Proceedings of the ASME 2011 Conference on Smart Materials, Adaptive Structures and Intelligent Systems

SMASIS2011
Proceedings of the ASME 2011 Conference on Smart 18-21,
September Materials, Adaptive
2011, Structures
Scottsdale, and
Arizona, USA
Intelligent Systems
SMASIS 2011
September 18-21, 2011, Scottsdale, Arizona, USA

SMASIS2011-
SMASIS2011-5165

THERMO-MECHANICAL MODELING OF A SHAPE MEMORY ALLOY HEAT ENGINE

Christopher B. Churchill∗ John Shaw

Aerospace Engineering Aerospace Engineering
University of Michigan University of Michigan
Ann Arbor, Michigan 48109 Ann Arbor, Michigan 48109
Email: [email protected]

ABSTRACT experimentally. Second, those relationships were used as inputs

Two thirds of the energy generated in the United States in a steady-state model of the heat engine, including both con-
is currently lost as waste heat, representing a potentially vast vective heat transfer and large-deformation mechanics. Finally,
source of green energy. Low Carnot efficiency is an inherent limi- the model was validated successfully against measurements of a
tation of extracting energy from low-grade thermal sources (tem- experimental heat engine built at HRL Labs.
perature gradients near or below 100C), and SMA heat engines
could be useful for those applications where low weight and
packaging are overriding considerations. Although many shape
INTRODUCTION
memory alloy (SMA) heat engines have been proposed to har-
vest this energy, and a few have been built and demonstrated in Since the discovery of Nitinol, there have been many ef-
past decades, they have not been commercially successful. Some forts [1] to construct a commercially viable shape memory alloy
of the barriers to commercialization include their perceived low (SMA) heat engine to harvest waste heat. While the theoretical
thermodynamic efficiency, high material cost, low material dura- efficiency of SMA heat engines is low [2], the supply of waste
bility, complexities when using fluid baths, and the lack of robust heat is vast, so one metric of a successful engine is to generate
constitutive models and design tools. Recent advances, however, enough power to justify the total cost of operation. In transporta-
in SMA longevity, reductions in materials costs (as production tions, engine packaging and mass are also critical factors. As
volumes have increased), and a better understanding of SMA be- much of the cost of an SMA engine is tied up in the production
havior have stimulated new research on SMA heat engines. of the SMA alloy, a reasonable metric to judge the success of
The Lightweight Thermal Energy Recovery System an engine is power per mass SMA. The ability to accurately pre-
(LighTERS) is an ongoing ARPA-E funded collaboration dict the true power output of an SMA heat engine, not just the
between General Motors, HRL Laboratories, Dynalloy, Inc., theoretical maximum, is critical to evaluating this metric.
and the University of Michigan. In the LighTERS engine (a The “Dr. Johnson” tension-loop heat engine was one of the
refinement of the Dr. Johnson engine), a closed loop SMA first patented engines [3], and was investigated in the most de-
spring element generates mechanical power by pulling itself tail by McDonnell Douglass in the 1970’s. Their work culmi-
between alternating hot and cold air regions. The first known nated in a 32 W experimental engine [4, 5], which used multiple
thermo-mechanical model for this type of heat engine was SMA spring loops running in parallel and series between hot and
developed in three stages. First, the constitutive and heat cold fluid baths. Though the power generation was significant,
transfer relationships of an SMA spring form were characterized the engine suffered from severe mixing of the hot and cold flu-
ids (which were splashed around by the spring), destroying the
thermal gradient and making scale-up problematic. To model
∗ Address all correspondence to this author.
1 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 the engine behavior, a mechanical design methodology was de- coiled spring geometry are determined using wind tunnel mea-
veloped to predict optimal gear ratios and SMA pre-strain, but surements. The SMA spring is thermo-mechanically character-
it was never coupled to the heat transfer properties of the SMA ized in a custom dead-load apparatus, and the results suggest a
spring. simple constitutive law governing the relationship between SMA
The LighTERS heat engine is of this same basic design, but spring temperature, deformation, and tension. A coupled set
distinguishes itself by using forced convection of gasses, rather of analytical equations for equilibrium, compatibility, and heat
than liquids, for heat transfer. While this results in a lower power transfer are solved for the time-dependent field variables of inter-
engine, it eliminates a major design complexity and enables a nal force, spring stretch ratio, and temperature. Finally, a simula-
more robust engine suitable for use in the transportation indus- tion of the LighTERS engine is presented using the SMA spring
try. The engine also takes advantage of the latest advances in properties measured in the preceding sections, and then com-
SMA alloys and joining techniques, using a material provided pared to measurements from the experimental engine [9]. Some
by Dynalloy, Inc. which has very low thermal hysteresis, high details of the engine operation and measurement have been with-
longevity, and a high quality joint. held to protect intellectual property of the project partners.
Our engine, in experimental form, consists of a single loop
of SMA spring wound from thin SMA wire. The loop is sus-
pended between two drive pulleys and one idler pulley, forming
an isoceles triangle. The drive pulleys are connected by a timing KINEMATICS AND EQUILIBRIUM
chain which couples their rotation rates. As the loop is heated
on only one side of the engine, a difference in tension develops Equilibrium
between the two sides, and given the asymmetry of the timing Figure 1 shows a diagram of the LighTERS engine, which
ratio, the engine turns. consists of a loop of SMA belt wound around three pulleys. The
Previous modeling efforts have addressed SMA heat en- top pulley is an idler pulley. The bottom pulleys (1 and 2) have
gines. Often, the engine is merely a toy problem to demonstrate a radius a, and rotation angles θ1 and θ2 , are in contact with the
a new constitutive model [6], and is not compared with exper- spring through a wrap angle φ , and are connected by an inexten-
iments. [Going further, Zhu et al [7] modeled the twin crank sional timing chain. A heating zone lies between points A and
engine of Iwanaga [8] with some success, though no justification B, and a cooling chamber between points C and D. An oppos-
was given for their choice of film coefficients for the SMA in a ing moment Mext is applied to pulley 1, representing an external
fluid bath, and the cooling section in the simulation was modeled load from, for example, an electric generator. Between points C
as water while the experiment used air. We have found no ex- and D (cooling chamber), the belt is under constant tension FCD ;
amples of a thermo-mechanical model for a “Dr. Johnson” style between points A and B (heating chamber) it is under a higher
tension-loop engine which includes heat transfer, let alone one tension FCD + ∆F. The timing sprockets have unequal radii b1
with experimental results for calibration or validation. and b2 . The timing chain is also under some tension Ft on the
This paper aims to develop a system-level model of the bottom side, and some higher tension Ft + ∆Ft on the top side.
LighTERS heat engine, with the chief unknown being the out- A sum of moments around pulleys 1 and 2 give the equilibrium
put power. To do this, the model must consider force equilib-
rium, compatibility, and energy equilibrium (heat transfer). The
assumption of steady-state operation allows the solution to be
obtained with reduced degrees of freedom, but the model still
calculates full temperature, deformation, and force fields for the CCD
SMA loop. The ultimate use of the model is not intended for
broad conceptual design of an SMA heat engine, but rather for
lt
further optimization of the LighTERS concept. The engine ge- Be Idler Pulley
M A
ometry and SMA dimensions have not been optimized, and the S . .
D θ1 θ2 C
values presented here should be interpreted as examples, not rec- a Timing Chain
ommendations.
Mext b2
First the operation of the LighTERS heat engine is described b1
φ
in detail, simultaneously developing the kinematic and equilib-
rium relationships which govern its steady-state motion. Next, CAB

