View metadata, citation and similar papers at core.ac.

uk brought to you by CORE

provided by Caltech Theses and Dissertations

A Constitutive Relation for Shape-Memory Alloys

Thesis by

Alex Kelly

In Partial Fulfillment of the Requirements

for the Degree of
Doctor of Philosophy

California Institute of Technology

Pasadena, California

2009
(Defended September 4, 2008)
 ii

c 2009
Alex Kelly
All Rights Reserved
 iii

Acknowledgements

It would cheapen the pure sentiment of gratitude that I wish to express in this section
to fill it with cringe- inducing metaphors about birds flying for the first time or pithy
quotes from philosophers of bygone centuries. Instead, I would like to thank those
people who have lent to me the best of themselves in spite of the fact that I did not
deserve it.
Michael Kelly Sheila Kelly
Evan Kelly Jared Kelly

Kaushik Bhattacharya Richard Murray

Jenni Buckley Alexis Cox

Meher Kiran Prakash Ayalasomayajula Patrick Dondl

Kaushik Dayal Samantha Daly

Amir Sadjadpour Lixiu Tian
Vikram Gavini Phanish Suryanarayana
Isaac Chenchiah Mathias Jungen
Sefi Givli Christian Lexcellent
Guruswami Ravichandran Sylvie Gertmenian
 iv

Abstract

The novel nonlinear thermoelastic behavior of shape-memory alloys (SMAs) makes

them increasingly desirable as components in many advanced technological applica-
tions. In order to incorporate these materials into engineering designs, it is important
to develop an understanding of their constitutive response. The purpose of this thesis
is to develop a constitutive model of shape-memory polycrystals that is faithful to the
underlying micromechanics while remaining simple enough for utility in engineering
analysis and design.
We present a model in which the material microstructure is represented macro-
scopically as a recoverable transformation strain that is constrained by the texture of
the polycrystal. The point of departure in this model is the recognition that the me-
chanics of the onset of martensitic transformation are fundamentally different from
those of its saturation. Consequently, the constraint on the set of recoverable strains
varies throughout the transformation process. The effects of constraint geometry
on the constitutive response of SMAs are studied. Several well known properties of
SMAs are demonstrated. Finally the model is simply implemented in a commercial
finite-element package as a proof of the concept.
 v

Contents

Acknowledgements iii

Abstract iv

Contents v

List of Figures vii

1 Introduction 1

2 Background 7
2.1 Review . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.1 Crystallography . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.1.2 Mathematical Background . . . . . . . . . . . . . . . . . . . . 9
2.1.3 Experiments . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
2.1.4 Homogenization Literature . . . . . . . . . . . . . . . . . . . . 13
2.1.5 Kinetics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
2.1.6 Constitutive Modelling . . . . . . . . . . . . . . . . . . . . . . 16
2.2 Motivation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19

3 Continuum Model 22
 vi
3.1 Kinematics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 22
3.2 Balance Laws . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
3.3 Energy . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 27
3.4 Initiation and Saturation . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.5 Driving Forces and Kinetics . . . . . . . . . . . . . . . . . . . . . . . 35

4 Some Features of the Model 41

4.1 One Dimension . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.1.1 Stress-Induced Martensite . . . . . . . . . . . . . . . . . . . . 41
4.1.2 Shape-Memory Effect . . . . . . . . . . . . . . . . . . . . . . . 46
4.2 Martensite Reorientation . . . . . . . . . . . . . . . . . . . . . . . . . 49

5 Demonstration and Parameter Study 51

5.1 Uniaxial Tension and Compression . . . . . . . . . . . . . . . . . . . 52
5.2 Simple Shear . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
5.3 Combined Uniaxial Extension and Shear . . . . . . . . . . . . . . . . 58
5.4 Anisotropic Materials . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
5.5 Proportional Loading in Different Directions . . . . . . . . . . . . . . 63
5.6 Nonproportional Loading . . . . . . . . . . . . . . . . . . . . . . . . . 68

6 Numerical Implementation and Example 73

7 Conclusions and Future Directions 80

Bibliography 83
 vii

List of Figures

1.1 A schematic representation of superelasticity in a specimen above Af . . 2

1.2 A schematic representation of the shape-memory effect . . . . . . . . . 4
1.3 A picture showing a) coronary stent (Nitinol Devices and Components
Inc.), b) schematic of a compressed stent inserted into a hollow structure
and expanded. (from http://openlearn.open.ac.uk). . . . . . . . . . . 5

2.1 Austenite and variants of the low symmetry martensite in the continuum 9
2.2 Polycrystalline specimens have the added wrinkle of intergrain com-
patability . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.3 Polarized Light Micrographs by Brinson et al. [19] at a) onset of trans-
formation and b) saturation . . . . . . . . . . . . . . . . . . . . . . . . 14

3.1 The volume fraction and the nominal effective strain are constrained to
lie in the identified set. . . . . . . . . . . . . . . . . . . . . . . . . . . . 25
3.2 A schematic representation of sets Gi,s . . . . . . . . . . . . . . . . . . 26
3.3 A schematic representation of the first two terms of W . . . . . . . . . 28
3.4 The degree of tension-compression asymmetry of the initiation and sat-
uration surfaces is adjusted through the parameter b. . . . . . . . . . . 30
 viii
3.5 The degree of eccentricity of the initiation and saturation surfaces is
varied through the parameter c. The effects of variation of this param-
eter are seen on a) a symmetric set, and b) a set with a high degree of
asymmetry. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
3.6 The direction of eccentricity is controlled through variation of the vector
ê. The effects of variation of this parameter are seen on a) a symmetric
set, and b) a set with a high degree of asymmetry. . . . . . . . . . . . 32
3.7 A suitable choice of parameters for the set of admissible transformation
strains (a) allows for recovery of the tension-compression asymmetry in
the biaxial experiments in the dual stress space (b). A more precise
fit to the experimental data was proposed by Lexcellent in d) and its
strain-space dual in c). . . . . . . . . . . . . . . . . . . . . . . . . . . . 34
3.8 The kinetic relation governing the phase transition in the stick-slip case. 39

4.1 Stress-induced martensite in one dimension shown in the volume fraction-

strain plane on the left and the stress-strain plane on the right. The
constraints on the volume fraction and nominal transformation strain
are indicated on the shaded set. . . . . . . . . . . . . . . . . . . . . . . 42
4.2 Shape-memory effect in one dimension shown in the the volume fraction-
strain plane on the left and the stress-strain-temperature space on the
right. The constraints on the volume fraction and nominal transforma-
tion strain are indicated on the shaded set. . . . . . . . . . . . . . . . . 46
4.3 The different eccentricities of the set Gi and Gs can lead to a significant
reorientation of the martensite. The nominal transformation strain at
different extents of transformation is shown. . . . . . . . . . . . . . . . 48
 ix
5.1 A proportional strain-controlled extension-shear test of an isotropic ma-
terial. (a) The resulting stress, (b) the resulting transformation strain
along with a verification of the numerical method, (c) uniaxial stress
vs. uniaxial strain (d) shear stress vs. shear strain and (e) equivalent
stress vs. equivalent strain. . . . . . . . . . . . . . . . . . . . . . . . . 60
5.2 Anisotropy of the initiation surface. Combined extension and shear of
an uniaxial material under strain control for various values of ci ranging
from 0.3-1.5. (a) The effective transformation strain trajectory. The
applied strain trajectory is indicated by the dashed line. (b) The stress
trajectory, (c) equivalent stress vs. equivalent strain (d) volume fraction
vs. equivalent strain. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 62
5.3 Anisotropy of the saturation surface. Combined extension and shear of
an uniaxial material under strain control for various values of cs rang-
ing from 0.3 to 1.5. (a) The effective transformation strain trajectory.
The applied strain trajectory is indicated by the dashed line. (b) The
stress trajectory, (c) equivalent stress vs. equivalent strain, (d) volume
fraction vs. equivalent strain. . . . . . . . . . . . . . . . . . . . . . . . 64
5.4 Anisotropy of the both surfaces. Combined extension and shear of an
uniaxial material under strain control for various values of ci , cs ranging
from 0.3 to 1.5 (in equal increments). (a) The effective transforma-
tion strain trajectory. The applied strain trajectory is indicated by the
dashed line. (b) The stress trajectory, (c) equivalent stress vs. equiva-
lent strain, (d) volume fraction vs. equivalent strain. . . . . . . . . . . 65
 x
5.5 Variation of direction applied strain. (a) The effective transformation
strain trajectory. The applied strain trajectories are indicated by the
dashed lines. (b) The stress trajectory, (c) equivalent stress vs. equiv-
alent strain, (d) volume fraction vs. equivalent strain. . . . . . . . . . . 66
5.6 Variation of direction applied stress. (a) The effective transformation
strain trajectory. (b) The stress trajectory, (c) equivalent stress vs.
equivalent strain, (d) volume fraction vs. equivalent strain. . . . . . . . 67
5.7 The experimental observations of Jung [36] show the emergence of sec-
ondary hysteresis in nonproportional tension-torsion. . . . . . . . . . . 68
5.8 Dual plateau in nonproportional loading. (a) The effective transfor-
mation strain trajectory. The applied strain trajectories are indicated
by the dashed lines. (b) The stress trajectory, (c) equivalent stress vs.
equivalent strain, (d) volume fraction vs. equivalent strain. . . . . . . . 69
5.9 Circular Loading Path. (a) The effective transformation strain trajec-
tory. The applied strain trajectories are indicated by the dashed lines.
(b) The stress trajectory, (c) equivalent stress vs. equivalent strain, (d)
volume fraction vs. equivalent strain. . . . . . . . . . . . . . . . . . . . 71
5.10 Square Loading Path. (a) The effective transformation strain trajec-
tory. The applied strain trajectories are indicated by the dashed lines.
(b) The stress trajectory, (c) equivalent stress vs. equivalent strain, (d)
volume fraction vs. equivalent strain. . . . . . . . . . . . . . . . . . . . 72

6.1 The Legendre transform of the dissipation potentials are shown in the
stick-slip case (black) for p = 2 and in the rate-independent case (red). 75
 xi
6.2 The effect of the polynomial constraint is seen. The intersection of the
16th degree polynomial shifts the intersection point with the quadratic
elastic energy slightly to the right of the hard (dashed) constraint. . . . 76
6.3 The smoothed constraint set is calculated at 10% applied strain. The
green dashed constraint is Gs , the red dashed constraint is the scaled
Gs at λ = λsat and the red solid set is the softened polynomial (N =
16) constraint at λsat . . . . . . . . . . . . . . . . . . . . . . . . . . . . 77
6.4 A flowchart of the numerical implementation . . . . . . . . . . . . . . . 78
6.5 An example of the finite-element implementation on a stent element. . 79
 1

Chapter 1

Introduction

Shape-memory alloys (SMAs) exhibit special macroscopic phenomena including su-

perelasticity and the shape-memory effect. In superelasticity, strains on the order of
a few percent can be induced by loading and are completely recovered on unloading.
In shape-memory, a sample deformed apparently permanently by a few percent strain
below a particular critical temperature returns to its original shape upon heating.
SMAs have been employed in a large number of engineering applications in fields
ranging from medicine to telecommunications. Due to their importance in estab-
lished and developing engineering fields, a vast literature has risen to study these
materials. Volumes by Olson and Owen [45], Otsuka and Wayman [46], and Bhat-
tacharya [12] as well as the review of Saburi and Nenno [52] provide an accessible
introduction to the development of experiments and theories used to characterize
these important alloys.
The special phenomena arise as a result of the existence of two solid phases in a
shape-memory alloy. These phases, rearrangements of the alloy’s crystal lattice, are
characterized by the lattice symmetry group. The high symmetry phase, austenite,
is generally stable at high temperatures; with the lower symmetry phase, martensite,
stable at lower temperatures. As an example in NiTiNOL, the most celebrated of
 2
σ 4

2 3

5
6

1
ε

Figure 1.1: A schematic representation of superelasticity in a specimen above Af .

the SMAs, the austenite is cubic with monoclinic martensite. The change between
the two phases is a first-order diffusionless transformation that can be activated
by changes in either temperature or stress. At zero stress, the phase transition
is characterized by four temperature points: the Martensite start (Ms ) at which
a sample that is completely austenite begins forming martensite, Martensite finish
(Mf ) is the temperature at which the austenite-martensite transformation saturates,
the Austenite start (As ) at which a completely martensitic sample begins forming
austenite, and the Austenite finish (Af ) is the temperature at which the martensite-
austenite transformation reaches completion. The possibility of large atomic scale
displacement in the sample contributes to highly nonlinear thermoelastic behavior
on the macro scale.
Superelasticity is the ability of the sample to recover large strains elastically.
The magnitude of elastic strains recovered in superelastic SMAs is quite impressive.
Whereas a typical metal recovers strains on the order of 0.2 %, a sample of NiTiNOL
demonstrates recovery of somewhere between 6% and 8%. For a constant temper-
ature above the austenite finish (Af ) the stress-strain behavior of a superelastic
material is shown schematically in figure 1.1.
The initial branch of the hysteresis loop (between points 1 and 2) is the elastic
 3
response of the austenite. When a threshold stress is reached (point 2) the formation
of martensite is induced. The atomic displacement that accompanies the change in
phase allows for large strains to accumulate as the austenite-martensite transforma-
tion continues. At point 3 the martensitic transformation has saturated and the
martensite responds elastically to the extent that it can before failure (point 4). Un-
loading sees the formation of the lower branch of the hysteresis loop. The martensite
unloads elastically until point 5. At this point, the free energy is low enough that
the transformation back to the stable austenite begins. When the transformation
back to austenite finishes (point 6), unloading finishes elastically (back to point 1).
It is worth noting that the entire amount of strain is recovered during the loading-
unloading cycle because the phase transformation is an reversible distortion of the
crystal lattice, and no breaking of bonds has occured (in a lattice free of defects).
The second characteristic phenomenon resulting from the martensitic phase trans-
formation is the shape-memory effect. In order for the crystal to remain intact after
it has undergone a phase transformation, recall that it must satisfy the Hadamard-
Legendre conditions at each interface between martensite variants with each other or
with the austenite. These jump conditions lead to the formation of self-accommodating
(no change in volume) microstructure in the SMA crystal as different variants of
martensite must mix coherently in order to preserve the integrity of the crystal. It
is this formation of microstructure that is responsible for the shape-memory effect.
As can be seen in figure 1.2, the considered transformation begins with a com-
pletely austenite reference configuration in the high temperature regime. As the
specimen cools and martensitic variants become more energetically favorable a self-
accomodating microstructure forms. As the cooled specimen is deformed, the phase
fraction of martensitic variants is changed via stress-induced transformation and the
resulting microstructure is different in the deformed configuration. Upon heating of
 4

