Modelling and Simulation in Materials Science and Engineering

Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 (15pp) doi:10.1088/0965-0393/22/8/085011

A three-dimensional non-isothermal
Ginzburg–Landau phase-field model for
shape memory alloys

R Dhote1,3 , M Fabrizio2 , R Melnik3 and J Zu1

1
Mechanical and Industrial Engineering, University of Toronto, 5 King’s College
Road, Toronto, ON, M5S3G8, Canada
2
Department of Mathematics, University of Bologna, Piazza di Porta S. Donato 5,
I-40126 Bologna, Italy
3
The MS2Discovery Interdisciplinary Research Institute & M2 NeT Laboratory,
Wilfrid Laurier University, Waterloo, ON, N2L3C5, Canada

E-mail: [email protected]

Received 12 March 2014, revised 18 August 2014

Accepted for publication 15 September 2014
Published 12 November 2014

Abstract
In this paper, a macroscopic three-dimensional non-isothermal model is
proposed for describing hysteresis phenomena and phase transformations in
shape memory alloys (SMAs). The model is of phase-field type and is based
on the Ginzburg–Landau theory. The hysteresis and phase transformations
are governed by the kinetic phase evolution equation using the scalar order
parameter, laws of conservation of the momentum and energy and a nonlinear
coupling of the stress, the strain and the order parameter in a differential form.
One of the important features of the model is that the phase transformation
is governed by the stress tensor, as opposed to the transformation strain tensor
typically used in the literature. The model takes into account different properties
of austenite and martensite phases based on the compliance tensor as a function
of the order parameter and stress. Representative numerical simulations on an
SMA specimen reproduce hysteretic behaviors observed experimentally in the
literature.

Keywords: shape memory alloy, non-isothermal, Ginzburg–Landau theory,

phase-field model
(Some figures may appear in colour only in the online journal)

0965-0393/14/085011+15$33.00 © 2014 IOP Publishing Ltd Printed in the UK 1

Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

1. Introduction

Over the past few decades, shape memory alloys (SMAs) have attracted increasing
attention from physicists, engineers and applied mathematicians because of their complex
microstructures and interesting thermomechanical hysteretic behaviors. The SMAs exhibit two
unique hysteresis behaviors, namely, a shape memory effect and pseudoelasticity at lower and
higher temperatures (with respect to the threshold temperature). These behaviors are caused
by the underlying atomic rearrangements from a symmetric configuration (called the austenite
(A) phase) to other configurations of lower symmetry (called the martensite (M) phases).
Under mechanical and thermal loadings, the atomic rearrangement results in a macroscopic
deformation of an SMA specimen due to diffusionless transformations. The simultaneous
presence of high-stress and high-strain properties for SMAs makes them suitable candidates
for actuator and sensor use in a wide range of products in automotive, aerospace, medicine
and bioengineering applications [1–6].
Several modeling approaches have been proposed for describing hysteretic behaviors in
SMAs. A comprehensive overview of different SMA models can be found in, e.g. [3, 7, 8].
The approaches based on phenomenology, phase diagrams, micromechanics, crystal plasticity,
phase-field models, etc. have been described in detail in [9–12] and the references therein.
In this paper, we focus on the phase-field (PF) model approach. This approach provides
a unified framework to describe temperature-induced and stress-induced transformations.
Several different PF models have been proposed in [13–30]. They differ in the free energy
description, selection of order parameters (OPs), model formulation and numerical approaches.
Inspired by the phase-field modeling of ferroelectric materials, Falk [13] applied the
Landau–Devonshire theory in order to describe martensitic transformations (MTs) in SMAs
by defining the shear strain as an OP. Later, Wang and Khachaturyan [31] proposed a three-
dimensional (3D) continuum stochastic-field kinetic model by defining the transformation-
induced elastic strain as OPs in order to predict the MTs. Curnoe and Jacobs [32], Lookman
et al [33] and Bouville and Ahluwalia [26, 27] used an approach of defining elastic strain
components as the OPs. Polynomial-based phenomenological descriptions of the free energy as
a function of OPs and their gradients are used to describe the dynamics of phase transformations.
One of the previous notable contributions to the PF theory is the Landau free energy proposed by
Levitas et al [19–21]. The free energy yields a description of the thermomechanical properties
of different phases using the tensorial OPs. The strain tensor is decomposed into an elastic
component and a transformational component, where the latter is a function of the OPs. The
model employs the same number of phase evolution equations as there are martensitic variants
considered during the phase transformation. Later, Mahapatra and Melnik [23, 24] derived a
non-isothermal model based on the free energy developed by Levitas et al [19–21] by modifying
the multivariant framework to obtain strongly coupled thermomechanical models with the
essential properties of frame indifference and material symmetry.
Recently, Berti et al [34, 35], Grandi et al [36], Maraldi et al [37] and Dhote et al [30]
developed a non-isothermal thermodynamic framework for modeling MTs in SMAs. The
macroscopic frameworks for 1D and 3D models have been developed on the basis of a simplified
version of the free energy proposed by Levitas et al [19]. Here, we are particularly interested
in the 3D model proposed by Dhote et al [30]. One of the important features of the 3D model
within the non-isothermal framework is the use of a scalar phase OP instead of the tensorial
OPs used earlier in the literature [19–21]. The application of this approach reduces the problem
size by limiting the number of phase evolution equations to 1, instead of considering separate
equations for each crystallographic variant [19–21]. This results in a simple model which
is amenable to an efficient numerical implementation. The other highlights of the model