heat transfer relationships are developed and characterized for Pulley 1 s Pulley 2
the SMA spring form. Specific and latent heat values for the A B
narrow hysteresis alloy are measured using a differential scan-
ning calorimeter. Convective heat transfer coefficients for the FIGURE 1: LighTERS ENGINE GEOMETRY

2 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 equations, respectively: be transformed between frames by

Mext + b1 ∆Ft = a∆F (1) Rw

dS = p dSw , (4)
b2 ∆Ft = a∆F (2) (πRs )2 + R2w
ds = λ dS, (5)
Using the equilibrium equations to eliminate ∆Ft , a single equa-
tion can be obtained coupling Mext to the difference in tension ∆F where λ is the spring’s axial stretch ratio of current pitch to ref-
and non-dimensional pulley ratio b̄ ≡ b1 /b2 , where 0 < b̄ < 1. In erence pitch λ = P/P0 . For example, the current length of the
other words, the SMA loop applies equal and opposite torques to SMA loop in Fig. 1 is L and the reference length (fully coiled) is
pulleys 1 and 2, but the timing chain applies a smaller torque to L0 . Since the idler pulley is centered between the drive pulleys,
pulley 1, with Mext taking up the difference. this gives the compatibility relations:

Mext = a∆F(1 − b̄). (3) L = CAB +CCD + 2aφ , (6)

L 1 L
Z
= λ0 = λ (s)ds, (7)
L0 L 0
SMA Spring Form
SMA springs used in the experimental engine (and in all
SMA characterizations presented) were provided by Dynalloy with λ0 as the “installation” stretch ratio of the spring. The
Inc, and are commercially available as part of the “Dr. John- springs considered in this paper have a wire radius Rw = 0.15 mm
son’s Clean Heat Engine” demonstration toy. They were primar- and spring radius Rs = 1.1 mm. Their maximum stretch ratio
ily chosen for their “off the shelf” convenience, and have not when fully extended (dSw = ds) is λmax = 23.06. During engine
been optimized for use in a heat engine. Because of the large de- operation, typical stretch ratios encountered are 3 < λ < 20 with
flections involved in the stretching of the spring, it is important installation stretch ratios 3 < λ0 < 15.
to first introduce three different coordinate systems. Figure 2a
shows the local frame of a cylindrical SMA wire of radius Rw Kinematics
and an axial coordinate Sw that runs along its centerline. The The engine model assumes steady state engine operation. By
wire is wound in a helix of radius Rs (from the wire centroid) and definition, the pulleys are not accelerating, (θ̈ = 0), and the SMA
axis S so that it is fully compressed, i.e. the reference pitch P0 is loop is not “piling up” on one side of the engine. This is enforced
quite close to 2Rw , to form the spring reference frame in Fig. 2b. by requiring the mass flow rate at any point be constant. Written
The spring’s current frame, Fig. 2c, allows it to be stretched out in both the current (velocity V ) and reference (velocity V0 ) spring
(still in a helix) to a current pitch P, and has a current axis s. coordinates, and then transforming the linear density dm dS between
If the wire frame Sw is considered inextensional, lengths can frames using Eqn. 5,

