2022 Marian
2022 Marian
2022 Marian
https://doi.org/10.1007/s00707-022-03236-0
O R I G I NA L PA P E R
Received: 8 January 2022 / Revised: 15 April 2022 / Accepted: 19 April 2022 / Published online: 6 July 2022
© The Author(s) 2022
Abstract This work presents a thermomechanical finite strain shape memory alloy model that utilizes a
projection method to deal with the incompressibility constraint on inelastic strains. Due to its finite strain
formulation, it is able to accurately predict the behavior of shape memory alloys with high transformation
strains. The key feature of this model is the thermomechanical modeling of the shape memory effect and
superelastic behavior by optimizing a global, incremental mixed thermomechanical potential, the variation of
which yields the linear momentum balance, the energy balance, the evolution equations of the internal variables
as well as boundary conditions of Neumann- and Robin-type. The proposed thermal strain model allows to
properly capture transformation induced volume changes, which occur in some shape memory alloys. A
finite strain dissipation potential is formulated, which incorporates the disappearance of inelastic strains upon
austenite transformation. This important property is consistently transferred to the time-discrete potential
using a logarithmic strain formulation. Yield and transformation criteria are derived from the dual dissipation
potential. The implementation based on an active set search and the algorithmically consistent linearization
are discussed in detail. The model is applied in three-dimensional simulations of a bistable actuator design to
explore its capabilities.
1 Introduction
Since the discovery of their unique properties, shape memory alloys are used in many medical and engineering
fields in various applications. Their frequent appearance stems from their unique features, like the shape
memory effect and superelasticity. Both effects emerge from the characteristic first-order phase transition
from the austenite to the martensite state and vice versa. While the occurring crystallographic effects involving
the detwinning of the martensite phase, which enables the shape memory effect, are well understood, the
thermomechanical modeling of the occurring effects is not straightforward. In recent times, there has been a high
effort to improve shape memory alloy models and to obtain fitting models for specific use cases (see, for reviews,
Section 4 of Lester et al. [27] or Cisse et al. [8]). They can be roughly categorized into three classes: models
based on statistical thermodynamics, models founded in micromechanics and phenomenological models.
The models based on statistical thermodynamics rely on finding the phase equilibrium through a minimiza-
tion of a three well potential energy (e.g., Seelecke and Müller [43] or Govindjee et al. [16]). Since these models
yield results that also consider the microstructure of the materials, they come mostly with a computational cost
S. Wulfinghoff
E-mail: swu@tf.uni-kiel.de
3060 M. Sielenkämper, S. Wulfinghoff
which is too high for large, structural simulations. Additionally, gathering required micromechanical material
parameters is sometimes an elaborate task.
On the other hand, models based on micromechanics usually consider the mechanics of shape memory
alloy (SMA) single-crystals. Many models are then extended into the regime of polycrystal modeling by usage
of homogenization techniques (e.g., see Patoor et al. [35] and Lagoudas et al. [25] or the more recent models
by Mirzaeifar et al. [29] and Yu et al. [55]). While these models consider the deep, underlying phenomena
of shape memory alloys, this advantage again comes with a high computational cost. This makes it really
challenging to use these models in large structural simulations of shape memory alloy actuators.
The third group of models is the class of phenomenological models.
Usually, they come with the advantage of only having macroscopic material parameters, which mostly are
obtained through tensile tests at different temperatures. In recent years, due to the plethora of shape memory
alloys effects and applications, many new shape memory alloy models were published, which try to include
more and more physical phenomenons. They can be divided into two subgroups: models which include a
geometrically linear theory (see, e.g., Auricchio et al. [3], Auricchio et al. [4] and Sedlak et al. [42]), and
models which include a geometrically nonlinear theory.
The two main ways to include a geometric nonlinearity in the shape memory alloy model is to either
employ a multiplicative split of the deformation gradient going back to [10,24,26] or to make use of an
additive split (see Nemat-Nasser [32]) of the rate of deformation, which are both well known from plasticity.
While models based on the additive split are computationally enticing, they only allow for small strains, while
still capturing large rotations well (see, e.g., Qidwai and Lagoudas [37], Müller and Bruhns [31] or Zhang
and Baxevanis [57]). On the other hand, models employing a multiplicative split are computationally more
elaborate, but can represent finite stretches well (see, e.g., Reese and Christ [38], Arghavani et al. [1] or Wang
et al. [49]). Additionally, there exist many new models considering geometric nonlinearities, which are limited
to superelasticity (see, e.g., Bellini et al. [5], Wang et al. [48] and Rezaee-Hajidehi et al. [39]). Furthermore,
some models also consider transformation induced plasticity (see, e.g., Hartl et al. [19], Xu et al. [53] and it’s
extension to partial phase transformations by Scalet et al. [41]).
Many of these models capture thermomechanics, martensite reorientation and detwinning as well as supere-
lasticity at different temperatures or allow for different elastic properties of the materials while being numeri-
cally efficient and robust. However, when trying to model bistable shape memory actuators (see the actuator
design in Arivanandhan [2]), one needs a numerically robust, fully thermomechanically coupled finite strain
model that also models thermal expansion and volumetric effects during phase transition (see, e.g., Potapov et
al. [36]). To our knowledge, there is no model in the literature fulfilling all of the aforementioned requirements,
which led to the model described in this paper. For example, the model of Wang et al. [49] considers a ther-
momechanically coupled finite strain theory which is capable of modeling the shape memory effect as well as
superelasticity, neglecting volumetric effects due to transformation as well as different expansion coefficients
of the SMA phases. The model at hand falls into the aforementioned category of phenomenological models.
It is embedded into the generalized standard materials framework developed by Halphen and Nguyen [17],
which was extended to thermomechanics by Yang et al. [54] and which allows to ensure thermodynamic
consistency. The energies as well as the dissipation potential can be seen as an extension of Sedlak et al. [42]
to the finite strain case. Because the satisfaction of the incompressibility of inelastic strains for finite strains
is not as straightforward as for the small strain case, a projection method developed for plasticity is incorpo-
rated into the model (see Hurtado et al. [21] and Sielenkämper et al. [44]). Further, due to the character of
the energies used in this model, special numerical treatment is necessary to solve the model equations using
a Newton scheme. Since our aim is to model microactuators in which the R-Phase is not present, it is not
incorporated into the model. Additionally, tension-compression anisotropy, which is an important effect in
many shape memory alloys (for experimental publications see, e.g., Gall et al. [15] or Wang and Zhu [47], and
for modeling approaches see, e.g., Zaki et al. [56] or Sedlak et al. [42]), is not included in the model.
In the past, numerous advancements to current microelectromechanical systems (MEMS) based on electro-
statics, magnetism and electrothermal principals have been made. For example, Hoffmann et al. [20] proposed
a microactuator based on electrothermal activation making use of bimetal effects. This, however, comes with
the downside of low actuation frequencies and a high power consumption. Han et al. [18] proposed an elec-
trostatic actuator-based micro-switch for photonics. Devices based on electrostatics usually can be actuated
with high frequencies and are adaptable to many applications. Devices using optomechanics were developed
by, e.g., Eichenfield et al. [11], which come with a rather small tuning range, but allow for a very high oper-
ation speed [9]. Despite their unique advantages, current MEMS devices are challenging to use in downsized
applications, where a high work output combined with a high power efficiency and bistability is crucial. The
A thermomechanical finite strain shape memory alloy model 3061
microactuator modeled in this paper is based on a concept published by Winzek et al. [50], which utilizes a
high temperature shape memory alloy with a large thermal hysteresis. This concept may overcome the afore-
mentioned weaknesses of other actuation principles, as shape memory alloys usually have a large work output
density and favorable downscaling capabilities (see, e.g., Kohl [22]). Additionally, downscaling this design
is expected to drastically increase the possible actuation frequency in comparison to other, electrothermally
activated actuators due to the decreasing masses and increasing thermal gradients. One further key aspect is,
that the proposed actuator design requires no power in the stable states.
The paper is structured as follows: First, the energies as well as the dissipation potential is derived. Then,
in Sect. 3, numerical strategies necessary to solve the model equations are discussed. The numerical results
are subsequently shown in Sect. 4 before a summary and outlook concludes the paper in Sect. 5.
Notation
Throughout this paper, a direct tensor notation is preferred. Scalars and scalar valued functions are typeset
using light-face italic characters, e.g., a or A. First and second-order tensors and tensor-valued functions are
represented by bold-face italic letters, e.g., a or A. Further, blackboard bold-faced letters are used to denote
fourth-order tensors, e.g., 𝕔 or C.
Additionally, the transpose of a second-order tensor is designated by AT , while the major transpose of a
fourth-order tensor is given by CT . The symmetric and deviatoric part of a second-order tensor A are denoted
by sym( A) = 21 ( A + AT ) and A = A − 13 tr( A)I, respectively. Here, I is the second-order identity tensor
and tr( A) denotes the trace of A. A double contraction of two tensors A and B is denoted by A : B, while
the dyadic product is denoted by a ⊗ b. A determinant of a tensor A is either designated by det( A) or by A’s
third invariant III A .
2.1 Kinematics
The deformation gradient F maps a line element from the reference configuration of a body with Volume V0
into the current configuration of a body with Volume V and is defined as
F = Grad(x(X, t)), (1)
where Grad(•) refers to the gradient with respect to the reference configuration while X and x are the position
vectors of a material point in the reference and current configuration, respectively. We consider a multiplicative
split of the deformation gradient in the form
F = Fe Fi Fθ (2)
(see Wang et al. [49]), where F e is the elastic, F i the isochoric part of the deformation due to transformation1
and F θ the part which describes the volume change due to thermal expansion and transformation2 of the
deformation gradient. This is fairly similar to the multiplicative split in plasticity going back to the works
of Eckart [10], Kröner [24] and Lee [26]. Further, we define the determinant J θ = J θ (θ, ξ ) = det(F θ ) to
be a function of the absolute temperature θ and the martensite volume fraction ξ ∈ [0, 1]. Since the thermal
deformation is assumed to be volumetric, we can express F θ in terms of J θ as
1
F θ = (J θ ) 3 I. (3)
As is commonly done and will be useful later, we define the elastic and inelastic left Cauchy-Green tensors as
be = F e F eT , bi = F i F iT . (4)
Likewise, we define the inelastic right Cauchy-Green tensor and the inelastic Green-Lagrange strain
1 i
C i = F iT F i , E i = (C − I). (5)
2
1 For brevity, F i is called the inelastic part of the deformation gradient throughout this paper.
2 For brevity, F θ is called the thermal part of the deformation gradient throughout this paper.
3062 M. Sielenkämper, S. Wulfinghoff
Motivated by the observation that the shape-memory effect is caused by an almost3 volume preserving trans-
formation of the crystal lattice, we assume C i to be volume preserving. Therefore, because det(F i ) = 1 has
to hold, the determinant of the deformation gradient is given by
J = det(F) = J θ J e , (6)
with Je = det(F e ). Additionally, we define the velocity gradient l and its symmetric part d as
l = Ḟ F -1 , d = sym(l). (7)
Finally, we define the inelastic ’velocity gradient’ Li and its symmetric part Di in analogy to l and d by
L = Ḟ F
i i i−1
, D = sym(L ).
i i
(8)
The quasistatic linear momentum balance for a body with volume V in the current configuration is given by
div(σ ) + ρb = 0 in V, (9)
where b is the body force and ρ is the mass density. Further, Cauchy’s lemma t = σ n as well as the angular
momentum balance σ = σ T are assumed to hold.