2
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

are (i) the rate-dependent constitutive equations coupling the stress, strain and phase order
parameter, (ii) the description of the phase transformation based on the stress tensor and (iii)
the phase-dependent properties obtained by incorporating the compliance tensor based on the
local phase value and stress.
In this paper, the 3D non-isothermal model is implemented and examples in a 2D setting
of study of the SMA behavior are provided. In our earlier publication [30], the numerical
experiments were conducted on the model by numerically solving the kinetic phase evolution
and constitutive equation in 1D and the pseudo-2D case driven by stress loading, but without
incorporating the conservation laws of momentum and energy. The model is now studied here
incorporating the full thermomechanical coupling and phase evolution equations. In addition,
the model is simulated with the material properties of the Ni55 Ti45 specimen [28, 38].
The paper is organized as follows. In section 2, a 3D non-isothermal model is described
using the kinetic phase evolution, nonlinear couplings of the stress, strain and order parameter
and laws of conservation of the momentum and energy. In section 3, representative simulations
on a rectangular SMA specimen are described in detail and studied under different loading
conditions. Finally, the conclusions are given in section 4.

2. The 3D non-isothermal phase-field model

The fully coupled thermomechanical PF model is developed to describe the nonlinear hysteretic
response of SMAs. We define the OP φ to describe the austenite (φ = 0) and martensite
(φ = 1) phases. Here we do not distinguish between different variants of martensites. This is
a different approach as compared to a multivariant OP approach (e.g. see [19–21]). It facilitates
the development of a simpler model which is computationally tractable, but at the expense of
distinction between different martensitic variants individually. In the following section, the
governing equations of the phase evolution and the conservation laws for momentum and
energy are described.

2.1. The phase evolution equation

In order to derive a phase evolution equation, we choose a free energy functional that has
minima at φ = 0 and φ = 1 with no distinction between martensitic variants. The free energy
functional  based on the Ginzburg–Landau potential is given by

  
κ 1  0 σ · σ
 = |∇φ|2 − (σ λ · σ ) + θ0 F (φ) + θ̂ − G (φ) , (1)
2 2 2  |σ |
where κ is the Ginzburg constant, σ is the stress tensor, λ is the compliance tensor,  is
the latent heat of the phase transition, θ0 and θ̂ are the temperatures and 0 is the equivalent
transformation strain. The potentials F (φ) and G (φ) are the 2–3–4 polynomial functions of
φ defined as


0 if φ < 0,
F (φ) = 2 φ − 3 φ + 4 φ + β(φ − φ), G (φ) = 2 φ − 4 φ
1 2 2 3 1 4 2 1 2 1 4
if 0
φ
1, (2)

1
4
if φ > 1.
Here, the constant β is a very small perturbing term added to accommodate slope variations
in the regime of instability (0 < β  1) as described in [35]. We define


θ − θA if θ > θA ,
θ̂ = (3)
0 if θ
θA ,
3
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

0.25
(φ)
(φ)

0.2
(φ)

0.15

-1
M −A A

w=
M
(φ),

0
w=
0.1
5 4 3 2 1

0.05 M

θ0 θM θA
0
-0.2 0 0.2 0.4 0.6 0.8 1
φ θ

Figure 1. Free energy E and |σ |0 –θ phase space plot.

where θA > θ0 . The temperature θM is defined as θM = θA − θ0 . The free energy potentials

F (φ) and G (φ), and the |σ |0 –θ phase space diagram, are plotted in figures 1(a) and (b),
respectively.
The two-well free energy functional E , defining the minimum at φ = 0 and φ = 1, is
mapped to the potentials F and G as
E = F + wG , (4)
where w is the function of temperature and stress defined as
 
1 0 σ · σ
w= θ̂ − . (5)
θ0  |σ |
The two-well free energy functional E is plotted for different values of w in figure 2. The
phase in the domain adheres to the following rules:
• w > 0: functional E has a minimum at φ = 0;
• w < −1: functional E has a minimum at φ = 1;
• −1 < w < 0: functional E has metastable states at φ = 0 and/or φ = 1.
Thus the lines w = −1 and w = 0 represent the critical threshold for the disappearance of the
minimum at φ = 0 and that at φ = 1, respectively [35]. The M, A and M–A regions are shown
in figure 1(b). The numbers
1 –
5 in figure 1(b) correspond schematically to the functional
E shown in figure 2(b).
We closely follow [30, 35] for deriving the governing equations. For consistency and
completeness, the highlights of the main derivation are summarized as follows.
The temporal evolution of the OP is described by the first-order kinetic time-dependent
Ginzburg–Landau (TDGL) equation. The TDGL equation is stated as follows:
 
∂φ δ δ
γ =− +∇ · , (6)
∂t δφ δ∇φ
where γ is the relaxation parameter and δ defines the functional derivative.
On substituting equation (1) into equation (6) and mathematical manipulation, we obtain
the phase evolution equation as