s dm dm
ṁ = V0 = V (s) = constant (8)
Sw dS ds
V (s) = V0 λ (s). (9)
S
As the SMA spring traverses pulley 1 or 2, it must undergo
P0
some change in both tension and stretch ratio, depending on the
friction conditions between spring and pulley. The assumptions
Dw
P in Fig. 3 were verified using high speed video to observe the
spring at pulley entrances and exits. At point D, the spring enters
Ds pulley 1 at some stretch ratio λ1 , “sticking” perfectly to the pul-
(a) (b) (c) ley as it traverses. As it leaves the pulley at point A, the spring
jumps in stretch as it enters the higher force heating zone. A
FIGURE 2: SPRING FRAMES OF REFERENCE: (a) WIRE similar process occurs on pulley 2: as the material enters the pul-
FRAME; (b) SPRING REFERENCE FRAME; (c) SPRING ley at point B, it “sticks” at λ2 as it traverses the pulley, then
CURRENT FRAME. contracts as it leaves pulley 2 at point C, jumping in force back

3 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 down to FCD . The effect is equivalent to that of a soft chain and Below is the heat equation, neglecting radiation, written in
rigid sprocket, if the sprocket could always match the pitch of the the wire frame (Fig. 2a) with internal contributions on the left
incoming chain. and external terms on the right. In the reference spring frame,
These assumptions lead to a convenient constraint on the the linear mass density is a constant, m0 = dS
dm
w
. Conduction along
stretch ratios λ1 and λ2 . Equation 8 can be written at points B the wire axis is ignored because of the slenderness of the wire.
and D as In the experimental engine, the aspect ratio of total wire length
Lw to wire diameter 2Rw is over 104 .
dm dm
ṁB = aθ̇1 /λ1 = ṁD = aθ̇2 /λ2 , (10)
dS dS m0 (Cw Ṫw + Λξ˙ ) = −2πRw H w (Tw − Ta ) (13)

where θ̇i are the rotation rates of each pulley. Additionally, From left to right, constants governing heat flow are the SMA
the timing chain creates a constraint between the rotation rates, specific heat Cw , SMA latent heat of transformation Λ (we use
which in turn couples λ1 and λ2 to the timing ratio b̄. a negative value), convective film coefficient H w , and ambi-
ent temperature Ta . The following sections describe each these
terms, as well as their treatment in the LighTERS model.
b1 θ̇1 = b2 θ̇2 , (11)
λ2 Internal Heat Flow
b̄ = . (12)
λ1 Heat capacity, stress-free transformation temperatures, and
stress-free latent heat of transformation were measured via dif-
ferential scanning calorimetry (DSC). Figure 4 presents the ther-
HEAT TRANSFER mogram of the SMA spring material, showing heat flow into the
As the SMA loop traverses the engine, it experiences three specimen vs. temperature, controlled at 10 ◦ C/min. The mea-
different heat transfer regimes.From A → B in Fig. 1, the heating sured heat capacity is 0.63 J/(g K), and the average latent heat of
chamber is modeled as a bath (air) at temperature Thot and veloc- transformation Λ = −7.65 J/g.
ity UAB . From C → D, the cooling chamber has a colder fluid Without a material level model (which is very difficult, due
temperature Tcold and velocity UCD . While on the power pulleys to the multi-axial nature of spring deformation), there is no in-
(B → C, D → A), no heat transfer is assumed. This assumption formation about the rate of martensite generation, ξ˙ . Instead, the
was tested on the experimental engine by using an infrared cam- martensite generation rate can be coupled to the stretch ratio,
era to measure the temperature of the SMA spring as it entered
and exited pulley 2. Under typical operating conditions, the tem-
λ̇
perature change was below 5 ◦ C. Adding this assumption to the ξ˙ = , (14)
kinematic assumption that λ does not change on the pulleys, it λM − λA
follows that the spring is at a single constant load FCD + ∆F from
A → C, and at FCD from C → A. ΔTp = 18 °C
1 Ap = 27 °C

Pulley 1 Pulley 2 dQ
|dT| Af = 52 °C
λ ΛMA = 8.2 J/g Cw = 0.63 J/(g-K)
λ2
(J/g-K)
λ1
0
AM
Λ = −7.1 J/g
F Mf = −30 °C
FCD FCD+ΔF
Mp = 9 °C
−1
−100 −50 0 50 100 150
D A B C
s T ( o C)