The Helmholtz free energy density ψ for this model is assumed to be a sum of elastic, chemical and hardening-
like contributions in the form
ψ = ψe (be , ξ ) + ψc (ξ, θ ) + ψh (C i , ξ ). (15)
The elastic energy ψe is assumed to be isotropic and to follow a modified Neo-Hookean formulation:
λ(ξ ) e 2 μ(ξ )
ψe (be , ξ ) = (J − 1 − 2 ln J e ) + tr(be ) − 3 − 2 ln J e . (16)
4 2
Here, λ(ξ ) and μ(ξ ) are the Lamé parameters in dependence of the martensite volume fraction ξ . Further, we
use a Reuss-like mixture rule to estimate the elastic constants, i.e.,
−1 −1
ξ 1−ξ ξ 1−ξ
μ(ξ ) = + , λ(ξ ) = + . (17)
μM μA λM λA
Here, and subsequently, the indices •A and •M refer to the austenite and martensite phase, respectively.
Manipulating Eq. (13), one can show that the ordinary form
∂ψe
τ = 2be (18)
∂ be
holds for elastic isotropy. Using Eqs. (14) and (18), one can show that
∂ψ iT ∂ψh iT
i = −2F i F = e − 2F i F , (19)
∂ Ci ∂C
i
:= h
where e = C e F e−1 τ F e−T is the Mandel stress with respect to the intermediate, elastically unloaded con-
figuration, which is symmetric due to the assumption of elastic isotropy.
For the chemical energy, we assume a standard relationship (see, e.g., Lexcellent et al. [28] or Panico and
Brinson [34]):
A AM θ
ψc = u 0 − θ s0 −ξ u
A
− θ s AM
+c θ − θ0 − θ ln
θ0
ψ0A ψ AM (20)
θ
= uA
0 − θ s0 + ξ (θ − θ0 ) s
A AM
+ c θ − θ0 − θ ln ,
θ0
where c is the specific heat capacity and s AM is the difference in specific entropy of the austenite and
martensite phase: s AM = s0A − s0M . Here, c is assumed constant (compare [42]). In Eq. (20), we made use
of the definition of the equilibrium temperature of austenite and martensite, which is θ0 = u AM /s AM .
Further, we assume for the equilibrium temperature As > θ0 > Ms , where As is the temperature where the
reverse transformation starts and Ms is the starting temperature of the forward transformation.
3064 M. Sielenkämper, S. Wulfinghoff
Since the inelastic strains vanish as ξ → 0, the inelastic strains E i are assumed to satisfy the relation E i = ξ E t
(see Otsuka and Ren [33]), where E t is a measure for the effective transformation strain. Now, the hardening-
like energy adapted from Sedlak et al. [42] for finite strains is assumed to be given by
t
Et
2
2 E
ψh = k E ξ
int
4
, E = 3
t
. (21)
1 − Et k
Here, k is the maximum transformation strain, E int is a hardening related parameter and E t is a modified von
Mises equivalent strain. Obviously, ψh → ∞ for E t → 1. This is desired, since it captures the martensite
becoming fully detwinned, which is assumed to cost additional energy.
Modeling thermal strains of multi-phase materials requires attention and special treatment. In this work, the
determinant of the thermal part of the deformation gradient is connected to the coefficients of thermal expansion
(CTEs) by
3
J θ = 1 + εθ (ξ, θ ) , εθ = ξ αM (θ − θ refM ) + (1 − ξ )αA (θ − θ refA ) . (22)
Here, εθ (ξ, θ ) is the thermal strain. Additionally, αM and αA are the CTEs of the martensite and austenite phase,
respectively. Further, θ refM and θ refA are the reference temperatures for austenite and martensite. Here, we
want to note that it is necessary to distinguish the two in order to properly represent the transformation-induced
volume change which is present in some shape memory alloys (see below).
Example
We assume a one-dimensional SMA rod with only one reference temperature. Further, we simplify the SMA
model by assuming that martensite and austenite transformation occur at the distinct temperatures θM and θA ,
and not at temperature ranges (see Fig. 1). The thermal strains are then given by εθ = αAM (θ − θ ref ), with
αAM = ξ αM + (1 − ξ )αA . Furthermore, we assume that we know the shape of our rod at the thermal annealing
temperature θ ref (see Fig. 1).
During forward transformation, the jump of εθ at θM can be analytically calculated and is given by
εθ = (θ ref − θM )(αA − αM ). (23)
Here, we can not only see that this jump is dependent on the reference temperature, but that it can be positive or
negative, depending on the reference temperature θ ref being larger or smaller than the forward transformation
temperature θM .
Using two reference temperatures θref M and θref A , we can not only circumvent this problem, but also model
the magnitude of this jump in compliance with, e.g., the experimental results of Potapov et al. [36], where they
calculated the jump V in volume from the lattice parameters for Ni49.8 Ti35.2 Hf15 to be 0.47%. However, we
want to emphasize that this effect does not play a significant role in many other shape memory alloys.
Fig. 1 Thermal strains when only considering one reference temperature θ ref
A thermomechanical finite strain shape memory alloy model 3065
The dissipative behavior of the shape memory alloy is assumed to be governed by the dissipation potential
⎧
⎪
⎨φM , if ξ̇ ≥ 0 ∧ tr( D ) = 0,
i
where Mf is the finish temperature of forward transformation and σ reo is the reorientation stress. Likewise, the
dissipation potential when going to a higher austenite volume fraction is defined as
2 reo ξ̇ i
i ξ̇ i
φA = ξ̇ s (θ0 − Af ) + ξ(Af − As ) + 3 σ
ξ ε + D − ξ ε .
AM
(26)
=:Q A (ξ )
Here, Af is the finish temperature of reverse transformation and ε i = 21 ln bi . The significance of using the
logarithmic inelastic strain ε i as a strain measure for the dissipation is explained below. The dissipation potential
represents a geometrically nonlinear generalization of the small strain potential proposed in Sedlak et al. [42],
which is based on the works of Bernardini and Pence [6], Panico and Brinson [34] and Moumni et al. [30].
It is assumed that the evolution of C i and ξ follows from the minimization problem
inf ψ̇ + φ, (27)
Di ,ξ̇
which is equivalent to the Legendre–Fenchel transformed problem (compare Eqs. (14), (19))
sup e − h : Di + q ξ̇ − φ(ξ̇ , Di , ξ ) =: φ ∗ ( i , q, ξ ), (28)
Di ,ξ̇
Due to the minimization in Eq. (27), there is a tendency that for ξ̇ < 0 we get Di − (ξ̇ /ξ )ε i = 0 as a result of
the last term in Eq. (26). In that case bi → I as ξ → 0 in time (for a proof, see Appendix A), i.e., for vanishing
martensite content, the inelastic strain vanishes, which is a physical necessity. With the reparametrization
Dt = Di − (ξ̇ /ξ )ε i (compare Eq. (26)) we find
⎧
⎪ sup i : Di − 2 σ reo
⎪
Di + q − Q M (ξ ) ξ̇ ,
⎪
⎪ 3
⎪ Di ,ξ̇ ≥0
⎪
⎨ i
tr( D )=0
∗
φ ( , q, ξ ) = sup
i i
⎪ t ε
⎪
⎪ sup : D − 3 σ
i t 2 reo
D + ξ : ε + q − Q A (ξ ) + 3 σ
1 i i 2 reo ξ̇ ,
⎪
⎪ ξ
⎪
⎩ Dt ,ξ̇t <0
tr( D )=0
(30)
3066 M. Sielenkämper, S. Wulfinghoff
where Q M and Q A are defined in Eqs. (25) and (26) and capture the transformation hysteresis due to temperature
change. Further reparametrizing D i / t = λN with λ ≥ 0 and the obvious solution
i
N = i , (31)
it follows that
⎧
⎨ sup λ f ( ) + ξ̇ gM (q)
i
⎪
∗ λ≥0,ξ̇ ≥0 0, if f ≤ 0 ∧ gM ≤ 0 ∧ gA ≥ 0,
φ ( , q, ξ ) = sup
i
= (32)
⎪
⎩ sup λ f ( ) + ξ̇ gA (q, )
i i ∞, else.
λ≥0,ξ̇ <0
Thus, we find the following transformation and yield criteria in the classical Karush-Kuhn-Tucker form as
well as evolution equations:
∂f
f ( i ) = i − 3σ
2 reo
≤ 0; λ f = 0; λ ≥ 0; Di = λ ;
∂ i
gM (q) = q − Q M (ξ ) ≤ 0; ξ̇ gM = 0; ξ̇ > 0 possible if gM = 0;
1 σ reo i
gA (q, i ) = i : ε i + q − Q A (ξ ) + 2 ε ≥ 0; ξ̇ _gA = 0; ξ̇ < 0 possible if gA = 0;
ξ 3 ξ
(33)
with the Macaulay bracket • = (•+|•|)/2 and its modified form • _ = (•−|•|)/2. It is then straightforward
to prove thermodynamic consistency (see Eq. (14)). The ’0’-branch in Eq. (32) implies that (see Eq. (28))
D = i : Di + q ξ̇ = φ. (34)
Thus, we find with Eqs. (4) and (10) the following alternative form of the energy balance:
θ ṡ = φ − Div( Q) + w. (35)
To simplify the discussion, we start by the isothermal case, i.e., in a first step, the temperature θ is considered
to be a given parameter. The rate potential and its time discretized form read
1
π = ψ̇ + φ ψ(F, C i , ξ, θ ) − ψn + φ . (36)
t
Here, ψn refers to the Helmholtz free energy at the previous time, t is the time step from tn to tn+1 4 and φ
is the time discretized version of the dissipation potential multiplied by t, defined by
σ reo α + Q(Cni , ξn+ 1 , sg (ξ ))ξ, if III i = 1,
φ = 2 (37)
∞, else,
with sg (•) referring to the sign function and ξ = ξ − ξn . Additionally, III i is the third invariant of C i , i.e.,
The constraint III i = 1 in Eq. (37) is consistent with the requirement tr( Di ) = 0 (see Eq. (24)). Additionally,
we compute the effective inelastic strain increment α as
1 i−1
2 1
α = 23 U
2 n r C i i−1
Un =: C i
3 2 r i−1 , (39)
Un
4 For brevity and simplicity, we dropped the index n + 1 whenever we deemed it to be not helpful for the presentation.
A thermomechanical finite strain shape memory alloy model 3067
√
where the inelastic right stretch is U i = C i , r C i = C i − Cri and the shorthand notation •Uni−1 =
i−1
U • Uni−1 is used. Further, we define Cri as
n
1+ ξ
Cri = Cni ξn −
. (40)
0
Now, for ξ going to zero in any given step, i.e., ξ = ξn + ξ = 0 ⇔ ξ = −ξn → Cri = Cni = I, Cri
goes back to unity again. Thus, the time discrete potential (37) based on definition (40) of Cri is the key to
ensure that the inelastic strain consistently disappears during the transformation from martensite to austenite.
