Professional Documents
Culture Documents
catheter jim.sagepub.com
Abstract
In this article, we present a method for optimizing the design of a shape memory alloy–actuated robotic catheter. Highly
maneuverable robotic catheters have the potential to revolutionize the treatment of cardiac diseases such as atrial fibril-
lation. To operate effectively, the catheter must navigate within the confined spaces of the heart, motivating the need for
a tight bending radius. The design process is complicated by the shape memory alloy’s hysteretic relationships between
strain, stress, and temperature. This article addresses the modeling and optimization of both a single-tendon and antago-
nistic tendon robotic catheter using COMSOL Multiphysics Modeling and Simulation software. Several design variables
that affect the actuator behavior are considered; these include the shape memory alloy tendon radius and its prestrain,
the shape memory alloy tendon offset from the neutral axis of the flexible beam, the flexible beam radius and elastic
modulus, and the thermal boundary condition between the shape memory alloy tendon and the beam. A genetic algo-
rithm is used to optimize the radius of curvature of the two catheter designs. Both a single-crystal and polycrystalline
models are implemented in COMSOL and are experimentally validated.
Keywords
actuator, optimization, shape memory
Figure 1. Conventional steerable ablation catheter from Boston Scientific (Natick, MA): (a) catheter tip and handle; (b) planar
bending.
control aspects; little attention has been paid to design 2004), smart inhalers (Paulsey and Seelecke, 2008), and
optimization. The design of SMA actuators is often smart dampers (Moallem, 2003).
iterative and complicated by the material’s complex The remainder of this article is organized as follows:
thermomechanical behavior. The limited work on SMA section ‘‘System model’’ presents the modeling of
design optimization has either focused on the material’s robotic catheters actuated by single and antagonistic
superelastic behavior (Masuda and Noori, 2002) or is SMA tendons. The SMA constitutive model and struc-
based on empirical models (Dumont and Kuhl, 2005; tural bending models are derived and implemented in
Lu et al., 2001). Masuda and Noori (2002) investigate COMSOL. Section ‘‘Design optimization’’ describes the
optimal hysteresis loop shapes in SMA-based energy design optimization problem, including objective func-
dissipation devices, which requires modifying the tions, design variables, and GA implementation. The
material’s chemical and physical properties (Sato experimental setup for validating the bending model is
et al., 1984). In Lu et al. (2001), the authors optimize presented in section 4.The model validation and design
the mass of an SMA-actuated corrugated plate sub- optimization results are presented in section ‘‘Results.’’
ject to specific performance constraints. Dumont and Conclusions are presented in section ‘‘Conclusion.’’
Kuhl (2005) optimize SMA spring actuators using an
empirical model (Ikuta, 1990) featuring an exponen-
tial function for the phase fractions. The spring geo- System model
metry is optimized, but the authors note that there is The robotic catheter uses SMA tendons for internal
no analytic function that quantifies the actuator’s actuation. Each bending segment features three main
capabilities. components: an axially stiff, laterally compliant central
In this article, we present a method for optimizing structure, one or more SMA tendons, and regularly
the design of a SMA-actuated catheter. The SMA’s spaced collets (Figure 2). The collets hold each SMA
thermomechanical behavior is modeled using
COMSOL Multiphysics Modeling and Simulation soft-
ware (COMSOL Inc., Burlington, MA), a finite ele-
ment analysis (FEA) software package that interfaces
with a genetic algorithm (GA) running in MATLAB
(The Mathworks, Natick, MA). The GA is used to
optimize the catheter’s radius of curvature, a measure
of the device’s maneuverability. GAs are ideally suited
to the optimization of smart material actuators, where
Jacobian’s models are difficult or impossible to
obtain.The GA is coded in MATLAB and uses stress
and strain estimates from the COMSOL model to eval-
uate a design objective function (the catheter’s radius
of curvature). While the focus here is on improving the
maneuverability of a robotic catheter for cardiac proce-
dures, the methods and results can be extended to a Figure 2. Robotic catheter prototype: (a) solid model and (b)
variety of applications including robotic hands (Lan photograph.
and Yang, 2009), fish actuators (Shinjo and Swain, SMA: shape memory alloy.
tendon a fixed distance from the neutral axis of the cen- Veeramani (2009) discusses the issues of antagonistic
tral structure, creating a constant moment over the slack in greater detail.
active beam length as the tendons contract.