  
∂φ  ∂F (φ) 0 σ · σ ∂G (φ)
γ =κ φ− θ0 + θ̂ − . (7)
∂t 2 ∂φ  |σ | ∂φ

4
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

w = -1.5
0.4
w = -1
0.3 w = -0.33
0.3 w = 0
w = 1 1
0.2 0.2
0.1
0 0.1
-0.1 2
-0.2 0 3
-0.3
-0.4 -0.1
1.5 4
1
1 -0.2
0.5
0
0.8 5
0.6
w -0.5
-1
0.4
φ 0 0.2 0.4 0.6 0.8 1
-1.5 0
0.2
φ

Figure 2. Two-well free energy function E .

2.2. The structural equations

The structural equations are described by using the kinematic relationship, the appropriate
constitutive equation and the law of conservation of momentum.

2.2.1. The kinematic relationship. The model is developed on the basis of isotropic material
properties and a small strain framework. The infinitesimal Cauchy–Lagrange strain tensor 
is defined as
1
= ∇ u + ∇ uT , (8)
2
where u is the displacement vector and x is the spatial coordinate vector.

2.2.2. The constitutive relationship. The relationship between the stress and the strain is
defined using the constitutive equations. In the case of the phase-changing systems, the
constitutive equations are modified to account for the phase transformation in order to describe
the hysteresis. Thus the constitutive equations not only relate the σ and , but also the φ. The
common methodology used to achieve this is the decomposition of strain into the elastic and
phase transformation components [14, 19–21, 23, 28, 37]. We too follow this methodology;
however, we take into account the different forms of the dependences of the compliance on φ
and σ . In this case, the constitutive equation is of differential form [30, 35], given as
∂ σ
˙ = λ1/2 (σ , φ) λ1/2 (σ , φ)σ + 0 Ġ(φ). (9)
∂t |σ |
The first and second terms on the right-hand side in equation (9) represent the elastic and
transformational components of the strain in a differential form. The compliance tensor λ
accommodates the properties of both austenite and martensite phases on the basis of the OP
value and the stress [19]. In the general 3D case, it is defined as
λ(σ , φ) = λ2 (φ) + λ3 (φ)σ + λ4 (φ)σ · σ , (10)
where λ2 , λ3 and λ4 are the tensors of fourth, sixth and eighth orders, respectively. A point to
be noted is that in equation (9) the phase transformation is governed by the stress tensor σ as
opposed to the transformation strain tensor described, e.g. in [14, 19–21, 23, 28, 37].

5
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

2.2.3. Conservation of the momentum. The structural balance is governed by the law of
conservation of momentum. It reads as
ρ ü = ∇ · σ + ρ f , (11)
where ρ is the density and f is the body force.

2.3. The thermal equation and thermodynamic consistency

The free energy ψ can be expressed in the terms of the internal energy e and entropy η as
ψ = e − θ η. (12)
According to the first law of thermodynamics, the balance of energy is written as
ρ ė(σ , φ, θ) = Pmi + Pφi − ∇ · q + r, (13)
where Pmi is the internal mechanical power, Pφi is the internal order structure power, q is the
heat flux defined as q = −k ∇θ and r is the external heat source. Pmi is defined as
Pmi = σ · ˙ , (14)
and Pφi is obtained by multiplying equation (7) by φ̇ and following the approach of [34]:

  
κ d  0 σ · σ
Pφ = γ φ̇ +
i 2
|∇φ| +
2
θ0 Ḟ (φ) + θ̂ − Ġ (φ) . (15)
2 dt 2  |σ |
In order to prove the consistency of the above model with the second law of
thermodynamics, the Clausius–Duhem inequality has to be satisfied. Thus,
q r
η̇  −∇ · + . (16)
θ θ
Using equations (12)–(15) and (16), this inequality is reduced to
q 1 d  
ψ̇ + θ̇ η
− · ∇θ + (λ (σ , φ)σ · σ ) + γ φ̇ 2 + κ ∇φ · ∇ φ̇ + θ0 Ḟ (φ) + θ̂ Ġ (φ).
θ 2 dt 2 2
(17)
Under the assumption that the free energy ψ is a function of the variables φ, ∇φ, σ , θ ,
equation (17) can be written as
     
∂ψ ∂ψ   ∂ψ
η+ θ̇ + − θ0 F  (φ) − θ̂ G  (φ) φ̇ + − κ ∇φ · ∇ φ̇
∂θ ∂φ 2 2 ∂∇φ
 
∂ψ q
+ − λ (σ , φ)σ · σ̇ − γ φ̇ 2 + · ∇θ
0. (18)
∂σ θ
The above inequality is satisfied for arbitrariness of φ̇, ∇ φ̇, σ̇ , θ̇ under the following
constitutive relations:
∂ψ ∂ψ   ∂ψ ∂ψ
η=− , = θ0 F  (φ) + θ̂ G  (φ) , = κ ∇φ, = λ (σ , φ)σ , (19)
∂θ ∂φ 2 ∂∇φ ∂σ
where F  and G  are derivatives with respect to the OP φ.
Substituting equation (19) in (18) leads to
k
|∇θ|2 + γ φ̇ 2  0, (20)
θ
thus proving the thermodynamic consistency with the positivity of the thermal conductivity k.
The relations in equation (19) enforce the following representation of the free energy:
1 κ  
ψ = ψ0 (θ ) + (λ (σ , φ)σ · σ ) + |∇φ|2 + θ0 F (φ) + θ̂ G (φ) , (21)
2 2 2
6
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