Further, the function Q(Cni , ξn+ 1 , sg (ξ )) is given by
2
⎧
⎨s AM (θ0 − Ms ) + ξn+ 1 (Ms − Mf ) , if ξ ≥ 0,
Q(Cni , ξn+ 1 , sg (ξ )) = 2
i (41)
2 ⎩s AM (θ0 − Af ) + ξ 1 (Af − As ) − 2σ reo ε , if ξ < 0,
n+ 2 3 ξn n
where ξn+ 1 = (ξ + ξn )/2 is the midpoint evaluation of ξ , which is employed to obtain a reasonable transfor-
2
mation when the material is not stressed (see Frost et al. [14] for a similar concept). Moreover, we used Cni and
ξn for the reverse transformation in Eq. (41) to circumvent the eigenvalue problem as well as its linearization
in every local Newton iteration. In this way, it suffices to solve the eigenvalue problem once per time step.
Using some further shorthand notations for constant terms in Q, we obtain
φ = σ reo α
⎧
⎪ Q M0 HM
⎪
⎪
⎪
⎪
⎨s AM (θ0 − Ms ) + ξn (Ms − Mf ) ξ + 21 s AM (Ms − Mf ) ξ 2 , if ξ ≥ 0
+ Q A0
⎪
⎪ HA
⎪
⎪ reo
⎪
⎩ s AM (θ0 − Af ) + ξn (Af − As ) − 2 σ ε i ξ + 1 s AM (Af − As ) ξ 2 , if ξ < 0
3 ξ n 2
n
M0
Q ξ + 21 H M (ξ )2 , if ξ ≥ 0
= σ reo α +
Q A0 ξ + 21 H A (ξ )2 , if ξ < 0
1
= σ reo α + Q(sg (ξ ))ξ + H (sg (ξ ))(ξ )2 ,
2
(42)
3 Numerical strategies
The potential π , as we formulated it in Eq. (48) carries two major numerical difficulties. First, C i is constrained
to be volume preserving, i.e., det(C i ) = 1, which is very important to be exactly satisfied to comply with physics
and not accumulate errors. Second, π is not differentiable at r C i = 0. The strategies employed to overcome
these and other numerical difficulties as well as general numerical approaches are presented in this section.
To deal with the constraint in C i , we employ a strategy using a projection of C i into the space of unimodular
tensors, which was introduced by Hurtado et al. [21] for crystal plasticity. Our approach here closely follows
the approach presented in Sielenkämper et al. [44].
First, we express C i in terms of the unconstrained inelastic auxiliary right Cauchy-Green tensor Ĉ i :
1
C i = III ˆ i = det(Ĉ i ).
ˆ i− 3 Ĉ i , III (49)
Thus, det(C i ) = 1 is automatically satisfied. This idea is borrowed from various formulations in hyperelasticity,
where similar approaches are used to decouple volumetric and deviatoric deformations (see, e.g., Flory and
Volkenstein [13] or Simo et al. [45]). Now, we replace C i by C i (Ĉ i ) in the minimization problem in Eq. (48):
However, while this removes the constraint from the minimization problem, π is now invariant with respect
ˆ i , and is therefore not uniquely solvable. For this reason, we add the regularization energy
to changes of III
A ˆi
ψr = ( III − B)2 (51)
2
ˆ i = B is exactly
to π . We want to emphasize that this has no effect on the solution. This is due to the fact that III
satisfied after converging to a solution, since ψ is the only term in π that is dependent on III ˆ i . Therefore, ψ
r r
is zero at any solution of the minimization problem. Further, the constants A and B can be chosen arbitrarily.
They do not have to be chosen particularly large. In this paper, we chose B = 1, but any other positive value
could be chosen and would lead to the exact same results. In this special case, C i = Ĉ i is exactly satisfied
once the solution algorithm is converged. However, we want to note that this equality does not hold in the not
yet converged state.
The following tensor, which is used in the residuals and stiffness terms later, is also known from hyperelastic
models:
i− 13 1 i ∂ Ci
Pi Ĉ = III
i ˆ I − Ĉ ⊗ Ĉ
s i−1
= , (52)
3 ∂ Ĉ i
where Is is the fourth-order identity on symmetric second-order tensors.
A thermomechanical finite strain shape memory alloy model 3069
3.2 Differentiability at r C i = 0
1
The ansatz C i = IIIˆ i− 3 Ĉ i renders the solution C i a priori volume preserving, but it does not solve the lack of
differentiability of φ for C i = Cri . However, this is achieved by the following reparametrization:
Ñ s
Ĉ i (γ , Ñ s ) = Cri + 2γ , γ ≥ 0, (53)
s
Ñ
Uni−1
where Ñ s is an unconstrained, symmetric 2nd-order tensor. The final unconstrained minimization problem
now reads
(γ , Ñ s , ξ ) = arg min π (F, C i , (Ĉ i (γ , Ñ s )), ξ, θ ). (54)
γ >0, Ñ s ,ξ
It is easy to show that π is invariant with respect to Ñ s . In order to render the solution unique, we
Uni−1
further modify ψr : 2
A ˆ
ψr = ( III i − 1)2 + Ñ s i−1 − 1 + ( Ñ s − Ñ s,tr )2 , (55)
2 Un 2
where the choice of A, with the same argument as in Sect. 3.1, still has no effect on the solution. Additionally,
the last term with a very small and constant is explained in Sect. 3.5. Now, the minimization problem from
Eq. (54) is computed by solving the stationary conditions:
∂π ∂π ∂π
= 0, = 0, = 0, (56)
∂γ ∂ Ñ s ∂ξ
where in general only a subset of these three equations is involved, depending on the ’active’ variables. The
active set of variables is determined by the activation or yield criteria presented in the sequel (see Sects. 3.3
and 3.10).
In a given time step, C i will only change (i.e., C i = Cni ⇔ γ > 0), if this decreases (minimizes) the potential.
In order to decide whether the couple (γ , Ñ s ) is activated for a given state (F, Cri , ξ, θ ), we evaluate the
algorithmic yield condition
∂
f := − inf π (F, C (Ĉ (γ , Ñ )), ξ, θ )
i i s . (57)
∂γ Ñ s γ →0
If f > 0, γ is activated, i.e., (γ , Ñ s ) are put into the active set. Otherwise, γ = 0 minimizes π ,
which corresponds to the case where f ≤ 0. If (γ , Ñ s ) are part of the active set, the related equations and
! !
concerning residuals are: ∂π /∂γ = 0 and ∂π /∂ Ñ s = 0. For a given F, ξ and θ , we obtain the explicit
form of f by variation of π: γ = 0 if π (F, C i , ξ, θ ) is minimized by C i = Cri . In other words, γ = 0 is
minimizer of π if
∂ Ci
δπ ≥ 0 ∀δC i = : (2δγ N s ), (58)
∂ Ĉ i
δ Ĉ i
with the shorthand notation
N s = Ñ s / Ñ s (59)
Uni−1
for an arbitrary Ñ s = 0 and δγ ≥ 0. Further, taking a closer look at the variation of π , we obtain (compare
Appendix B)
∂ψ
δπ = i i : Pi (Cri ) : (2δγ N s ) + 23 σ reo δγ Uni−1 N s Uni−1 . (60)
∂ C C =Cr
i
=1
3070 M. Sielenkämper, S. Wulfinghoff
With the definitions of N s in Eq. (59) and DEVCri (•) in Eq. (61) as well as making use of the fact that Uni−1
and Cri−1 are coaxial, we can simplify Eq. (60) to
⎛ ⎞
∂ψ U i−1 Ñ s U i−1
δπ = ⎝ Uni DEVCri 2 i i i Uni : n n
+ 23 σ reo ⎠ δγ . (62)
∂ C C =Cr i−1 s i−1
Un Ñ Un
Now, since inequality (58) must hold for arbitrary symmetric Ñ s , we get
∂ψ
δπ ≥ −
DEVCri 2 ∂ C i C i =C i i − 3 σ
2 reo
δγ ≥ 0 ∀δγ ≥ 0. (63)
r Un
However, note that for ξ = 0, and thus Cri = Cni , we have a simplified yield criterion
i i ∂ψ
f = Rn 2Un iT
− 3σ
i 2 reo
Un Rn
∂ C i C i =Cni (65)
(19) e,tr
= − h,tr − 23 σ reo ,
where the superscript ’tr’ denotes the trial state and Ri = F i U i−1 . This result again shows the consistency
with the time-continuous theory (see Eq. (33)).
In analogy to Sect. 3.3, the activation criteria for forward (’M’, ξ ≥ 0) and reverse (’A’, ξ < 0) transfor-
mations read
∂ ∂
gM = − π , gA = − π , (66)
∂ξ ξ →0+ ∂ξ ξ →0−
where 0+ and 0− denote the right- and left-hand limit (recall that sg (ξ ) is an argument of Q in Eq. (41)).
In general, the direction of inelastic flow Ñ s is determined by the minimization of π through a Newton
scheme, which requires a reasonable initial guess when γ is activated. Coincidentally, an analytical solution
for γ → 0 exists for Ñ s :
Cni DEVCri −2 ∂∂ψ
C i C i =C i
Cni
Ñ =
s,tr .
r
(67)
DEVC i −2 ∂ψi
r ∂ C C i =C i i
r Un
The proof is given in Appendix B.
A thermomechanical finite strain shape memory alloy model 3071
It is noted that for γ = 0, it follows that C i = Cri , which is independent of Ñ s . That means that any
choice of Ñ s minimizes π , i.e., there is no unique solution. For very small γ , the sensitivity of π with
respect to Ñ s is also small, which can lead to a bad condition of the nonlinear equation system, which needs
to be solved to minimize π . To stabilize the solution process, the last term in Eq. (55) is added to π, since for
γ → 0 it is known that Ñ s = Ñ s,tr minimizes π . In theory, can be chosen arbitrarily small, in practice a
finite value is necessary due to the limited numerical accuracy. Thus, the last term in Eq. (55) is used to ’guide’
the algorithm toward the right solution when it is no longer able to find it by itself.
The derivatives of α are numerically tough to obtain for γ → 0, due to an almost zero denominator of
M̃C i (see Eq. (112) and preceding equations). To overcome this issue, we derived the derivatives of γ with
respect to ξ , Ñ s and γ separately for γ → 0, which are then numerically feasible to obtain. However,
this seems to be a rather theoretical as we never observed the case that γ was too close to zero to obtain the
derivatives using the usual way shown in Appendix C. Therefore, and for brevity, we do not discuss them here.
3.7 Regularization of ψh
Numerically, ψh is challenging, since with E t → 1, ψh and ∂ψh /∂ E t go to infinity. This is very demanding,
because values of E t ≥ 1 can occur during the iterative solution process of the Newton scheme, leading to a
loss of convergence. The term (compare to Eq. (21))
2
Et
ϕ( E t ) = 4
(68)
1 − Et
is illustrated in Fig. 2 in black.
To stabilize the solution process, we partially replace the function ϕ by a linear approximation in the region
where E t > c, where c is close to 1 (see Fig. 2 in green). This regularization approach is adopted from
crystal plasticity, where it successfully improved the numerical treatment of the power law in Wulfinghoff
and Böhlke [52]. In order to prevent the solution E t from taking nonphysical solutions beyond 1, we add
a penalty-type energy as illustrated in Fig. 3. We penalize all states ( E i , ξ ) (see Fig. 3) below the dashed
half-line starting at ξ0 by the energy
⎧
⎨ 1 Hpl 2 ξ < E i + ξ0
F
ψp = 2 c
(69)
⎩0 else
where lF is the minimum distance of the point ( E i , ξ ) from the half-line and Hp is a large penalty parameter.