FIGURE 3: STRETCH RATIO AND FORCE CONDITIONS FIGURE 4: DSC THERMOGRAM FOR LOW HYSTERESIS
AROUND PULLEYS 1 AND 2 SMA SPRING

4 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 where λM and λA are maximum (fully martensite) and min-
imum (fully austenite) stretch ratios, determined by experi- 12
ments on the springs used in the engine (see Section THERMO-
Nu
MECHANICAL RESPONSE OF AN SMA SPRING). 8

Forced Convection 4
The primary mode of heat transfer to the SMA spring is con-
vective heat transfer. While there is an abundance of literature on 0
convective heat transfer of straight cylinders in crossflow [10,11], 0 200 400 600 0 5 10 15
there is no literature for helical geometries, so the properties were Re Rs /Rw
measured experimentally. Figure 5 shows a photo of the setup, (a) (b)
essentially a large scale hot wire anemometer. A steel spring
of similar geometry to the SMA spring (0.125 < Rs < 0.15 mm, 12
0.8 < Rs < 1.98 mm) was held at a prescribed stretch ratio λ
and angle ψ to the airflow (ψ = 0 is parallel to the flow). Steel Nu
8
springs could be substituted for SMA springs because, with a rel-
atively smooth surface, convective heat transfer is dominated by
the geometry of the boundary (spring surface) and the properties 4
of the fluid [10]. The spring was heated via joule heating, and the
temperature of the surface Tw inferred using the constant temper- 0
ature coefficient of resistance. A parametric study was performed 0 3 6 9 0 30 60 90
to investigate the influence of wire temperature, air velocity, wire λ ψ (°)
diameter, spring diameter, angle of attack, and stretch ratio. (c) (d)
Figure 6 shows the change in effective Nusselt number Nu =
Lc H w /k f (k f = 0.025 W/(m · K) for air) with respect to selected FIGURE 6: SELECTED VALUES OF Nu FOR A HELICAL
dimensionless parameters. The characteristic length (for Re, Nu) SPRING WITH RESPECT TO (a) REYNOLDS NUMBER,
was chosen as Lc = 2Rw . For all cases, the ambient temperature (b) SPRING ASPECT RATIO, (c) STRETCH RATIO, AND
(d) FLOW ANGLE

was constant at 21 ◦ C, the spring temperature between 70 and

100 ◦ C. Though the investigation was not exhaustive (no varia-
voltmeter lead tion in fluid properties other than velocity), it was still valuable to
identify how different spring parameters effected heat transfer. In
power lead the top left, Fig. 6a shows the change in Nu with Reynolds num-
ber Re. The air velocity U was changed from 5 to 26 m/s, hold-
pitot-static probe
ing constant Rw = 0.15mm, Rs = 1.6mm, λ = 1.68, and ψ = 90o .
adjustable mount The relationship follows the expected power law correlation [10],
ψ
with the best fit being
steel spring

Nu = 0.2309Re0.59 . (15)

Figure 6b shows the effect of three different spring radii Rs , hold-

U ing constant Rw = 0.29mm, Re = 720, ψ = 90o , and λ = 8. The
change in Nu followed no monotonic trend, so spring radius was
(a) (b) not considered to be a significant factor in the range tested. Fig-
ure 6c shows the influence of stretch ratio for Re = 447. For this
FIGURE 5: CONVECTIVE HEAT TRANSFER MEASURE- instance, (and many others not shown for brevity), Nu was rel-
MENT SETUP: (a) PHOTOGRAPH OF SETUP IN 0.61x0.61m atively constant above λ = 2.5. Since values of λ below 3 was
WIND TUNNEL (b) TOP VIEW DIAGRAM SHOWING AN- below the typical range of experimental engine operation, λ was
GLE OF ATTACK ψ not a significant player with respect to convective heat transfer.

