Está en la página 1de 121

CONSTITUTIVE MODELING AND SIMULATION

OF THE SUPERELASTIC EFFECT


IN SHAPE-MEMORY ALLOYS

Panos Papadopoulos
Department of Mechanical Engineering, University of California, Berkeley

Acknowledgments
T. Duerig, NDC
V. Imbeni, SRI
Y. Jung, PU
J.M. McNaney, LLNL
R.O. Ritchie, UCB
H.-R. Wenk, UCB
SMA – p.1/108
Outline

OUTLINE OF LECTURES

Part I: Phenomenology and crystallographics

Part II: Constitutive modeling

Part III: Algorithmics

SMA – p.2/108
Contents-I

C ONTENTS OF PART I

Superelasticity and the shape-memory effect

Technological applications

Crystallography of solid-solid phase transitions

Texture

Experimental results

SMA – p.3/108
Superelasticity

stress superelastic plateau

strain

The superelastic effect (at constant temperature)


SMA – p.4/108
Superelasticity

500

450

400

350

300
σeq (MPa)

250

200

150

100

Tension 6.0%
50 Tension 6.0%
Tension 6.0%
Tension 6.0%
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
ε
eq

Uniaxial stress-strain plot of polycrystalline superelastic Ni-Ti


SMA – p.5/108
Superelasticity

Superelasticity is a property of crystalline materials.

Typical examples of known superelastic materials:

Ni-Ti
Nitinol (acronym for Nickel Titanium Naval Ordnance
Laboratory) is the most widely used superelastic material.

Cu-Zn-Al

Cu-Al-Ni

Co-Ni-Al

Fe-Mn-Si
SMA – p.6/108
Applications

(Courtesy: NDC Inc.)

Typical biomedical devices made of Nitinol


SMA – p.7/108
Crystallography

The superelastic effect is due to the displacive,


diffusionless, reversible, solid-solid transformation between
an austenitic (highly structured) phase and a martensitic
(less structured) phase.

Examine in detail the Ni-Ti crystal:

Ti
Ni

Austenite (body-centered cubic crystal)


SMA – p.8/108
Crystallography

The superelastic effect is due to the displacive,


diffusionless, reversible, solid-solid transformation between
an austenitic (highly structured) phase and a martensitic
(less structured) phase.

Examine in detail the Ni-Ti crystal:

Ti
Ni

Austenite (body-centered cubic crystal)


SMA – p.8/108
Crystallography

The superelastic effect is due to the displacive,


diffusionless, reversible, solid-solid transformation between
an austenitic (highly structured) phase and a martensitic
(less structured) phase.

Examine in detail the Ni-Ti crystal:

Ti
Ni

Martensite (monoclinic crystal)


SMA – p.8/108
Crystallography

The interface between austenite and martensite is called


the habit plane.

How does martensite grow inside an undistorted austenite


matrix?

austenite

martensite

SMA – p.9/108
Crystallography

The interface between austenite and martensite is called


the habit plane.

How does martensite grow inside an undistorted austenite


matrix?

Austenite

Lattice distortion (incompatible!)


SMA – p.9/108
Crystallography

The interface between austenite and martensite is called


the habit plane.

How does martensite grow inside an undistorted austenite


matrix?

Austenite

Single martensite with slip (irreversible process)


SMA – p.9/108
Crystallography

The interface between austenite and martensite is called


the habit plane.

How does martensite grow inside an undistorted austenite


matrix?

Austenite

M2
M1
M2
M1

Twin martensite variants (reversible process)


SMA – p.9/108
Crystallography

How does martensite grow under imposed shear-like


deformation?

austenite

1 2 12 1 2 1 2 1

austenite

Austenite initially transforms to twinned martensite, which,


in turn, gives way to single variant martensite (detwinning).
SMA – p.10/108
Superelasticity

stress A −> M1,2 M1,2 −> M

A M

A <− M
A

strain

Microstructural interpretation of the superelastic effect

SMA – p.11/108
Shape-memory effect

deformation

ements

Mf Ms As Af
temperature

Microstructural interpretation of shape-memory effect

SMA – p.12/108
Material

Consider a thin-walled Nitinol tube, which is the starting


component for manufacturing various biomedical devices.

Composition: Ti 44.5 wt.%, Ni 55.5 wt.%

Heat-treatment at 485◦ C in air (5 min), followed by


water-quenching to produce a microstructure that
enhances the superelastic properties.

Transformation temperatures:

As = − 6.36◦ C , Af = 18.13◦ C
Ms = − 51.55◦ C , Mf = − 87.43◦ C
SMA – p.13/108
Material




Optical micrograph of a portion of the cross section of a
thin-walled Nitinol tube (note the polycrystalline structure)
SMA – p.14/108
Material

Nitinol, thin tube (as received)

2500 B2(110)

2000
I (counts)

1500

1000

B2 (200) B2 (211)
500
B19' B2 (310)
B2(100)

0
20 30 40 50 60 70 80 90 100 110 120

X-ray diffraction analysis of as received Ni-Ti material


SMA – p.15/108
Texture

In polycrystalline solids, the overall mechanical response


may depend strongly on the preferred lattice orientation of
the crystals, i.e., on the texture of the polycrystal.

In superelastic materials, texture may affect the


mechanical behavior in two ways:

By controlling the phase transformation process (i.e.,


by enabling nucleation of certain variants and not
others).
By inducing anisotropy in the continuum-level elastic
response.

Texture measurements can be taken using X-ray or


electron diffractometry.
SMA – p.16/108
Texture

Texture measurements are represented using


stereographic projections of the spherical poles
corresponding to the normal to a given crystal plane, when
the plane is assumed to pass through the center of the
sphere (pole figures).

SMA – p.17/108
Texture

The tube manufacturing process induces primarily


h111i{110}-type sheet texture “wrapped” around the
cylindrical surface, such that the h111i austenite lattice
direction is aligned with the longitudinal axis of the tube.