To further improve the numerical robustness, we added an artificial viscosity-like term for the martensite
volume fraction. Since it is a dissipative term, we add this term to the discretized dissipation potential:
1
φη = ηξ ξ 2 , (70)
2t
where ηξ is a small, positive constant.
Like the displacements, the temperature is an unknown for our model. Both fields are coupled through the
thermomechanical energy and through the dissipative terms, which lead to a heating of the material. Addi-
tionally, the temperature changes the material behavior and leads to thermal strains. Both fields are coupled
using a potential-based monolithic approach similar to the one presented in Yang et al. [54]. Further, we
assume Fourier’s heat conduction law with heat conductivity κ. The thermomechanic quasistatic time-discrete
potential then reads
θ t θ
π = ψ − (u n − θ sn ) + φ − κ Grad(θ )2 + t 2 κ Grad(θn )2 , (71)
θn 2θn θn
where we neglect the influence of body forces. The integral form of the local potential π is then given by
¯ t t
= π dV − t̂ · udS − Q̄θ dS + θ wdV, (72)
θn θn
V0 ∂ V0t V0Q V0
where w denotes the heat source and ¯t̂ as well as Q̄ denote prescribed tractions and normal heat flux at the
Neumann-type boundaries ∂ V0t and ∂ V0Q , respectively. Now, the classical weak form of the linear momentum
balance is obtained by variation of Eq. (71) with respect to u:
δu = τ : dδ dV − ¯t̂ · δudS =! 0, (73)
V0 ∂ V0t
with dδ = sym(Grad(δu) F -1 ). Likewise, we obtain the weak form of the energy balance by variation of
Eq. (71) with respect to θ :
A thermomechanical finite strain shape memory alloy model 3073
∂ψ 1 κt
δθ = + sn + φ δθ − Grad(θ ) · Grad(δθ )
∂θ θn θn
V0
δθ t t !
+t κ Grad(θ n )2
+ w δθ dV − Q̄δθ dS = 0. (74)
θn2 θn θn
V0 Q
where κu = {u : u = ū on ∂ V0u } is the set of admissible displacements satisfying the Dirichlet boundary
conditions imposed on the boundary ∂ V0u and κθ = {θ : θ = θ̄ on ∂ V0θ } is the set of admissible temperatures
satisfying the Dirichlet boundary conditions imposed on the boundary ∂ V0θ . Additionally, z is the vector of
internal variables, i.e., z = (ξ, C i ).
Manipulating Eq. (74) using the definition s = −∂ψ/∂θ as well as applying Gauss theorem, we obtain
1 κt κt κt
δθ = −s + φ + Div(Grad(θ )) − 2 Grad(θ ) · Grad(θn ) + 2 Grad(θn )2
θn θn θn θn
V0
t Grad(θ ) · N t
+ w δθ dV + −tκ δθ dS − Q̄δθ dS = 0, (76)
θn θn θn
∂ V0Q ∂ V0Q
where N is the external normal on the boundary ∂ V0 in the reference configuration. To show consistency with
the time-continuous theory, we take a look at the integrand over the volume integral in Eq. (76). By multiplying
with θn /t, we get
s 1 κ
θn = φ − Div(−κGrad(θ )) + w + Grad(θn ) · (Grad(θn ) − Grad(θ )). (77)
t t θn
Here, we want to note that this requires κ to be independent of the temperature θ . However, one possibility is
to consider κ = κ(θn ), which could circumvent this limitation.
Now, for t → 0, we get (also see Eq. (35))
θ ṡ = φ − Div(−κGrad(θ )) + w. (78)
Hence, we arrived at the energy balance (Eq. (35)) with Fourier’s law Q = −κGrad(θ ), which proves the
consistency with the time-continuous theory. Further, the surface integrals in Eq. (76) imply the Neumann
boundary condition Q̄ = −κGrad(θ ) · N on ∂ V0Q . Alternatively, one can include Robin-type boundary
conditions into the model by replacing the integrand t Q̄θ/θn in Eq. (72) by 1/2th(θ − θs )2 /θn , where h is
the heat convection coefficient and θs is the temperature of the surrounding medium. In that case, the variation
of yields the boundary condition −κGrad(θ ) · N = h(θ − θs ) on ∂ V0Q .
When solving the set of Eq. (56), one has to decide which variables will evolve using the activation criteria
given in Eqs. (64) and (66). They will then be put into the active array of variables A. If then, at a later state,
an activation criterion is inactive, the variable is taken from A again. This is done using the active set search
algorithm outlined in Algorithm 1.
The algorithm is structured as follows: First, if A = ∅ or no variable from A was rescaled (see the end
of this subsection for an explanation of rescaling) during the last iteration, we need to evaluate the activation
criteria. With them at hand, we determine our new set A j+1 . The details for determining which variable to
activate are given in Algorithm 2. Subsequently, we check if the active set changed from the last iteration. If
this is not the case, the solution from the last iteration is confirmed as solution of the equation system. In that
case, we compute the algorithmic tangent (see Appendix E for details), save the history variables and exit the
material routine. Otherwise, we solve the minimization problem for the current active set A j+1 . This is done
3074 M. Sielenkämper, S. Wulfinghoff
Algorithm 1 Active set algorithm solving the minimization problem in Eq. (56).
loop
j ← j +1
if no rescaling done in last step or A j = ∅ then
determine A j+1 (see Algorithm 2)
end if
if A j+1 = A j then
compute 𝕔a , ∂π ∂ π ∂τ
2
∂θ , ∂θ 2 , ∂θ Eqns. (165), (171), (172)
save history variables
exit
end if
ξ0 = ξ j , γ0 = γ j , Ñ s 0 = Ñ s j
loop
k ←k+1
get ∂πj+1 , ∂ j+1
2π
s
2 based on ξk+1 ,γk+1 and Ñ k+1
Eqns. (87), (119)-(124)
∂A ∂A
if k > maxIter or ∂π /∂A j+1
< tolNwtn then
exit
else
calculate ξk+1 , γ , Ñ s
k+1 if in A
j+1
k+1
if ξ j + ξk+1 < 0 then
, γ
rescale ξk+1 s
k+1 and Ñ k+1 s.t. ξk+1 = ξ min
hard
end if
update ξk+1 = ξk + ξk+1 , γ s s s
k+1 = γk + γk+1 , Ñ k+1 = Ñ k + Ñ k+1 according to A
j+1
end if
end loop
if ξk+1 goes in dir. not matching gA /gM then
, γ
rescale ξk+1 s
k+1 and Ñ k+1
xiJustDeactivated ← true
else
xiJustDeactivated ← false
end if
ξ j+1 = ξ j + ξk+1 , γ j+1 = γ j + γ s j+1 = Ñ s j + Ñ s
k+1 and Ñ k+1
end loop
using a Newton scheme. First, we compute our residual and stiffness concerning active set A j+1 , i.e., only
derivatives with respect to the active set are computed. If the maximum iterations are exceeded or the norm of
the residual is lower than the tolerance, we exit the loop. Otherwise, we check if updating ξ j+1 would result
in a negative ξ when applying the Newton step. If this is the case, we rescale, i.e., multiply the increments
by a scalar such that ξk+1 hard , which is 5 × 10−5 in this work. Subsequently, we compute ξ
is set to ξ min k+1 ,
s
γ j+1 and Ñ j+1 . In fact, we do not update ξ j+1 , γ j+1 or Ñ s j+1 yet. We update them only after
exiting the Newton scheme and being sure that ξ incremented in a direction matching gA and gM , which were
calculated before the Newton loop. If, due to a bad starting solution, this is not the case, we rescale the Newton
step such that ξ j+1 = ξn . If this happens, we make sure that in the next iteration of the active set algorithm,
we won’t activate ξ again.
4 Numerical results
In this section, the previously presented model is tested for different examples. First, to test the model’s time
convergence behavior and to show the superelastic as well as martensite reorientation behavior at high and low
temperatures, respectively, thermomechanical Gauss point evaluations are conducted. Finally, a full actuator
model is simulated. To cope with the thin structures occurring in the actuator, the SMA model at hand is
embedded into a hexahedral element formulation with reduced integration and hourglass stabilization for the
displacement, while the heat conduction terms are fully integrated.
For the Gauss point evaluations, thermal expansion as well as the transformation induced volume change,
which is discussed in Sect. 2.3.4, are neglected for simplicity. The simulations are conducted using a reduced
integration hexahedral element with hourglass stabilization, which is embedded into the finite element program
FEAP [46]. The material constants used in the Gauss-point are given in Table 1.
Here, it is noted that θ refA and θ refA are chosen such that V = 0.47% (see Eq. (23)), matching the
findings reported in Potapov et al. [36]. The numerical parameters are summarized in Table 2.
Figure 4 shows a tensile test at a temperature of 270 ◦ C for 49 and 40,000 time steps.
Clearly, convergence with regard to time step width is not an issue, as both time step widths yield accurate
results.
To demonstrate that the model results indeed do not depend on the numerical parameters, we compare
results for two different sets of numerical parameters in a tensile test. The first set of numerical parameters is
the one given in Table 2, which is used in the remainder of this paper. The second set is defined in Table 3 just
Elastic constants
λA [GPa] λM [GPa] μA [GPa] μM [GPa]
23.076 13.461 34.615 20.192
Dissipation potential parameters
θ0 [◦ C] Ms [◦ C] Mf [◦ C] As [◦ C] Af [◦ C] σ reo [MPa]
80 70 50 120 140 160
Chemical and hardening parameters
s AM [MPaK−1 ] E int [MPa] k [–]
0.383 30 0.12
Thermal constants
αM [10−6 K−1 ] αA [10−6 K−1 ] c [mJ mm−3 K−1 ] κ [W m−1 ] θ refM [◦ C] θ refA [◦ C]
6.6 11.0 2.9 8.6 588.28 500.0
3076 M. Sielenkämper, S. Wulfinghoff
Numerical parameters
c [–] Hp [MPa] A [MPa] B [–] ξ0 [–] [MPa] ηξ [MPas−1 ]
Fig. 4 Tensile test for 49 and 40,000 time steps. No error due to large load steps is visible
for this comparison. The results are shown in Fig. 5. Clearly, the numerical parameters do not have a noticeable
effect on the results.
Next, we modeled tensile tests at 50 ◦ C, 100 ◦ C, 160 ◦ C and 200 ◦ C. The resulting stress–strain curves are
shown in Fig. 6.
For low temperatures, i.e. 50 ◦ C and 100 ◦ C, the model predicts a remaining martensite reorientation. On
the other hand, for rather high temperatures, the model captures superelastic material behavior when unloading.
However, we want to emphasize that these results do not consider any damage or plastic deformations which
might already occur.
Finally, the model captures the shape memory effect well. This is shown in Fig. 7, where we applied different
stresses (160 MPa, 300 MPa, 400 MPa and 500 MPa) and then started a thermal cycle.