5 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 Finally, Fig. 6d shows the effect of the angle of the fluid velocity mal hysteresis for a shape memory alloy. Depending on the load,
to spring axis, ψ. Above angles of 30o , heat transfer is relatively the thermal hysteresis is between 5 and 20 °C. It is lowest in
insensitive to ψ. Below 30o , likely because of self-shielding, Nu the range of forces at which the specimen was cycled in the ex-
drops as much as 40 %. For the LighTERS model, experimental perimental engine, between 1.5 and 3 N. Second, according to
correlations were made for the effect of Re, wire temperature, the stretch measurement, the transformation to Martensite (cool-
and wire diameter (through Lc ). ing) never saturated within the temperature range we were able
to achieve.It came closer to saturation in the transformation to
Austenite (heating), especially at lower loads. Finally, at and
THERMO-MECHANICAL RESPONSE OF AN SMA above 3.92 N the hysteresis loop failed to close, i.e. the stretch
SPRING ratio at the end of the cycle was larger than the start.
Experimental characterization
The final input to the model is the “constitutive behavior” Fitting
of the SMA spring. Because the spring experiences a (piece- The material stresses within a spring are inherently multi-
wise) constant tension force between pulleys, a custom appara- axial, with combined bending and torsion. To avoid the com-
tus was constructed to perform constant load characterizations of plexity of a material-level, multiaxial SMA model, the authors
the SMA spring under temperature-controlled cooling and heat- sought to fit the experimental curves in Fig 7 directly, and found
ing cycles. The load was provided by a hanging mass, and tem- good agreement using a cosecant curve. This was inspired by
perature control by a liquid nitrogen cooled thermal chamber. the well-known solution of the classical buckling problem of a
Experiments were performed on a single specimen with ref- vertical rigid bar-torsion spring system with a slight angular im-
erence length L0 = 13.5 mm, or 45 coils. It was excised from perfection under a vertical compressive dead load.
a larger loop that had been cycled in the experimental engine
for approximately 8,000 cycles, enough that repeatable behav-    
ior had been well established. Figure 7 shows the results of 8 λ − λA − λ0 λ − λA
T (λ ) = T0 − π T1 csc π . (16)
experiments on a single specimen, starting at 4.9 N and ending λM − λA λM − λA
with 4.4 N. The initial hot (predominantly austenite), stress-free
stretch ratio of the cycled spring was 1.68. For each cycle, the Once optimal values are chosen for the five fit parameters,
temperature started at a high value, and was then lowered at a Eqn. 16 can describe any one of the curves in Fig. 7 with ex-
rate of about 10 °C/min. Once the low temperature was reached, cellent accuracy. Figure 8 shows an example fit of the heating
the chamber was heated back to the starting high temperature. response from the 1.96 N experiment. The fit is well-behaved
The experiments show several interesting characteristics. beyond the temperatures measured– at high temperatures the
Consistent with Fig. 4, the material has an especially low ther- stretch asymptotically approaches a minimum stretch λA , while
at low temperatures it (more slowly) asymptotically approaches a
maximum stretch λM . T0 is a reference temperature, T1 is related
20 to the maximum slope dT dλ
, and λ0 governs the curvature.
Figure 9 shows their values for each of the 5 fit parameters
16 for each dead load FCD , in both heating and cooling directions.
Each parameter can be easily fit with a polynomial to become a
λ
12 F (N) function of load, i.e. λA = λA (F), and substituted into Eqn. 16 to
4.413 create a complete description of the thermomechanical behavior:
3.922
3.432
8
2.942
2.450
 
λ − λA (F) − λ0 (F)
1.961 T (λ , F) = T0 (F)−π T1 (F)
4 1.471
λM (F) − λA (F)
(17)
0.980
 
λ (F) − λA (F)
0.490 csc π .
0 λM (F) − λA (F)
−100 −50 0 50 100 150 200
T ( o C) This constitutive law is not able to model “inner loops”, as it sim-
ply jumps from the cooling side fit to the heating side according
FIGURE 7: SPRING STRETCH RATIO–TEMPERATURE RE- to sgn(Ṫ ). However, because inner loops are bounded by outer
SPONSE OF THE CYCLED SMA SPRING AT SEVERAL loops, modeling errors are bounded by the narrow temperature
DEAD LOADS hysteresis, 5 − 10 ◦ C.

6 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 20
λ
λ
15 λM A+
T0
10 FCD+ΔF
FCD
5 λ1 D B
λA λ2
0 C+
−100 −50 0 50 100 150 200
T (°C)
T1 T2
T
FIGURE 8: FIT (THICK LINE) VS. EXPERIMENT (THIN
LINE) DURING HEATING AT A 1.47 N DEAD LOAD FIGURE 10: THE SMA SPRING STATE AT POINTS B (λ2 , T2 ,
FCD + ∆F) AND D (λ1 , T1 ,FCD ) DETERMINE THE ENTIRE
STRETCH-TEMPERATURE CYCLE