 

 







 


SMA – p.18/108
Texture

Pole figures for Nitinol tubes (R.D. horizontal, T.D. vertical)


SMA – p.19/108
Experiments

25 mm

Test section
r = 1.5 mm

thickness = 0.20mm
25 mm

outer radius = 2.15mm

Gripping sections
25 mm
thickness = 0.37mm
outer radius = 2.32mm

Schematic illustration of the NiTi specimen (not to scale)


SMA – p.20/108
Experiments

500

450

400

350

300
σeq (MPa)

250

200

150

100

Tension 6.0%
50 Tension 6.0%
Tension 6.0%
Tension 6.0%
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
ε
eq

Equivalent stress-strain plot showing repeatability of tension tests


SMA – p.21/108
Experiments

Nitinol, thin tubes, loaded/unloaded in tension


4000

B2(110)
3500

3000

2500
I (counts)

2000

1500
B2 (211)
1000
B2 (200)
B19' B2 (310)
500

0
20 30 40 50 60 70 80 90 100 110 120

X-ray diffraction after complete tension cycle


SMA – p.22/108
Experiments

tensile strain tensile strain


shear strain shear strain

2
strain

strain
#

1
"






!

.-,

.0/

.43

765
(a) Tension followed by torsion
%$

)*
&'(
(b) Torsion followed by tension

98

>
==
:;<
tensile strain
D shear strain
strain

.CB
?

.A@

(c) Simultaneous tension-torsion


GF

KL
KK
HIJ

Strain/time sequence for T YPE I, II, III loading SMA – p.23/108


Experiments

The experimental setup provided measurements of the


equivalent Cauchy (true) stress σeq versus the equivalent
Lagrangian strain εeq , namely
q r
2 2 2 4 2
σeq = σt + 3σs , εeq = ε t + εs ,
3
where σt and σs denote the tensile and shearing stress,
while εt and εs denote the tensile and shearing strain.

Strain rates were imposed in the 10−5 /s to 10−4 /s were


used for both the tensile and torsional loading.

SMA – p.24/108
Experiments

500

450

400

350

300
σeq (MPa)

250

200

150

100

50
Tension 0.0%, Torsion 2.0%
Tension 0.0%, Torsion 2.0%
0
0 0.005 0.01 0.015 0.02 0.025 0.03
ε
eq

Equivalent stress-strain plot showing repeatability of torsion tests


SMA – p.25/108
Experiments

N itin o l, p u re to rs io n

10000
B2(110
9000

8000

7000

6000
I ( co u n t s )

5000
B2 (211)

4000

3000

2000
B2 (200) B19'
B2 (310)
1000

0
20 30 40 50 60 70 80 90 100 110 120

X-ray diffraction after complete torsion cycle


SMA – p.26/108
Experiments

500

450

400

350

300
σeq (MPa)

250

200

150

100 Tension 0.7%, Torsion 2.0%


Tension 1.05%, Torsion 2.0%
Tension 1.5%, Torsion 2.0%
50 Tension 2.0%, Torsion 2.0%
Tension 3.0%, Torsion 2.0%
Tension 6.0%, Torsion 2.0%
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
ε
eq

Equivalent stress-strain plot of tension followed by torsion (T YPE I)


SMA – p.27/108
Experiments

500

450

400

350

300
σeq (MPa)

250

200

150

100

Tension 1.05%, Torsion 2.0%


50 Tension 1.05%, Torsion 2.0%
Tension 3.0%, Torsion 2.0%
Tension 3.0%, Torsion 2.0%
0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035 0.04
ε
eq

Equivalent stress-strain plot showing repeatability of T YPE I tests


SMA – p.28/108
Experiments

500

450

400

350

300
σeq (MPa)

250

200

150

100

Torsion 2.0%, Tension 0.7%


50 Torsion 2.0%, Tension 1.05%
Torsion 2.0%, Tension 3.0%
Torsion 2.0%, Tension 5.8%
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
ε
eq

Equivalent stress-strain plot of torsion followed by tension (T YPE II)


SMA – p.29/108
Experiments

500

450

400

350

300
σeq (MPa)

250

200

150

100

Tension 0.7%, Torsion 2.0%


50 Tension 1.5%, Torsion 2.0%
Tension 3.0%, Torsion 2.0%
Tension 6.0%, Torsion 2.0%
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
ε
eq

Equivalent stress-strain plot of simultaneous tension-torsion (T YPE III)


SMA – p.30/108
Closure

In superelastic materials, austenite-to-martensite


phase transformation occurs by nucleation of
martensite twins, followed by detwinning.

The stress-strain response is rich and complex,


especially under non-proportional loading.

The response to torsion defies the classical metal


plasticity paradigm.

The modeling of the mechanical response of


superelastic materials leads to a multiscale problem.

Martensite laminates, crystal grains, continuum


SMA – p.31/108
References

1. J.M. Ball and R.D. James, Arch. Rat. Mech. Anal., 100, pp. 13-52,
(1987) [Non-linearly elastic theory of martensitic
transformation].

2. K.F. Hane and T.W. Shield, Acta Mater., 47, pp. 2603-17, (1999)
[Exhaustive description of the microstructure in NiTi phase
transition].

3. K. Gall and H. Sehitoglu, Int. J. Plast., 15, pp. 69-92, (1999)


[Role of texture in NiTi].

4. T.W. Duerig, A. Pelton and D. Stöckel, Mater. Sci. Engrg.,


A273-275, pp. 149-160, (1999) [A review of biomedical
applications of Nitinol].

5. J.M. McNaney, V. Imbeni, Y. Jung, P. Papadopoulos and R.O.


Ritchie, Mech. Mat., 35, pp. 969-986, (2003) [Multiaxial
experiments on polycrystal NiTi].
SMA – p.32/108
Contents-II

C ONTENTS OF PART II

Continuum modeling at the crystal grain level

Loading and transformation criteria

Connection between continuum and microstructural


modeling

A constrained optimization problem

Twinned and detwinned martensite

SMA – p.33/108
Nomenclature

Nomenclature:

F : deformation gradient
E : Lagrangian strain tensor
U : Right stretch tensor
 : Infinitesimal strain tensor
P : 1st (unsymmetric) Piola-Kirchhoff stress tensor
S : 2nd (symmetric) Piola-Kirchhoff stress tensor
σ : Infinitesimal stress tensor
Q : general rotation tensor
I : second-order identity tensor

SMA – p.34/108
Modeling

Consider constitutive modeling at the crystal grain level.