Each catheter bending segment has two degrees of
freedom (it can bend in two orthogonal planes), and the SMA constitutive model
motion can be decoupled into two planar bending prob- The nonlinear, hysteretic behavior of SMA is due to the
lems. The bending angle u can be decoupled into uxz (the presence of three crystalline phases: one austenite (A)
bending angle projection in the XZ plane) and uyz (the and two martensite variants (M + and M ). Austenite is
projection in the YZ plane). More details regarding this induced at high temperatures, while the martensite
kinematic decoupling and its utilization in control synth- phases are stress induced. M + occurs under tensile
esis are provided in Veeramani et al. (2008b). Here, the stresses, and M occurs under compressive stresses.
goal is to model and optimize the planar design, resulting Because an SMA tendon cannot experience compres-
in optimal bending performance for each catheter seg- sive forces, its low-temperature phases are assumed to
ment. Furthermore, we analyze two separate planar cases: be detwinned martensite at high stresses (M + ) and
a single SMA actuator and antagonistic SMA actuators. twinned martensite at low stresses (M + = , a blend of
Schematics of both cases are shown in Figure 3. both martensite phases). At high temperatures, the
A fully maneuverable robotic catheter requires SMA tendon exists entirely in its austenite phase
antagonistic actuation; however, the single-tendon (Figure 4).
actuation case is presented first to isolate the essential Constitutive models of SMA can be categorized into
complexities encountered in SMA actuation. The opti- physical models and empirical (phenomenological)
mization methods presented here can be easily extended models. Physical SMA models (often referred to as free
to other single-tendon applications (like those found in energy models) date back to Muller and Wilmanski
Zhou and Lloyd (2009) and Yang and Gu (2002) or (1980) and Achenbach (1989) with further work by
designs requiring antagonistic actuation for increased Seelecke and coauthors (Heintze et al., 2003; Heintze
maneuverability and bandwidth (Han et al., 2003; and Seelecke, 2005; Muller and Seelecke, 2001;
Moallem, 2003; Peirs et al., 1998). Antagonistic actua- Seelecke, 2002; Seelecke and Muller, 2004) and Huo
tion adds to the complexity of modeling and control- (1989). Smith et al. have done extensive work on the
ling SMA-actuated systems by introducing additional efficient implementation of polycrystalline (PC) ver-
nonlinearities. When one SMA tendon (the protago- sions of these models, known as homogenized energy
nist) is actuated, the antagonistic (inactive) tendon models (Smith and Braun, 2006; Smith and Massad,
strains, increasing the system’s bending stiffness. When 2005). Free energy models determine the equilibrium
the protagonist is deactivated, slack may develop in the states of phase fractions based on the Gibb’s energy
antagonistic tendon. This slack must first be recovered landscape. Empirical models, like those developed by
before the beam will bend in the antagonistic direction. Liang and Rogers (1990) and Brinson (1993),
frequently use trigonometric or exponential functions the latent heat of phase transformation, and j(t) is the
to represent the crystalline phases. Joule heating input.