temperature of its cycle T1 , having just emerged from the cooling

Heating 20 chamber and jumped to the hot side higher force FCD + ∆F. The
Cooling λM force FCD + ∆F determines the subsequent stretch history, i.e.,
λ
which of the dotted heating paths the material follows. As the
T 150 T0 15
SMA travels through the heating chamber, it heats up to its maxi-
( o C) mum temperature T2 at point B. Following the kinematic and heat
100 10 transfer assumptions of Fig. 3, the spring state (λ , T , FCD + ∆F)
λA
T1 is constant between B and C. At point C+ as the spring leaves
50 5 the pulley, the force jumps down to FCD and the stretch jumps
λ0 down according to the constitutive law. In the cooling chamber
between C and D, the temperature drops and the spring stretches
0 0
0 2 4 6 0 2 4 6 out; at point D the spring is at its lowest temperature T1 , the same
F (N) F (N) temperature it started at, and the stretch ratio is at λ1 . As with
pulley 2, the temperature is assumed constant between D and A.
(a) (b)
While the constitutive law couples temperature and stretch
FIGURE 9: (a) TEMPERATURE AND (b) SPRING STRETCH ratio, the heat equation must be solved to find how temperature
RATIO PARAMETERS AS A FUNCTION OF THE SMA evolves with time. Consider a material point traversing a sin-
SPRING TENSION F gle section of the engine, from A to B. Within this section, the
starting and ending temperatures are the identified degrees of
freedom, and the temperature is guaranteed to increase mono-
SMA ENGINE SOLUTION tonically from T1 to T2 . With a constant tension FCD + ∆F, the
For the LighTERS engine model, equilibrium and the heat stretch ratio λ (T ) is given by the constitutive fit. What remains
equation have well-known piecewise analytical solutions, which is to determine the evolution of temperature (and by extension,
can be folded into the compatibility equation. This results in a set the stretch ratio) with respect to time, which can be obtained from
of nonlinear algebraic equations that can be solved numerically energy balance (the heat equation). Ignoring conduction, the heat
for unknown temperatures, stretch ratios, and forces at discrete equation is the ordinary differential equation:
locations along the spring loop. The steady-state stretch, tem-
perature, and force path of a material point for a given engine
configuration can be obtained in terms of the material state at
m0 (Cw Ṫw + Λξ˙ ) = −2πRw H AB
w (Tw − Thot ). (18)
points B and D. This is illustrated in Fig. 10, which shows a
hypothetical example engine cycle in temperature-stretch space,
overlaid on constant force paths generated by Eqn. 17. Starting The assumption in Eqn. 14 couples the evolution of ξ to λ ,
just to the right of point A (A+ ), the material is at the coldest and the heat equation can be further simplified if λ is coupled

7 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 to the temperature evolution. The relationship between T and λ Verification
is nonlinear, but it can be considered proportional if the changes As a check on the validity of the model, results were com-
in both are sufficiently small. Consider a small numerical sub- pared to the experimental engine constructed at HRL Laborato-
increment in temperature and stretch ratio, from Ti to Ti+1 , and ries, whose geometry and baseline configuration are presented
λi to λi+1 . If temperature and stretch ratio change proportionally in Table 1. A more complete description of experimental efforts
over the small increment, the latent heat term can be rewritten can be found in a companion paper, which is published in these
in terms of the temperature change using Eqn. 14 and the con- proceedings [9].
stitutive fit parameters. Now, temperature is the only remaining Some adjustments were made to the expected film coeffi-
variable evolving with time. Terms with i and i + 1 subscripts cients H w in each section, over those predicted from the wind
change with each sub-increment: tunnel correlations, to account for the fluid conditions within the
experimental engine. In the heating chamber, airflow was paral-
lel to (along the axis of) the spring (ψAB = 0). Figure 6d shows

λi+1 − λi Ṫw
 that parallel flow reduced Nu by 30 %, so H AB w was reduced by
m0 Cw Ṫw − Λ = −2πRw H w (Tw − Ta ). the same ratio from the ψ = 90o correlation. On the cold side,
λM − λA Ti+1 − Ti
(19) flow was oriented in the opposite direction as the hot chamber.
The heat equation has now been reduced to a 1st order ordi- No adjustment to H CD w was made for ψCD , as it was above 30 .
o

nary differential equation, and can be solved analytically to find In addition, the true flow velocity as observed from a material
the time sub-increment δti = ti+1 − ti , and by extension T (t) and point on the spring is (summing vectors) u(s) = UAB − V0 λ (s).
λ (t): While the model does not account for this variation in fluid veloc-
ity with position, some correction can be made for the difference
by using the average spring velocity, V0 λ0 . Since the SMA ve-
m0
   
Ti+1 − Ta λi+1 − λi locity is not known a priori, the solution was iterated upon, using
δti = − ln Cw − Λ
Ti − Ta λM − λA 2π Rw hw the previous iteration’s λ0 in the next, until convergence. Only
(20) two iterations were required to converge on a fluid velocity (in
Practically speaking, there is no need for more than 20 total sub- the material frame) within a tolerance of 0.1 m/s. For this con-
increments; solutions presented in this paper use 60. In a material figuration, the cold and hot side fluid velocities from the material
point’s cycle through the engine, time starts at point A (t(A) = frame (lower case) were uAB = 4.5 m/s and uCD = 3.4 m/s. The
0). The stretch history for each increment is λAB (t) and λCD (t). film coefficients in Table 1 reflect these adjustments.
Seven governing equations are required for the seven degrees of The following describes our baseline case, Simulation 1, us-
freedom: V0 , λ1 , λ2 , T1 , T2 , FCD , and ∆F. They are as follows: ing the parameters of Table 1. Figure 11 shows the steady-state
fields (λ (s), T (s), F(s)) of the SMA loop as a function of the
current spring axis s. Figure 12 shows the same results, but this
Mext = a ∆F(1 − b̄), (21)
time in terms of the stretch-temperature cycle of a material point
λ2 as it travels around the engine. Critical points from Fig 10 are
b̄ = , (22)
λ1
T2 = T (λ (B), FCD + ∆F), (23)
TABLE 1: SIMULATION 1 PARAMETERS
T1 = T (λ (D), FCD ), (24)
Z tB
Mechanical Thermal
CAB = V0 λAB (t) dt, (25)
0 CAB 387 mm UAB 4.8 m/s
Z tD
CCD = V0 λCD (t) dt, (26) CCD 614 mm UCD 3.7 m/s
tC
L0 = V0 (tB + tD − tC ) + aφ (λ1−1 + λ2−1 ). (27) φ 132o Thot 200°C
λ0 7.5 Tcold 20°C
The first equation is force equilibrium from Eqn. 3, the second b̄ 0.6 Λ −7.65 J/g
comes from mass conservation of Eqn. 12, the third and fourth
from the constitutive behavior fit of Eqn 17, and the remaining a 20.3 mm Cw 0.63 J/(g K)
ones from global compatibility, Eqn. 7. If no solution can be Mout 4 N mm H AB 205 W/(m2 K)
w
found, this indicates the engine configuration will not run, or runs
in an unsteady “pulsing” mode, which has been observed in the Mfrict 2.5 N mm H CD
w 270 W/(m2 K)
experimental engine on occasion.