where we choose ψ0 as
c 2
ψ0 (θ ) = − θ . (22)
2θc
Now, the expression for the internal energy e in equation (12) can be simplified as follows:
c 2 1 κ  
e= θ + (λ(σ , φ)σ · σ ) + |∇φ|2 + θ0 F (φ) + (θ̂ − θ θ̂  )G (φ) , (23)
2θc 2 2 2
where θ̂  is the derivative with respect to the temperature θ .
On differentiating equation (23) with respect to time and equating it to the right-hand side
of equation (13), we obtain the heat equation as
c  
θ θ̇ − θ G (φ)θ̂  θ̇ + θ̂  Ġ (φ) − γ φ̇ 2 = k θ + r. (24)
θc 2
Defining θ̂  and θ̂  as the Heaviside function H and the Dirac-delta function δd , the
equation (24) can be written explicitly as
c   
θ θ̇ − θ G (φ)δd θ̇ + H Ġ (φ) − γ φ̇ 2 = k θ + r. (25)
θc 2

2.4. The system of equations and boundary conditions

Now, we summarize all the governing partial differential equations and constitutive relations
for the above model as

  
∂φ  ∂F (φ) 0 σ · σ ∂G (φ)
γ =κ φ− θ0 + θ̂ − , (26.1)
∂t 2 ∂φ  |σ | ∂φ
ρ ü = ∇ · σ + ρ f , (26.2)
c   
θ θ̇ − θ G (φ)δd θ̇ + H Ġ (φ) − γ φ̇ 2 = k θ + r. (26.3)
θc 2
The kinematic relations and constitutive equation are described as follows:
1 
= ∇ u + ∇ uT , (27.1)
2
∂ σ
˙ = λ1/2 (σ , φ) λ1/2 (σ , φ)σ + 0 Ġ (φ), (27.2)
∂t |σ |
along with the boundary conditions (refer to figure 3 for the boundary nomenclature)
    
n · ∇φ Γ = 0, u = 0, n · σ  = 0, n · σ  = σ̄ or u = ū,
1
 2,3 4 4

n · ∇θ Γ = h(θ − θext ), (28)

where h is the heat transfer coefficient, θext is the external environment temperature and σ̄
and ū are the stress-based and displacement-based ramp loadings, respectively. The initial
conditions are defined as
φ(x, 0) = φ0 (x), u(x, 0) = u0 (x), u̇(x, 0) = u̇0 (x), θ (x, 0) = θ̃0 (x) = θext (x).
(29)

7
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

Γ3
Γ1 Γ4
tn
x2 n
σii
ui x1 Γ2

t
load unload

Figure 3. Schematic of boundary nomenclature and the ramp loading and unloading.

2.5. The weak formulation

Let  ⊂ Rd be an open set in the d-dimensional space (d = 2, 3) defined by the coordinate

system x. The boundary is denoted by Γ and its outward normal by n. As the constitutive
equations are assumed in differential form in equations (27.1) and (27.2), we solve them
along with equations (26.1)–(26.3). The weak formulations of equations (26.1)–(26.3) and
equations (27.1) and (27.2) are derived by multiplying the equations with weighting functions
{, U , Σ, } and transforming them by using integration by parts. Let X denote both the
trial solution and the weighting function spaces, which are assumed to be identical. Let (·, ·)
denote the L2 inner product with respect to the domain . The variational formulation is stated
as follows. Find the solution S = {φ, u, σ , θ } ∈ X such that ∀W = {, U , Σ, } ∈ X, we
have B(W , S ) = 0, with
   
 ∂F  ∂G
B(W , S ) = , γ φ̇  + (∇ · , κ∇ · φ) + , θ0 + , θ̂
2 ∂φ  2 ∂φ 
 
σ · σ ∂G
− , 0 + (U , ρ ü) + (∇ · U , σ ) − (U , ρ f ) + (Σ, ˙ )
|σ | ∂φ 
       
∂ σ θ 
− Σ, λ1/2 λ1/2 σ − Σ, 0 Ġ (φ) + Θ, c θ̇ − , θ θ̇δd G
∂t |σ | θc  2
    

− , θ H Ġ − , γ φ̇  + (∇ · , k∇ · θ) − (, r)
2
2 
− (, κ ∇φ · n) − (U , σ · n) − (, k ∇θ · n) . (30)

Equations (30) have been implemented in a weak finite element formulation in the Comsol
Multiphysics software [39].