While empirical models provide greater computa- Equations (1) to (5) completely describe the constitu-
tional efficiency compared to physical models, they tive behavior of an SMA tendon and can be simulated
require a series of conditional statements to capture using COMSOL (Li, 2006). The resulting COMSOL
hysteretic behavior. Furthermore, these models do not models can serve as objective function calls from
capture the rate dependence of phase fractions. Recent MATLAB optimization algorithms. COMSOL is capa-
increases in the computational power of ordinary PCs ble of solving partial differential equations of the gen-
have enhanced the suitability of free energy models for eral form
optimization problems; the associated ordinary differ-
ential equations (ODEs) and algebraic equations can be ∂2 u ∂u
ea + da + rG = F ð6Þ
readily implemented into multiphysics FEA packages ∂t2 ∂t
such as COMSOL. For this reason, the free energy where ea is considered the mass matrix, da is the damp-
model described in Heintze and Seelecke (2005) is ing matrix, G is the flux matrix, and F is the source
utilized. term. By appropriately specifying the coefficient
The free energy model of SMA uses phase frac- matrices (ea , da , G, and F), the COMSOL model (6)
tions x (0 < x < 1) for each of the three crystalline can be used to represent the SMA phase fractions (1)
phases: xA is the austenite phase fraction, xM + is the and (2), its constitutive relationship (4), as well its heat
tension-induced martensite phase fraction, and xM is transfer dynamics (5). After converting the constitutive
the compression-induced martensite phase fraction. relationships into specific volume form, the coefficient
The dynamics of the xM + phase fraction are governed matrices are
by
ea = 0 ð7Þ
x_ M + ðtÞ = p + A xM + ðtÞ + pA + xA ðtÞ ð1Þ 2 3
0 0 0 0
and the xM phase fraction is governed by 60 1 0 0 7
6
da = 4 7 ð8Þ
0 0 1 0 5
x_ M ðtÞ = pA xM ðtÞ + pA xA ðtÞ ð2Þ 0 rsma H rsma H rsma cv
2 3
Since the sum of all three phase fractions is unity, s ðt Þ 0 0 0
the austenite phase fraction is given by 6 0 0 0 0 7
G=6
4 0
7 ð9Þ
0 0 0 5
xA ðtÞ = 1 xM + ðtÞ xM ðtÞ ð3Þ 0 0 0 kT ðtÞ
The phase transition probabilities pij are determined and
from barriers in the Gibb’s energy landscape; these are
2 3
described in greater detail in Heintze (2004) and Smith 0
(2005). 6 p + A xM + ðtÞ + pA + xA ðtÞ 7
6 7
Due to the presence of these crystalline phases, the F =6 pA xM ðtÞ + pA xA ðtÞ 7 ð10Þ
4 1 5
stress–strain relationship is nonlinear ðhAs ðT ðtÞ T‘ Þ + jðtÞÞ
Ac Lsma
eðtÞ eT ðxM + ðtÞ xM ðtÞÞ
s ðt Þ = xA ðtÞ xM + ðtÞ + xM ðtÞ
ð4Þ The state variables in equation (6) are
EA + EM
u = ½ ux ð t Þ xM + ðtÞ xM ðtÞ T ðt Þ T ð11Þ
where s(t) represents stress in the SMA tendon and e(t)
is its strain. EA is the elastic (Young’s) modulus of the In these matrices, ux (t) represents axial displacement
austenite phase, and EM is the modulus of the marten- of the SMA tendon, k is its conduction heat transfer
site phases. coefficient, Ac is its cross-sectional area, and rsma is its
The thermodynamic behavior is described using a density. The tendon stress s(t) is given by the constitu-
first-order, lumped parameter model tive model (4).
In addition to these model equations, appropriate
mcT_ ðtÞ = hAs ðT ðtÞ T‘ Þ + H x_ M + ðtÞ + H x_ M ðtÞ + jðtÞ boundary conditions must be specified in COMSOL.
ð5Þ For the mechanical boundary conditions, one end of
the SMA tendon is fixed (a Dirichlet boundary condi-
Here m is the mass of the SMA tendon, c is its spe- tion, which specifies the value of the state variable at
cific heat, h is the convective heat transfer coefficient the boundary) while the other end is free. The thermal
between the SMA tendon and ambient air at tempera- boundary condition can either be isothermal (Dirichlet)
ture T‘ , As is the surface area of the SMA tendon, H is or adiabatic (a Neumann boundary condition, which
where Psma is the force produced by the SMA actuator, Ebeam Ibeam
a is its offset from the neutral axis of the beam, and Kbeam = ð19Þ
a 2 Ac
ssma is its stress. The bending radius is related to the tip
bending angle by The COMSOL model for a spring-coupled SMA
actuator is similar to the model previously presented.