8 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 A B C D T (D) = 38.2 ◦ C.
16 160 T 4 F Figure 12 shows the same history in stretch-temperature
space. The vertical portions of the path take place at points A
Pulley 2

Pulley 1
λ λ ( oC) (N) and C, where the spring encounters a step change in tension. The
12 120 3 overall shape of the path shows that this engine configuration
tends to use only a portion of the available “stroke” of the mate-
rial. In order to make better use of the material, the simulation
8 F 80 2 suggests better performance may be achieved by using either a
material with a higher transformation temperature, or an engine
with sub-ambient cooling.
4 40 1 The configuration in the example output turns pulley 1 at
T θ̇1 = 23.2/s, generating 93 mW of power and completing a full
Tcold material cycle every 3.13 s. The measured experimental value
0 0 0 was significantly higher, θ̇1 = 31.6/s (302 RPM) or 126 mW.
0 0.5 1 While the speed of the simulation does not match the experi-
s (m) ment well, the force value does, as the predicted cold side ten-
sion FCD = 1.71N is nearly identical to the measured value of
FIGURE 11: SIMULATION 1 – STRETCH RATIO, TEM-
FCD = 1.70N. The discrepancy in speed could be a combination
PERATURE, AND FORCE DISTRIBUTIONS OF THE SMA of many effects not included in the model: changing heat transfer
SPRING LOOP conditions through the heating/cooling chambers; heat transfer
while on the three pulleys; slippage of the SMA spring while on
the pulleys; and/or friction between the SMA spring and pulley.
16
Using this single experimental measurement as a calibration
λ(A+) = 13.6
λ point, an additional “correction factor” of 1.39 was applied to
12 H AB CD
w and H w to match the power output of that particular ex-
λ1 = 10.1 periment. With the new film coefficients, the model was then
compared to experimentally measured torque curves, with Mout
8 varied between 0 and 10 Nmm. In addition, the hot side velocity
λ2 = 6.1

4 λ(C+) = 4.3 50 HRL Experiment

T1 = 38.2 ºC T2 = 106.8 ºC Model
0 θ̇1 40
0 40 80 120 160
T ( o C) (s-1) Calibration Exp.
30
FIGURE 12: SIMULATION 1 – STRETCH RATIO VS. TEM-
PERATURE CYCLE
UAB (m/s)
20
4.8

indicated. Note that the material never reaches the fluid tem- 3.5
peratures. Following a material point as it moves through the 10 2.2
engine, the material starts at point A+ having just exited pulley 1 0.74
at the higher force FCD + ∆F = 2.51 N and highest stretch ratio
λ (A+ ) = 13.6. As it goes through the heating chamber, the tem- 0
perature rises to T1 = 106.8 ◦ C, and the SMA spring contracts to 0 2 4 6 8 10
λ (B) = λ2 = 6.1. The spring traverses pulley 2 without chang- Mout (N mm)
ing state. As it leaves pulley 2, the spring jumps to the lower
force FCD = 1.71 N of the colder side and drops in stretch ratio FIGURE 13: EXPERIMENT VS. SIMULATION – ROTATION
to λ (C+ ) = 4.3. Between points C and D, the SMA stretches RATE VS. OUTPUT TORQUE AT PULLEY 1 FOR FOUR
out again, reaching λ (D) = λ1 = 10.1 and its lowest temperature HEATING CHAMBER AIR FLOW SPEEDS