3. Numerical experiments

We exemplify the SMA hysteretic behavior in a 2D setting. A rectangular specimen of domain

 = [0, lx ] × [0, ly ] is chosen for the numerical simulations. We assume the plane stress
formulation, considering that the thickness is small as compared to the other dimensions. For
simplicity, we adopt the quasistatic approximation under the assumption that the timescales
of the thermal dynamics and phase evolution phenomenon are larger than the stress wave
timescale [28, 37]. We neglect the nonlinear effect of the λ3 and λ4 compliance tensors.
We assume that no external body and thermal loads are applied during the simulation. The

8
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

Table 1. Material parameters of polycrystalline Ni55 Ti45 .

λA λM  0 θA µ
2.5 × 10−11 Pa−1 3.57 × 10−11 Pa−1 106 Pa K−1 0.14 288.5 K 0.33

γ θM θ0 θc c k κ
−1 −1 −1
80 × 10 Pa s
5
273 K 212.7 K 296 K 3.2 × 10 Pa K 6
18 W m K 0.15 N

1 338 1000
Homogeneous
336 Non−homogeneous
0.8 800
334

332

σ11 (MPa)
0.6 600
θ (K)
φ

330
0.4 400
328

326
0.2 200
Homogeneous 324 Homogeneous
Non−homogeneous Non−homogeneous
0 322 0
0 10 20 30 40 0 10 20 30 40 50 0 0.02 0.04 0.06 0.08
t (s) t (s) 11

Figure 4. Average φ, θ evolution and σ11 –11 curves for phase-dependent compliance
properties.

phase-dependent compliance tensor λ2 takes the form

 
(1 + φ) −µ (1 + φ) 0
λ2 = λA −µ (1 + φ) (1 + φ) 0 , (31)
0 0 (1 + µ) (1 + φ)
where  = (λM − λA )/λA . The material properties of an Ni55 Ti45 specimen [28, 38] are
summarized in table 1. Some of the parameter values have been assumed to reproduce the
hysteretic properties of the material. The equivalent transformation strain 0 = 0.14 during
the simulations is selected. Although the 0 value falls in the finite strain regime, we have not
considered it in the current formulation. The interested readers are referred to [40, 41] and
references therein for further details on the finite strain phase-field models. The governing
equations are first rescaled and then implemented in the weak formulation. The results are
later converted back to the dimensional form.

3.1. The SMA behavior under stress-controlled loadings

In the following subsections, we describe the results of the simulations that have been carried
out on a rectangular domain with lx = 0.1 m and ly = 0.008 m, to show the ability of the
model to reproduce the SMA hysteretic behavior under the stress-controlled loading. It should
be noted that the stress-controlled loading often leads to uniform nucleation in a specimen,
which is different from the nonuniform nucleation observed during a displacement-controlled
loading [28, 38]. Hence, a stress drop is not observed during the phase nucleation in a stress-
controlled loading [37]. In this paper, we do not consider the defects and their associated
oscillating stress fields. The interested readers are referred to [42, 43] for more details.

3.1.1. The phase-dependent properties. First, a simulation was carried out to show the effect
of phase-dependent properties on the hysteretic behavior of SMAs. Two cases are considered:
9
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

Figure 5. Plot of (a) the 11 –θ–φ loop and the central axial line arc length (x̂) extrusion
plot for the (b) θ and (c) displacement u1 evolution.
370 1000
1 θ̃0 = 350 K
θ̃0 = 340 K
0.9 360 θ̃0 = 323 K
θ̃0 = 296 K
0.8 800
350
0.7
340
0.6 600

σ11 (MPa)
θ (K)
φ

0.5 330

0.4 400
320
0.3
θ̃0 = 350 K 310
0.2 θ̃0 = 350 K
θ̃0 = 340 K 200
0.1 θ̃0 = 323 K
θ̃0 = 340 K
300
θ̃0 = 323 K
0 θ̃0 = 296 K
θ̃0 = 296 K
290 0
0 5 10 15 20 25 30 35 40 0 5 10 15 20 25 30 35 40 0 0.02 0.04 0.06 0.08
t (s) t (s) 11

Figure 6. Average φ, θ evolution and σ11 –11 curve starting with different values of θ̃0 .

first, an SMA specimen with equal elastic compliances of the two phases (λM = λA ); and
second, an SMA specimen with local phase-dependent compliance of the austenite and
martensite phases (λM = λA ). The loadings are carried out with σ̇11 = 46.47 MPa s−1 starting
with the initial temperature θ̃0 = 323 K for both cases. The average φ, θ evolution and σ11 –11
curve calculated over the area of the specimen are shown in figure 4. The φ and θ evolve
identically in the two cases. However, the effect of local phase-dependent compliance is
apparent on the σ11 –11 curve. Figure 5(a) indicates the thermal hysteresis θ –11 and phase
evolution φ–11 loops for the local phase-dependent compliance case. The central axial line
(refer to the dashed line on the domain in figure 3) arc length (x̂) extrusion plots of θ and
the axial displacement u1 are shown in figures 5(b) and (c). Note that the phase-dependent
compliance has been reported experimentally [1, 3, 38]. Due to the higher compliance of the
martensite phase, as compared to the austenite phase, the area under the σ11 –11 curve is
large, thus causing more energy dissipation in the local phase-dependent properties. In all
of the subsequent simulations, we use the local phase-dependent compliance properties of
the phases.