Numerical parameters
c [–] Hp [MPa] A [MPa] B [–] ξ0 [–] [MPa] ηξ s−1
Fig. 5 Investigation of the numerical parameters influence (set 1 is given in Table 2, set 2 in Table 3)
A thermomechanical finite strain shape memory alloy model 3077
τ11 = 160 MPa τ11 = 300 MPa τ11 = 400 MPa τ11 = 500 MPa
14
12
10
E11 [%]
8
6
4
2
0
50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300 50 100 150 200 250 300
Temperature [°C] Temperature [°C] Temperature [°C] Temperature [°C]
Fig. 7 Thermal cycling tests with different stresses (160 MPa, 300 MPa, 400 MPa and 500 MPa)
A 0.2
¯t̂
¯t̂
1
4 z
x
y
Fig. 8 Sketch of the plate with a hole subjected to tension (dimensions in mm)
For 160 MPa, one obtains only twinned martensite, i.e., E t = 0 in the context of this work. Therefore, one
only gets a small hysteresis due to the different elastic constants of austenite and martensite. For the larger
prestresses, the austenite is transformed into detwinned martensite, i.e., E t = 0. This leads to the typical
shape memory effect.
In this example, we simulate a plate with a cylindrical hole (see Fig. 8), which is at first loaded by a traction
¯t̂, then unloaded and subsequently heated, which lets it recover the initial shape.
The material and numerical parameters are unchanged from Sect. 4.1, except the reference temperatures
are now θrefA = 80 ◦ C and θrefM = 87.66 ◦ C. Due to the symmetry, using appropriate symmetry conditions,
only one-eighth of the entire plate is simulated.
During the loading by the traction, the temperature at the upper and lower end is held constant at the initial
temperature of θ̄ = 80 ◦ C (Ms < θ̄ < As ). Additionally, in the initial state, the plate is fully austenite. First,
the traction ¯t̂ is increased linearly to a maximum of 330 MPa in longitudinal direction over the duration of 50s.
Subsequently, the traction ¯t̂ is decreased to zero over the duration of 50s. Finally, the temperature at both ends
of the plate is increased to 200 ◦ C over the course of 20s.
The mesh used in the eighth of the plate consists of 2907 uniformly distributed elements and is shown in
Fig. 9.
For the entire simulation, 293 time steps were required, taking a total CPU-time of 2524s.
3078 M. Sielenkämper, S. Wulfinghoff
(a)
(b)
(c)
Fig. 10 Temperature and martensite volume fraction at different loading stages (see text)
At first, during the loading, almost the entire plate is transformed to martensite. The plate at maximum
traction is shown in Fig. 10a. Here, the E x x strain at the edge of the hole reaches roughly up to 18% (see
Fig. 11).
Additionally, the plate heats up slightly due to the latent heat and mechanical dissipation (Fig. 10a).
However, most of the heat is conducted out of the plate at the temperature Dirichlet boundaries due to the small
size of the plate and the long simulated time5 . Subsequently, after unloading, the plate does not transform
back, which is shown in Fig 10b. Finally, when increasing the temperature at both ends, the martensite is
transformed back to austenite, leading to a recovery of the initial shape (see Fig. 10c). Further, Fig. 12
shows the temperature and displacement of point A (see Fig. 8) over time. Here, the increase in slope of
the displacement at roughly 15s is the point where we have a forward transformation in the vicinity of the
hole. Then, the second change in slope of the displacement at roughly 36s is the rest of the plate starting to
undergo forward transformation. At 50s, when releasing the traction, the displacements decrease. Then again,
due to thermal expansion, the displacement increases slightly after increasing the temperature at the ends at
100s. However, when the temperature reaches the backward transformation temperatures, the material almost
5 A separate simulation with “zero heat flux” boundary conditions led to a temperature increase in approximately 50 ◦ C.
A thermomechanical finite strain shape memory alloy model 3079
recovers the initial shape. The nonzero displacement at t = 120s stems from thermal expansion and the V
effect.
Overall, the examples have been chosen such that the stresses in the vicinity of the hole reach unphysical
values, at which a real material is expected to show irrecoverable strains or even fracture. Thus, this simulation
rather serves as an example showing that the implementation is able to find a solution, even under severe loads.
In this section, we model a bistable shape memory microactuator using in the finite element program FEAP [46].
The actuator concept was published by Winzek et al. [50] for large structures and is built with three layers: a
bottom layer of molybdenum, a middle layer of NiTiHf and a top layer of polymethyl methacrylate (PMMA)
(Winzek et al. [50] used PMMA layers on both sides). For simplicity, a hexahedral element formulation with
reduced integration and hourglass stabilization for the displacements is used for all materials. The material
and numerical parameters for the shape memory alloy remain unchanged (see Tables 1 and 2).
The actuator works through interlacing of the large hysteresis of NiTiHf with the polymers hysteresis. This is
shown in Fig. 13.
It is actuated by joule heating to specific temperatures and cooling back to room temperature. Depending
on the heat cycle, the martensite state or austenite state is held in place by the polymer. This is understood best
when looking at the following example, which is depicted in Fig. 14, where we neglected the CTE of PMMA
Fig. 13 Schematic of the interlaced hysteresis of the NiTiHf-molybdenum films and the polymer. Dashed line for the polymer
stiffness, solid line for the SMA film stress
3080 M. Sielenkämper, S. Wulfinghoff
Fig. 14 A bimorph with SMA and molybdenum (left) in comparison to the bistable actuator with an additional layer of PMMA
(right). The temperatures are indicated by different thermometer colorings
Fig. 15 Zoom in into the layers of the actuator (molybdenum in red, NiTiHf in blue and PMMA in orange)
for simplicity. Here, we compare a bimorph of SMA and molybdenum on the left to the proposed actuator on
the right. When heating from ambient temperature (blue, Fig. 14a) to a temperature above Af (red), the glass
transition temperature is reached, which drastically decreases the stiffness of the PMMA. Subsequently, the
SMA in the bimorph as well as the bistable actuator reach reverse transformation temperatures, which makes
them bend up (Fig. 14b). Now, upon cooling, the actuator reaches θg , which ’freezes’ it’s current shape. Thus,
when subsequently reaching forward transformation temperatures, unlike the bimorph, it will not revert into
the original shape at room temperature (Fig. 14c). Now, we can heat the actuator to a temperature above θg but
below As (orange in Fig. 14d) to soften the polymer, which makes the actuator adopt the deformation of the
bimorph. When now cooling back to room temperature again, the actuator is in it’s initial stable state again and
the actuation cycle could be started again. The two stable states at room temperature are given by Figs. 14c, e.
For the actuation behavior, it is important to have a homogeneous temperature profile in the actuator. Therefore,
a wing-shape actuator developed in Arivanandhan et al. [2] is used to obtain rather homogeneous temperatures
in the double beam cantilever.
To rapidly optimize the actuators properties, it turned out that building and modeling only the beam section
of the actuator is advantageous. Therefore, we simulated just the beam part with suitable boundary conditions
instead of simulating the entire wing structure including the attached wafer material. This thin film actuator
with a zoom into the layers is shown in Fig. 15, where the Dirichlet boundary condition for the displacements
and the Robin boundary condition for the temperature are indicated as well.
A thermomechanical finite strain shape memory alloy model 3081
Since the interest in modeling the actuator lies rather in the states at θ1 , θ2 and θ3 , and less in the states
between them, we chose a very simple polymer model. It is governed by a thermally coupled viscoelastic
Maxwell model for finite strains (Young’s modulus and Poisson’s ratio are 500 MPa and 0.4, respectively),
where the viscosity is high (107 MPas) at low temperatures and low (1 MPas) at high temperatures. Therefore,
it has almost no stiffness at high temperatures while it behaves almost elastically at low ones. Additionally,
the glass transition temperature is given by θg = 77◦ C. The molybdenum is modeled with a thermally coupled
Neo-Hookean elastic model. For the molybdenum, Young’s modulus, Poisson’s ratio, the thermal expansion
coefficient as well as the reference temperature are E = 65 × 103 MPa, ν = 0.31, α = 5 × 10−6 K−1 and
θref = 500 ◦ C, respectively.
The actuator considered is shown in Fig. 15. It is clamped on the left side. It has a length of 10 mm, a width of
5 mm. The layer thicknesses of the molybdenum and TiNiHf are for this work 20 um and 10 um, respectively.
The polymer layer thickness is 160 um for now, before different layer thicknesses are compared in Sect. 4.3.4.
For the initial conditions, we assume zero displacements as well as a temperature of 500 ◦ C, i.e., the temperature
at which the actuator is thermally annealed in a flat state. Furthermore, we assume the material to be in its
austenite state at t = 0, which directly implies that initially C i = I. Subsequently, it is cooled down to room
temperature at 20 ◦ C, which bends the actuator due to the mismatch in the coefficients of thermal expansions
and difference in cell volume between the austenite and martensite phase.
The heating cycle is realized through applying a heat source in the Mo and NiTiHf material. The heat
source magnitude is modeled by a sine function, which is cut off when below 0. It is given by
The polymer layer thickness plays a key role for this actuator design—if it is too thin, the high temperature
shape cannot be held at room temperature by the polymer layer. On the other hand, it should not be too thick,
since body forces and manufacturing problems come up in that case. Further, since the polymer insulates
3082 M. Sielenkämper, S. Wulfinghoff
Fig. 16 Stroke (left y-axis) and temperature (right y-axis, dashed in black) over time for different amounts of elements over the
length of the actuator
Fig. 17 Stroke (left y-axis) and temperature (right y-axis, dashed in black) over time for different amounts of elements over the
thickness of each material layer
Fig. 18 Stroke and temperature over time for V = 0.49% and V = 0.0%
the actuator thermally to one side, thinner polymer layers lead to the possibility of faster actuation cycles.
Therefore, we tested several polymer layer thicknesses, the results are shown in Fig. 19.
First, at
1 (see Fig. 19), the actuator is heated. At first, the larger CTE of the polymer makes the actuator
bend down. Then, when reaching θg , the polymer softens up and the actuator relaxes. Subsequently, at , 2 the
actuator is cooled down to room temperature by the surrounding air. The polymer hardens again and the device
reaches stable configuration A at room temperature. Now, when heating again, the actuator bends down due to
the large CTE of the PMMA at . 3 As soon as the temperature reaches θg , the polymer softens up. Afterward,
at 4 the shape memory alloy reaches Af and the martensite is transformed back to austenite, which makes
the actuator deflect up. Finally, when we remove the source at , 5 the polymer hardens before Mf is reached.
Thus, depending on the polymer thickness, the polymer may hold the actuators shape or release some of it’s
stroke. Finally, the actuator reaches it’s second stable state B at room temperature, which is also it’s initial
state.
A thermomechanical finite strain shape memory alloy model 3083
Fig. 19 Stroke (left y-axis) and temperature (right y-axis, dashed in black) over time for different PMMA layer thicknesses
For many actuator designs, one wants to maximize the achievable stroke while keeping the device as small
and power efficient as possible. Therefore, we try to maximize the difference in stroke between states
A and
.
B In turn, one must choose a sufficient polymer thickness, such that the actuator does not release too much
stroke at .
5 For example, 0.1mm of PMMA is not thick enough to hold the shape, while 0.3 mm holds the
shape perfectly (see Fig. 19 at ).
5 Depending on the actuation frequency in mind and necessary achievable
stroke, either value in between might be a suitable choice.
In this paper, a new thermomechanical shape memory alloy model for finite strains is presented. It uses a
projection method to fulfill the incompressibility constraint for the inelastic stretches. Further, the model is
realized in the generalized standard material formulation being extended to thermomechanics. The optimiza-
tion of the global, incremental mixed thermomechanical potential by variation yields the mechanical balance
principles as well as the evolution equations of the internal variables and boundary condition integrals. The
presented model employs a thermal strain formulation for the shape memory alloy which allows to describe
the transformation induced volume changes found in some shape memory alloys. Using a logarithmic strain
formulation, a finite strain dissipation potential incorporates vanishing inelastic strains upon austenite transfor-
mation in a manner consistent with the time-continuous case. Additionally, yield and transformation criteria
as well as the algorithmically consistent tangent for the coupled problem are given and discussed. Due to
numerical difficulties, a regularization of the hardening energy term is implemented.