The total deformation induced by the martensitic


transformation can be expressed in the form:

F = F 1 F 2 R3 ,
where

F1 : lattice deformation
F2 : twinning deformation
R3 : rigid rotation

Of the three deformation components, F1 is dominant.


SMA – p.35/108
Modeling

There are three basic modeling choices at the crystal grain


level:

Model the transformation of austenite to detwinned


martensite.

austenite

austenite 1

austenite

SMA – p.36/108
Modeling

There are three basic modeling choices at the crystal grain


level:

Model the transformation of austenite to twinned


martensite.

austenite

austenite 1 2 12 1 2 1 2

austenite

SMA – p.36/108
Modeling

There are three basic modeling choices at the crystal grain


level:

Model the transformation of austenite first to twinned


martensite, then to detwinned martensite.

austenite

austenite 1 2 12 1 2 1 2 1

austenite

SMA – p.36/108
Modeling

Consider first modeling the transformation of austenite to


detwinned martensite. In the case of Ni-Ti (cubic to
monoclinic transformation), there are a total of 12
martensite “lattice correspondence variants”, i.e.,

F1,i = R(Qi UQTi ) , i = 1, . . . , 12 .

In the above polar decompositions, R is the rotation tensor,


U is the unique stretch tensor, and Qi are the 12 rotations
corresponding to the 12 ways in which a cube maps onto
itself.

The tensor U can be deduced from the crystallography of


the austenite-martensite transformation.

SMA – p.37/108
Modeling

Consider next modeling the transformation of austenite to


twinned martensite.

In the case of Ni-Ti (cubic to monoclinic transformation),


one may expect up to 66x8 = 528 martensite “habit plane
variants”.

Compatibility requirements reduce this number to 192


feasible habit plane variants. These variants can be
obtained by an energy minimization method.

Of the 192 feasible habit plane variants, only 24 are


experimentally observed to be dominant.

SMA – p.38/108
Modeling

In the case of a typical habit plane variant α, the


microscopic deformation gradient takes the form

F12,α = I + gmα ⊗ nα ,

where g is the trans-


formation displacement, gmα twinned martensite
mα is the unit vector in the
transformation direction,
and nα is the outward unit
normal to the habit plane.
PSfrag replacementsnα

habit plane
SMA – p.39/108
Modeling

Proceed with modeling by analogy to the theory of


elastic-plastic materials.

A plastic (resp. phase-tranforming) material can be views


as an elastic material whose response is parametrized by
the plastic (resp. transformation) variables.

Two key differences:

Martensitic transformation is fully reversible.


Loading conditions in multi-surface plasticity are
determined per surface, while in martensitic
transformation they are cumulative.

SMA – p.40/108
Modeling

Micromechanically-motivated constitutive assumptions:

1. Existence of a Lagrangian transformation strain Et


depending on the martensitic volume fractions

n nv
X o
{ξα } = (ξ1 , ξ2 , · · · , ξnv ) | ξβ ≤ 1 , ξ β ≥ 0 ,
β=1

such that Et = Êt ({ξα }), where nv stands for the


number of potentially present variants.

Homogeneity condition: Êt ({0}) = 0.

Note the analogy (and contrast) with plasticity theory.


SMA – p.41/108
Modeling

2. Admittance of a Helmholtz free energy Ψ = ρ0 ( − ηθ),


where Ψ = Ψ̂(E, {ξα }, θ). Here, ρ0 is the mass density,
 is the internal energy per unit mass, η is the entropy,
and θ the absolute temperature.

Recall the local form of the energy equation

ρ0 ˙ = ρ0 r − Divq0 + S · Ė ,
where r is the heat supply per unit mass and q0 the
referential heat flux vector.

Also recall the Clausius-Duhem inequality


q0 · Grad θ
ρ0 η̇θ ≥ ρ0 r − Divq0 + .
θ
SMA – p.42/108
Modeling

For any homothermal superelastic process,


! nv
∂ Ψ̂ X ∂ Ψ̂ ˙
S− · Ė − · ξα ≥ 0 .
∂E α=1
∂ξα

Since E and {ξα } can be varied independently, a


standard process leads to

∂ Ψ̂
S = ,
∂E

while the Clausius-Duhem inequality further implies


that
nv
X ∂ Ψ̂ ˙
Ḋ = (− ) ξα ≥ 0 .
α=1
∂ξα
SMA – p.43/108
Modeling

3. Existence of functions Ŷαf and Ŷαr in the form

Yαf = Ŷαf (E, {ξβ }, θ) , Yαr = Ŷαr (E, {ξβ }, θ)

associated with the forward and reverse transformation


of variant α, respectively, where Ŷαf < Ŷαr .

Yαf = 0 ⇔ variant α active in forward transformation


Yαr = 0 ⇔ variant α active in reverse transformation

Forward and reverse transformation active sets:

J f (E, θ) = {α | Ŷαf (E, {ξβ }, θ) = 0, ξα > 0} ,


J r (E, θ) = {α | Ŷαr (E, {ξβ }, θ) = 0, ξα > 0} .
SMA – p.44/108
Modeling

Yield surfaces in stress space

dg ef

ij

k
h
h
lo mn

qr
q
p
p
(b)
(a)
(e) (c)

(d)
0
vu

xy
st

x
w
w
}|


z{


~
~

(c)
ONPM

(c)
OVWU
(b)

(a) (b)
(a)

(d)
(d)
(e)
(e)
Q

\[Z

7cb
X
STR

^]
` _
a

Y
SMA – p.45/108
Modeling

Persistency of a forward-active variant α,

∂ Ŷαf X
Ẏαf = · Ė − Qαβ ξ˙β = 0 ,
∂E fβ∈J
f
∂ Ŷ
where Qfαβ = − α are the components of the forward
∂ξβ
coupling matrix (assumed invertible) which quantifies
the coupling between variants during forward transformation.