3.1.2. The SMA behavior at different initial temperatures. Next, the hysteretic response of an
SMA specimen is studied starting with different initial temperatures θ̃0 ranging from 296 K
to 350 K, using the axial stress rate σ̇11 = 46.47 MPa s−1 . The average φ, θ evolution and
σ11 –11 curve are shown in figure 6. At the lower initial temperature (θ̃0 = 296 K), the shape
memory effect is observed with remnant strain at the end of the unloading. The pseudoelastic
behavior, with fully recoverable strain, is observed at higher θ̃0 . As θ̃0 increases, higher axial
stress is required for the start of the phase transformation, thus offsetting the σ11 –11 curve
toward a higher value. The area under the σ11 –11 curve remains constant. These results are
consistent with the results reported in the literature [28, 37, 38].

10
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

338 1000
1

0.9 336

0.8
800
334
0.7
332
600

σ11 (MPa)
0.6

θ (K)
φ

0.5 330

0.4 400
328
σ̇11 = 4.647 MPa/s σ̇11 = 4.647 MPa/s
0.3
σ̇11 = 9.857 MPa/s σ̇11 = 9.857 MPa/s σ̇11 = 4.647 MPa/s
326 σ̇11 = 9.857 MPa/s
0.2 σ̇11 = 14.082 MPa/s σ̇11 = 14.082 MPa/s
200 σ̇11 = 14.082 MPa/s
σ̇11 = 46.47 MPa/s σ̇11 = 46.47 MPa/s
0.1 324
σ̇11 = 46.47 MPa/s
σ̇11 = 98.57 MPa/s σ̇11 = 98.57 MPa/s σ̇11 = 98.57 MPa/s
0 σ̇11 = 140.82 MPa/s σ̇11 = 140.82 MPa/s σ̇11 = 140.82 MPa/s
0 50 100 150 200 250 300 350 400
322
0 50 100 150 200 250 300 350 400 450
0
0 0.02 0.04 0.06 0.08 0.1
t (s) t (s) 11

Figure 7. Average φ, θ evolution and σ11 –11 curve for different loading rates σ̇11 .

(i) 1 338
(v) (v)
(vi) 336 (vi)
(iv) (iv)
(ii) 0.8
334
(iii) 0.6 332

θ (K)
φ

330 (iii)
(iv) 0.4 (iii)
328
(v) 326
0.2
(ii) 324 (i) (ii)
(vi) 0
(i)
322
0 10 20 30 40 0 10 20 30 40
t (s) t (s)

Figure 8. Bending loading evolution of (a) φ, (b) average φ and (c) average θ. Markers
in parts (b) and (c) correspond to different time t instants in (a). Displacements u are
superimposed on φ in (a) for better clarity. The phase evolution plots (i)–(v) in (a)
correspond to the loading path and (vi) is for near the end of unloading.

3.1.3. The SMA behavior at different stress rates. Next, simulations were conducted on
an SMA specimen with different axial σ̇11 loading rates ranging between 4.647 MPa s−1 and
140.82 MPa s−1 and starting with the initial temperature θ̃0 = 323 K. The average φ, θ evolution
and σ11 –11 curve are shown in figure 7. With increase in σ̇11 , the phase transformation starts
at higher stress values. As the loading rate increases, the heat generated during the exothermic
(A → M) process causes the internal temperature of the specimen to increase due to there
being insufficient time for heat transfer to the environment. During the (M → A) process, the
heat is absorbed due to the endothermic nature of the phase transformation. The dissipation
energy (the area of the hysteresis loop in the σ11 –11 curve) increases with increasing σ̇11 . It
is observed that with increase in σ̇11 , the slope of phase transformation increases. The above
behaviors have been experimentally observed [38].

3.1.4. The SMA behavior under bending loading. A simulation is conducted on an SMA
specimen by applying the constraint u = 0 on the boundary 1 and applying tangential ramp
stress loading on the boundary 4 in the x2 direction corresponding to σ̇yy = 19.9 MPa s−1 .
The initial temperature at the start of the simulation is θ̃0 = 323 K. The phase evolution φ
snapshots are plotted at different times in figure 8(a). The blue and red colors represent A and
M phases, respectively. During the loading phase, the tensile stress is generated on the top fiber
and the compressive stress on the bottom fiber of the specimen. Nonhomogeneous regions of
A and M domains, separated by domain walls, are formed. As the model formulation does not
account for different martensite variants, the M regions on the top and bottom coalesce near the
end of loading. The average φ and θ plots are presented in figures 8(b)–(c). The formation of

11
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

(i) φ 1
(v)
1 (iv)
(ii) 0.8
0.9
(iii) 0.8 (iii)
0.7 0.6
(iv) 0.6

φ
(ii)
0.5 0.4
(v)
0.4
(vi) 0.3 0.2
0.2 (vi)
(vii) 0.1 (i) (vii)
0 (viii)
0
(viii) 0 100 200 300 400
t (s)

336
(v)
(v) 800
334
(iv)
332
(iii) 600
σ11 (MPa)

330
(K)

(i)
(ii) (iv)
328 400 (iii)
(ii)
326 (vi) (vi)
200 (vii)
324 (vii)
(viii)
(i)
322
0 100 200 300 400 0 (viii)
t (s) 0 0.02 0.04 0.06
11

Figure 9. Displacement-controlled loading: evolution of (a) φ, (b) average φ, (c)

average θ and (d) average σ11 –11 . Markers in parts (b), (c) and (d) correspond to
different time t instants in (a). The phase evolution plots (i)–(v) in (a) correspond to
the loading and plots (vi)–(viii) to the unloading path.