9 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org

 UAB was varied between 4.8 and 0.74 m/s, with a separate torque success, it can help optimize the engine geometry once a heat
curve generated at each velocity. Figure 13 shows θ̇1 of both the source has been chosen.
simulated and experimental engines. At each velocity, the speed
of both the simulated and experimental engines decrease mono-
tonically with Mout . The agreement is good at high flow veloci- REFERENCES
ties, and the trends with output moment are captured well. How- [1] Goldstein, D. M., and McNamera, L. J., eds., 1978. Pro-
ever, as UAB decreases to 0.74 m/s, agreement between model ceedings of the NITINOL Heat Engine Conference.
and experiment is worse. There is likely some additional phys- [2] Ziótkowski, A., 1993. “Theoretical analysis of efficiency
ical effect responsible for the different behavior with changing of shape memory alloy heat engines (based on constitu-
velocity. In our opinion, the most likely sources are the physi- tive models of pseudoelasticity)”. Mechanics of Materials,
cal measurement of air velocity (difficult at low speeds), natural 16(4), pp. 365 – 377.
convection (not modeled), or the breakdown of experimental cor- [3] Johnson, A. D., 1977. Memory alloy heat engine and
relations of H AB
w at very low Re (no measurements were taken method of operation - US patent 4055955, November.
below 5 m/s in Fig. 6). [4] McNichols, J., Brookes, P., Ginell, W., and Cory, J., 1980.
Prototype nitinol heat engine. Tech. Rep. MDCG9290, Mc-
Donnell Douglas Astronautics Company, November.
CONCLUSIONS [5] Ginell, W., McNichols, J., and Cory, J., 1978. “One horse-
A thermo-mechanical model for the LighTERS waste heat power thermoturbine nitinol engine”. In Proc. of the NITI-
engine was presented and then calibrated against a working ex- NOL Heat Engine Conference, D. M. Goldstein and L. J.
perimental engine. Equilibrium, kinematic, compatibility, and McNamera, eds., pp. 2–1.
heat transfer relationships were derived for an idealized version [6] Boyd, J., and Lagoudas, D., 1996. “A thermodynamical
of the proposed SMA heat engine. Special attention was given to constitutive model for shape memory materials. part 1. the
the large stretch ratios inherent to the SMA springs used in the monolithic shape memory alloy”. International Journal of
engine, and their effect was considered in compatibility, kine- Plasticity, 12(6), pp. 805–842.
matic, and heat transfer relationships. [7] Zhu, J. J., Liang, N. G., Huang, W. M., and Liew, K. M.,
The SMA spring used in the experimental engine is a com- 2001. “Energy Conversion in Shape Memory Alloy Heat
mercially available narrow hysteresis spring provided by Dy- Engine Part II: Simulation”. Journal of Intelligent Material
nalloy, Inc. The spring element was subjected to a full range Systems and Structures, 12(2), pp. 133–140.
of thermo-mechanical characterizations. First, its latent heat of [8] Iwanaga, H., Tobushi, H., and Ito, H., 1988. “Basic re-
transformation and heat capacity were measured using a DSC search on the output power characteristics of a shape mem-
thermogram. Next, convective heat transfer characteristics of ory alloy heat engine (1st report)”. Nippon Kikai Gakkai
spring form wires were measured using a hot wire anemometer Ronbunshu, A Hen/Transactions of the Japan Society of
setup in a wind tunnel. Finally, the narrow hysteresis spring was Mechanical Engineers, Part A, 54(497), pp. 177 – 180.
mechanically characterized by a series of dead-load experiments [9] Keefe, A. C., McKnight, G. P., Herrera, G. A., and Be-
performed in a thermal chamber. The results were fit to a cose- degi, A. P., 2011. “Development of a shape memory aloy
cant curve, which was used to create a constitutive law governing heat engine through experiment and modeling”. In Pro-
the relationship between spring tension, temperature, and stretch ceedings of the ASME 2011 Conference on Smart Materi-
ratio. als, Adaptive Structures and Intelligent Systems (SMASIS
An example simulation produced by the LighTERS model 2011), pp. SMASIS2011–5212.
was presented and the output compared to that of an experimen- [10] Zukauskas, A., and Ziugzda, J., 1985. Heat transfer of a
tal engine constructed at HRL Laboratories. This single solution cylinder in crossflow. Hemisphere Publishing,New York,
was treated as a calibration point to adjust the convective film NY.
coefficients by +39 %. The calibrated model was then compared [11] Sanitjai, S., and Goldstein, R., 2004. “Forced convection
to a set of rotation rate–torque curves measured on the experi- heat transfer from a circular cylinder in crossflow to air and
mental engine at various heating chamber velocities. Agreement liquids”. International journal of heat and mass transfer,
between experiment and model was good except for the lowest 47(22), pp. 4795–4805.
flow velocity. Research is ongoing to identify additional factors
that affect the operation of the engine, hopefully reducing the
discrepancy between simulation and experiment.
By providing accurate power estimates for a given thermal
gradient and engine geometry, not only can the model provide
guidance on which waste heat source offers the best chance of

10 Copyright © 2011 by ASME

Downloaded From: http://proceedings.asmedigitalcollection.asme.org/pdfaccess.ashx?url=/data/conferences/smasis2011/70449/ on 07/13/2017 Terms of Use: http://www.asme.org