It follows that during forward transformation of variant α,


f
X ∂ Ŷ
ξ˙α = f −1 β
Qβα · Ė .
f
∂E
β∈J

This is a rate-type, rate-independent equation (note


connection with rate-independent plasticity). SMA – p.46/108
Modeling

4. Stipulate of forward and reverse transformation criteria.

From a state of a forward transformation (J f 6= ∅):


 > 0 ⇔ forward transformation


f
X ∂ Ŷ
Wαf α ·Ė = 0 ⇔ neutral forward transformation .
f
∂E 

α∈J
< 0 ⇔ elastic unloading

From a state of a reverse transformation (J f = ∅, J r 6= ∅):


 > 0 ⇔ elastic reloading


r
X ∂ Ŷ
Wαr α ·Ė = 0 ⇔ neutral reverse transformation .
α∈J r
∂E 

< 0 ⇔ reverse transformation

SMA – p.47/108
Modeling

To determine the weight function Wαf , note that during


forward transformation
X
f ∂ Ψ̂ ˙
(−Yα − ) ξα ≥ 0 .
f
∂ξα
α∈J

Recalling the equation for ξ˙α , it follows that


X X f −1 f ∂ Ψ̂ ∂ Ŷαf
Qαβ (−Yβ − ) · Ė ≥ 0 ,
f f
∂ξβ ∂E
α∈J β∈J

which implies that


X f −1 ∂ Ψ̂
Wαf = Qαβ (−Yβf − ).
∂ξβ
β∈J f

A similar analysis applies to Wαr .


SMA – p.48/108
Modeling

Consider an alternative option: postulate a single forward


transformation function of the form

f ∂ Ψ̂
Y = k− k q − Fc ,
∂ξα
where k(·)kq denotes the standard vector q-norm.

PSfrag replacements

q=1
q=2
q=∞

Note analogy to yield functions in plasticity theory.

Why use non-smooth transformation criteria?


SMA – p.49/108
Modeling

How to incorporate microstructural information into the


crystal grain constitutive model?

Homogenize the kinematics and kinetics of the


microstructure over a representative referential volume
element (RVE) V that characterizes the physics of the
martensitic transformation.

austenite

austenite 1 2 12 1 2 1 2

austenite

SMA – p.50/108
Modeling

“Descend” into the limiting case of infinitesimal


deformations and let the mechanical part of the Helmholtz
free energy be of the form

nv+1
X 1
ψ(, {ξα }) = ξα ( − tα ) · C( − tα ) + ψm ({ξα }) ,
2 | {z }
|α= {z } mixing
single phases

where C is the elasticity tensor (assumed phase-invariant).

Subsequently, write the mechanical part of the Gibbs free


energy of the mixture as
nv nv
X 1 X
γ(σ, {ξα }) = σ · C−1 σ + ξα σ · tα + . . .
α=1
2 α=1
SMA – p.51/108
Modeling

Assume a spatially homogeneous stress σ 0 over the


ensemble and employ a Legendre transformation to write
Z Z
 0 0

sup −γ(σ , {ξα }) + σ ·  dV = ψR (, {ξα }) dV ,
σ0 V V

where ψR is the Reuss (lower) bound to the mechanical


part of the free energy.

It follows from the above that the Reuss bound is


nv nv
1 X X
ψR (¯, {ξα }) = (¯ − ξα tα ) · C(¯ − ξα tα )
2 α=1 α=1
Z
1
in terms of the volume-averaged strain ¯ =  dV .
vol(V) V
SMA – p.52/108
Modeling

Alternatively, assume a spatially homogeneous strain 0


over the ensemble and again homogenize the elastic part
of the Helmholtz free energy to find that

nv nv nv
1 X X 1 X
ψV (¯, {ξα }) = (¯− ξα tα )·C(¯− ξα tα )− σ̃·ξα tα ,
2 α=1 α=1
2 α=1

where σ̃ is the self-equilibrated zero-mean fluctuation


stress.

One may estimate the contribution of the fluctuation stress,


although the existing estimates are generally not very
accurate.

SMA – p.53/108
Modeling

Motivated by the micromechanical model, let the Helmholtz


free energy be written as

Ψ = Ψ̂(E, {ξα }, θ) =
nv
1   X
E − Et · C E − Et + B(θ − θ0 ) ξα .
|2 {z } α=1
mechanical | {z }
chemical

where Et is the cumulative transformation strain, B is a


chemical energy constant (latent heat of transformation),
and θ0 is the equilibrium temperature.

Note that the strains are moderate (typically, less than


6-10%).
SMA – p.54/108
Modeling

The cummulative transformation strain is now defined as


nv
X
Et = ξα Etα ,
α=1

where the variant transformation strain Etα is given by

1
Etα = g(mα ⊗ nα + nα ⊗ mα + gnα ⊗ nα ) ,
2

in terms of the microscopic gm twinned martensite

deformation gradient

Fα = I + gmα ⊗ nα . n

habit plane
SMA – p.55/108
Modeling

Further, admit critical thermodynamic force-based


transformation functions
∂ Ψ̂ ∂ Ψ̂
Yαf = − − Fc , Yαr = − + Fc ,
∂ξα ∂ξα

to characterize forward and reverse transformation,


respectively.

Given the particular form of the Helmholtz free energy,


nv
X  
Yαf = Ŷαf (E, {ξβ }, θ) = C E− ξβ Etβ · Etα − B(θ − θ0 ) + F c
,
β=1
nv
X  
Yαr = Ŷαr (E, {ξβ }, θ) = C E− ξβ Etβ · Etα − B(θ − θ0 ) − F c
.
β=1

SMA – p.56/108
Modeling

During elastic loading or forward transformation, the


loading conditions simplify to
X
ξ˙α ≥ 0 ,
α∈J f

namely, the total production of martensite is non-negative.