M regions on the top and bottom edges has been experimentally reported by Rejzner et al [44]
for SMAs subjected to pure bending loading.

3.2. The SMA behavior under displacement-controlled loading

In the previous sections, we studied the behavior of an SMA specimen under stress-controlled
loading. Here, we now load the SMA specimen (lx = 0.1 m and ly = 0.01 m) in the x1
direction with a displacement-controlled ramp loading. The specimen, initially at θ̃0 = 323 K,
is constrained on the boundary 1 with u = 0 and a ramp displacement corresponding to
˙11 = 3.3 × 10−4 s is applied on the boundary 4 .
Time evolution snapshots of phase φ at different times are plotted in figure 9(a). The
martensite nucleates at different locations in the domain, causing nonhomogeneous regions of A
and M. As the loading progresses, different martensite regions coalesce to form a homogeneous
M region. At the end of unloading, the domain returns back to the A phase. Average
φ–t, θ–t and σ11 –11 results obtained during the loading–unloading process are presented
in figures 9(b)–(d). Exothermic and endothermic natures of the phase evolution are observed
during the (A → M) and (M → A) transformations, respectively.

12
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

4. Conclusions

A macroscale non-isothermal 3D model has been developed for describing nonlinear phase-
dependent hysteretic behaviors in SMAs. The model is thermodynamically consistent with a
rate-dependent constitutive relationship, and the conservation laws of thermal and mechanical
physics, in conjunction with a kinetic phase evolution equation. In its description of the phase
transformations, the model is based on the stress tensor and utilizes a scalar order parameter.
Representative numerical simulations for the stress-controlled loadings illustrate that
the phase-dependent compliance properties improve the pseudoelastic nonlinear hysteretic
description. The tensile tests on SMA specimens with different initial temperatures, as well
as loading rates, elucidate the model’s ability to capture efficiently the thermomechanical
behavior and phase kinetics. The model successfully reproduces experimentally observed
SMA behaviors reported in the literature.

Acknowledgment

RD and RM were supported by NSERC and CRC programs, Canada. RM also acknowledges
TÜBITAK for support.

References

[1] Otsuka K and Wayman C 1998 Shape Memory Materials (Cambridge: Cambridge University Press)
[2] Kohl M 2004 Shape Memory Microactuators (Berlin: Springer)
[3] Lagoudas D 2008 Shape Memory Alloys: Modeling and Engineering Applications (London:
Springer)
[4] Miyazaki S, Fu Y Q and Huang W M 2009 Thin Film Shape Memory Alloys: Fundamentals and
Device Applications (Cambridge: Cambridge University Press)
[5] Ozbulut O, Hurlebaus S and DesRoches R 2011 Seismic response control using shape memory
alloys: a review J. Intell. Mater. Syst. Struct. 22 1531–49
[6] Elahinia M, Hashemi M, Tabesh M and Bhaduri S 2012 Manufacturing and processing of NiTi
implants: a review Prog. Mater. Sci. 57 911–46
[7] Khandelwal A and Buravalla V 2009 Models for Shape memory alloy behavior: an overview of
modelling approaches Int. J. Struct. Changes Solids—Mech. Appl. 1 111–48
[8] Lagoudas D, Brinson L and Patoor E 2006 Shape memory alloys: II. Modeling of polycrystals
Mech. Mater. 38 430–62
[9] Birman V 1997 Review of mechanics of shape memory alloy structures Appl. Mech. Rev.
50 629–45
[10] Smith R 2005 Smart material systems: model development vol 32 (Philadelphia, PA: SIAM)
[11] Paiva A and Savi M A 2006 An overview of constitutive models for shape memory alloys. Math.
Prob. Eng. 56876 1–30
[12] Mamivand M, Zaeem M and El Kadiri H 2013 A review on phase field modeling of martensitic
phase transformation Comput. Mater. Sci. 77 304–11
[13] Falk F 1980 Model free energy, mechanics, and thermodynamics of shape memory alloys Acta
Metall. 28 1773–80
[14] Khachaturian A 1983 Theory of Structural Transformations in Solids (New York: Wiley)
[15] Melnik R, Roberts A and Thomas K 1999 Modelling dynamics of shape-memory-alloys via
computer algebra Proc. SPIE Math. Control Smart Struct. 3667 290–301
[16] Melnik R, Roberts A and Thomas K A 2002 Computing dynamics of Copper-based SMA via center
manifold reduction models Comput. Mater. Sci. 18 255–68
[17] Artemev A, Jin Y and Khachaturyan A G 2001 Three-dimensional phase field model of proper
martensitic transformation Acta Mater. 49 1165–77