The numerical results show, that the model is capable of producing reasonable results in tensile tests. The
dependency on the temperature at which the tensile tests are carried out is captured accurately. Furthermore,
it enables the solution of thin film problems.
Using the model, we are now able to estimate viable layer thicknesses and device sizes as well as fitting
joule heating parameters. Additionally, it turned out that only applying polymer to one side increases the
maximum actuation frequency while also introducing more deflection between the stable states and requiring
more power. Moreover, we found that the V effect as well as the thermal expansion of the SMA, due to the
design being very sensitive to volume changes, is fairly important for the actuators stroke and therefore needs
to be modeled accurately as well. Furthermore, the model results show that the bistability at room temperature,
which is enabled by the polymer locking in the deformation, is achievable for the shown actuator concept.
In the future, it remains interesting to find ways of solving the model problem without any penalty terms,
which are not reasoned physically and are numerically challenging to deal with. A possible solution to this
problem could be to make use of the Fischer-Burmeister [12] complementary function. In fact, this has been
done in Auricchio et al. [4] for shape memory alloys and in Brepols et al. [7] for damage-plasticity with great
success. Furthermore, there is still space for improvements to the algorithms convergence behavior which could
increase the speed of the proposed material model, especially under severe loading conditions. Additionally,
an inclusion of functional fatigue properties, e.g., a shifting of the TiNiHf transformation temperatures over
several transformation cycles into the model would enable to predict the long-term behavior of the actuators.
3084 M. Sielenkämper, S. Wulfinghoff
Acknowledgements Financial support of subproject A3 Cooperative Actuator Systems for Nanomechanics and Nanophotonics:
Coupled Simulation of the Priority Programme SPP 2206 by the German Research Foundation (DFG) (Grant WU847/3-1) is
gratefully acknowledged by the authors.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use,
sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original
author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other
third party material in this article are included in the article’s Creative Commons licence, unless indicated otherwise in a credit
line to the material. If material is not included in the article’s Creative Commons licence and your intended use is not permitted
by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To
view a copy of this licence, visit http://creativecommons.org/licenses/by/4.0/.
Declarations
Conflict of interest The authors have no competing interests to declare that are relevant to the content of this article.
Appendix
which is independent of Ñ s .
With the definition of the tensor
∂ Ns 1 s
PNs = = I − N s ⊗ Cni−1 N s Cni−1 , (84)
∂ Ñ s
Ñ s
Uni−1
C First derivatives of π
To minimize the potential using the active set algorithm including the Newton scheme, we need to obtain first
and second-order derivatives of the potential with respect to the set of internal variables and temperature. The
differential of the discretized potential with respect to the internal variables (u and θ are fixed) is given by
(53),(59) ∂(ψe + ψh + ψr + ψp ) ∂ Cri ∂α ∂(ψe + ψh + ψc + ψp )
dπ = + : + σ reo + + Q(ξn , ξ )
∂ Ĉ
i ∂ξ ∂ξ ∂ξ
∂ Cri H− (ξ )
= ln(Cni )Cri , (89)
∂ξ ξn
where H− (ξ ) is the double negative Heaviside function: H− (ξ ) = −H (−ξ ). The remaining occurring
derivatives are given by
∂(ψe + ψh + ψp ) (50) ∂(ψe + ψh + ψp )
= P
i : , (90)
∂ Ĉ i ∂ Ci
3086 M. Sielenkämper, S. Wulfinghoff
∂ψe ∂ψe
= −C i−1 F T e FC i−1 , (91)
∂C i
∂b
∂ψe (49) λ(ξ ) μ(ξ ) e−1 μ(ξ )
= det(b e
) − 1 − b + I, (92)
∂ be 4 2 2
∂ψe 1 ∂λ 1 ∂μ
= det(be ) − 1 − 2 ln J e + tr(be ) − 3 − 2 ln J e , (93)
∂ξ 4 ∂ξ 2 ∂ξ
−2
∂λ 1−ξ ξ −1 1
=− + + , (94)
∂ξ λA λM λA λM
∂μ 1−ξ ξ −2 −1 1
=− + + , (95)
∂ξ μA μM μA μM
∂ψc (20)
= (θ − θ0 )s AM , (96)
∂ξ
∂ψh (21) ∂ψh ∂ E t ∂ E i ∂ E i
= : (97)
∂ Ci ∂ Et ∂ Ei ∂ Ei ∂ C i
1 1 ∂ψh ∂ E i
= , (98)
2 ξ ∂ Et ∂ Ei
t
∂ψh (68) k E int ξ 2 E1−+2 Et 5
E t ≤ c,
= E t 4
(99)
∂ Et k E int ξ a1 E t > c,
∂ Ei 1 Ei
= 2 , (100)
∂ Ei k 3 Ei
t 2
∂ψh ∂ψh ∂ E t k E int 1−EE t 4 E t ≤ c,
= + (101)
∂ξ ∂ E t ∂ξ k E int (a1 E t + a0 ) E t > c,
∂ Et − Ei
= , (102)
∂ξ ξ2
1 1 ∂ψp ∂ E i
= : , (106)
2 ξ ∂ Et ∂ Ei
∂ψp HplF √ ξ2 ξ ≥ ξ0 − c E i ,
= c +1 (107)
∂ Et Hp ξ 2 E t else,
⎧
∂ψp ⎨ HplF √E t −c
ξ ≥ ξ0 − c E i ,
= c2 +1 (108)
∂ξ ⎩ H (ξ − ξ + ξ E t 2 ) else,
p 0
A thermomechanical finite strain shape memory alloy model 3087
∂α 1 ∂ Cri
= 2
M̃C i : (Pi − Is ) : , (109)
∂ξ 3 2 ∂ξ
∂α
= 2
3 M̃C i : Pi : N s , (110)
∂γ
∂α
= 3 γ P N s
2
: P
i : M̃C i , (111)
∂ Ñ s
∂ r C i U i−1 Cni−1 r C i Cni−1
M̃C i = n
= . (112)
∂ Ci r C i i−1
Un
Here, a1 is the slope of the linear approximation in Fig. 2 at E t = ck and a0 is the value of the function ϕ at
that same point. Further, the terms related to ψp are set to zero if ξ > E i /c + ξ0 .
D Second derivatives of π
The second derivatives of the potential π are necessary to compute the algorithmic tangent. They are briefly
given in this appendix. First, we define the symmetrizing box product as
s
A B : C = Asym(C)B, (113)
where A and B are arbitrary second-order tensors. Now, we compute the differential of the operator P N s :
d(P
N s : B) = D N s (B) : d Ñ + P N s : dB,
s (114)
where B is an arbitrary symmetric second-order tensor and
1
D N s (B) = − 3 f (B) ⊗ Cni−1 Ñ s Cni−1 + Cni−1 Ñ s Cni−1 ⊗ f (B)
s
Ñ i−1
Un
(115)
1 s
− 2 ( Ñ : B)Cn Ñ Cn ⊗ Cn Ñ Cn + ( Ñ : B)Cn Cn
s i−1 s i−1 i−1 s i−1 s i−1 i−1
.
s
Ñ i−1
Un
Here, B is defined as
1
f (B) = B − 2 ( Ñ s : B)Cni−1 Ñ s Cni−1 . (116)
s
Ñ i−1
Un
Similarly, we define the derivative of Pi via
d(P
i : B) = Di (B) : d Ĉ + Pi : dB,
i (117)
with
1 ˆ i− 13 1 i 1 i
Di (B) = − III B − (Ĉ : B)Ĉ i−1
⊗ Ĉ i−1
+ Ĉ i−1
⊗ B − (Ĉ : B)Ĉ i−1
3 3 3
(118)
s 1
− (Ĉ i : B) Ĉ i−1 Ĉ i−1 − Ĉ i−1 ⊗ Ĉ i−1 .
3
Subsequently, we find for the derivatives of π :
∂ 2 π ∂ Cri ∂(ψe + ψh + ψp ) ∂ 2 (ψ + ψ + ψ )
e h p ∂ 2ψ
r ∂ Cri
= : D i + Pi : : Pi + :
∂ξ 2 ∂ξ ∂ Ci ∂ C i2 ∂ Ĉ i
2 ∂ξ
3088 M. Sielenkämper, S. Wulfinghoff
∂ 2 π ∂ 2 (ψ + ψ + ψ )
∂ 2ψ ∂ Cri reo ∂ α
2
= 2N s : P
e h p r
i : : P i + : + σ
∂ξ ∂γ ∂ C i2 ∂ Ĉ i
2 ∂ξ ξ γ
(122)
∂ 2 (ψ + ψ + ψ )
+ 2N s : P
e h p
i : ,
∂ξ ∂ C i
∂ 2 π ∂ (ψe + ψh + ψp ) ∂ 2 ψr ∂ Cri ∂ 2 α
2
= P N s : 2γ Pi : : Pi + : + σ reo
∂ξ ∂ Ñ s ∂ Ci 2
∂ Ĉ i
2 ∂ξ ∂ξ ∂ Ñ s
(123)
∂ (ψe + ψh + ψp )
2
+ 2γ P N s : Pi : ,
∂ξ ∂ C i
∂ 2 π reo ∂ α ∂ (ψe + ψh + ψp ) ∂ 2 ψr
2 2
=σ + P N s : 2γ Pi : : Pi + : 2N s
∂γ ∂ Ñ s ∂γ ∂ Ñ s ∂ C i2 ∂ Ĉ i
2
(124)
∂(ψ e + ψ h + ψ p ) ∂ψ r
+ 2 Pi : + .