Recalling further the form of the forward transformation


criterion, it follows that during elastic loading or forward
transformation
X ∂ Ψ̂ 
c ˙
− − F ξα = 0 .
∂ξα
α∈J f

An analogous result applies to reverse transformation.


SMA – p.57/108
Modeling

Note that here


∂ 2 Ψ̂
= Qαβ = Etα · CEtβ ,
∂ξα ∂ξβ
where the nv × nv coupling matrix [Qα β] is symmetric and
of rank at most 6. This implies that for nv > 6, the matrix
[Qαβ ] is necessarily positive semi-definite.

Recalling again that


X ∂ Ψ̂ 
c ˙
− − F ξα = 0 ,
∂ξα
α∈J f

it follows that the identification of active variants and


volume fractions can be cast as a constrained optimization
problem at fixed strain and temperature.
SMA – p.58/108
Modeling

Specifically, in the case of forward transformation, the


martensitic content {ξα } satisfies

{ξα } = argmin Φf ,
{ξγ }∈J f

subject to inequality constraints


nv
X
−ξα ≤ 0 , ξα ≤ 1 .
α=1

where the functional Φf is defined as


X
f
Φ =Ψ+ F c ξα ,
α∈J f

and may attain multiple minima. A similar conclusion


applies to reverse transformation for the functional
P
Φ = Ψ − α∈J r F c ξα .
r
SMA – p.59/108
Modeling

How to model detwinning?

The twinned and detwinned variants may coexist, so that

Ψ = Ψ̂(E, {ξαhab , ξβlat }, θ) =


nv
1 t
 t
 X
E − E · C E − E + B(θ − θ0 ) ξα ,
2 α=1

where now
nvh
X nvl
X
Et = ξαhab Etα + ξβlat Etβ .
α=1 β=1

In the above, nvh and and nvl denote the total number of
habit and lattice correspondence variants, respectively.
SMA – p.60/108
Closure

Plasticity theory can be used as a guide toward the


development of superelastic constitutive models.

It is possible to motivate continuum formulations of


superelasticity using micromechanics.

It is sensible to consolidate the loading criteria, while


keeping the transformation criteria separate.

Under certain assumptions, superelasticity can be cast


as a constraint minimization problem.

The role of interaction energy between variants


remains an open question.
SMA – p.61/108
References

1. J.G. Boyd and D.C. Lagoudas, Int. J. Plast., 12, pp.


805-842, (1996) [Phenomenological modeling of phase
transformation].

2. S. Leclercq and C. Lexcellent, Int. J. Mech Phys Sol., 44, pp.


953-980, (1996) [Phenomenological modeling of phase
transformation including detwinning].

3. M. Huang and L.C. Brinson, J. Mech. Phys. Sol., 46, pp.


1379-1409, (1998) [Micromechanically motivated model
for single crystal]

4. N. Siredey, E. Patoor, M. Berveiller and A. Eberhardt, Int. J.


Sol. Struct., 36, pp. 4289-4315, (1999)
[Micromechanically motivated multivariant modeling]

SMA – p.62/108
Contents-III

C ONTENTS OF PART III

Operator-split algorithms at the crystal grain level

Optimization-based algorithms at the crystal grain level

Modeling of textured polycrystals

Experiments and simulation

An application to the design of biomedical devices

SMA – p.63/108
Algorithmic treatment

Within the finite element paradigm, the algorithmic problem


for a single crystal is set up as follows:

“Given the state at time tn , and the total displacement un+1


at time tn+1 , determine the state at time tn+1 .”

The preceding problem is iteratively nested inside the


global momentum balance problem.

material
global (i)
momentum
un+1 (Gauss pt.)
balance
i←i+1
PSfrag replacements i−th iterate
local state
i−th iterate
global state
assemble

SMA – p.64/108
Algorithmic treatment

Two potential approaches:

Unified approach
Determine the phase state by solving a constrained
optimization problem.
Unified treatment of transforming and non-transforming
states

Operator-split approach
Analogy with computational plasticity for
rate-independent materials
Different treatment of transforming and
non-transforming states

SMA – p.65/108
Unified approach

Treat elastic loading/forward transformation in a unified


manner.
σeq
PSfrag replacements
ξ¯u = 1

¯ P
ξl = ξα

0 eq
Express the constraint conditions as
nv
X
−ξα ≤ 0 , ξ¯l − ξα ≤ 0 , ξ¯u ≤ 1 ,
α=1

where ξ¯l and ξ¯u are lower and upper values of the total
martensitic volume fraction. SMA – p.66/108
Unified approach

Treat elastic unloading/reverse transformation in a unified


manner.
σeq
PSfrag replacements ¯ P
ξu = ξα

ξ¯l = 0

0 eq
Express the constraint conditions as
nv
X
−ξα ≤ 0 , ξα − ξ¯u ≤ 0 , 0 ≤ ξ¯l ,
α=1

where ξ¯l and ξ¯u are lower and upper values of the total
martensitic volume fraction. SMA – p.66/108
Unified approach

Forward transformation Reverse transformation


Š ‰
„

ˆ ‡
„
† …
„

Ž 
„
ƒ ‚
„

Œ ‹
„
The polytope constraints for nv = 3

SMA – p.67/108
Unified approach

Recall that the forward transformation problem can be cast


in the form
{ξα } = argmin Φf ,
{ξγ }∈J f

subject to the preceding polytope constraints.

Introduce Lagrange multipliers {λα , α = 1, . . . , nv}, λl , and


λu , and write
∂Φf
+ λ α + λ l − λu = 0 ,
∂ξα
where
nv
X nv
X nv
X
λα (−ξα ) + λl (ξ¯l − ξα ) + λ u ( ξα − ξ¯u ) = 0 .
α=1 α=1 α=1
SMA – p.68/108
Unified approach

Three key algorithmic tasks:

Identify the state as “loading” (forward transforming) or


“unloading” (reverse transforming).

Identify the set C of potentially active variants.


Select up to 6 such variants from the set of all nv habit
plane variants.