Cool

Heat Deform

Figure 1.2: A schematic representation of the shape-memory effect

the sample, austenite once again becomes the energetically favorable phase, and as
a result the reference configuration is restored as the crystal returns to the austenite
lattice. The sample has recovered its reference configuration through the change in
phase, hence shape ”memory”.
Shape-memory alloys are used in many devices. Biomedical devices such as stents
and braided catheters take advantage of the superelasticity of SMAs. The ability of
the alloy to recover large strains elastically allows for expandable structures to be
compressed down to very small sizes and delivered into an obstructed passage. allows
for the obstructions to be circumvented without recourse to invasive surgeries. The
stent, shown in figure 1.3, is a cylindrical wire mesh structure that is compressed
to fit around an insertion tube or catheter. The compressed stent is inserted on
the catheter into an obstructed (by arterial plaque) blood vessel and expanded via
 5

a) b)

Figure 1.3: A picture showing a) coronary stent (Nitinol Devices and Components
Inc.), b) schematic of a compressed stent inserted into a hollow structure and ex-
panded. (from http://openlearn.open.ac.uk).

balloon. Because the stent is able to recover the large strain required for insertion,
it is able to hold the blocked vessel open.
Shape-memory is applied in the aerospace field, particularly in deep-space actu-
ators such as latches where traditional moving parts would be a riskier method of
achieving the desired actuation than the thermomechanical capability of the SMA
([42], for example, shows one such example in the case of unfolding solar panels on the
Hubble Space Telescope). For further information on a wide variety of applications,
the reader is referred to [34] and the references within.
As the applications of SMAs grow more sophisticated, there is a real need to
develop a faithful yet easy-to-use constitutive model that can be used as a design
tool. This is the motivation for this thesis. The plan for the thesis is as follows:
In chapter 2, the existing literature in both mathematical theory and experiments
is examined. This literature review is undertaken with the purpose of underscoring
some of the important open issues in the understanding of shape-memory alloys. The
examination of these open issues results in the proposal of a heuristic that defines the
mechanics of onset and saturation of martensitic transformation in shape-memory
 6
polycrystals as distinctly different processes.
Chapter 3 contains the incorporation of this heuristic into a continuum model.
The onset and saturation heuristics are manifest in this model as a pair of constraints
on the set of admissible transformation strains resulting from the formation of mi-
crostructure in the polycrystal. These constraints are vital to the formulation of
a polycrystalline Helmholtz free energy that, along with an appropriate dissipation
potential, yields the proposed continuum model with few internal variables.
The ability of the proposed continuum model to duplicate the well-known charac-
teristics of SMAs is evaluated in chapter 4. The continuum model in the third chapter
is shown to reproduce superelasticity and the shape-memory effect. Experimental
observation of martensitic reorientation and retention of austenite are accounted for
as well.
The constraints that form the backbone of the proposed model undergo parameter
studies in numerical simulations in chapter 5. When the shape of the constraint sur-
faces is varied parametrically, a large variability in the resultant constitutive response
demonstrates the potential of the model to be fit to a large number of important
experiments. An explanation is offered for the curious observations of Jung [36] in
non proportional tension-torsion experiments on Nitinol tubes.
Finally, in chapter 6 the model is incorporated into a commercial finite element
package as an ABAQUS UMAT. This is done in order to demonstrate the relative ease
of implementation of this model as a feasible design tool for engineering applications.
Chapter 7 summarizes this thesis and provides a discussion of open issues and
future research.
 7

Chapter 2

Background

2.1 Review

2.1.1 Crystallography

The crystal lattice of a shape-memory alloy is assumed to undergo transformation

from a high symmetry austenite phase (stable at higher temperatures) to a lower
symmetry martensite (stable at lower temperatures). Mathematically the crystal is
described as a Bravais lattice with basis vectors {e1 , e2 , e3 }. For simplicity’s sake, it
is assumed that these basis vectors coincide with the high symmetry austenite phase.
Assume that the martensite (low symmetry) phase is described by a different set of
basis vectors {f1 , f2 , f3 }. The two bases are related by a matrix transformation A:

[f1 , f2 , f3 ] = A [e1 , e2 , e3 ] . (2.1)

Recalling that the matrix A can be decomposed by the Polar Decomposition Theo-
rem into a rotation Q and a symmetric positive definite matrix U such that A = QU.
The rotation is irrelevant to the energy landscape of the lattice due to the require-
 8
ment that it be materially frame indifferent. This being the case, the martensite is
characterized by the Bain or transformation matrix U. The variants of martensite
are then easily found by symmetry through application of point group of the austen-
ite phase (for example the cubic point group in nickel-titanium). For each rotation
Ri in the point group of the austenite, the transformation matrix for the martensitic
variant Ui is found by:
Ui = RT
i URi . (2.2)

It is worth noting that application of rotations Ri that are also in the point group of
the martensite (monoclinic in the case of NiTi) do not produce a new variant. The
Pa
number of unique variants then can be shown to be Pm
where Pa is the cardinality
of the point group of the austenite, and Pm that of the martensite.
The Helmholtz free energy for the crystal takes the form of a multiwell energy
with wells at the unique austenite (I) as well as each of the martensite Ui lattice
deformations. This multiwell structure goes back to Ericksen [29]. The relative
heights of the austenite and martensite wells change generally with temperature to
reflect the stability of each phase with respect to some critical temperature. The
transition is usually of first order in the language of Landau free energies.
The connection between the lattice picture and the continuum is made through
the Cauchy-Born hypothesis [9]. Matrix deformations of the bravais lattice basis
vectors are assumed to be written in the continuum as deformation gradients. The
austenite phase is represented as the deformation gradient I. The martensites are
represented by the corresponding deformation gradients Ui . The multiwell Helmholtz
free energy for the single crystal in the continuum can be represented schematically
as Wsc (F) in figure 2.1. In the small strain approximation, the austenite is at zero
strain (ε = 0), and the wells associated with each variant are εi = Ui − I.
 9

Wsc(F)
Ui

Uj

F I
Ui I Uj Uk

Figure 2.1: Austenite and variants of the low symmetry martensite in the continuum

2.1.2 Mathematical Background

The most striking feature of the multiwell Helmholtz Free Energy described above is
its nonconvexity. The nonconvexity of the energy introduces complications in under-
standing of the material’s macroscopic response. Ball and James [8] recognized that
microstructures form in these multiwell materials as a consequence of their noncon-
vex energies. This was independently concluded by Chipot and Kinderlehrer [24].
These microstructures, essentially mixtures of the low energy phases arranged in such
a way as to reduce the energy penalties associated with kinematic incompatability,
lead to overall ”relaxed” energies that are lower than the multiwell ones. However,
precise characterizations of these relaxed energies is a difficult analytical problem.
The proper analytical setting for the relaxation of energies is weak lower semi-
continuity. It has been known for some time [44] that a necessary and sufficient
condition for weak lower semi-continuity is quasiconvexity. Quasiconvexity is a non-
 10
local property. Consequently, it is in general difficult to calculate the quasiconvex
hulls of these multiwell free energies. It is possible to more easily bound the quasi-
convex hulls from above and below. Ball [6] developed the notion of polyconvexity,
and the polyconvex hull is an upper bound to the quasiconvex hull. From below,
the quasiconvex hull is bounded by the rank-one convex hull. The relation between
rank-one convexity and quasiconvexity is not completely understood. For functions
in space dimensions greater than two Sverak [63] has constructed a rank-one convex
function that is not quasiconvex. The problem of whether such a function exists in
the plane is still open.
It is worth mentioning that some special cases of quasiconvex hulls have been
found. The two-well problem has been fairly thoroughly examined. Ball and James
([9]) found the zero energy set of two nonlinear, compatible wells. Kohn addressed
the case of two linear wells [37] of equal modulus, as did Pipkin [49]. Chenchiah [23]
extended these results to the case of two linear wells of possibly unequal moduli.
Bhattacharya [10] found the hull of multiple, pairwise compatible, linear wells of
equal modulus. Govindjee et al. [33] estimated a lower bound in the case of many
wells by constructing a pairwise quasiconvex hull.
More difficulty arises in the case of the martensitic polycrystal. One has to
account for the relaxation of the multiwell energy within each of the grains as well as
intergrain compatablility problems that arise from the orientations of the different
grains. The work of Bruno et al.[21] focuses on the effects of this intergranular
interaction. By treating each grain as an isolated circular inclusion transforming
in an elastic medium, the elastic field of each transforming grain is found using
Eshelby [30] solutions. These solutions are easily superposed. The elastic response
of the ”polycrystal” is examined by finite dimensional minimization of the energy of
each grain. A similar line of inquiry that attempts to incorporate more of the grain
 11

Figure 2.2: Polycrystalline specimens have the added wrinkle of intergrain compata-
bility

geometry was conducted by Patoor and co-workers [47]. The grains in this model
are no longer treated as isolated from each other in their analysis. The analysis is
still conducted by using Eshelby inclusions. The elastic energy contrinuted by the
transforming inclusions is found by superposition. The result of these calculations is
a quadratic transformation energy that is easily incorporated into an FEM model.
These Eshelby-type calculations, though attractive for their relative computational
simplicity, serve only as a starting point for polycrystalline analysis. They generally
skirt the complex issues of microstructural formation and granular interactions.
 12
2.1.3 Experiments

The mathematical analysis of microstructures in shape-memory polycrystals is ma-

turing quickly. It remains, however, largely incapable of a precise characterization
of their macroscopic response in general. A parallel pursuit has been to attempt to
characterize the mechanisms of shape-memory and superelasticity experimentally. A
fairly thorough review of the historical development of this experimental literature
in single crystals is in the book of Otsuka and Wayman [46]. This thesis is more
interested in the recent systematic experimental examination of these phenomena in
polycrystals. Experimentally it has been known for some time that the texture of a
polycrystal drastically effects its elastic response. This was demonstrated in 1990 [31]
in the case of rolled sheets of nickel-titanium in uniaxial tension tests. The results of
Daly et al. [26] reinforce this strong textural dependence in thin sheets of NiTiNOL.
The failure of the resolved shear stress criterion for polycrystals, observed in their
experiments, indicates that the mechanics of onset of tranformation in polycrystals
is highly sensitive to load orientation with respect to the sample texture.
A systematic examination of the response of shape-memory polycrystals in mul-
tiple loading modes is required to completely explain the effect of texture. Lexcellent
et al. [41, 39] conducted a detailed investigation of the onset of transformation in
Nickel-Titanium, Cu-Zn-Al, Cu-Al-Ni and Cu-Al-Be alloys in biaxial proportional
loading. These investigations allowed for the formulation of yield surface for trans-
formation in the stress space. Sittner’s experiments [59] in 1995 similarly illuminated
the anisotropy of onset mechanics in combined loading. The experiments of Shaw
and Kyriakides [56] among others ([32],[26]) indicate that the onset of transforma-
tion is a local mechanism. The observed domains of martensite in thin sheets and
thin-walled tubes tend to nucleate in bands that widen and propagate throughout
 13
the sample until transformation saturates.
The experimental examination of the onset of transformation is interesting. An
understanding of the saturation of transformation is required to complete the story.
The experimental observations of Jung et al. ([36, 43]) on mixed-mode loading of
NiTiNOL tubes reveal that there are interesting mechanics still to be resolved after
the onset of transformation. The observation of secondary hysteresis loops in tension
after saturation of the transformation in tension indicated that the saturation process
was also anisotropic. In addition, the saturation does not proceed until the sample is
completely martensitic (unlike in the single crystal case); loading in torsion was able
to increase the total amount in martensite in the sample after the transformation
was saturated in tension.
The in situ microscopy experiments by Brinson et al. ([19, 55]) on polycrystalline
Nickel-Titanium lend some insight into the mechanics of phase transitions from on-
set all the way to saturations. These experiments confirm the local nature of the
initiation of transformation. Islands of martensite are seen nucleating in isolated
grains in the elastic austenite. The saturation of the transformation is arrested with
large interlocking ”fingers” of martensite oriented in the same direction. At satura-
tion these experiments clearly demonstrate that the intergranular constraints arrest
the transformation process before completion. The saturation occurs at martensitic
phase fractions between 60% and 70 % in these tests.