L0 The state variables (equation (11)) remain unchanged,
u= ð14Þ
r while the only difference in the coefficient matrices is
where L0 is the length of the flexible beam experiencing found in the flux term
the applied moment. Combining (12) and (13), the tip 2 3
~ ðt Þ
s 0 0 0
bending angle becomes 6 0 0 0 0 7
G=6
4 0
7 ð20Þ
aAc L0 ssma 0 0 0 5
u= ð15Þ 0 0 0 kT ðtÞ
Ebeam Ibeam
Alternatively, the bending angle can be determined ~ (t) relates the equivalent stiffness of
where the stress s
geometrically by the beam (18) to the constitutive model for SMA (4)
Ebeam Ibeam
r= ð23Þ
aAc sðtÞ
2 3 ð28Þ
s ðt Þ 0 0 0
6 0 0 0 0 7 The thermal boundary condition can be either iso-
G1 = 6
4 0
7 ð25Þ
0 0 0 5 thermal or adiabatic. The displacement condition at
0 0 0 kT ðtÞ Boundary 1 is the Dirichlet condition
u1 ð0Þ = 0 ð29Þ
∂
s ðL 0 Þ = 0 ð30Þ
∂x
The final displacement condition at Boundary 3
depends on the prestrain in the SMA tendons and fixes
the displacement at a distance
Figure 6. Simplified one-dimensional model of a flexible In contrast to the single-tendon case, the radius of
catheter actuated by antagonistic SMA tendons. curvature is determined from the stress difference at the
SMA: shape memory alloy. midpoints of each tendon
Table 1. Design variables and bounds. are permitted for the flexible beam, yielding the area
moment of inertia that appears in equation (35).
Variable Description Lower Upper Units
bound bound
Experimental setup
Ebeam Beam elastic modulus 10 137 GPa
rbeam Beam radius 0.05 1.0 mm A single-tendon test rig (Figure 8) was designed and fab-
rsma SMA actuator radius 0.02 1.0 mm ricated to validate the bending model (equation (23)). It
a Actuator offset from 0.37 3.9 mm consists of a 0.457-mm-diameter superelastic Nitinol wire
neutral axis as the central flexible beam and a 0.127-mm-diameter
eP SMA actuator 0.5 5.3 %
prestrain FLEXINOL wire (Dynalloy Inc., Tustin, CA) as the
b Thermal boundary 0 1 — actuator. The SMA tendon is secured at a fixed distance
condition from the neutral axis with collets fabricated from acrylic
plastic (rapid prototyping). The locations of the catheter
SMA: shape memory alloy.
tip and base (distal and proximal locations, respectively)
are measured using a trakSTAR DC magnetic tracking
system (Ascension Technology Corporation, Burlington,
to equation (5). This boundary can either be adiabatic VT). Current is applied to the SMA actuator using a
(0) or be isothermal (1), and it only applies to the ends programmable power supply (Agilent E3615A, Agilent
of the actuator, as the length of the actuator experiences Technologies, Santa Clara, CA).
convective cooling. The test rig is capable of measuring the radius of
The radius of curvature curvature for four different SMA tendon offsets
4 (0.64, 1.05, 1.42, and 1.92 mm) and eleven different
Ebeam prbeam prestrains (0.5%–5.5% in 0.5% increments). The off-
F ð~
xÞ = r = ð35Þ
2
4a prsma ssma set distance is controlled by inserting pins into the
collets that hold the actuator a fixed distance from
is used for the objective function because it adequat- the neutral axis (Figure 8(b)), and the prestrain is
ely quantifies the catheter’s tip maneuverability. controlled by adjusting a tensioning bolt at the prox-
Manipulators such as endoscopes and catheters often imal end of the catheter.
have to navigate in confined spaces, where a tight radius The test rig used to validate the antagonistic actuation
of curvature is necessary. Similarly, large bending angles model (Figure 9) is similar to the one used for single-
can be achieved by increasing the length of the structure. tendon actuation. However, since both tendons need to
In equation (35), the stress is found using COMSOL and be held at fixed distances from the neutral axis, four dif-
the models presented in section ‘System model’. The con- ferent test rigs are built, each utilizing fixed collets (for
straints are automatically satisfied, as the COMSOL each SMA tendon offset: 0.8, 1.05, 1.42, and 1.92 mm).
models automatically handle the constitutive behavior Because of nonlinear issues such as tendon slack,
and boundary conditions. Only circular cross sections all antagonistic actuation experiments involve three
Figure 8. Experimental test rig: (a) photograph of entire setup and (b) illustration of collets showing slots to control actuator
offset from the neutral axis.
SMA: shape memory alloy.
Figure 10. Transient response of the antagonistic test rig: (a) input current and (b) bending angle.
Figure 11. Stress–strain validation for a the SC model: (a) 24°C, (b) 45°C, (c) 75°C, and (d) 95°C.