∂ Ci ∂ Ĉ i
Here, for the occurring derivatives we find:
∂ 2 ψe 1 ∂ 2λ 1 ∂ 2μ
= det(be
) − 1 − 2 ln J e
+ tr(be ) − 3 − 2 ln J e , (125)
∂ξ 2 4 ∂ξ 2 2 ∂ξ 2
∂ 2λ 1−ξ ξ −3 −1 1 2
= 2 + + , (126)
∂ξ 2 λA λM λA λM
∂ 2μ 1−ξ ξ −3 −1 1 2
=2 + + , (127)
∂ξ 2 μA μM μA μM
∂ 2 ψe ∂ 2 ψe
= −Bei : , (131)
∂ξ ∂ C i ∂ξ ∂ be
∂ 2 ψe 1 ∂λ 1 1 ∂μ
= J − e
e e e−1
J b + (I − be−1 ), (132)
∂ξ ∂ be 4 ∂ξ J 2 ∂ξ
∂ 2 Cri H− (ξ ) i 2 i
= C ln (Cn ), (133)
∂ξ 2 ξn2 r
∂ 2 ψr ∂ 2 ψr ˆ i ∂ψr ˆ i Ĉ i−1 ⊗ Ĉ i−1 − ∂ψr III ˆ i Ĉ i−1 s Ĉ i−1 ,
= III + III (134)
2
ˆ i2 ˆi
∂ III ˆ
∂ III i
∂ Ĉ i ∂ III
⎛ ⎛ ⎞ ⎞
∂ 2ψ r ⎜ ⎜
s 1 ⎟ 1 ⎟
= A ⎝Cni−1 Cni−1 ⎝1 − ⎠+ Cni−1 N s Cni−1 ⊗ Cni−1 N s Cni−1 ⎠ , (135)
2 s
∂ Ñ s Ñ i−1
Ñ s
Un Uni−1
2
∂ 2 ψh ∂ 2 ψh ∂ E t ∂ 2 ψh ∂ Et ∂ψh ∂ 2 E t
= 2 + + , (136)
∂ξ 2 ∂ E t ∂ξ ∂ξ ∂ Et
2 ∂ξ ∂ E t ∂ξ 2
⎧
⎨k E int 2 E +2t
t Et 5
∂ 2ψ h (1− E 4 )2 E t ≤ c,
= (137)
∂ E t ∂ξ ⎩ int
k E a1 E t > c,
⎧
⎨k E int ξ 2+24 E t+64 3E
t 4 t 8
∂ 2 ψh (1− E )
E t ≤ c,
= (138)
∂ Et
2 ⎩
0 E t > c,
∂ 2 Et 2 Ei
= , (139)
∂ξ 2 ξ3
∂ 2 ψh 1 ∂ 2 ψh ∂ 2 ψh E i 1 ∂ψh ∂ Ei
= − − , (140)
∂ C ∂ξ
i 2ξ ∂ E ∂ξ
t
∂ Et ξ
2 2 ξ ∂ Et ∂ Ei
∂ 2 ψh 1 ∂ 2 ψh 1 ∂ E i ∂ Ei ∂ψh 1 ∂ 2 E i
= ⊗ + , (141)
∂ C i2 ∂ Et ξ ∂ E ∂ Ei ∂ Et ξ ∂ Ei2
4 2 2 i
∂ 2 Ei 1 Ei Ei
= 2 Is − ⊗ , (142)
∂ Ei2
3
k Ei Ei Ei
⎧ 2
⎪
⎨ Hp √E−c
t
∂ 2 ψp ξ ≥ ξ0 − c E i ,
= 1+c2 (143)
∂ξ 2 ⎪
⎩ H (1 + E t 2 ) else,
p
∂ 2 ψp 1 ∂ 2 ψp 1 ∂ E i ∂ Ei ∂ψp 1 ∂ 2 E i
= ⊗ + , (144)
∂ C i2 4 ∂ Et 2 ξ 2 ∂ Ei ∂ Ei ∂ Et ξ ∂ Ei2
3090 M. Sielenkämper, S. Wulfinghoff
∂ 2 ψp 1 ∂ 2 ψp ∂ 2 ψp E i 1 ∂ψp ∂ Ei
= − − , (145)
∂ C ∂ξ
i 2ξ ∂ E ∂ξ
t
∂ Et ξ
2 2 ξ ∂ Et ∂ Ei
⎧ 2
∂ 2 ψp ⎨ Hp √ ξ
ξ ≥ ξ0 − c E i ,
= 1+c2 (146)
∂ Et
2 ⎩
Hp ξ 2 else,
⎧ t
⎨ Hp √E −c √ ξ
∂ 2ψ p
+ lF √ 1
ξ ≥ ξ0 − c E i ,
1+c2 1+c2 1+c2
= (147)
∂ξ ∂ E t ⎩
2Hp ξ E t else,
∂ 2 α ∂ Cri 1 ∂ M̃ i 1 ∂ Cri
C
= 2
: Pi − I : s
: Pi − I s
+ D i ( M̃ i ) :
∂ξ 2 ∂ξ
3 2 ∂ Ci 2 C ∂ξ
(148)
1 ∂ 2 Cri
+ M̃C i : (Pi − Is ) : ,
2 ∂ξ 2
∂ 2 α 1 1 ∂ M̃C i
= 2
3 2N
s
: Di M̃ i + Pi : : Pi : 2N s , (149)
∂γ 2 2 C 2 ∂ Ci
∂ 2 α 1
= 2γ D N s P
2
: M̃ i
∂ Ñ s
2 3 i
2 C
(150)
2 1 1 ∂ M̃ C i
+ (2γ ) P N s : Di M̃ i + Pi : : Pi : P N ,
s
2 C 2 ∂ Ci
∂ 2 α 1 ∂ Cri 1 ∂ M̃C i ∂ Cri
= 2
3 2N
s
: Di M̃ i : + P
i : : (Pi − Is ) : , (151)
∂ξ ∂γ 2 C ∂ξ 2 ∂C i ∂ξ
∂ 2 α 1 ∂ Cri 1 ∂ M̃C i ∂ Cri
= 2
3 2γ P N s : Di M̃ i : + P : : (Pi − I s
) : , (152)
∂ξ ∂ Ñ s 2 C ∂ξ i
2 ∂ Ci ∂ξ
∂ 2 α 1 1 1 ∂ M̃C i
= 2
2P N s : Pi : M̃C i + 2γ P N s : Di M̃C i + Pi : : Pi : 2N s
, (153)
∂γ ∂ Ñ s 3 2 2 2 ∂ Ci
∂ M̃C i 1 s
= Cni−1 Cni−1 − M̃C i ⊗ M̃C i . (154)
∂C i
Ñ s
Uni−1
A thermomechanical finite strain shape memory alloy model 3091
E Consistent tangent
When using the finite element method to solve boundary value problems, we make use of the consistent tangent
operator. To derive it with respect to dd from the potential, we start with the virtual work of the internal forces
τ : dδ dV. (155)
V0
Here, dδ is defined as
dδ = sym(Grad(δu) F -1 ). (156)
lδ
Now, we need to calculate the differential of virtual work of the internal forces:
d(τ : dδ )dV, (157)
V0
where the first term, i.e., l d τ l δT is related to the so-called geometric tangent and the second term arises from
the incremental material stiffness. Further, l d and dd are defined in analogy to l δ and dδ . Here, we neglected,
in a first step, the influence of perturbations dθ of the temperature (for the full linearization, see below). Now,
we can split 𝕔a into the elastic and inelastic parts:
s ∂S s ∂(F SF T ) ∂A
𝕔a : dd = F F T : : F T F : dd + : : dE
∂E ∂A ∂E (160)
∂τ ∂A T
= 𝕔 : dd +
e
: : F dd F,
∂A ∂ E
where (with a small abuse of notation) A denotes the vector of active variables and may contain ξ , γ and
Ñ s . Additionally, the elastic tangent 𝕔e is given by
∂π
=0 (162)
∂A
holds in the converged state. Now, calculating the differential of Eq. (162), we obtain
∂π
0=d
∂A
∂ π
2 ∂ 2 π
= dA + : dE = 0 (163)
∂A2 ∂A∂ E
∂ 2 π ∂S T
= dA + : dE = 0.
∂A2 ∂A
3092 M. Sielenkämper, S. Wulfinghoff
∂τ ∂τ
=2 : N s, (168)
∂γ ∂ Ĉ i
∂τ ∂τ
= 2γ : PNs , (169)
∂ Ñ s ∂ Ĉ i
2
∂τ s ∂ψe ∂ψe s s s ∂ ψe
= −2 I + e
I + I be
+ be
I :
∂ Ĉ i ∂ be ∂ b ∂ be 2 (170)
s s
T
:FF :C i−1
C
i−1
: Pi .
Additionally, the algorithmic consistent tangent with respect to θ as well as the mixed contributions can be
computed through linearizing the weak form in Eq. (74):
∂π κθ
d δθ − Grad(θ ) · Grad(δθ ) dV
∂θ θn
V0
−1
∂ 2 π ∂ 2 π ∂ 2 π c ∂τ κt
= δθ − − dθ + : dd − Grad(δθ ) Grad(dθ ) dV. (171)
∂A∂θ ∂A2 ∂A∂θ θ ∂θ θn
V0
with T
∂ 2 π σ reo ∂α Q H ξ σ reo ∂α σ reo ∂α T
= s AM
+ + + ; ; . (173)
∂A∂θ θn ∂ξ θn θn θn ∂γ θn ∂ Ñ s
Now, using standard formulations for the element ansatz functions N and its gradients B, one can assemble
the element stiffness matrix (see, e.g., Wriggers [51]).
References
1. Arghavani, J., Auricchio, F., Naghdabadi, R.: A finite strain kinematic hardening constitutive model based on Hencky strain:
general framework, solution algorithm and application to shape memory alloys. Int. J. Plast. 27(6), 940–961 (2011). https://
doi.org/10.1016/j.ijplas.2010.10.006
2. Arivanandhan, G., Li, Z., Curtis, S., Velvaluri, P., Quandt, E., Kohl, M.: Temperature homogenization of co-integrated shape
memory-silicon bimorph actuators. Proceedings (2020). https://doi.org/10.3390/IeCAT2020-08501
3. Auricchio, F., Reali, A., Stefanelli, U.: A three-dimensional model describing stress-induced solid phase transformation with
permanent inelasticity. Int. J. Plast. 23(2), 207–226 (2007). https://doi.org/10.1016/j.ijplas.2006.02.012
4. Auricchio, F., Bonetti, E., Scalet, G., Ubertini, F.: Theoretical and numerical modeling of shape memory alloys accounting
for multiple phase transformations and martensite reorientation. Int. J. Plast. 59, 30–54 (2014). https://doi.org/10.1016/j.
ijplas.2014.03.008
5. Bellini, C., Berto, F., Di Cocco, V., Iacoviello, F.: A cyclic integrated microstructural-mechanical model for a shape memory
alloy. Int. J. Fatigue 153, 106473 (2021). https://doi.org/10.1016/j.ijfatigue.2021.106473
6. Bernardini, D., Pence, T.J.: Models for one-variant shape memory materials based on dissipation functions. Int. J. Non-Linear
Mech. 37(8), 1299–1317 (2002). https://doi.org/10.1016/S0020-7462(02)00020-3
7. Brepols, T., Wulfinghoff, S., Reese, S.: Gradient-extended two-surface damage-plasticity: micromorphic formulation and
numerical aspects. Int. J. Plast. 97, 64–106 (2017). https://doi.org/10.1016/j.ijplas.2017.05.010
8. Cisse, C., Zaki, W., Ben Zineb, T.: A review of constitutive models and modeling techniques for shape memory alloys. Int.