2.1.4 Homogenization Literature

The important class of ”yield surface” constitutive models were introduced purely
phenomenologically. An important question to ask when considering their connec-
tion to underlying micromechanics is whether they can be reconciled with existing
 14

a)

b)

Figure 2.3: Polarized Light Micrographs by Brinson et al. [19] at a) onset of trans-
formation and b) saturation
 15
ideas in the mathematical literature, and whether these ideas can be used to im-
prove the models. The total recoverable strain of a polycrystal is something that
has been examined in some detail in the homogenization literature. The work of
Kohn and Bhattacharya [13] attempted to determine the the total recoverable strain
of polycrystals with certain microstructures and textures. They demonstrate that
the Taylor (or constant strain) bound is usually a good approximation of the total
recoverable strain. This work was a novel use of the Taylor bound that has its origin
in the crystal plasticity literature (for example Kohn and Little [38] and the refer-
ences therein). Bhattacharya and Shu [58] expanded on the framework set out by
Kohn and Bhattacharya, examining the effect of polycrystalline texture, in the case
of compatible and incompatible martensitic variants, on shape-memory in multiaxial
loading cases. The importance of the Taylor and constant stress (Sachs) bound on
the response of polycrystal has led Schlömerkemper [54] and Schlömerkemper and
Bhattacharya [15] to characterize them in various cases. Among the interesting rev-
elations of these works was the striking difference in the shape of the Taylor and
Sachs bounds for a given microstructure in many cases.
An alternative to placing bounds on the total recoverable strain of the polycrystal
by determination of the Taylor set was proposed by Smyshlyaev and Willis [60]. The
formulation of a Hashin-Shtrikman variational principle is presented in terms of grain
statistics (particularly the two-point statistics of the polycrystal). An upper bound
of the total recoverable strain can be derived from the Hashin-Shtrikman variational
principle in the case of the statistically uniform polycrystal.
 16
2.1.5 Kinetics

The experimental observations of Shaw and Kyriakides [56] and Brinson et al. [19]
indicate that the martensitic transformation is a process that begins locally and
advances through the propogation of one or many phase boundaries. In order to close
a constitutive model in the sense of [3], a kinetic law(s) for the volume fraction of
martensite (or each variant of martensite) is required. In an attempt to understand
how such kinetic laws might be derived, the study of the advancement of phase
boundary in continuua has been a subject of keen interest. The propagation of
phase boundaries is thought to be due to a combination of metastability and pinning
([2, 11, 14]). The effect of pinning was investigated in some detail by Dondl ([28])
who found a universal power law between phase boundary speed and driving traction.
Dayal and Bhattacharya [27] had some success in numerically deriving a stick-slip
kinetics in a double well material through the use of peridynamics.

2.1.6 Constitutive Modelling

In order to incorporate these experimental observations into the design process, it is

necessary to derive constitutive model for their material response. Several attempts
have been made historically to incorporate the key mechanisms of shape-memory into
a constitutive framework. Abeyaratne and Knowles [4] developed a generic thermo-
mechanical model for a phase-transforming solid. This model featured a Helmholtz
free energy dependent on the martensitic phase fraction, a nucleation condition for
martensitic growth, and a kinetic law for the transformation process in one dimen-
sion. More continuum models followed in this general mold. Sun and Hwang [61]
developed a micromechanics inspired free energy to reproduce the superelasticity and
shape-memory effect under uniaxial loading. Boyd and Lagoudas [17] developed a
 17
Helmholtz free energy and dissipation potential for isotropic shape-memory alloys.
A final significant uniaxial, micromechanical model worthy of examination is that
of Huang and Brinson [35]. This uniaxial model in three dimensions considers the
formation of self-accomodating groups of martensite variants as a mode of energy
minimization. Tracking the formation of these energy-minimizing groups, the model
is able to accurately capture the uniaxial response of single crystals in uniaxial load-
ing. Brinson et al. [20] also developed a three-dimensional constitutive model based
on the notion of microplanes.
An alternative approach in the three-dimensional case was introduced by Auric-
chio and Taylor [5]. The model, incorporating an idea from the plasticity literature,
uses a ”yield surface” in the space of stress for the onset of phase transformation.
The yield surface is determined by fitting several uniaxial experiments in different
directions. As such it is not derived from a coherent micromechanics. However, it
has been quite successful in simplifying the computational implementation of SMA
response in commercial finite-element packages [1]. Lexcellent and his collaborators
([39, 41]) developed a micro-macro constitutive model in this vein. Their numeri-
cal and experimental determination of the onset yield surfaces is quite accurate for
CuAlBe, CuAlZn, and CuAlNi, but is not accurate for NiTi. Sadjadpour [53] ex-
amined the constitutive response in terms of the set of admissible transformation
strains. The key idea was to simplify the micromechanical considerations by treating
the transformation strain as an ensemble average of all of the variants appearing in
the microstructured sample. The definition of the set of transformation strains is the
convex dual of the onset stress surface incorporated by Auricchio and Lexcellent in
their models. Lexcellent and Laydi [40] examine this duality in detail. They describe
the surface transport between stress and strain spaces as well as the convexity condi-
tions for the class of surfaces they derived previously. It is important to notice that
 18
these single-surface models are not panacea. In each of the cases mentioned above,
the determination of the yield surfaces and their duals is accomplished through the
fitting of many uniaxial data points. This is natural as the definition of the yield
surface must have an associated normality rule. The downside of this in the polycrys-
talline case is that the polycrystal is replaced effectively with a single crystal. Many
of the important experimental observations in polycrystals cannot be incorporated
in this way. In particular, martensite reorientation and transformation saturation
are not incorporated in these models.
Some polycrystalline models attempt to reproduce the important characteristics
that are not accounted for in the yield surface models. Thamburaja and Anand [64]
model the polycrystalline sample by incorporating a single-crystal model on the grain
scale in a finite-element mesh. This model is able to capture the effects of texture
on several different types of load cycle. It is very computationally demanding as
the constitutive modelling is done on the level of the grain as opposed to true poly-
crystalline constitutive modelling. The computational model of Patoor et al. [47]
incorporates the grain structure of the sample through Eshelby solutions that are
ensemble averaged across the grains to determine a quadratic form for the energy
associated with the transformation strain. The modelling approach of Brinson and
her collaborators, that saw a good deal of success in the single crystal through the
incorporation of self-accomodating groups of martensite variants, was extended to
the polycrystal. Brinson and Panico [18] are able to capture the martensitic reori-
entation in three-dimensional polycrystals. The transformation strains associated
with each variant are tracked as internal variables as well as the strains associated
with the possible pairs of twinned martensitic variants in the polycrystal. Tracking
this (possibly large) roster of internal variables and the associated driving forces al-
lows for the microstructure in the sample to change during the loading cycle. This
 19
innovation shows one way of incorporating the heretofore elusive phenomenon of
martensitic reorientation.

2.2 Motivation
The objective of this thesis is to develop a constitutive relation for shape-memory
alloys that is simple and robust, but also capable of describing diverse phenomena
under complex thermomechanical loading so that it can be an effective tool in engi-
neering analysis and design.
The key heuristic and the point of departure of this work is the recognition that
the mechanics of initiation and saturation of the martensitic transformation in poly-
crystals are two essentially different processes. Consider a polycrystal completely
in the austenite state above its transformation temperature, and subject it to an
increasing stress. The initiation of transformation is governed by those grains that
are best oriented to the applied load. In particular, Brinson et al. [19] used in-
situ optical microscopy to observe that the first appearance of the martensite occurs
in the form of isolated regions in well-oriented grains. This is supported by the
mescoscale observations of Daly et al. [26] that deviations in linearity of the stress-
strain curve occurs well before the formation of macroscopic transformed regions.
Finally, Schlömerkemper and Bhattacharya [16] have recently proved in an idealized
setting with uniform modulus that the transformation begins in isolated grains, and
consequently the Sachs or Reuss constant stress bound accurately describes the ini-
tiation of the transformation. Thus, the essential mechanics of initiation is described
by treating the grains as essentially non interacting, isolated bodies in the elastic
austenite.
As the best oriented grains begin to transform, the grain boundary interactions
 20
begin to increase in importance. The transformed regions lead to inhomogeneous
stress and this in turn causes other grains to transform. Gradually the driving force
is large enough for the poorly oriented grains to start transforming as well. However,
they have smaller transformation strain in the direction of loading, and thus they
begin to quickly saturate and ”lock” together eventually leading to a network of fully
transformed grains. Thus, the saturation of the transformation is governed by the
poorly oriented grains. Indeed, Brinson et al. [19] observed substantial regions of
untransformed austenite even when the macroscopic stress-strain curve had turned
around to indicate macroscopic saturation of the transformation. Further, various
theoretical and computational analysis have shown that the constant strain Taylor
or Voigt bound gives a good description of the overall effective strains. [13, 58, 62].
Thus, the essential physics of saturation is described by looking at the poorly oriented
grains, and one has retained austenite when the transformation has macroscopically
saturated.
Finally, an important implication of the fact that the initiation and saturation of
the martensitic transformation in polycrystals are two essentially different processes
is that the critically resolved shear stress criterion or the Clausius-Clapeyron relation
fails in a polycrystal (see [26]), though it is known to hold well in a single crystal
(for example, [57]).
The discussion above has focussed on stress-induced transformation. The situ-
ation is slightly different in thermally induced martensite. Consider a polycrystal
completely in the austenite state above the transformation temperature . As it is
cooled, it transforms to martensite with two characteristic features. First, it is self-
accommodating so that there is no macroscopic change in shape. In other words,
even though individual unit cells change shape as a result of the transformation, the
grains form such a microstructure of the different variant that there is no macroscopic
 21
change in shape. Second, the transformation proceeds to completion. In other words,
there is no retained austenite. Now deform this self-accommodated martensite. The
microstructure changes to the extent it can under intergranular constraints to accom-
modate the applied load. However, as before, this deformation is constrained by a
network of poorly oriented grains. Consequently the constant strain Taylor or Voigt
bound gives a good description of the available strain through martensitic reorien-
tation [13, 58, 62]. Finally, when the polycrystal in the deformed state is heated, it
transforms back to the austenite accompanied with strain recovery.
 22

Chapter 3

Continuum Model

3.1 Kinematics
We seek to incorporate the heuristics described in section 2.2 into a continuum model.
We do so taking a multi scale view so that each material point of our continuum cor-
responds to a representative volume of material with numerous grains with possible
fine-scale microstructure in each grain. We denote the strain as ε and the tempera-
ture as θ.
We then introduce two key kinematic or internal variables to incorporate the
heuristic considerations described above. The first is the volume fraction λ of the
martensite. The second is the nominal transformation strain of the martensite εm
which we define as follows. Consider the representative volume that is partially
transformed and average the transformation strain of every region of martensite in
every grain in this representative volume. This is the nominal transformation strain.
Note that this is not the overall or effective transformation strain which is given
by λεm . We refer the reader to Sadjadpour [53] for a detailed discussion of these
variables.
These kinematic variables are subject to some constraints. The constraint on λ
 23
is obvious:
λ ∈ [0, 1]. (3.1)

The constraint on the nominal transformation strains is also relatively simple: it has
be a possible average of microstructures of martensite average over grains:
( )
X X
εm ∈ Gi := ξ:ξ= µi,n QTn εi Qn , µi,n ≥ 0, µi,n = 1 (3.2)
i,n i,n

where µi,n may be the volume fraction of the ith variant of martensite in the nth
grain and Qn is the rotation that describes the crystallographic orientation of the nth
grain. We call the set Gi the set of nominal transformation strains. Notice that this
set considers all possible arrangements of martensite with no regard to compatibility.
Therefore this set will play an important role when we consider initiation of trans-
formation. Finally consider the constraint on the effective transformation strain:
we require it be a value that one can obtain by making a kinematically compatible
microstructure of martensite:

λεm ∈ Gs := {ξ : ξ = average strain of a compatible microstructure of martensite.}

(3.3)
We call the set Gs the set of effective transformation strains. It is identical to the set
of recoverable strains defined by Bhattacharya and Kohn [13]. Notice that this set
limits itself to compatible microstructures and thus is limited by networks of poorly
oriented grains. Therefore this set will play an important role when we consider sat-
uration of the transformation. From the definitions it is clear that the set of nominal
 24
transformation strains is larger than the set of effective transformation strains:

Gi ⊃ Gs . (3.4)

Before we proceed, it is instructive to look at these constraints in a one-dimensional

situation. Here, εm is a scalar and the sets Gi and Gs are nested intervals:

Gi,s = [εci,s , εti,s ], εci ≤ εcs ≤ 0εts ≤ εti . (3.5)

The three constraints are plotted in figure 3.1, Note that when the material is in
the austenite and λ = 0, εm can explore the entire interval Gi . This is consistent
with the physics that the initiation of stress-induced transformation is controlled
by the best oriented grains. However, when the material is in the martensite and
λ = 1, it can explore only a smaller interval Gs because of the constraint on the
effective transformation strain λεm . This is consistent with the observation that the
deformation of the thermally induced martensite is constrained by the compatibility
of the grains. Similarly, note that the volume fraction can range from zero to one
when we have a self-accommodated martensite (εm = 0) as during cooling. However,
it can only explore a smaller region when εm is large as in the stress-induced situation.

3.2 Balance Laws

We postulate the usual balance laws of continuum mechanics. In local form, the
balance of linear momentum and energy may be stated as

ρutt = div σ, (3.6)

 25

λ λ
1 1

λεm≤ɛ t,c
s

εm≤ɛ t,c
i

εm λεm
ɛ ci ɛ ti

Figure 3.1: The volume fraction and the nominal effective strain are constrained to
lie in the identified set.