SC: single crystal.
simulations and experimental data, the optimal pre- Antagonistic actuation. A comparison between the antag-
strain is the maximum recoverable strain eT . onistic actuation model and experimental data is
Figure 14 highlights the dependence of radius of cur- presented in Figure 15. The SC model, the PC model,
vature on actuator offset from the neutral axis. The and the experimental data are shown for the four val-
mean and two standard deviations are again shown for ues of the offset a. The mean and two standard
the experimental data. deviations are indicated on each graph. Like the
The dependence on actuator offset is nonlinear, and single-actuator test results, the experimental data
the optimal actuator offset for the test rig is between more closely match the models at higher prestrains,
the two extremes of the simulated and experimental likely due to inaccuracies in determining zero
data. Furthermore, for lower prestrains, the PC and SC prestrain.
models predict slightly different optimal offsets (as The main difference between these antagonistic ten-
determined by the minima of the curves in Figure 14). don actuation results and the single-tendon results
This nonlinear behavior is one reason why design opti- (Figure 14) is that the optimal prestrain is no longer
mization is necessary for smart materials such as SMA. the maximum recoverable eT . A high prestrain can be
Both the SC model and PC capture the trends observed detrimental to performance, and the effect is more pro-
experimentally. Furthermore, the models lie within two nounced for larger offsets. Close-ups of the models and
standard deviations of the experimental data and are experimental data for the 1.42 and 1.92 mm offsets are
more accurate at higher prestrains, which is the likely shown in Figure 16. For these larger offsets, the experi-
operating region for single-tendon actuators. ments and models indicate an optimal prestrain of
Figure 12. Stress–strain validation for a the PC model (N = 25): (a) 24°C, (b) 45°C, (c) 75°C, and (d) 95°C.
PC: polycrystalline.
approximately 3.5%–4.5%, as measured by the minima accuracy. The GA is randomly initialized with 60 indi-
of the curves. viduals. Binary encoding is used for the designs, and
The same experimental data are presented in Figure uniform crossover is performed on the population with
17 to show the dependence of radius of curvature on a crossover rate of 90%. For crossover, one parent is
the offset a. chosen randomly from the top 30% of the population,
Similar to the single-tendon case, the results in and the other parent is chosen randomly from the top
Figure 17 indicate that the optimal offset is somewhere 50%. Single point mutation is used, where the rate
between the extremes for the offsets investigated. The is 5%.
dependence is similar to the single-tendon model, and The convergence criteria for the SC model consist
the optimal value varies with prestrain but is around 1 of a homogeneity criterion (the population has the
mm, as determined by the experimental data and the same fitness throughout) and a generation limit (50),
PC model. whichever comes first. After the GA converges
using the SC simulation, the population number is
decreased due to increases in computational time
Design optimization results of the PC model; furthermore, the population is
For the optimization of SMA actuators, the GA uses slightly perturbed to introduce new design informa-
both the SC model and the PC model (N = 25). The SC tion. The convergence criteria for the PC simulation
model is used to quickly find an optimal design, at are the same, except the generation limit is reduced
which point the PC model is used for enhanced to 10.
Figure 13. Single-tendon actuation comparison between the radius of curvature for the SC model, PC model, and the
experimental data for four actuator offsets: (a) 0.67, (b) 1.05, (c) 1.42, and (d) 1.92 mm.
SC: single crystal; PC: polycrystalline.
Single-tendon actuation. The optimization results for 7.6 mm after the PC simulation. The design variables
single-tendon actuation are shown in Figure 18. The for these two designs are listed in Table 2. It is interest-
average fitness of the population is shown in Figure ing to note that the SC and PC models produce the
18(a), and the fittest individual in the population is same optimal design; the only difference is the radius of
shown in Figure 18(b). The SC simulation is used dur- curvature associated with the design. These results sug-
ing the first 21 generations, at which point the optimi- gest that the PC model may be unnecessary for the opti-
zation algorithm switches to the PC model for four mization algorithm and only necessary for determining
generations. the performance of the optimal design.