Identify the variants from the set C that are active.

All three tasks are deformation-dependent (albeit


history-independent) and generally need to be performed
in each loading step!
SMA – p.69/108
Unified approach

Recall that the functional Φf is defined as


nv
1   X X
Φf = E − Et · C E − Et B(θ − θ0 ) ξα + F c ξα ,
2 α=1 f α∈J

∂ 2 Φf
hence is quadratic in {ξα }, and the 24 × 24 Hessian
∂ξα ∂ξβ
is at most of rank 6. This leads to a semi-definite quadratic
programming problem.

To “relax” this problem, identify a set C of 6 candidate


active variants at each step.

Subsequently, a standard active set strategy of


positive-definite quadratic programming is applied to
determine the active variants and volume fractions.
SMA – p.70/108
Unified approach

There are several ways to identify a “reasonable” set C of 6


candidate active variants. One possibility is to first
determine the extrema of Φf with respect to each variant
separately, i.e., solve
f
∂ Φ̂
Ŷαf (E, ξα , θ) = = 0,
∂ξα
for each ξα , α = 1, . . . , 24. Subsequently, select the 6
variants that correspond to the lowest values of
Φf = Φ̂f (E, ξα , θ).

Alternatively, one may perform a pair-wise minimization to


determine 3 pairs of potentially active variants, as
motivated by experiments.

The set of active variants depends crucially on the load.


SMA – p.71/108
Unified approach

Given the set C of candidate variants, express the discrete


counterpart of the first-order conditions
∂Φf
+ λ α + λ l − λu = 0 ,
∂ξα
nv
X nv
X nv
X
λα (−ξα ) + λl (ξ¯l − ξα ) + λ u ( ξα − ξ¯u ) = 0
α=1 α=1 α=1

in matrix form as

[Q][ξn+1 ] + [Πn+1 ]T [λn+1 ] = [cn+1 ] ,


[Πn+1 ][ξn+1 ] = [hn+1 ] .

This system possesses a unique solution, as [Q] is


positive-definite and [Πn+1 ] has full row-rank.
SMA – p.72/108
Unified approach

Use an active set strategy starting from any feasible initial


guess of volume fractions and active variants (working set).

1. Solve the first-order equations and determine new


values of [ξn+1 ] and [λn+1 ].

2. If one or more of the inactive constraints are violated,


then add the most offending constraint to the working
set and go to step 1.

3. If one or more of the Lagrange multipliers become


negative, then drop from the working set the constraint
that corresponds to the lowest value over all multipliers
and go to step 1.

Slow, but reliable!


SMA – p.73/108
Unified approach

The discrete counterparts of the loading criteria from a


forward transformation state at t = tn+1 are




 > 0 ⇔ forward transformation
X f 
(ξα,n+1 −ξα,n ) = 0 and λl = 0 ⇔ neutral forward transformation ,


α∈C  = 0 and λl > 0 ⇔ elastic unloading

where ξα,n+1
f
is calculated by constrained minimization of Φf .

Note that the latter two conditions are modified in order to


account for the explicit enforcement of the constraint
¯ P
ξl − α∈C ξα ≤ 0.

SMA – p.74/108
Unified approach

The discrete counterparts of the loading criteria from a


reverse transformation state at t = tn+1 are




 < 0 ⇔ reverse transformation
X 
r
(ξα,n+1 −ξα,n ) = 0 and λu = 0 ⇔ neutral reverse transformation ,


α∈C  = 0 and λu > 0 ⇔ elastic reloading

where ξα,n+1
r
is calculated by constrained minimization of Φr .

Note that the first two conditions are modified in order to


account for the explicit enforcement of the constraint
P ¯u ≤ 0.
α∈C ξ α − ξ

SMA – p.75/108
Unified approach

In the unified approach, there is no distinction in the


treatment of elastic or transforming states. This implies
that one only needs to check whether the state is “forward”
or “reverse”.

Use a flag to handle the characterization (set initially to


“forward”). and resolve all possible combinations of
forward/reverse loading:
H
HH ∆ξnf
H + 0
∆ξnr HH
H

− !flag ‘reverse’

0 ‘forward’ flag
SMA – p.76/108
Unified approach

The unified approach is also applicable for problems that


lead to non-linear programming. For example, assume

nv
1 t
 t
 X
Ψ = E−E · C̄ E − E + B(θ − θ0 ) ξα ,
2 α=1

where, using the rule of mixtures,


nv
! nv
X X
C̄ = 1 − ξ α Ca + Cm .
α=1 α=1

Use non-linear programming with active set strategy or


reduce the problem to one of quadratic programming by
using time-lagging estimates of C̄.
SMA – p.77/108
Operator-split approach

Examine the operator-split approach in connection with the


preceding model.

Given the state at tn (i.e., {ξβ,n }), the temperature θn+1 ,


(i)
and the i−th iterate of the displacement un+1 :

(i)
1. Assume that the total strain En+1 is elastic, i.e., there is
no phase forward transformation.
(i)
2a. If Ŷαf (En+1 , {ξβ,n }, θn+1 ) < 0 for all variants α, then the
process in (tn , tn+1 ] is elastic.
(i)
2b. If Ŷαf (En+1 , {ξβ,n }, θn+1 ) ≥ 0 for some variant(s) α, then
there is some phase transformation.

SMA – p.78/108
Operator-split approach

As in the unified approach, one again needs to determine:

The set C of potentially active variants.

The variants from the set C that are active.

The operator-split approach is popular in computational


plasticity because of the simplicity of the elastic predictor
and the physical interpretation of the plastic corrector.

Are these attractive features preserved in the case of


micromechanically motivated phase transition models?

SMA – p.79/108
Operator-split approach

The operator-split approach has two complications:

1. It is not readily obvious how to effect the “plastic” (i.e.,


transformation) correction. It is sometimes assumed
that
f
∂Y
ξ˙α = κ α
,
∂ξα
which corresponds to an “associated” flow rule.