˙ = −∇q + r + σ : ε̇, (3.7)

where u is the displacement, ρ is the (referential) mass per unit length, σ is the
stress,  is the internal energy density, q the heat flux and r the radiative heating.
We also use the local form of the second law of thermodynamics,

q∇θ
−Ẇ − η θ̇ + σ : ε̇ − ≥ 0, (3.8)
θ

where W =  − θη is the Helmholtz free energy density, η the entropy density and θ
the (absolute) temperature.
 26

ɛ2

Gi

Gs ɛ1

Figure 3.2: A schematic representation of sets Gi,s

 27
3.3 Energy
We assume that the Helmholtz free energy density of the system is given by

1
W (ε, εm , λ, θ) = (ε − λεm ) : C (ε − λεm )
2  
θ
+λω (θ) − cp θ ln (3.9)
θ0
+Gi (εm ) + Gs (λεm ) .

The first term on the right hand side is the elastic energy density. Here it is assumed
that the total strain ε is the summation of the elastic component and the effective
transformation strain λεm . The material is assumed to have a constant elastic mod-
ulus tensor C in both phases for the sake of simplicity. To make the modulus a
function of phase fraction would not unduly complicate matters. A simple C (λ)
could be chosen by simply taking a weighted average of the austenite and martensite
moduli as the phase fraction varies from 0 to 1. Another, more exotic variation of
the modulus tensor with the martensitic phase fraction can be chosen. As long as it
is monotone, it should not change the tenor of the calculations to be undertaken.
The second term in the energy represents the change in chemical energy density
between the austenite and martensite phase at the given transformation temperature,
and can be written in the form

(θ − θc )
ω(θ) = L ,
θc

where L is the latent heat of transformation and θc is the thermodynamic transfor-

mation temperature (where the austenite and martensite phases are equally stable).
The third term is the contribution of heat capacity and cp is the specific heat which
 28

W
M
a
r
t
M e
n
i s
x i
A t
u t
s u e
t r
e e
n
i εM
t
e λεM

ε
Figure 3.3: A schematic representation of the first two terms of W

is assumed to be equal in both the austenite and the martensite.

Finally the last two terms in the energy describe the increased energy with in-
creasing transformation strain due to inhomogeneous stresses and also enforce the
constraints (3.1), (3.2), and (3.3). We discuss these presently.
 29
3.4 Initiation and Saturation
We postulate that the energy contributions Gi and Gs simply enforce the constraints
and consequently have the simple form:

 0 ξ ∈ Gi,s ,
Gi,s (ξ) = (3.10)
 +∞ else,

where

2 1
Gi,s = {ξ : tr ξ = 0 and (ξ · ξ)3/2 + bi,s det (ξ) + ci,s |eˆi,s · ξ eˆi,s |3 − gi,s 5 0},(3.11)
3 3

and bi,s , ci,s , gi,s are material parameters. Here, we assume that the material is trans-
versely isotropic about a direction ê. Therefore, the set can be described as a function
of the three principle invariants of ξ and the elongation along ê. However, we have
assumed self-accommodation so that the trace of all transformation strains are zero.
Thus, the set can be described as functions of |ξ|2 , det ξ and ê  ξê. We choose the
form above for ease and to have uniform powers of ξ.
The effect of the three parameters is as follows: gi,s determines the size of the
initiation and saturation surfaces respectively, bi,s determines the degree of tension
compression asymmetry in the sample response due to the determinant term (fig-
ure 3.4), and ci,s imbues the specimen with a degree of uniaxial eccentricity in the
direction êi,s (figures 3.5 and 3.6).
It is worth noting that the uniaxial eccentricities varied through the terms ci and
cs are certainly not the only anisotropies that could be feasibly incorporated into
this model. In theory the only restrictions imposed on the sets Gi and G∫ are those
of convexity and energetic frame indifference. The intent of the uniaxial eccentricity
 30

ε23

0.08

−0.08 0.08 ε33

b= 0
b= 3.5
−0.08 b= 5.0
b= 5.5
b= 6.0
b= -6.0

Figure 3.4: The degree of tension-compression asymmetry of the initiation and sat-
uration surfaces is adjusted through the parameter b.
 31

ε23 ε23

a) b)
0.08 0.08

−0.08 0.08
ε33 −0.08 0.08 ε33

c= 0

c= 1.0

c= 2.0
c= 4.0 −0.08 −0.08
c= 8.0
c= 16.0

Figure 3.5: The degree of eccentricity of the initiation and saturation surfaces is
varied through the parameter c. The effects of variation of this parameter are seen
on a) a symmetric set, and b) a set with a high degree of asymmetry.
 32

a) ε23 b) ε23

0.08 0.08

−0.08 0.08 ε33 −0.08 0.08

e2
α= 0 ê
α= ∏/2

α= ∏/3
α= ∏/4 −0.08 −0.08
α= ∏/6
α
α= 2∏/3

e3

Figure 3.6: The direction of eccentricity is controlled through variation of the vector
ê. The effects of variation of this parameter are seen on a) a symmetric set, and b)
a set with a high degree of asymmetry.
 33
is merely to account for circumstances, such as those previously mentioned experi-
ments on the effect of rolling direction on the response of thin sheets of NiTI, which
introduce a highly directional anisotropy in the material response.
The proposed functions for Gi and Gs are intentionally simple in their formula-
tion. In general, taken as they are, these functions may not be sufficient to fit a given
set of experiments. However, they should be able to incorporate the striking exper-
imental phenomena of tension-compression asymmetry and directional anisotropy.
An important example of this is in the isotropic but asymmetrical experiments of
Lexcellent et al. [39]. The surface of transformation onset stresses is determined
experimentally in the case of biaxial tension and compression. As the investigators
state, a yield surface that is a level set of the type of function that has been proposed
(a linear combination of norm and determinant terms) is inadequate to fit the given
experiments. They propose a surface of the type:
 
arcos(1 − a(1 − y))
F(y) = {cos ≤ f }, (3.12)
3

with a ∈ (-1,1) and y related to the determinant of the deviatoric stress.

27 det (devσ)
y=
32 . (3.13)
2 3
devσ : devσ
2

The distinction is drawn between the stress yield surface F(σ) and its convex dual
Gi (εM ) that is of interest to this model. On first observation it seems clear that the
dual of the type of surface described in 3.12 cannot be generally fit by those described
in 3.11. With a suitable choice of parameters, the general shape can be recovered.
Quantitatively the experimental data could, of course, be fitted in a least squares
manner. A more convenient choice is to select those values for the parameters bi
 34

a) 22
εΜ b) σ22(MPa)

600

0.06

200

0.02

11
−600 −200 200 600
σ (MPa)

−0.06 −0.02 0.02 0.06 ε11Μ

−200

−0.02

−600
−0.06
0.08

22
c) εΜ d) σ22(MPa)

600

0.06

200

0.02

11
−600 −200 200 600
σ (MPa)
11
−0.06 −0.02 0.02 0.06 ε Μ

−200
−0.02

−0.06 −600

Figure 3.7: A suitable choice of parameters for the set of admissible transformation
strains (a) allows for recovery of the tension-compression asymmetry in the biaxial
experiments in the dual stress space (b). A more precise fit to the experimental data
was proposed by Lexcellent in d) and its strain-space dual in c).
 35
and gi which allows for the incorporation of the key phenomenon presented. In this
case, since in the sequel we are concerned with tension-compression and tension-shear
tests, the choices of bi and gi are made to recover the tension-compression asymmetry
of the experiments. In figure 3.7 the parameters are set to bi = −1.85 and gi = 2.05
x 10−4 in order to meet this end. It is worth noting that the proposed F(σ) is convex
(and so must be its dual) and frame indifferent. It could be incorporated into the
model. This would be an excellent course of action if the biaxial load case was the
specific point of investigation of this thesis. The mathematical complexity that this
would add to the formulation might prove more cumbersome than it is worth in a
general investigation of how the nature of initiation and saturation surfaces effect
the macroscopic response of the alloy.

3.5 Driving Forces and Kinetics

With the free energy density specified, we can use the dissipation inequality and
arguments following Coleman and Noll [25] to obtain constitutive relations for the
stress and entropy:

σ = C(ε − λ εm ), (3.14)
  
L θ
η = λ − cp 1 + ln . (3.15)
θcr θ0

We also obtain the following driving forces as the thermodynamic conjugates to the
internal variables λ and εm respectively:

∂W
dλ = −
∂λ
∂Gs
= (ε − λεM ) : (C)εM − ω (θ) − : εM (3.16)
∂ (λεM )
 36
∂Gs
= = σ : εM − ω (θ) − : εM ,
∂ (λεM )
∂W
dεm = −
∂εM
∂Gi ∂Gs
= λC (ε − λεM ) − −λ (3.17)
∂εM ∂ (λεM )
∂Gi ∂Gs
= λσ − −λ .
∂εM ∂ (λεM )

In writing these relations, we have assumed that the functions Gi,s are smooth. In
the non-smooth situation (i.e., in the case in which the energies Gi,s are zero-infinity
wells), the differentiation that must be done in order to find the driving forces is
∂Gi ∂Gs
slightly more subtle. The tensorial derivatives ∂εM
and ∂(λεM )
must be understood as
subdifferentials, ∂εM Gi and ∂λεM Gs . This is a classical idea in the literature of convex
analysis (see for example [51]). Inside the set G, the subdifferential is identically 0.
On the boundary of G the subdifferential is multivalued (i.e., consisting of multiple
subgradients). The directional derivative of the constraint energy G on the boundary
of G is defined by the supremum:

G0 (ε; δx) = sup g : δx. (3.18)

g∈∂G(ε)

This convex program can be incorporated into a descent algorithm to minimize the
proposed energy numerically. However this can be an onerous process, amounting to
numerically solving constrained minimization problems with Lagrange multipliers.
Minimization is made much easier by a smoothing of the constraint energies. The
particular form of smoothing used in simulations is discussed in a later chapter.
We are now in a position to specify the evolution laws, or kinetic relations, for our
internal variables λ and εM . We assume that the martensitic variants can rearrange
 37
much more easily than the phase transformation can proceed, and thus εM has much
faster kinetics than λ. In fact, we take this to an extreme and insist on equilibriating
εM at each time. In the non-smooth case this reduces to:

εM = max σ : εM . (3.19)
εM ∈∂(Gi ∩λGs )

We assume that the phase transformation evolves according to the kinetic rela-
tion:
i) Stick-Slip
   −1
p

λ̇+
1 + 1
dλ ≥ d+
λ and λ ≤ 1

 +
(dλ −dλ )



   −1
λ̇ = p
(3.20)
λ̇−
1 + −
1
dλ ≤ d−
λ and λ ≥ 0
(dλ −dλ )






 0 else.

ii) Rate Independent Kinetics




 +∞ dλ ≥ d+
λ and λ ≤ 0

λ̇ = −∞ dλ ≤ d−
λ and λ ≥ 0
(3.21)



 0 else.

The rate-independent kinetic law reduces to a simultaneous equilibration of the vari-

ables λ and εM . When the driving force (σ : εM − ω(θ)) remains in (d− +
λ , dλ ) no

change in λ occurs. When the driving force threshold is exceeded, the new value of
λ can be determined in the non-smooth case by finding the equilibrium value of εM
 38
for each value of λ as in 3.19. The equilibrium equation for λ is given by:

λ= min |λ − λ0 | . (3.22)
σ:ε(λ)−ω(θ)∈(d− +
λ ,dλ )

It is important that the kinetic relation chosen incorporates some characteristics

observed in the experimental literature and the theory. Shaw and Kyriakides [56] ob-
served the stick-slip behavior of the martensite transformation. The growth of bands
of martensite was triggered when a critical tensile stress was exceeded and stopped
when the stress was relaxed below that level. Careful investigation of the hysteresis
associated with these transformation processes [7] showed that the energy under the
hysteresis loop did not change depending on the rate at which loading and unloading
occured. Traditionally, many microscopic models of phase transformations are mod-
eled with viscous kinetics which are linear near zero driving force. Chu, James and
Abeyaratne [2] and Bhattacharya [11] were able to reconcile the discrepancy between
the microscopically viscous kinetics and the stick-slip, rate-independent behavior ob-
served macroscopically. The defects present in a ”wiggly” energy landscape were
shown to locally arrest phase-boundary motion. The effect of this local pinning on
the microscopic viscous kinetics on the whole was a rate-independent stick-slip kinet-
ics. Dondl [28] extended this work to two- dimensional systems. Stick-slip behavior is
incorporated in both kinetic laws through the condition λ̇ = 0 for dλ ∈ (d− +
λ , dλ ). The

rate independence, the indeterminate transformation speed when the critical driving
force is exceeded, is incorporated by requiring a vertical tangent in both curves at
the critical driving force.
The important difference between the two kinetic laws is in the regime of high
driving forces. The rate-independent model makes the simplifying assumption that
the phase fraction of martensite equilibrates essentially infinitely quickly. The stick-
 39

λ+

-
dc
+
dc dλ

λ-

Figure 3.8: The kinetic relation governing the phase transition in the stick-slip case.
 40
slip condition makes an assumption, slightly more appealing to common sense, that
in the limit of high driving forces the rate of transformation must approach a limiting
speed asymptotically. This assumption of limiting speed, ostensibly the sound speed
of the material, follows the work of Purohit [50] for example. It is important to note
that the rationale for this assumption is a dynamic one. The setting of this problem
is quasistatic. Derivation of kinetic relations in the dynamic setting is still largely
an open issue.
 41

Chapter 4

Some Features of the Model

We now demonstrate a few features of the model.

4.1 One Dimension

We specialize the model to one dimension so that
 
1 θ
W (ε, εm , λ, θ) = C |ε − λεm |2 + λω (θ) − cp θ ln + Gi (εm ) + Gs (λεm ) , (4.1)
2 θ0

where the sets Gi,s are given by (3.5). We further assume that kinetics of εm is
extremely fast so that it minimizes the energy at each instant of time and that λ
follows a strict rate-independent stick-slip kinetics:

λ̇ = 0 for dλ ∈ (d− +
λ , dλ ), dλ ∈ [d− +
λ , dλ ], dλ λ̇ ≥ 0. (4.2)

4.1.1 Stress-Induced Martensite

Consider an isothermal strain-controlled experiment where the temperature is held

constant at a value significantly higher than the transformation temperature so that
 42
λ Gs(λεM)
σ
Gi(εM)

1 2

3 2

4 3

ε ε
4 1

Figure 4.1: Stress-induced martensite in one dimension shown in the volume fraction-
strain plane on the left and the stress-strain plane on the right. The constraints on
the volume fraction and nominal transformation strain are indicated on the shaded
set.