13
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

[18] Chen L 2002 Phase field models for microstructure evolution Annu. Rev. Mater. Res. 32 113–40
[19] Levitas V and Preston D 2002 Three-dimensional Landau theory for multivariant stress-induced
martensitic phase transformations: I. austenite ↔ martensite Phys. Rev. B 66 1–9
[20] Levitas V and Preston D 2002 Three-dimensional Landau theory for multivariant stress-induced
martensitic phase transformations: II. Multivariant phase transformations, stress space analysis
Phys. Rev. B 66 1–15
[21] Levitas V, Preston D and Lee D 2003 Three-dimensional Landau theory for multivariant stress-
induced martensitic phase transformations: III. Alternative potentials, critical nuclei, kink
solutions, and dislocation theory Phys. Rev. B 68 1–24
[22] Ahluwalia R, Lookman T and Saxena A 2006 Dynamic strain loading of cubic to tetragonal
martensites Acta Mater. 54 2109–20
[23] Mahapatra D and Melnik R 2006 Finite element analysis of phase transformation dynamics in shape
memory alloys with a consistent Landau–Ginzburg free energy model Mech. Adv. Mater. Struct.
13 443–55
[24] Mahapatra D and Melnik R 2007 Finite element approach to modelling evolution of 3D shape
memory materials Math. Comput. Simul. 76 141–8
[25] Wang L and Melnik R 2007 Finite volume analysis of nonlinear thermo-mechanical dynamics of
shape memory alloys Heat Mass Transfer 43 535–46
[26] Bouville M and Ahluwalia R 2008 Microstructure and mechanical properties of constrained shape
memory alloy nanograins and nanowires Acta Mater. 56 3558–67
[27] Bouville M and Ahluwalia R 2009 Phase field simulations of coupled phase transformations in
Ferroelastic–ferroelastic nanocomposites Phys. Rev. B 79 094110
[28] Grandi D, Maraldi M and Molari L 2012 A macroscale phase-field model for shape memory
alloys with non-isothermal effects: influence of strain rate and environmental conditions on the
mechanical response Acta Mater. 60 179–91
[29] Dhote R, Melnik R and Zu J 2012 Dynamic thermo-mechanical coupling, size effects in finite shape
memory alloy nanostructures Comput. Mater. Sci. 63 105–17
[30] Dhote R, Fabrizio M, Melnik R and Zu J 2013 Hysteresis phenomena in shape memory
alloys by non-isothermal Ginzburg–Landau models Commun. Nonlinear Sci. Numer. Simul.
18 2549–61
[31] Wang Y and Khachaturyan A 1997 Three-dimensional field model and computer modeling of
martensitic transformations Acta Mater. 45 759–73
[32] Curnoe S and Jacobs A 2000 Twin wall of proper cubic–tetragonal ferroelastics Phys. Rev. B
62 11925–8
[33] Lookman T, Shenoy S R, Rasmussen K, Saxena A and Bishop A R 2003 Ferroelastic dynamics and
strain compatibility Phys. Rev. B 67 24114
[34] Berti V, Fabrizio M and Grandi D 2010 Phase transitions in shape memory alloys: a non-isothermal
Ginzburg Landau model Physica D 239 95–102
[35] Berti V, Fabrizio M and Grandi D 2010 Hysteresis and phase transitions for one-dimensional and
three-dimensional models in shape memory alloys J. Math. Phys. 51 062901
[36] Grandi D, Maraldi M and Molari L 2012 A macroscale phase-field model for shape memory
alloys with non-isothermal effects: influence of strain rate and environmental conditions on the
mechanical response Acta Mater. 60 179–91
[37] Maraldi M, Molari L and Grandi D 2012 A non-isothermal phase-field model for shape memory
alloys: numerical simulations of superelasticity and shape memory effect under stress-controlled
conditions J. Intell. Mater. Syst. Strct. 23 1083–92
[38] Zhang X, Feng P, He Y, Yu T and Sun Q 2010 Experimental study on rate dependence of macroscopic
domain and stress hysteresis in NiTi shape memory alloy strips Int. J. Mech. Sci. 52 1660–70
[39] Comsol Multiphysics Software, Version 4.2. (www.comsol.com)
[40] Levitas V 2013 Phase-field theory for martensitic phase transformations at large strains Int. J. Plast.
49 85–118
[41] Levin V, Levitas V, Zingerman K and Freiman E 2013 Phase-field simulation of stress-induced
martensitic phase transformations at large strains Int. J. Solids Struct. 50 2914–28

14
Modelling Simul. Mater. Sci. Eng. 22 (2014) 085011 R Dhote et al

[42] Levitas V and Lee D 2007 Athermal resistance to interface motion in the phase-field theory of
microstructure evolution Phys. Rev. Lett. 99 245701
[43] Lloveras P, Castán T, Porta M, Planes A and Saxena A 2010 Thermodynamics of stress-induced
ferroelastic transitions: influence of anisotropy and disorder Phys. Rev. B 81 214105
[44] Rejzner J, Lexcellent C and Raniecki B 2002 Pseudoelastic behaviour of shape memory alloy beams
under pure bending: experiments and modelling Int. J. Mech. Sci. 44 665–86

15