PSfrag replacements

Y1f = 0
Yf2 = 0

There is no constitutive justification for this assumption!


SMA – p.80/108
Operator-split approach

2. The evolution of the variants is subject to the polytope


constraints.

ag replacements

f ξ<0
Y >0

Yf =0 ξ=0

One needs to perform constrained projections in


multi-dimensional space. In some cases, the
projections indicate unsuitable selection of potentially
active variants.
SMA – p.81/108
Polycrystal modeling

Texture can potentially play a pivotal role in the mechanical


response of polycrystalline superelastic alloys.

Two ways to model the textured polycrystal structure:

Direct simulation
Resolve the polycrystal structure using individual finite
elements. This is a simple, but potentially expensive
approach.

Two-scale analysis
Solve fine-scale boundary-value problem on a
representative domain and extract volume-averaged
stress to be used in a continuum-level simulation. This
approach can radically reduce the computational cost.
SMA – p.82/108
Polycrystal modeling

To conduct either direct simulation or two-scale analysis,


one needs to utilize the texture information.

As an example, consider the Nitinol tubes described


earlier. Here, the manufacturing process induces primarily
h111i{110}-type sheet texture “wrapped” around the
cylindrical surface, such that the h111i austenite lattice
direction is aligned with the longitudinal axis of the tube.
’”“

™
–

žŸ œ

 
žŸ

š
›


¢
’‘

—
¡
˜

SMA – p.83/108
Polycrystal modeling

Pole figures for Nitinol tubes (R.D. horizontal, T.D. vertical)


SMA – p.84/108
Polycrystal modeling

Generally, pole figure data are directly processed by the


software that handles crystallographic data acquisition.

Relevant software: BEARTEX, LaboTex, popLA, etc.

Typical software output:

(1) (1) (1)


ω1 ω2 ω3 q (1)
(2) (2) (2)
ω1 ω2 ω3 q (2) ,
··· ··· ··· ···

(i)
where ωI , I = 1, 2, 3, are orientation-defining angles for
the i-th bin and q (i) is the corresponding diffraction
intensity.
SMA – p.85/108
rag replacements
Polycrystal modeling

q0 = q00 q

y
z x 0
00
r =z p
ω2
ω1
p
ω3 r = r0

q000 = y

Matthies-Roe angle convention (p, q, r) → (x, y, z)


SMA – p.86/108
Polycrystal modeling

Also, recall that the austenite is wrapped around the tube.

q
p

PSfrag replacements ω0
r

This induces an initial rotation ω0 of the cubic crystals


relative to the r axis.

SMA – p.87/108
Polycrystal modeling

In summary, the total rotation Q that characterizes the


texture can be expressed as

Q = Q 3 Q2 Q1 Q0 ,
where
Q0 : rotation by ω0 relative to r ,
Q1 : rotation by ω1 relative to r ,
Q2 : rotation by ω2 relative to q0 ,
Q3 : rotation by ω2 relative to r00 .

To obtain a matrix representation of the rotation, recall the


Rodrigues formula for rotation in {p, q, r} by ω with respect
to p:
Qr = p ⊗ p + cos ω(q ⊗ q + r ⊗ r) − sin θ(q ⊗ r − r ⊗ q) .
SMA – p.88/108
Polycrystal modeling

For any given grain, the angles ωI , I = 1, 2, 3, are picked


from the pole figure data using a random processes
weighted by the diffraction intensities.

The angle ω0 is computed directly as a function of the


location of the grain on the tube.

Once the cumulative rotation Q is known, the habit plane


transformation strains {Etα } of a typical crystal grain are
expressed as

Eta,tex = QEtα QT ,
where Etα are the transformation strains relative to the
austenite (cubic) frame.
SMA – p.89/108
Polycrystal modeling

The direct simulation approach is straightforward.

The two-scale approach requires the solution of a


boundary-value problem at each Gauss point:

Use the macro-scale deformation as input for the fine-scale


problem, determine the mean stress, and use it to define
the constitutive behavior in the macro-scale problem.
SMA – p.90/108
Polycrystal modeling

Use Taylor assumption, i.e., impose constant deformation


gradient field F̄ = F in the fine-scale problem.

Recalling the Hill-Mandel averaging theorem, compute and


output the mean 1st Piola-Kirchhoff stress P̄.

Interesting issues:

Size of the fine-scale problem (convergence of the


microstructure)

Implicit global solution options (Newton’s method and


its variants)

Other consistent averaging options

SMA – p.91/108
Numerical simulations

Material parameters:

Austenite Young’s modulus : E = 38.0 GPa


Martensite Young’s modulus : E = 10.0 GPa
Poisson’s ratio : ν = 0.3
Chemical energy constant : B = 0.607 MPa/◦ C
Temperature : θ − θ0 = 22.3◦ C
Critical force : F c = 7.5 MPa.

The crystallographic vectors are taken to have components

[n] = [−0.88888, 0.21523, 0.40443]T


[m] = [0.43448, 0.75743, 0.48737]T ,

relative to the austenite lattice, while g = 0.13078. SMA – p.92/108


Numerical simulations

500

450

400

350

300
Teq (MPa)

250

200

150

100

Experment 1
50 Experiment 2
Experiment 3
Simulation
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Eeq

Equivalent stress-strain plot for pure tension


SMA – p.93/108
Numerical simulations

1
Variant 2
Variant 3
0.9 Variant 11
Variant 19
Variant sum
0.8

0.7
Volume fraction

0.6

0.5

0.4

0.3

0.2

0.1

0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
E
eq

Volume fraction vs. equivalent strain for pure tension


SMA – p.94/108
Numerical simulations

_________________
STRESS 3

2.86E+08
3.02E+08
3.19E+08
3.35E+08
3.52E+08
3.68E+08
3.85E+08
4.01E+08
4.18E+08
4.34E+08
4.50E+08
4.67E+08
4.83E+08

Time = 6.00E+01

T33 stress distribution for pure tension at 6 % strain


SMA – p.95/108
Numerical simulations

500

450

400

350

300
Teq (MPa)

250

200

150

100

50
Tension 0.0%, Torsion 2.0%
Simulation
0
0 0.005 0.01 0.015 0.02 0.025 0.03
Eeq

Equivalent stress-strain plot for pure torsion


SMA – p.96/108
Numerical simulations

0.12
Variant #15
Variant #23
Variant #11
Variant #19
0.1 Variant Sum

0.08
Vol. Fraction

0.06

0.04

0.02

0
0 0.01 0.02 0.03
εeq

Volume fraction vs. equivalent strain for pure torsion


SMA – p.97/108
Numerical simulations

600

500

400
Teq (MPa)

300

200

100

Exp (Tension 2.0%, Torsion 2.0%)