Based on the fact that the SC simulation terminates Comparing the optimal design (Table 2) to the
after 21 generations, the homogeneity criterion is bounds on the design variables (Table 1) reveals that
reached. At this point, the population number is the stiffness of the flexible beam lies on the lower bound
reduced, the population is slightly perturbed, and the (as determined by elastic modulus and radius). The
GA switches to the PC model. The PC simulation results demonstrate the inherent trade-off between
increases both the average fitness and the fitness of the bending flexibility (trackability) and axial stiffness (a
best individual in the population, an outcome seen in characteristic closely related to the ‘‘pushability’’ of
the model validation section. catheters (Bloss et al., 2003)), and motivate the need for
When the SC simulation terminates, the optimal multiobjective optimization that addresses these trade-
radius of curvature is 7.0 mm; this value increases to offs. Both the radius of the SMA tendon and the offset
Figure 14. Single-tendon actuation comparison between the radius of curvature for the SC model, PC model, and the
experimental data for four prestrains: (a) 4.0%, (b) 4.5%, (c) 5.0%, and (d) 5.5%.
SC: single crystal; PC: polycrystalline.
from the neutral axis lie on the lower bound, while the prestrain. The optimal prestrain is lower (3.56% versus
prestrain is on the upper bound, as expected. 4.01%) based on the PC model than the SC model.
This is also lower than the 5.3% optimal prestrain for
the single actuator, but it is not half that value as might
Antagonistic actuation. The single-objective optimization be intuitively expected.
results for antagonistic actuation are shown in Figure The results of the antagonistic actuation case again
19. The average population fitness is shown in Figure highlight the trade-offs in bending flexibility, as the
19(a), and the fitness of the top individual is shown in beam elastic modulus and radius are at the lower
Figure 19(b). Unlike the single-tendon results, the SC bounds. Ultimately, the design variable bounds may
simulation reaches the generation limit (50) before the have to be refined based on observed experimental
optimization algorithm switches to the PC simulation. results. Furthermore, material availability and cost
The PC simulation reaches the time limit (24 h) in nine were not considered when determining the design vari-
generations. able bounds.
The optimal design variables (based on the SC and
PC models) are listed in Table 3. In the single-actuator
Conclusion
results, the design variables are exactly the same; only
the predicted radius of curvature changes. In the PC This article demonstrates the use of COMSOL and
results, the design variables are identical except for the MATLAB to model and optimize the design of SMA
Figure 15. Antagonistic actuation comparison between the radius of curvature for the SC model, PC model, and the experimental
data for four actuator offsets: (a) 0.8, (b) 1.05, (c) 1.42, and (d) 1.92 mm.
SC: single crystal; PC: polycrystalline.
Figure 16. Close-up of the radius of curvature for two offsets: (a) 1.42 and (b) 1.92 mm.
Figure 17. Antagonistic actuation comparison between the radius of curvature for the SC model, PC model, and the experimental
data for four prestrains: (a) 2.3%, (b) 2.74%, (c) 3.62%, and (d) 4.06%.
SC: single crystal; PC: polycrystalline.
Figure 18. GA results for the single-objective optimization of a single SMA actuator: (a) average fitness of the population and (b)
fitness of the best individual in the population.
GA: genetic algorithm; SMA: shape memory alloy.
Figure 19. GA results for the single-objective optimization of antagonistic SMA actuation: (a) average fitness of the population and
(b) fitness of the best individual in the population.
GA: genetic algorithm; SMA: shape memory alloy.
Table 2. Optimal design variables and objective function for Table 3. Optimal design variables and objective function for
the single-objective optimization of a single-tendon catheter. the single-objective optimization of antagonistic actuation.
Variable SC optimal value PC optimal value Units Variable SC optimal value PC optimal value Units
SC: single crystal; PC: polycrystalline. SC: single crystal; PC: polycrystalline.
actuators. Specifically, the article concentrates on the However, the optimal antagonistic actuator design
use of SMA actuators as bending actuators for flexible changes for the SC and PC model, with the PC model
robotic structures, where the design objective is mini- predicting a slightly lower prestrain.
mum radius of curvature. The constitutive model of The optimization results indicate that axial stiffness
SMA exhibits nonlinear, hysteretic dependence on needs to be considered, as the optimal designs lie at the
stress and temperature, which complicates the design lower bound of stiffness. In fact, ‘‘pushability’’ is an
process. The optimization methods presented have the important characteristic of cardiac catheters. Future
potential to greatly expedite the design process by iden- work will focus on quantifying and optimizing ‘‘push-
tifying suitable parameters through simulation. While ability’’ using multiobjective optimization.
the goal here is improved articulation of a robotic
catheter for cardiac procedures, the methods and results Funding
can easily be extended to other systems utilizing SMA
This study was funded by the National Heart, Lung and
actuators.