ω(θ) > −d−

λ > 0. Consider the specimen at rest at zero stress so that it is in the

austenite state with λ = 0 and εm ∈ [εci , εti ] indeterminate. Now subject the specimen
to a monotonically increasing overall strain tensile strain ε(t). For very small times,
the stress is given by σ(t) = Cε(t). Since this is positive, the nominal transformation
strain εm , takes its maximal tensile value εti and the driving force for transformation,
dλ = σεti − ω is too small begin transformation (i.e., in the range (d− +
λ , dλ ) so that

λ̇ = 0). Thus the volume-fraction, stress and strain begin from zero and traverse
toward the point marked 1 in figure 4.1.
As the applied strain increases, so do the stress and the driving force dλ until we
reach the point marked 1 in the figure where dλ = Cε − ω = d+
λ and transformation

begins. At this point, the strain and stress are given as

d+ +ω d+ +ω
εtM S = λ t , t
σM S = λ t . (4.3)
Cεi εi

At this point transformation begins and proceeds in such a manner to keep dλ and
consequently the stress σ constant so that we traverse from the point marked 1
 43
towards the point marked 2 in figure 4.1 as the applied load increases. As λ increases,
the overall transformation strain λεm eventually saturates the constraint λεm ≤ εts
at the value
tεts
λ = t. (4.4)
εi
This is indicated by the point 2 in the figure, and we have

d+
λ +ω d+
λ +ω
εtM F = + εts , t
σM F = . (4.5)
Cεti εti

The transformation is now saturated, and further loading does not further lead to
any further transformation. Thus, the stress increases linearly to the applied strain
as indicated by the increasing branch of the stress-strain curve in figure 4.1.
Now start unloading the specimen by monotonically decreasing the applied tensile
strain. There is no transformation initially and we unload elastically (the stress
decreasing linearly with strain) until we reach the point marked 3 when the driving
force has reached a low enough value to start reverse transformation, dλ = σεti − ω =
d−
λ so that
d−
λ +ω d−
λ +ω
εtAS = t
+ εts , t
σAS = . (4.6)
Cεi εti
Reverse transformation now begins as λ begins to decrease as unloading proceeds.
The driving force and stress remain constant and we traverse from point 3 to point
4 on figure 4.1. The reverse transformation is complete at the point 4 when λ = 0,
and
d−
λ +ω d−
λ +ω
εtAF = , t
σAF = . (4.7)
Cεti εti
The material responds elastically on further onloading and we return to the origin.
We have an analogous situation in a compressive loading cycle with the analogous
quantities obtained by replacing the superscript t with c in the formulae (4.3) to (4.7)
 44
above.
A series of comments are in order. First, note that the transformation from
austenite to martensite does not go to completion at a maximum volume fraction of
λt,c . This is consistent with the observations of Brinson et al. [19].
Second, tension and compression are different. This is consistent with various
observations going back to Burkart and Read [22].
Third, the value of stress at which the transformation begins and completes or the
reverse transformation begins and completes depend on temperature through ω. If
ω depends linearly in temperature as in most common models ω = L(θ − θc ) where L
is the latent heat, then these stresses depend linearly on temperature consistent with
the Clausius-Clapeyron relation in a particular deformation mode. Further, we can
invert these relations to obtain the values of temperature where the transformation
begins etc at zero stress:

d+
λ d−
λ
Ms = Mf = θc − , As = Af = θc + . (4.8)
L L

Fourth we can infer the values of transformation strain, the stress hysteresis and
mean-value of stress in the hysteresis to be

1 t,c
εt,c
trans := (ε − εt,c t,c t,c
M S + εAS − εAF ) = εs ,
t,c
(4.9)
2 MF
−
t,c 1 t,c t,c t,c t,c d+λ − dλ
σhyst := (σM S + σM F − σ AF − σ AS ) = , (4.10)
2 εt,c
i
−
1 t,c t,c t,c t,c 1 d+λ + dλ + 2ω
σ t,c := (σM S + σM F + σAF + σ AS ) = . (4.11)
4 2 εt,c
i

Note that each of them can be independently constitutively prescribed and that
there is no universal relation amongst them. This is a manifestation of the lack
 45
of any ”resolved stress” criterion or Clausius-Clapeyron relation across deformation
modes.
t,c t,c t,c t,c
Fifth, we see above that σM S = σM F , σAS = σAF and Ms = Mf . These are

all manifestations of i) rate-independent or strictly stick-slip kinetics (3.21) and ii)

isothermal conditions. If we assume that the material has a rate-dependent kinetic
t t
relation as in (3.20), then we would have σM F > σM S with the difference depending on

the loading rate. We would also have these if we assume non-isothermal situation. To
elaborate on this, let us assume rate-independent kinetics and adiabatic conditions.
Then, we have to solve (3.7) with q = r = 0. If we further assume that ω = L(θ −θc ),
we can show that the contribution of the latent heat to the driving force will require
an increase in stress to sustain and saturate the martensitic transformation. Recall
that
dλ = σ : εt,c t,c
i − ω = σ : εi − L(θ − θc ). (4.12)

In order to initiate transformation, the driving force dλ must equal the critical driving
force d+
λ . The stress required to initiate transformation is written

d+λ L(θ0 − θc )
σM S = t,c − . (4.13)
εi εt,c
i

In the isothermal case, the temperature remains constant at θ0 throughout the trans-
formation. In the adiabatic case, the constitutive assumption

cp θ̇ = θλ̇L (4.14)

must be integrated in order to yield the temperature change throughout the trans-
formation process.  
L(λ(t) − λ(0))
θ(t) = θ0 exp (4.15)
cp
 46

Gi(εM) Gs(λεM) λ
3 4
2
5
4

3
2 ε
5

7
ε
1 6
7

Figure 4.2: Shape-memory effect in one dimension shown in the the volume fraction-
strain plane on the left and the stress-strain-temperature space on the right. The
constraints on the volume fraction and nominal transformation strain are indicated
on the shaded set.
 
Lλsat
The temperature at saturation is going to be θf = θ0 exp cp
. Plugging this into
the driving force equality yields the stress as the saturation of transformation

d+λ L(θf − θc )
σM F = t,c − , (4.16)
εi εt,c
i
   
Lθ0 Lλsat
σM F − σM S = t,c exp −1 ; (4.17)
εi cp

yields the extent of hardening in the adiabatic situation.

4.1.2 Shape-Memory Effect

We now subject our one dimensional specimen to a controlled (infinitely slow) temperature-
load cycle to explore the shape-memory effect. We begin, as before with an unloaded
 47
specimen at a temperature above Af . This is marked as the point 1 in figure 4.2. We
lower the temperature below Ms and Mf so that the material transforms to marten-
site and λ goes to 1 as we traverse from point 1 to point 2 in the figure. We now keep
the temperature constant and subject the material to a strain controlled loading un-
loading cycle. The moment the strain and consequently the stress becomes positive,
the nominal transformation strain εm takes the value εti to optimize the energy and
we traverse from point 2 to point 3 in the figure. Now the overall strain is the trans-
formation strain εts . Note that this happens at a value of stress equal to 0+ since we
assume that the kinetics of εm is infinitely fast. Further loading takes us to point 4,
and unloading brings us to 5 (with coincides with 3 in the stress-strain-temperature
space). We now keep the stress zero and heat the specimen. There is no change in
strain or transformation and traverse from point 5 to point 6 (which coincide on the
left) in the figure. At point 6, the temperature has reached As so that dλ = ω = d−
λ,

the reverse transformation begins and the volume fraction decreases. We traverse
from point 6 to point 7 (which happen to coincide on the right) on the figure. As the
volume fraction decreases at this fixed temperature, it eventually reaches the point
λt at which point the constraint on the effective transformation strain is activated.
Now, any further reduction in volume fraction can only occur with a reduction in
εm and strain recovery. Therefore we traverse from point 6 to point 7 in the figure.
Further heating brings us back to the starting point.
Note that in the thermal cycle, there is no restriction on the amount of transfor-
mation, and the specimen is able to transform fully to martensite. Once again, rate
and heat transfer effects would make the curves rounded.
 48

λ=0 λ=0.1

ɛ23 ɛ23

ɛ33 ɛ33
Gi
Gi

1 Gs
λ

λ=0.25 λ=0.56

ɛ23 ɛ23

ɛ33 1 Gs ɛ33
Gi
Gi λ

1 Gs
λ

Figure 4.3: The different eccentricities of the set Gi and Gs can lead to a signifi-
cant reorientation of the martensite. The nominal transformation strain at different
extents of transformation is shown.
 49
4.2 Martensite Reorientation
An important aspect of the interaction that we seek to incorporate in our model is the
fact that initiation and saturation are controlled by different aspects in a polycrystal.
Therefore it is possible for the martensite to reoirent as the transformation proceeds,
i.e., the variants and proportions that are present at initiation may be quite different
from those present at saturation. We now demonstrate how this arises in our model
as a result of different eccentricities of the sets Gi and Gs . We note that the sets
Gi and Gs can have different eccentricities even when they have consistent material
symmetry since the ratios of the parameters ai , bi , ci can be different from those of
as , bs , cs . Indeed, Schlömerkemper [54] has recently pointed out that the eccentricities
of the Sachs and Taylor estimates of the set of recoverable strains of an isotropic
polycrystal made of a material undergoing cubic-orthorhombic transformation can
be quite different.
We now consider an idealized setting where the stress and strain are two dimen-
sional vectors. We assume that the set Gi is elongated along the (εm )1 direction while
the set G∫ is elongated in the (εm )2 direction as noted in figure 4.3. We consider a
situation when the crystal is above Af , and subject it to a steady dead load in the
direction σ sufficiently large to eventually initiate and saturate the transformation.
As λ starts at zero and evolves until it eventually saturates at some value λmax ,
the nominal transformation strain εm has to be optimized subject to the constraint
that it is restricted to the set Gi ∩ λ1 Gs . This is done by finding the point in the set
Gi ∩ λ1 Gs that has a maximum projection in the direction σ. This is demonstrated in
figure 4.3 for various values of λ. As λ changes, εm changes from a direction where it
is optimized in Gi to a direction where it is optimized in Gs . Due to the fact that the
initiation and saturation constraints have markedly different eccentricities, these di-
 50
rections are quite different resulting a significant reorientation as the transformation
proceeds.
 51

Chapter 5

Demonstration and Parameter

Study

We now turn to a systematic demonstration and parameter study of the model

presented in chapter 3.
For the sake of consistency we specify a set of base parameters which shall be
used in the absence of statements to the contrary. In order to replicate the tension-
compression asymmetry seen in the onset surface of Lexcellent [41], mentioned pre-
viously, the initiation parameters are set as bi = −1.85 and gi = 2.05 x 10−4 . No
asymmetry is assumed in general for the saturation surface (bs = 0). As the satura-
tion surface must be contained within the initiation set, gs is set to 2.25 x 10−5 . No
anisotropy is assumed as a base for either surface. The Young’s modulus is chosen
as 115GPa.
For the thermal parameters we utilize the same base set as Sadjadpour [53]. The
latent heat L = 79 M J
, the heat capacity cp = 5.4 mM3JK , the Martensite start



m3

Ms = 217.60K, and Austenite start As = 266.79K are chosen as base parameters.

 52
These are used to select the kinetic parameters.
   
As − Ms J
d+
λ = −d−
λ =L = 8x10 6
(5.1)
As + Ms m3

As +Ms
Using the fact that the critical temperature, θcr = 2
, the isothermal loading tests
are conducted at a temperature chosen to make ω(θ) a round number (θ = 333K):
 
L J
ω(θ) = (θ − θcr ) = 1x107 . (5.2)
θcr m3

Finally, the stick-slip exponent p is chosen to be 2.

It is assumed that the temperature is well above the transformation temperature.
In the absence of stress, the austenite is the stable phase.
This demonstration focuses on the interplay between the onset and saturation
constraints. Particularly we are concerned with how the geometry of each constraint
can give rise to some of the important features discussed in the previous chapter.

5.1 Uniaxial Tension and Compression

We consider the isotropic situation (ci = cs = 0) with strict rate-independent stick-
slip kinetics (3.21) for the volume fraction. We consider a specimen well above
transformation temperature so that λ = 0 and subject it to an uniaxial tension or
compression. The resulting stress-strain curve is as shown on the right-hand side of
figure 4.1 as we presently see. Assume that the applied load is along the e3 axis so
that the deviatoric stress is

σ
σ̂ = diag{−1, −1, 2}.
3
 53
As the load increases from zero, the nominal effective strain εm takes the value that
optimizes its projection along the applied load (εm : σ̂) amongst all allowable values
in the set Gi . Since we are in the isotropic case, symmetry dictates that the optimal
strain will be of the form
x
εm = diag{−1, −1, 2},
3
for some known x. Substituting this in the formula (3.11) for Gi , we conclude that
the optimal x satisfies,

  13
  32  
2 2 2 2 3 gi
x ± bi x = gi =⇒ xt,c = ±   , (5.3)
3 3 t,c 27 t,c
5
2 2
± 2
b
3 27 i

with the choice of positive sign for tension and negative for compression. We sub-
stitute this in the equation for the driving force, and find that the stress is initially
too small to induce any transformation. So the material responds elastically. The
transformation begins when the driving force it reaches its critical value d+
λ.