Sim (Tension 2.0%, Torsion 2.0%)
0
0 0.005 0.01 0.015 0.02 0.025 0.03 0.035
Eeq

Equivalent stress-strain plot for tension followed by torsion (T YPE I)


SMA – p.98/108
Numerical simulations

500

450

400

350

300
Teq (MPa)

250

200

150

100

50
Experiment
Simulation
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Eeq

Equivalent stress-strain plot for simultaneous tension/torsion (T YPE III)


SMA – p.99/108
Numerical simulations

500

450

400

350

300
Teq (MPa)

250

200

150

100

50 Experiment
Simulation with correct texture
Simulation with incorrect texture
0
0 0.01 0.02 0.03 0.04 0.05 0.06 0.07
Eeq

Effect of texture on equivalent stress-strain response


SMA – p.100/108
Numerical simulations

cut

original tube fine mesh

Schematic of stent manufacturing

SMA – p.101/108
Numerical simulations

_________________
MARTENSITE

0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00
0.00E+00

r = 4Ro

Martensite distribution on a stent in compression/tension cycle


SMA – p.102/108
Numerical simulations

_________________
MARTENSITE

0.00E+00
6.77E-03
1.35E-02
2.03E-02
2.71E-02
3.39E-02
4.06E-02
4.74E-02
5.42E-02
6.09E-02
6.77E-02
7.45E-02
8.13E-02

r = 3Ro

Martensite distribution on a stent in compression/tension cycle


SMA – p.102/108
Numerical simulations

_________________
MARTENSITE

0.00E+00
3.93E-02
7.86E-02
1.18E-01
1.57E-01
1.97E-01
2.36E-01
2.75E-01
3.14E-01
3.54E-01
3.93E-01
4.32E-01
4.72E-01

r = 2Ro

Martensite distribution on a stent in compression/tension cycle


SMA – p.102/108
Numerical simulations

_________________
MARTENSITE

0.00E+00
6.81E-02
1.36E-01
2.04E-01
2.72E-01
3.41E-01
4.09E-01
4.77E-01
5.45E-01
6.13E-01
6.81E-01
7.49E-01
8.17E-01

r = Ro

Martensite distribution on a stent in compression/tension cycle


SMA – p.102/108
Numerical simulations

_________________
MARTENSITE

0.00E+00
6.71E-02
1.34E-01
2.01E-01
2.68E-01
3.35E-01
4.03E-01
4.70E-01
5.37E-01
6.04E-01
6.71E-01
7.38E-01
8.05E-01

r = 2Ro

Martensite distribution on a stent in compression/tension cycle


SMA – p.102/108
Numerical simulations

_________________
MARTENSITE

0.00E+00
2.76E-02
5.52E-02
8.28E-02
1.10E-01
1.38E-01
1.66E-01
1.93E-01
2.21E-01
2.48E-01
2.76E-01
3.03E-01
3.31E-01

r = 3Ro

Martensite distribution on a stent in compression/tension cycle


SMA – p.102/108
Numerical simulations

16

14

12

10
Pressure (MPa)

Pressure vs. radial displacement


0
0 0.5 1 1.5 2 2.5 3
(4Ro−r)/Ro

Pressure-radial displacement plot for a stent


SMA – p.103/108
Numerical simulations

16

14

12

10
Pressure (MPa)

Stent with correct texture


Stent with incorrect texture
0
0 0.5 1 1.5 2 2.5 3
(4Ro−r)/Ro

Effect of texture on stent response


SMA – p.104/108
Numerical simulations

16
Thickness = 0.375mm
Thickness = 0.3mm
Thickness = 0.2mm
14 Thickness = 0.1mm

12

10
Pressure (MPa)

0
0 0.5 1 1.5 2 2.5 3
(4Ro−r)/Ro

Effect of wall-thickness on stent response


SMA – p.105/108
Numerical simulations

16
Height = 5.0mm
Height = 7.0mm
Height = 8.0mm
14 Height = 10.5mm

12

10
Pressure (MPa)

0
0 0.5 1 1.5 2 2.5 3
(4Ro−r)/Ro

Effect of cascade-height on stent response


SMA – p.106/108
Closure

Superelasticity may be algorithmically tackled either as


a multi-phase elasticity problem or as a plasticity-like
problem.

The multi-phase elasticity problem (unified approach)


makes no algorithmic distinction between elastic
loading and forward transformation.

The plasticity-like problem introduces an element of


new physics in the flow-like rule for the evolution of the
variant volume fractions.

Texture can be incorporated either via direct simulation


or by two-scale modeling.
SMA – p.107/108
References

1. M.A. Qidwai and D.C. Lagoudas, Int. J. Numer. Meth.


Engrg., 47, pp. 1123-1168, (2000) [Plasticity-like
algorithmic development for phenomenological theory].

2. S. Govindjee and C. Miehe, Comp. Meth. Appl. Mech.


Engrg., 191, pp. 215-238, (2001) [Plasticity-like
algorithmic development for microstructural theory].

3. Y. Jung, P. Papadopoulos and R.O. Ritchie, Int. J. Numer.


Meth. Engrg., 60, pp. 429-460, (2004) [Elasticity-based
algorithmic development for microstructural theory].

4. U.F. Kocks, Chapter 2 of Texture and Anisotropy,


Cambridge UP, (2000) [Background on texture
measurement].

SMA – p.108/108