (2004) 37–61.
[11] P. Popov, D.C. Lagoudas, A 3-D constitutive model for shape memory alloysincorporating pseudoelasticity and detwinning of self-accommodated mar-tensite, International Journal of Plasticity 23 (2007) 1679–1720.
[12] F. Thiebaud, C. Lexcellent, M. Collet, E. Foltete, Implementation of a modeltaking into account the asymmetry between tension and compression, thetemperature effects in a finite element code for shape memory alloysstructures calculations, Computational Materials Science 41 (2007) 208–221.
[13] Z. Moumni, W. Zaki, Q.S. Nguyen, Theoretical and numerical modeling ofsolid–solid phase change: application to the description of the thermo-mechanical behavior of shape memory alloys, International Journal ofPlasticity 24 (2008) 614–645.
[14] J. Arghavani, F. Auricchio, R. Naghdabadi, A. Reali, S. Sohrabpour, A 3-Dphenomenological constitutive model for shape memory alloys undermultiaxial loadings, International Journal of Plasticity 26 (2010) 976–991.
[15] R. Mahnken, S. Wilmanns, Simulation of asymmetric effects for shapememory alloys by decomposition of transformation strains, ComputationalMaterials Science 42 (2008) 295–305.
[16] Y. Wang, Z. Yue, J. Wang, Experimental and numerical study of thesuperelastic behaviour on niti thin-walled tube under biaxial loading,Computational Materials Science 40 (2007) 246–254.
[17] F. Auricchio, R.L. Taylor, Shape-memory alloys: modelling and numericalsimulations of the finite-strain superelastic behavior, Computer Methods inApplied Mechanics and Engineering 143 (1997) 175–194.
[18] F. Auricchio, A robust integration-algorithm for a finite-strain shape-memory-alloy superelastic model, International Journal of Plasticity 17 (2001) 971–990.
[19] A. Ziolkowski, Three-dimensional phenomenological thermodynamic modelof pseudoelasticity of shape memory alloys at finite strains, ContinuumMechanics and Thermodynamics 19 (2007) 379–398.
[20] S. Reese, D. Christ, Finite deformation pseudo-elasticity of shape memoryalloys—constitutive modelling and finite element implementation, Interna-tional Journal of Plasticity 24 (2008) 455–482.
[21] D. Christ, S. Reese, A finite element model for shape memory alloysconsidering thermomechanical couplings at large strains, InternationalJournal of Solids and Structures 46 (2009) 3694–3709.
[22] V. Evangelista, S. Marfia, E. Sacco, A 3D SMA constitutive model in theframework of finite strain, International Journal for Numerical Methods inEngineering 81 (6) (2010) 761–785.
[23] J. Arghavani, F. Auricchio, R. Naghdabadi, A. Reali, S. Sohrabpour, A 3D finitestrain phenomenological constitutive model for shape memory alloysconsidering martensite reorientation, Continuum Mechanics and Thermo-dynamics 22 (2010) 345–362, doi:10.1007/s00161-010-0155-8.
[24] J. Arghavani, F. Auricchio, R. Naghdabadi, A. Reali, On the robustness andefficiency of integration algorithms for a 3D finite strain phenomenologicalSMA constitutive model, International Journal for Numerical Methods inEngineering, accepted for publication, doi:10.1002/nme.2964.
[25] P. Thamburaja, A finite-deformation-based phenomenological theory forshape-memory alloys, International Journal of Plasticity 26 (8) (2010)1195–1219.
[26] C. Muller, O. Bruhns, A thermodynamic finite-strain model for pseudoelastic shapememory alloys, International Journal of Plasticity 22 (2006) 1658–1682.
[27] F. Auricchio, L. Petrini, Improvements and algorithmical considerations on arecent three-dimensional model describing stress-induced solid phasetransformations, International Journal for Numerical Methods in Engineering55 (2002) 1255–1284.
[28] F. Auricchio, L. Petrini, A three-dimensional model describing stress-temperature induced solid phase transformations: thermomechanical cou-pling and hybrid composite applications, International Journal for NumericalMethods in Engineering 61 (2004) 716–737.
[29] F. Auricchio, L. Petrini, A three-dimensional model describing stress-temperature induced solid phase transformations: solution algorithm andboundary value problems, International Journal for Numerical Methods inEngineering 61 (2004) 807–836.
[30] V.A. Lubarda, Elastoplasticity Theory, CRC Press, 2001.[31] P. Haupt, Continuum Mechanics and Theory of Materials, Springer Publica-
tion House, 2002.[32] A.L. Eterovic, K.-J. Bathe, A hyperelastic-based large strain elasto-plastic
constitutive formulation with combined isotropic-kinematic hardening usingthe logarithmic stress and strain measures, International Journal forNumerical Methods in Engineering 30 (1990) 1099–1114.
[33] C. Miehe, Exponential map algorithm for stress updates in anisotropicmultiplicative elastoplasticity for single crystals, International Journal forNumerical Methods in Engineering 39 (1996) 3367–3390.
[34] I.N. Vladimirov, M.P. Pietryga, S. Reese, On the modelling of non-linearkinematic hardening at finite strains with application to springback—com-parison of time integration algorithms, International Journal for NumericalMethods in Engineering 75 (2008) 1–28.
[35] I.N. Vladimirov, M.P. Pietryga, S. Reese, Anisotropic finite elastoplasticity withnonlinear kinematic and isotropic hardening and application to sheet metalforming, International Journal of Plasticity 26 (5) (2010) 659–687.
[36] J. Arghavani, Thermomechanical behavior of shape memory alloys undernonproportional loading: constitutive modelling and numerical inplementationin small and finite strains, Ph.D. Thesis, Sharif University of Technology, Iran, 2010.