We conclude that transformation initiates at the following stress:

!  5 ! 13
t,c 3 ω (θ) + d+
c 2 2 2
σM S =± 1 ± bi , (5.4)
2 gi
3 3 27

with the choice of positive sign for tension and negative for compression. Since the
response is linear up to this point, the onset strain is obtained by dividing these
with the Young’s modulus. Since we have a strict rate-independent kinetics, the
transformation proceeds at constant stress until we effective transformation strain
saturates the constraint λεm ∈ Gs . This leads to equations similar to (5.3) with
x replaced by λx, and bi , gi with bs , gs . We conclude that the saturation volume
 54
fractions of the martensite are given by


25  13
2 2  31
±

b
27 i  gs
λt,c
sat = 3

52 , (5.5)
2
± 2
b gi
3 27 s

with the choice of positive sign for tension and negative for compression. The associ-
ated transformation strain, the change in strain during the transformation, is given
by

  13
gs
εt,c
trans =

5  , (5.6)
2 2 2
3
± b
27 s

with the choice of positive sign for tension and negative for compression.
The transformation is now saturated. Thus, as the load increases the material
responds elastically. Now start unloading. The driving force is too large to have any
reverse transformation. The reverse transformation begins when the driving force
reaches d−
c and this corresponds to an applied stress of

!  5 ! 31
t,c 3 ω (θ) + d−
c 2 2 2
σAS =± 1 ± bi , (5.7)
2 gi 3 3 27

with the choice of positive sign for tension and negative for compression. The volume
fraction of martensite decreases at constant stress until it reaches zero, and the change
in strain during this process is identical to transformation strain (5.6). As the load
decreases further and to zero, the material unloads elastically to the origin.
Comparing (5.4) and (5.7), we find that the stress hysteresis during this cycle is
 55
given by

!  5 ! 13
−
t,c 3 d+
c + dc 2 2 2
σhyst =± 1 ± bi , (5.8)
2 gi3 3 27

with the choice of positive sign for tension and negative for compression.
We now make a series of comments about these results and the various parameters
t,c
of the model. We first consider the critical stress for transformation σM S given in

(5.4). This stress depends on temperature through the function ω(θ). If this is linear
as is commonly assumed with the slope related to the latent heat, then the critical
stress for transformation will also change linearly with the temperature with the slope
related to the latent heat. This is consistent with various experimental observations
[48]. We also see that at any given temperature, the parameter gi controls the value
of this critical stress for transformation in both tension and compression, while the
parameter bi independently controls the difference in the value between tension and
compression. Thus, parameter bi is responsible for the asymmetry in stress between
tension and compression. We now turn to the transformation strain described in
(5.6). Note that this depends on the parameters gs and bs , and is thus completely
independent of the critical stress for transformation. The parameter gs controls the
magnitude of this strain in both tension and compression while the parameter bs
controls the difference between tension and compression. Finally, the parameters d±
c

control the amount of stress hysteresis, see (5.8). In summary, the model has enough
freedom to assign the critical stress, the transformation strain, and stress hysteresis
independently, and independently in tension and compression.
 56
5.2 Simple Shear
We now turn to simple shear, and show that the parameters bi and bs have a more
subtle effect on the transformation strains and cause the transformation strain to be
nonparallel to the applied stress even in the isotropic setting. As before we work at
a constant temperature well above the transformation temperature and start with
λ = 0. We apply an increasing stress of the form
 
0 0 0
 
σ= 0 0 τ .
 
 
0 τ 0

As the stress increases, the nominal transformation strain increases to its optimal
value with the set of constraints. By symmetry, it has the general form
 
−x − z 0 0
 
εm =  z y .
 
0
 
0 y x

To find the specific value, we need to find the point where the normal to ∂Gi is
parallel to the applied stress, or ∂Gi /∂m is parallel to σ. One has to be careful in
this differentiation since it has to be carried out an interpreted in the space of trace-
free matrices. Thus, if one differentiates these functions in the space of matrices, one
has to project the derivative on to the space of trace-free matrices, or
 
∂Gi
P || σ,
∂m
 57
where P is the projection to the space of trace-free matrices. In our case, the deriva-
tive of the determinant in the space of matrices gives the cofactor, and the cofactor
of trace-free matrices need not be trace-free. Working through these details, we find
that this condition is satisfied if and only if the parameters x, y, z satisfy:

2 2 1 1
2Ez − zx − x2 + y 2 + z 2 = 0, (5.9)
3 3 3 3
2 1 2 1 2 2 2
2Ex − zx + x + y − z = 0.
3 3 3 3

where E = |εm | bi . Subtracting the second equation above from the first yields the
condition

2E(z − x) − x2 + z 2 = 0 ⇐⇒ −2E(x − z) = (x + z)(x − z).

This leaves two possibilities, (x − z) = 0 and (x + z) = −2E. The latter possibility

implies that ∂Gi /∂εm is identically zero and is hence rejected. We conclude that
x=z  
−2x 0 0
 
εm =  x y ,
 
0
 
0 y x

Substituting it back into (5.9) and doing some manipulation yields

   
24 4 8 2 1
2
−1 x + 2 + x2 y 2 − y 4 = 0.
bi bi 3 9

We have to solve this simultaneously with the condition that εm ∈ ∂Gi or

2
3
6x2 + 2y 2 2 + bi 2xy 2 − 2x3 − gi = 0,


3
 58
for x, y.
Unfortunately, we are unable to solve this in closed form. Yet, the above equations
reveal two important features. First, except in the case bi = 0, applied simple shear
gives rise to nontrivial normal components in the nominal transformation strain even
in the isotropic situation. This is akin to the Poynting effect, and implies that the
normal boundary constraint is critical in studying torsion. Second, notice that these
conditions are impervious to the change of sign of σ. This is true in all aspects, and
thus there is no asymmetry with respect to sign unlike in uniaxial extension.
Once we know the optimal nominal transformation strain ε0m , we can use it to
determine driving force and in turn the stress at which the onset of transformation
begins. As the transformation proceeds and λ increases until the effective transfor-
mation strain λε0m saturates the constraint Gs . However, depending on the values
bi and bs , λε0m may not be the matrix that is optimal in Gs for the applied load. If
it is not, the martensite reorients and begins the nominal transformation strain εm
changes until λεm is optimal in Gs for the applied load. Thus, a misorientation of Gi
and Gs can lead to realignment of martensite.

5.3 Combined Uniaxial Extension and Shear

We now turn to strain-controlled experiments combining uniaxial extension and
shear. This loading protocol is motivated by the combined tension-torsion of a
tube. As before we consider an isotropic situation well above the transformation
temperature and begin with a specimen completely in the austenite. We load it until
transformation begins and saturates, and then unload it completely.
The algebra becomes too difficult, and we therefore resort to numerical studies.
Since hard constraints are difficult to implement numerically, we replace the hard
 59
constraints, (3.10), (3.11) with the following high order polynomials:

!16
2
3
(ξ · ξ)3/2 + bi,s det (ξ) + 13 ci,s |eˆi,s · ξ eˆi,s |3
Gi,s (ξ) = . (5.10)
gi,s

We shall have a opportunity to evaluate the influence of this constraint presently

We begin in the isotropic situation (ci = cs = 0) with a strain-control test where
we apply a strain of the form
 
0 0 0
 
ε =  0 0 γ(t)  , (5.11)
 
 
0 γ(t) ε(t)

with γ(t) = cε(t). In this case c is chosen to be equal to 1.

Figure 5.1(a) shows the resulting trajectory of the stress where σ denotes the
uniaxial (33) component and τ the shear (23) component. Note that the stress
components (σ-τ ) are not proportional although the applied strain components (-γ)
are. This is consistent with the experimental observations of Anand and Thamburaja
and Jung [64], [43].
The underlying reason for this non-proportionality is the reorientation of marten-
site, and this is clear in the space of transformation strains shown in figure 5.1(b).
This figure shows the trajectory of the normal and shear components of the effective
transformation strain in blue bold line, the constraint Gi on the nominal transforma-
tion strain as the dashed black arc and the constraint Gs on the effective transforma-
tion strain as the dotted red arc. It also shows the imposed strain direction as the
purple dashed line. The effective transformation strain starts out at zero and grad-
ually increases in a direction that is optimal for the applied load. Thus the original
 60
τ(MPa)

300

200

100

100 200 300 σ(MPa)

(a)
γΜ

0.07

0.05

0.03

0.01

0.02 0.04 0.06 0.08 0.10

εΜ
(b)
σ(MPa) τ(MPa) σ eq (MPa)
300 400 600

200

100 200

0.02 0.06 ε 0.02 0.06 γ 0.05 0.10 εeq

(c) (d) (e)

Figure 5.1: A proportional strain-controlled extension-shear test of an isotropic ma-

terial. (a) The resulting stress, (b) the resulting transformation strain along with a
verification of the numerical method, (c) uniaxial stress vs. uniaxial strain (d) shear
stress vs. shear strain and (e) equivalent stress vs. equivalent strain.
 61
direction depends on the nominal transformation strain and thus on the shape of the
set Gi . Therefore it can be-and is so in this case-different from the direction of the
applied total strain. The effective transformation strain increases along this direc-
tion as the transformation proceeds until it encounters the boundary of the set Gs .
It then tries to optimize itself within this latter set and this results in reorientation
of the martensite, and this is clearly visible in this case. It also causes the observed
reorientation in the direction of stress.
Note that we have reorientation of martensite in this case even though it is
isotropic. The determinant-dependant terms (those with the coefficients bi,s ) that
were responsible for the tension-compression asymmetry enable the initiation and
the saturation surfaces to be eccentric.
Figure 5.1(c-e) show the stress-strain curve in components and in equivalent form.
Note that these curves can look quite different in the normal and the shear directions.
In particular, the shear curve appears to have a negative slope. This is because the
reorientation of martensite gives rise to a significant change in the shear component
of the transformation strain. However, the negative slope does not necessarily signal
an instability.
We conclude with a comment on the softened polynomial constraint (5.10). Fig-
ure 5.1(b) plots the surfaces according to the hard constraint, and we see that the
trajectory computed with the soft constraint tracks is quite closely. We also see a
little overshoot in the stress-strain curves due to the numerical method.

5.4 Anisotropic Materials

The discussion has until now focused on isotropic materials. We now consider uni-
axial or transversely isotropic materials subjected to combined extension along the
 62
γΜ τ(MPa)
a) b)
0.04 300

200

0.02

100

εΜ σ(MPa)
0.02 0.06 100 300 500
σeq (MPa) λ
c) d)
700 0.70

500 0.50

300 0.30

100 0.10

0.02 0.06 0.10

εeq 0.02 0.06 0.10
εeq

Figure 5.2: Anisotropy of the initiation surface. Combined extension and shear of
an uniaxial material under strain control for various values of ci ranging from 0.3-
1.5. (a) The effective transformation strain trajectory. The applied strain trajectory
is indicated by the dashed line. (b) The stress trajectory, (c) equivalent stress vs.
equivalent strain (d) volume fraction vs. equivalent strain.

axis of symmetry and shear. In other words, we set ei = es = e3 with ci , cs possibly

non-zero and apply a strain of the form (5.11).
Figure 5.2 shows the results for increasing anisotropy of the initiation surface
or various values of ci ranging from 0.3 to 1.5 (by equally spaced increments) for
cs = bi = bs = 0. When ci = 0, the initiation and saturation surfaces are coaxial
and aligned. Thus there is no reorientation of martensite and the transformation
strain and stress are proportional and is close to the direction of applied strain (the
small difference depends on the elastic modulus). When ci is non-zero, the initiation
 63
surface is eccentric. Therefore the optimal nominal transformation strain deviates in
direction from that of the applied strain. Thus, the effective strain trajectory starts
out on a different direction, though the stress appears to maintain the same direction.
Further, the initiation and saturation surfaces are no longer aligned. Thus, when the
effective transformation strain is large enough, we see reorientation of the martensite,
and consequently non-proportional transformation strain and stress. Further, for
high values of ci , we see that the equivalent stress-equivalent strain curve shows a
downward slope. However, this does not indicate any instability. Finally, we see that
the transformation is not complete even for large stresses.
Figure 5.3 shows the results of anisotropy in the saturation surface various values
of cs ranging from 0.3 to 1.5 in equal increments for ci = bi = bs = 0. Since the
initiation surface is held fixed, each trajectory of the effective transformation strain
starts out in the same direction. However, the saturation and reorientation depends
on the value of cs . The amount or reorientation happens to be less dramatic than in
the previous situation.
Figure 5.4 shows the results of anisotropy in both the initiation and saturation
surfaces. It proportionally varies the parameters ci , cs in the range 0.3 to 1.5. The
trajectory of the effective transformation strain shows features of the earlier cases,
the directions start out differently and the reorientation is quite significant.