J. Plast. 76, 244–284 (2016). https://doi.org/10.1016/j.ijplas.2015.08.006
9. Du, H., Chau, F.S., Zhou, G.: Mechanically-tunable photonic devices with on-chip integrated mems/nems actuators. Micro-
machines 7(4), 69 (2016). https://doi.org/10.3390/mi7040069
10. Eckart, C.: The thermodynamics of irreversible processes. IV. The theory of elasticity and anelasticity. Phys. Rev. 73(4), 373
(1948). https://doi.org/10.1103/PhysRev.73.373
11. Eichenfield, M., Camacho, R., Chan, J., Vahala, K.J., Painter, O.: A picogram-and nanometre-scale photonic-crystal optome-
chanical cavity. Nature 459(7246), 550–555 (2009). https://doi.org/10.1038/nature08061
12. Fischer, A.: A special newton-type optimization method. Optimization 24(3–4), 269–284 (1992). https://doi.org/10.1080/
02331939208843795
13. Flory, P.J., Volkenstein, M.: Statistical mechanics of chain molecules. Biopolymers 8(5), 699–700 (1969). https://doi.org/
10.1002/bip.1969.360080514
14. Frost, M., Benešová, B., Sedlák, P.: A microscopically motivated constitutive model for shape memory alloys: formulation,
analysis and computations. Math. Mech. Solids 21(3), 358–382 (2016). https://doi.org/10.1177/1081286514522474
15. Gall, K., Sehitoglu, H., Chumlyakov, Y.I., Kireeva, I.V.: Tension-compression asymmetry of the stress-strain response in
aged single crystal and polycrystalline niti. Acta Mater. 47(4), 3 (1999). https://doi.org/10.1016/S1359-6454(98)00432-7
16. Govindjee, S., Mielke, A., Hall, G.J.: The free energy of mixing for n-variant martensitic phase transformations using
quasi-convex analysis. J. Mech. Phys. Solids 51(4), I–XXVI (2003). https://doi.org/10.1016/S0022-5096(03)00015-2
17. Halphen, B., Nguyen, Q.S.: Sur les matériaux standard généralisés. (1975)
18. Han, S., Seok, T.J., Quack, N., Yoo, B.-W., Wu, M.C.: Large-scale silicon photonic switches with movable directional
couplers. Optica 2(4), 370–375 (2015). https://doi.org/10.1364/OPTICA.2.000370
19. Hartl, D.J., Chatzigeorgiou, G., Lagoudas, D.C.: Three-dimensional modeling and numerical analysis of rate-dependent
irrecoverable deformation in shape memory alloys. Int. J. Plast. 26(10), 1485–1507 (2010). https://doi.org/10.1016/j.ijplas.
2010.01.002
20. Hoffmann, M., Bezzaoui, H., Voges, E.: Micromechanical cantilever resonators with integrated optical interrogation. Sens.
Actuators A 44(1), 71–75 (1994). https://doi.org/10.1016/0924-4247(94)00776-4
21. Hurtado, D., Stainier, L., Ortiz, M.: The special-linear update: an application of differential manifold theory to the update of
isochoric plasticity flow rules. Int. J. Numer. Methods Eng. 97(4), 298–312 (2014). https://doi.org/10.1002/nme.4600
22. Kohl, M.: Shape Memory Microactuators. Springer, Berlin (2004)
23. Kohl, M., Krevet, B., Just, E., Sma microgripper system. Sens. Actuators A Phys. 97–98, 646–652. https://doi.org/10.1016/
S0924-4247(01)00803-2
24. Kröner, E.: Allgemeine Kontinuumstheorie der Versetzungen und Eigenspannungen. Arch. Ration. Mech. Anal. 4(1), 273
(1959). https://doi.org/10.1007/BF00281393
25. Lagoudas, D.C., Entchev, P.B., Popov, P., Patoor, E., Brinson, L.C., Gao, X.: Shape memory alloys, part II: modeling of
polycrystals. Mech. Mater. 38(5), 430–462 (2006). https://doi.org/10.1016/j.mechmat.2005.08.003
26. Lee, E.H.: Elastic-plastic deformation at finite strains. J. Appl. Mech. 36(1), 1–6 (1969). https://doi.org/10.1115/1.3564580
27. Lester, B.T., Baxevanis, T., Chemisky, Y., Lagoudas, D.C.: Review and perspectives: shape memory alloy composite systems.
Acta Mech. 226(12), 3907–3960 (2015). https://doi.org/10.1007/s00707-015-1433-0
28. Lexcellent, C., Boubakar, M., Bouvet, C., Calloch, S.: About modelling the shape memory alloy behaviour based on the
phase transformation surface identification under proportional loading and anisothermal conditions. Int. J. Solids Struct.
43(3), 613–626 (2006). https://doi.org/10.1016/j.ijsolstr.2005.07.004
3094 M. Sielenkämper, S. Wulfinghoff
29. Mirzaeifar, R., DesRoches, R., Yavari, A., Gall, K.: A micromechanical analysis of the coupled thermomechanical superelastic
response of textured and untextured polycrystalline niti shape memory alloys. Acta Mater. 61(12), 4542–4558 (2013). https://
doi.org/10.1016/j.actamat.2013.04.023
30. Moumni, Z., Zaki, W., Nguyen, Q.S.: Theoretical and numerical modeling of solid-solid phase change: application to the
description of the thermomechanical behavior of shape memory alloys. Int. J. Plast. 24(4), 614–645 (2008). https://doi.org/
10.1016/j.ijplas.2007.07.007
31. Müller, C., Bruhns, O.: A thermodynamic finite-strain model for pseudoelastic shape memory alloys. Int. J. Plast. 22(9),
1658–1682 (2006). https://doi.org/10.1016/j.ijplas.2006.02.010
32. Nemat-Nasser, S.: On finite deformation elasto-plasticity. Int. J. Solids Struct. 18(10), 857–872 (1982). https://doi.org/10.
1016/0020-7683(82)90070-1
33. Otsuka, K., Ren, X.: Physical metallurgy of ti-ni-based shape memory alloys. Prog. Mater Sci. 50(5), 511–678 (2005). https://
doi.org/10.1016/j.pmatsci.2004.10.001
34. Panico, M., Brinson, L.: A three-dimensional phenomenological model for martensite reorientation in shape memory alloys.
J. Mech. Phys. Solids 55(11), 2491–2511 (2007). https://doi.org/10.1016/j.jmps.2007.03.010
35. Patoor, E., Lagoudas, D.C., Entchev, P.B., Brinson, L.C., Gao, X.: Shape memory alloys, part I: general properties and
modeling of single crystals. Mech. Mater. 38(5), 391–429 (2006). https://doi.org/10.1016/j.mechmat.2005.05.027
36. Potapov, P., Shelyakov, A., Gulyaev, A., Svistunov, E., Matveeva, N., Hodgson, D.: Effect of hf on the structure of ni-ti
martensitic alloys. Mater. Lett. 32(4), 247–250 (1997). https://doi.org/10.1016/S0167-577X(97)00037-2
37. Qidwai, M., Lagoudas, D.: On thermomechanics and transformation surfaces of polycrystalline niti shape memory alloy
material. Int. J. Plast. 16(10), 1309–1343 (2000). https://doi.org/10.1016/S0749-6419(00)00012-7
38. Reese, S., Christ, D.: Finite deformation pseudo-elasticity of shape memory alloys—constitutive modelling and finite element
implementation. Int. J. Plast. 24(3), 455–482 (2008). https://doi.org/10.1016/j.ijplas.2007.05.005
39. Rezaee-Hajidehi, M., Tůma, K., Stupkiewicz, S.: Gradient-enhanced thermomechanical 3d model for simulation of trans-
formation patterns in pseudoelastic shape memory alloys. Int. J. Plast. 128, 102589 (2020). https://doi.org/10.1016/j.ijplas.
2019.08.014
40. Rockafellar, R.T.: Convex Analysis. Princeton University Press, Princeton, 2015. ISBN 9781400873173. https://doi.org/10.
1515/9781400873173
41. Scalet, G., Karakalas, A., Xu, L., Lagoudas, D.: Finite strain constitutive modelling of shape memory alloys considering
partial phase transformation with transformation-induced plasticity. Shape Memory Superelast. 7(2), 206–221 (2021). https://
doi.org/10.1007/s40830-021-00330-5
42. Sedlák, P., Frost, M., Benešová, B., Ben Zineb, T., Šittner, P.: Thermomechanical model for niti-based shape memory alloys
including r-phase and material anisotropy under multi-axial loadings. Int. J. Plast. 39, 132–151 (2012). https://doi.org/10.
1016/j.ijplas.2012.06.008
43. Seelecke, S., Müller, I.: Shape memory alloy actuators in smart structures: modeling and simulation. Appl. Mech. Rev. 57(1),
23–46 (2004). https://doi.org/10.1115/1.1584064
44. Sielenkämper, M., Dittmann, J., Wulfinghoff, S.: Numerical strategies for variational updates in large strain inelasticity with
incompressibility constraint. Int. J. Numer. Methods Eng. https://doi.org/10.1002/nme.6855
45. Simo, J., Taylor, R., Pister, K.: Variational and projection methods for the volume constraint in finite deformation elasto-
plasticity. Comput. Methods Appl. Mech. Eng. 51(1), 177–208 (1985). https://doi.org/10.1016/0045-7825(85)90033-7
46. Taylor, R.L.: FEAP—finite element analysis program. http://projects.ce.berkeley.edu/feap/ (2017)
47. Wang, B., Zhu, S.: Cyclic tension-compression behavior of superelastic shape memory alloy bars with buckling-restrained
devices. Constr. Build. Mater. 186, 103–113 (2018). https://doi.org/10.1016/j.conbuildmat.2018.07.047
48. Wang, J., Moumni, Z., Zhang, W.: A thermomechanically coupled finite-strain constitutive model for cyclic pseudoelasticity
of polycrystalline shape memory alloys. Int. J. Plast. 97, 194–221 (2017). https://doi.org/10.1016/j.ijplas.2017.06.003
49. Wang, J., Moumni, Z., Zhang, W., Xu, Y., Zaki, W.: A 3d finite-strain-based constitutive model for shape memory alloys
accounting for thermomechanical coupling and martensite reorientation. Smart Mater. Struct. 26(6), 065006 (2017). https://
doi.org/10.1088/1361-665x/aa6c17
50. Winzek, B., Schmitz, S., Rumpf, H., Sterzl, T., Hassdorf, R., Thienhaus, S., Feydt, J., Moske, M., Quandt, E.: Recent
developments in shape memory thin film technology. Mater. Sci. Eng. A 378(1), 40–46 (2004). https://doi.org/10.1016/j.
msea.2003.09.105
51. Wriggers, P.: Nonlinear Finite Element Methods. Springer, Berlin (2008)
52. Wulfinghoff, S., Böhlke, T.: Equivalent plastic strain gradient crystal plasticity—enhanced power law subroutine. GAMM-
Mitteilungen 36(2), 134–148 (2013). https://doi.org/10.1002/gamm.201310008
53. Xu, L., Solomou, A., Baxevanis, T., Lagoudas, D.: Finite strain constitutive modeling for shape memory alloys considering
transformation-induced plasticity and two-way shape memory effect. Int. J. Solids Struct. 221, 42–59 (2021). https://doi.
org/10.1016/j.ijsolstr.2020.03.009
54. Yang, Q., Stainier, L., Ortiz, M.: A variational formulation of the coupled thermo-mechanical boundary-value problem for
general dissipative solids. J. Mech. Phys. Solids 54(2), 401–424 (2006). https://doi.org/10.1016/j.jmps.2005.08.010
55. Yu, C., Kang, G., Sun, Q., Fang, D.: Modeling the martensite reorientation and resulting zero/negative thermal expansion of
shape memory alloys. J. Mech. Phys. Solids 127, 295–331 (2019). https://doi.org/10.1016/j.jmps.2019.03.015
56. Zaki, W., Moumni, Z., Morin, C.: Modeling tensile-compressive asymmetry for superelastic shape memory alloys. Mech.
Adv. Mater. Struct. 18(7), 559–564 (2011). https://doi.org/10.1080/15376494.2011.605016
57. Zhang, M., Baxevanis, T.: An extended three-dimensional finite strain constitutive model for shape memory alloys. J. Appl.
Mech. 88(11), 08 (2021). https://doi.org/10.1115/1.4051833.111010
Publisher’s Note Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional
affiliations.