5.5 Proportional Loading in Different Directions

In the previous set of simulations, we saw that the geometry of initiation and sat-
uration surfaces have an significant effect on the overall hysteretic behavior of the
material as well as the reorientation of martensitic transformation strain. We now
examine the role of loading direction. We fix the initiation surface by fitting it to the
 64

γΜ τ(MPa)
a) b)
0.04

200

0.02
100

0.02 0.06
εΜ 100 300 500
σ(MPa)
σeq (MPa) λ
c) d)
0.70

600

0.50

400

0.30

200

0.10

0.02 0.06 0.10

εeq 0.02 0.06 0.10
εeq

Figure 5.3: Anisotropy of the saturation surface. Combined extension and shear of
an uniaxial material under strain control for various values of cs ranging from 0.3 to
1.5. (a) The effective transformation strain trajectory. The applied strain trajectory
is indicated by the dashed line. (b) The stress trajectory, (c) equivalent stress vs.
equivalent strain, (d) volume fraction vs. equivalent strain.
 65

γΜ τ(MPa)
a) b)

0.04

200

0.02
100

0.02 0.06
εΜ 0.02 0.04
σ(MPa)
c)
σeq (MPa) d) λ
0.70

600

0.50

400

0.30

200

0.10

0.02 0.06 0.10

εeq 0.02 0.06 0.10
εeq

Figure 5.4: Anisotropy of the both surfaces. Combined extension and shear of an
uniaxial material under strain control for various values of ci , cs ranging from 0.3 to
1.5 (in equal increments). (a) The effective transformation strain trajectory. The
applied strain trajectory is indicated by the dashed line. (b) The stress trajectory,
(c) equivalent stress vs. equivalent strain, (d) volume fraction vs. equivalent strain.
 66
γΜ τ(MPa)
a) b)
0.10 500

300

−0.10 0.10
εΜ

−0.10
−600 −200 200
σ(MPa)
c)
σeq (MPa) d) λ
0.70

600

0.50

400

0.30

200

0.10

0.02 0.06 0.10

εeq 0.02 0.06 0.10
εeq

Figure 5.5: Variation of direction applied strain. (a) The effective transformation
strain trajectory. The applied strain trajectories are indicated by the dashed lines.
(b) The stress trajectory, (c) equivalent stress vs. equivalent strain, (d) volume
fraction vs. equivalent strain.

experimental results of Lexcellent and the saturation surface. Specifically, bi = −1.8,

gs = 2.05 x 10−4 . A slight eccentricity (coaxial) and asymmetry have been added to
the saturation surface to underscore the effect of these parameters on the martensitic
reorientation: bs = 1.0, cs = 1.5, and gs = 2.25 x 10−5) .
Figure 5.5 shows the response of the material for proportional applied strain with
different ratios of extension and shear. The direction of applied strain in the various
cases are indicated by the dashed lines, and the resulting trajectory of the effective
transformation strain is shown in the continuous line of the same color in figure
5.5(a). The two surfaces as well as the trajectory of the effective transformation
 67

a)
γΜ b) γ
0.10

0.06

−0.10 0.10
εΜ

−0.10
−0.05 0.05
ε
c) σeq (MPa) d) λ
700 0.70

500 0.50

300 0.30

100 0.10

0.02 0.06 0.10

εeq 0.02 0.06 0.10
εeq

Figure 5.6: Variation of direction applied stress. (a) The effective transformation
strain trajectory. (b) The stress trajectory, (c) equivalent stress vs. equivalent
strain, (d) volume fraction vs. equivalent strain.

strain are also shown. We see that the amount of deviation between the direction
of the applied strain and the initial direction of the transformation strain (which
arises due to the large distortion of the initiation surface normal vectors due to
the considerable determinant term added to incorporate the tension-compression
asymmetry) as well as the amount of reorientation of martensite varies considerably
with loading direction. The resulting equivalent strain-equivalent stress curves also
vary significantly. Importantly, the amount of transformation or the volume fraction
of martensite on saturation depends significantly on the loading direction. This will
turn out to be very important when we consider nonproportional tests.
 68

Figure 5.7: The experimental observations of Jung [36] show the emergence of sec-
ondary hysteresis in nonproportional tension-torsion.

Figure 5.6 shows the corresponding response in stress control. The observed
features are similar.

5.6 Nonproportional Loading

The fact that the response of the material can be different in different directions mo-
tivates us to study nonproportional loads. An interesting case are the experimental
observations of Jung [36] reproduced in figure 5.7. They were studying the combined
tension and torsion of thin walled tubes, and observed a dual plateau for when they
applied a uniaxial extension followed by torsion. Thin-walled NiTi tubes (0.37 mm
wall thickness) were loaded slowly (10−5 s−1 to 10−4 s−1 ) avoid rate effects. The load-
ing program followed was to load to tensile strain of 6% followed by a torsional strain
 69

γΜ τ(MPa)
a) b)
0.08 500

300

εΜ
−0.10 0.15
100

σ(MPa)
100 300 500
−0.06 −100

σeq (MPa) λ
c) d)
600 0.50

400
0.30

200
0.10

0.02 0.06 0.10

εeq 0.05 0.15
εeq

Figure 5.8: Dual plateau in nonproportional loading. (a) The effective transformation
strain trajectory. The applied strain trajectories are indicated by the dashed lines.
(b) The stress trajectory, (c) equivalent stress vs. equivalent strain, (d) volume
fraction vs. equivalent strain.
 70
of 2%. The unloading was in the reverse order (torsion unloads before tension). Un-
fortunately, they do not provide enough details for us to fit our model completely.
Thus, we are left to examine the situation qualitatively. Figure 5.8 shows the results
of a simulation that reproduces this dual plateau. The initiation and saturation
surfaces are chosen to display marked asymmetry. As the applied strain increases
in the normal direction, the transformation begins and saturates. Thus the stress
shows a plateau and then begins to increase. Now, we apply shear. The martensite
begins to reorient with the stress at a more or less constant volume fraction until
it reaches a highly curved section. Sufficient reorientation now makes it possible to
have additional transformation, and thus gives rise to the second plateau.
This reasoning also indicated why, when the thin tubes were loaded in torsion
first and then tension, dual plateaus were not observed. In the shear direction, the
transformation saturates at a higher value of λ. When the tensile strain is applied, the
headroom in the phase fraction that was available in the previous loading program
is not available. No more stress relief can take place due to transformation. A
secondary plateau is unable to emerge.
We conclude the parameter study by displaying the results of a circular and square
loading path in figures 5.9 and 5.10 respectively. In these results, the initiation and
saturation surfaces are set to be those in the previous subsection.
 71

a) γΜ b) τ(MPa)
0.10 400

−0.10 0.10
εΜ 200

−0.10
−200 200 600
σ(MPa)
c) σeq (MPa) d) λ
500 0. 70

0.50
300

0.30

100
0.10

0.02 0.06
εeq 0.02 0.06
εeq

Figure 5.9: Circular Loading Path. (a) The effective transformation strain trajec-
tory. The applied strain trajectories are indicated by the dashed lines. (b) The
stress trajectory, (c) equivalent stress vs. equivalent strain, (d) volume fraction vs.
equivalent strain.
 72

a)
γΜ b) τ(MPa)
0.10

200

εΜ 100
−0.10 0.10

−100 100 300

σ(MPa)
−0.10

c) σeq (MPa) d) λ
0.70

600

0.50

400

0.30

200

0.10

0.04 0.08
εeq 0.02 0.06
εeq

Figure 5.10: Square Loading Path. (a) The effective transformation strain trajectory.
The applied strain trajectories are indicated by the dashed lines. (b) The stress tra-
jectory, (c) equivalent stress vs. equivalent strain, (d) volume fraction vs. equivalent
strain.
 73

Chapter 6

Numerical Implementation and

Example

In the bigger picture, in order for the model to be of use in an engineering context,
it should be implementable in a finite-element package. To utilize the model on the
scale of engineering applications, an UMAT for the ABAQUS package was developed.
Given the strain control field t , the fields tM and λt are found by the implicit
minimization scheme at each time step and node:

1 t+1
λt+1 , εt+1 t+1 t+1 t+1 t+1 t+1



M = arg min { ε − λ ε M : C ε − λ ε M + (6.1)
εt+1 s 2
t+1 εt+1 ∈G
M ∈Gi ,λ M
 t+1
λ − λt

t+1 t+1 t+1 t+1 ∗



λ ω(θ) + Gi εM + Gs λ εM + ∆tϕ }.
∆t

Here the function ϕ∗ (λ̇) is the Legendre transform of the dissipation potential asso-
∂ϕ(dλ )
ciated with each kinetic law (i.e., λ̇ = ∂dλ
). Examples of these potentials can be
written in the stick-slip case and the rate-independent one.
 74
i) Stick-slip (the case p = 2)
 r !
2  
λ̇3 λ̇ λ̇+ λ̇2 λ̇2 λ̇2
+ 21 +

λ̇d+
c +
+
+ ln +


λ̇2+ −λ̇2 2 λ̇2+ −λ̇2 λ̇2+ −λ̇2 λ̇2+ −λ̇2





 r 2  

λ̇2 λ̇2
−λ̇+ + λ̇ ≥ 0


λ̇2+ −λ̇2 λ̇2+ −λ̇2
  
∗
ϕ λ̇ = r 2  
!
−λ̇3 λ̇− −λ̇− λ̇2 λ̇2 λ̇2

−λ̇d−
c + + ln + + 1
+


λ̇2− −λ̇2 2 2
λ̇− −λ̇2 2 2
λ̇− −λ̇2 2
λ̇− −λ̇2





 r 2  

λ̇2 λ̇2
λ̇ ≤ 0


 +λ̇− λ̇2− −λ̇2
+ λ̇2− −λ̇2
(6.2)
ii) Rate-independent kinetics

   d+ λ̇ λ̇ ≥ 0
c
ϕ∗ λ̇ = (6.3)
 d− λ̇ λ̇ ≤ 0
c

These potentials can be seen in figure 6.1. Here, for the sake of convenience, the
simplifying assumption that d− +
c = −dc has been made.

In order to make the constraints differentiable, they are implemented using stan-
dard penalties:

!N
2
(ε
3 M
· εM )3/2 + bi det (εM ) + 31 ci |eˆi · εM eˆi |3
Gi = , (6.4)
gi

!N
2
3
(λεM · λεM )3/2 + bs det (λεM ) + 13 cs |eˆs · λεM eˆs |3
Gs = , (6.5)
gs

where N is some large, positive, even exponent; in this case taken to be 16.
The effect of this smoothing constraint is to generally overestimate the set of
admissible transformation strains. As is seen in the figure 6.2, the intersection of the
polynomial constraint and the quadratic elastic energy is shifted slightly outward
 75

·
φ*(λ)

•
λ
•
-1 1 λ+

Figure 6.1: The Legendre transform of the dissipation potentials are shown in the
stick-slip case (black) for p = 2 and in the rate-independent case (red).
 76
G(εM )

soft
GS(εM) GS (εM)

Figure 6.2: The effect of the polynomial constraint is seen. The intersection of the
16th degree polynomial shifts the intersection point with the quadratic elastic energy
slightly to the right of the hard (dashed) constraint.

from the hard constraint (dashed well). Naturally, increasing the degree N of the
constraining polynomial will make the value of strain at which this intersection occurs
closer to that of the hard constraint. The trade-off is that the derivative of the
constraint polynomial becomes increasingly steep with N, making it harder to deal
with in a conjugate gradient scheme.
The result of this softening can be seen in terms of the total transformation strain
in the polycrystal in figure 6.3. The dashed green set represents the ideal saturation
constraint Gs . The red dashed curve is the scaled constraint set at λsat . The solid
red set represents the softened (N = 16) constraint for 10% strain applied in each
 77
γΜ λ
soft
GS ( λ sat εM)

λ sat
GS(εM)

GS( λ sat εM)

εΜ

εeq

Figure 6.3: The smoothed constraint set is calculated at 10% applied strain. The
green dashed constraint is Gs , the red dashed constraint is the scaled Gs at λ = λsat
and the red solid set is the softened polynomial (N = 16) constraint at λsat .

direction in the tension-shear plane. Even at large values of applied strain, the error
introduced by softening the constraint does not seem very significant.
With each component in the energy smoothed, the implicit minimization of the
nodal energy is conducted using a nested conjugate gradient scheme in the (λ, εM )
space. With the transformation strain and volume fraction fields attained through
minimization, the nodal stress values (σ = C : ( − λεM )) are computed trivially.
The linearized strain assumption also allows for simple computation of the tangent
modulus ∂σ


∂ε
. It should be noted that the contribution of the initiation and satu-
∂ε
ration surfaces is incorporated into the tangent modulus through the ∂εM
term, the
value of which is calculated by exploiting the fact that the energy density is at a
stationary point (δW = 0) and implicitly differentiating.
Because one of the current areas of interest in the application of shape memory
alloys for engineering is in the biomedical field, this area was selected as a suffi-
ciently fashionable one for a demonstration of the finite-element implementation of
 78

Given: ε t+1

Iterate

Minimize: W(ε,εM,λ)
t+1 t+1
Obtain: εM , λ

t+1
σ t+1=C(ε t+1-λ t+1εM )

Tangent Modulus: ∂σ
∂ε

Figure 6.4: A flowchart of the numerical implementation

 79
the model. A beam element of a NiTiNOL stent was examined in an ABAQUS
simulation in pure bending with symmetry boundary conditions as shown below.

Figure 6.5: An example of the finite-element implementation on a stent element.

 80

Chapter 7

Conclusions and Future Directions

Shape-memory alloys demonstrate interesting nonlinear thermoelastic behaviors in-

cluding superelasticity and the shape-memory effect. These behaviors find their
origin in the formation of microstructure. We have proposed a constitutive model of
these materials in which the strain associated with microstructural formation (εM )
is constrained to lie within two interacting sets. The two constraint sets arise from
the fundamental difference in mechanics between the initiation of martensitic trans-
formation and its saturation: the initiation constraint being a Sachs (or Reuss) type
constant stress bound on the set of admissible transformation strains, and the satu-
ration constraint a Taylor (or Voigt) constant strain bound. The interaction of these
two constraints in the model have been shown to result in several well-known features
of SMAs including superelasticity, shape-memory, reorientation of martensite, and
retention of austenite.
The geometry of the constraint sets has a great effect on the response of the SMA
polycrystal. This was examined in detail in parameter studies in proportional and
nonproportional loading, in stress and strain controlled simulations. Varying the
constraint shapes had a large effect on the tension-compression asymmetry, reorien-
tation of the martensite, and the amount of retained austenite. In nonproportional
 81
loading, the geometry of the constraints could be manipulated to give rise to the
curious secondary stress plateau observed experimentally. The parameter studies
indicate that, as an alternative to deriving the Sachs and Taylor bounds of the poly-
crystal, they can be recovered by fitting an appropriate set of functions to known
experiments.
The development of the model with only one constrained deviatoric tensorial
variable (εM ) for recoverable strain, and one scalar (λ) for martensitic phase frac-
tion, results in a low-dimensional numerical implementation which can be easily
incorporated into commercial finite-element packages (in this case ABAQUS). The
development of a UMAT for the proposed model serves as a proof of the concept in
a currently relevant design software.
Although this constitutive model reproduces many experimental characteristics
of SMAs in a compact and robust way, it is unable to reproduce the well-known phe-
nomena of localization or the formation of martensite transformation bands. Initially
it was hoped that banding phenomena could be reproduced in sets of sufficiently dif-
ferent eccentricity; rapid reorientation giving rise to macroscopic instability. This no
longer seems to be a fruitful direction for inquiry as all numerical investigations have
failed to yet yield such an instability. Instead a possible future pursuit would be
to attempt to explain banding as a mesoscale process; for instance as a grain-scale
percolation phenomenon.
Another limitation of the current model is that it is not easily extended to the
regime of finite deformation. The general notion of interacting constraint sets can
be incorporated in a finite deformation framework. The meaning of the Sachs and
Taylor bounds are no longer as clear cut however. The constant strain bound es-
pecially no longer serves as a good approximation of the total recoverable strain of
the polycrystal. The further complication that the strains are no longer additive
 82
(ε = εelast + λεM ), but multiplicative in their composition (F = Fe FM ) means that
the effect of λ on the transformation strain can no longer be assumed to be linear.
Future investigation of these bounds on the nonlinear polycrystal is an obvious av-
enue of future interest. In addition a more ad hoc extension of the model to finite
deformations might be useful for design purposes in regimes where experimental data
exists.
 83

Bibliography

[1] ABAQUS. Umat and vumat routines for the simulation of nitinol. Answer ID
1658, ABAQUS Inc. , Pawtucket, RI.

[2] R. Abeyaratne, C. Chu, and R. D. James. Kinetics of materials with wiggly

energies: Theory and application to the evolution of twinning microstructure in
a cu-al-ni alloy. Phil. Mag. A, 73:457–497, 1996.

[3] R. Abeyaratne and J. K. Knowles. On the driving traction on a surface of a

strain discontinuity in a continuum. J. Mech. Phys. Sol., 38:345–360, 1990.

[4] R. Abeyaratne and J. K. Knowles. A continuum model of a thermoelastic solid

capable of undergoing phase-transitions. J. Mech. Phys. Sol., 41:541–571, 1993.

[5] F. Auricchio and R. I. Taylor. Shape-memory alloys: modelling and numeri-

cal simulations of the finite-strain superelastic behavior. Comp. Methods Appl.
Mech. Eng., 143:175–194, 1997.

[6] J. Ball. Convexity conditions and existence theorems in nonlinear elasticity.

Arch. Rat. Mech. Anal., 63:337–403, 1977.

[7] J. Ball, C. Chu, and R. D. James. Hysteresis during stress-induced variant

rearrangement. J. Phys. IV France, 5:245–251, 1995.
 84
[8] J. Ball and R. D. James. Fine phase mixtures as minimizers of energy. Arch.
Rat. Mech. Anal., 100:13–52, 1987.

[9] J. Ball and R. D. James. Proposed experimental tests of a theory of fine mi-
crostructure and the two-well problem. Phil. Trans. R. Soc. Lond. A, 338:389–
450, 1992.

[10] K. Bhattacharya. Comparison of the geometrically nonlinear and linear theories

of martensitic transformation. Cont. Mech. and Therm., 5:205–242, 1993.

[11] K. Bhattacharya. Phase boundary propogation in heterogeneous bodies. Proc.

Roy. Soc. Lond. A, 455:757–766, 1999.

[12] K. Bhattacharya. Microstructure of Martensite: Why it Forms and How it Gives

Rise to the Shape-Memory Effect. Oxford University Press, 2004.

[13] K. Bhattacharya and R.V. Kohn. Energy minimization and the recoverable
strains of polycrystalline shape-memory alloys. Arch. Rational Mech. Anal.,
139:99–180, 1997.

[14] K. Bhattacharya, P. Purohit, and B. Craciun. The mobility of twin and phase
boundaries. J. Phys. IV France, 112:163–166, 2003.

[15] K. Bhattacharya and A. Schlömerkemper. Transformation yield surface of shape

memory alloys. J. Phys. IV France, 115:155–162, 2004.

[16] K. Bhattacharya and A. Schlömerkemper. Stress-induced phase transformations

in shape-memory polycrystals. In preparation, 2008.
 85
[17] J. G. Boyd and D. C. Lagoudas. A thermodynamic constitutive model for shape
memory material: 1. the monolithic shape memory alloy. Int. J. Plast., 12:843–
873, 1996.

[18] L. C. Brinson and M. Panico. A three-dimensional phenomenological model

for martensite reorientation in shape memory alloys. J. Mech. Phys Solids,
55:2491–2511, 2007.

[19] L.C. Brinson, I. Schmidt, and R. Lammering. Stress-induced transformation

behavior of a polycrystalline niti shape memory alloy: micro and macrome-
chanical investigations via in situ optimcal microscopy. J. Mech. Phys. Solids,
52:1549–1571, 2004.

[20] M. Brocca, L. C. Brinson, and Z. Bazant. Three-dimensional constitutive model

for shape memory alloys based on microplane. J. Mech. Phys Solids, 50:1051–
1077, 2002.

[21] O. P. Bruno, F. Reitich, and P. H. Leo. The overall elastic energy of polycrys-
talline martensitic solids. J. Mech. Phys. of Solids, 44:1051–1101, 1996.

[22] M. W. Burkart and T. A. Read. Diffusionless phase change in the indium-

thallium system. Trans. Amer. Inst. Mining and Metall. Eng., 197:1516–1254,
1953.

[23] I. V. Chenchiah. Energy-minimizing microstructures in multiphase elastic solids.

PhD thesis, California Institute of Technology, 2004.

[24] M. Chipot and D. Kinderlehrer. Equilibrium configurations of crystals. Arch.

Rat. Mech. and Anal., 103:237–277, 1988.
 86
[25] B. D. Coleman and W. Noll. The thermodynamics of elastic materials with heat
conduction and viscosity. Arch. Rat. Mech. Anal., 13:167–178, 1963.

[26] S. Daly, K. Bhattacharya, and G. Ravichandran. Stress-induced martensitic

phase transformation in thin sheets of nitinol. Acta Mater., 55:3593–3600, 2008.

[27] K. Dayal and K. Bhattacharya. Kinetics of phase transformations in the peri-

dynamic formulation of continuum mechanics. Preprint, pages 1–34, 2005.

[28] P. W. Dondl. Structure and evolution of martensitic phase boundaries. PhD

thesis, California Institute of Technology, 2007.

[29] J. L. Ericksen. Some phase transitions in crystals. Arch. Rat. Mech and Anal.,
73:99–124, 1980.

[30] J. D. Eshelby. The determination of the elastic field of an ellipsoidal inclusion,

and related problems. Proc. Royal Soc. London A, 241:376–396, 1957.

[31] S. Eucken and J. Hirsch. The effect of textures on shape memory behaviour.
Mat. Sci. Forum, 56-58:487–492, 1990.

[32] P. Feng and Q. P. Sun. Experimental investigation on macroscopic domain

formationand evolution in polycrystalline niti microtubing under mechanical
force. J. Mech. Phys. Sol., 54:1568–1603, 2006.

[33] S. Govindjee, A. Mielke, and G. J. Hall. The free energy of mixing for n-variant
martensitic phase transformations using quasiconvex analysis. J. Mech. Phys.
Sol., 51:I–XXVI, 2003.

[34] D. E. Hodgson, M. H. Wu, and R. J. Biermann. Shape memory alloys. Online

at http://www.sma-inc.com/SMAPaper.htm, 2003.
 87
[35] M. Huang and L. C. Brinson. Multivariant model for single crystal shape mem-
ory. J. Mech. Phys. Sol., 46:1379–1409, 1998.

[36] Y. Jung. Thin tube experiments and numerical simulations of micromechanical

multivariant constitutive modeling in superelastic nitinol. PhD thesis, University
of California Berkeley, 2003.

[37] R. Kohn. The relaxation of a double-well energy. Cont. Mech, Thermodyn.,

3:193–236, 1991.

[38] R. V. Kohn and T. D. Little. Some model problems of polycrystal plasticity

with deficient basic crystals. SIAM J. Appl. Math., 59:172–197, 1998.

[39] C. Lexcellent and P. Blanc. Phase transformation yield surface determination

for some shape memory alloys. Acta Mat., 52:2317–2324, 2004.

[40] C. Lexcellent and M. R. Laydi. Yield criteria for shape memory materi-
als:convexity conditions and surfaces transport. Math. and Mech. of Solids,
To Appear, 2008.

[41] C. Lexcellent, A. Vivet, C. Bouvet, S. Calloch, and P. Blanc. Experimental and

numerical determinations of the initial surface of phase transformation under
biaxial loading in some polycrystalline shape-memory alloys. J. Mech. Phys.
Solids, 50:2717–2735, 2002.

[42] AMT: Advanced Materials and Technologies. http://www.amtbe.com, 2008.

[43] J. M. McNaney, V. Imbeni, Y. Jung, P. Papadopoulos, and R. O. Ritchie. An

experimental study of the superelastic effect in a shape-memory nitinol alloy
under biaxial loading. Mechanics of Materials, 35:969–986, 2003.
 88
[44] C. B. Morrey. Quasiconvexity and the semicontinuity of multiple integrals. Pac.
J. Math., 2:25–53, 1952.

[45] G. B. Olson and W. S. Owen, editors. Martensite. ASM, 1992.

[46] K. Otsuka and Wayman C. M. Shape Memory Materials. Cambridge University

Press, 1998.

[47] B. Peultier, B. T. Zineb, and E. Patoor. A simplified micromechanical constitu-

tive law adapted to the design of shape memory applications by finite element
methods. Mat. Sci. and Eng. A, 481-482:384–388, 2008.

[48] R. Peyroux, A. Chrysochoos, C. Licht, and M. Lobel. Thermomechanical cou-

plings and pseudoelasticity of shape memory alloys. Int. J. Eng. Science, 36:489–
509, 1998.

[49] A. C. Pipkin. Elastic materials with two preferred states. Quart. J. Appl. Math.,
44:1–15, 1991.

[50] P. Purohit. Dynamics of phase transitions in strings, beams, and atomic chains.
PhD thesis, California Institute of Technology, 2002.

[51] R. T. Rockafellar. Convex Analysis. Princeton University Press, 1970.

[52] T. Saburi and S. Nenno. The shape memory effect and related phenomena. In
H. I. Aaronson, D. E. Laughlin, R. F. Sekerka, and C. M. Wayman, editors,
Proc. Int. Conf. on Solid-Solid Phase Transformations, pages 1455–1479, New
York, 1981. The Metall. Soc. AIME.

[53] A. Sadjadpour. A micromechanics inspired constitutive model for shape-memory

alloys: the one-dimensional case. Smart Mater. Strut., 16:S51–S62, 2007.
 89
[54] A. Schlömerkemper. On a sachs bound for stress-induced phase transformations
in polycrystalline shape memory alloys. Proc. Appl. Math. Mech., Preprint,
2006.

[55] I. Schmidt, L. C. Brinson, and R. Lammering. Micro and macromechanical

observation of polycrystalline niti using in situ microscopy. J. Phys. IV France,
112:655–658, 2003.

[56] J. A. Shaw and S. Kyriakides. On the nucleation and propagation of phase

transformation fronts in a niti alloy. Acta Metall., 45:683–700, 1997.

[57] T.W. Shield. Orientation dependence of the pseudoelastic behavior of single-

crystals of cu-al-ni in tension. J. Mech. Phys. Solids, 43:869–895, 1995.

[58] Y.C. Shu and K. Bhattacharya. The influence of texture on the shape-memory
effect in polycrystals. Acta Mater., 46:5457–5473, 1998.

[59] P. Sittner, Y. Hara, and M. Tokuda. Experimental study on the thermoelas-

tic martensitic transformation in shape memory alloy polycrystal induced by
combined external forces. Metall. Mat. Trans., 26:2923–2935, 1995.

[60] V. P. Smyshlaev and J. R. Willis. A ”non-local” variational approach to the

elastic energy minimization of martensitic polycrystals. Proc. Roy. Soc. Lond.
A, 454:1573–1613, 1998.

[61] Q. P. Sun and K. C. Hwang. Micromechanics modeling for the constitutive be-
havior of polycrystalline shape memory alloys: 1. derivation of general relations
2. study of individual phenomena. J. Mech. Phys. Sol., 41:1–17, 19–33, 1993.
 90
[62] P.M. Suquet and K. Bhattacharya. Model problem concerning recoverable
strains of shape-memory polycrystals. Proc. Roy. Soc. Lond. A, 461:2797–2816,
2005.

[63] V. Sverak. Rank-one convexity does not imply quasiconvexity. Proc. Royal Soc.
Edinburgh A, 120:185–189, 1992.

[64] P. Thamburaja and L. Anand. Polycrystalline shape-memory materials: Effect

of crystallographic texture. J. Mech. Phys. Solids, 49:709–737, 2001.