Blood Institution (NHLBI) of the National Institutes of
Both SC and PC SMA models are presented and
Health (NIH), grant number 5 R44 HL095227-03.
experimentally validated for two different cases: single-
tendon and antagonistic actuation. The PC model con-
sistently predicts a larger (lower performance) radius of References
curvature than the SC model. Furthermore, the optimi- Achenbach M (1989) A model for an alloy with shape mem-
zation results for the single-tendon case reveal that the ory. International Journal of Plasticity 5: 371–395.
optimal design does not change for the SC and PC Auricchio F, Taylor R and Lubliner J (1997) Shape-memory
model; only the predicted radius of curvature changes. alloys: macromodelling and numerical simulations of the
superelastic behavior. Computer Methods in Applied Liang C and Rogers C (1990) One-dimensional constitutive
Mechanics and Engineering 146: 281–312. relations for shape memory materials. Journal of Intelligent
Bazaraa M, Sherali H and Shetty C (2006) Nonlinear Pro- Material Systems and Structures 1: 207–234.
gramming: Theory and Algorithms. Hoboken, NJ: Wiley- Lu T, Hutchinson J and Evans A (2001) Optimal design of a
Interscience. flexural actuator. Journal of the Mechanics and Physics of
Bertsimas D and Tsitsiklis J (1997) Introduction to Linear Solids 49: 2071–2093.
Optimization. Belmont, CA: Athena Scientific. Masuda A and Noori M (2002) Optimization of hysteretic
Bloss P, Rothe W, Wünsche P, et al. (2003) Investigations of characteristics of damping devices based on pseudoelastic
the pushability behavior of cardiovascular angiographic shape memory alloys. International Journal of Non-Linear
catheters. Bio-Medical Materials and Engineering 13: 327– Mechanics 37: 1375–1386.
343. Matthews R (1993) The use of genetic algorithms in cryptana-
Brinson L (1993) One-dimensional constitutive behavior of lysis. Cryptologia 17(2): 187–201.
shape memory alloys: thermomechanical derivation with Moallem M (2003) Deflection control of a flexible beam using
non-constant material functions and redefined martensite shape memory alloy actuators. Smart Materials and Struc-
internal variable. Journal of Intelligent Material Systems tures 12: 1023–1027.
and Structures 4: 229–242. Muller I and Seelecke S (2001) Thermodynamic aspects of
Calkins H, Bruguda J, Packer DL, et al. (2007) Expert con- shape memory alloys. Mathematical and Computer Model-
sensus statement on catheter and surgical ablation of atrial ling 34: 1307–1355.
fibrillation: recommendations for personnel, policy, proce- Muller I and Wilmanski K (1980) A model for a pseudoelastic
dures, and follow-up. Europace 9: 335–379. body. Nuovo Cimento della Societa Italiana di Fisica B 57:
Crews J (2011) Development of a shape memory alloy actuated 283–318.
robotic catheter for endocardial ablaton: modeling, design Paulsey M and Seelecke S (2008) Multifunctional SMA-based
optimization, and control. Doctoral Dissertation, North smart inhaler system for improved aerosol drug delivery—
Carolina State University, Raleigh, NC. design and fabrication. In: Proceedings of the SPIE confer-
Deurig T, Pelton A and Stockel D (1999) An overview of niti- ence on active and passive smart structures and integrated
nol medical applications. Materials Science and Engineer- systems, San Diego, CA, 10 March 2008, vol. 6928,
ing A 273–275: 149–160. p.69280T.
Dumont G and Kuhl C (2005) Finite element simulation for Peirs J, Reynaerts D and Van Brussel H (1998) Design of a
design optimisation of shape memory alloy spring actua- shape memory actuated endoscopic tip. Sensors and Actua-
tors. Engineering Computations: International Journal tors A: Physical 70: 135–140.
for Computer-Aided Engineering and Software 22(7): 835– Purser MF, Richards AL, Cook RC, et al. (2009) Evaluation
848. of a shape memory alloy reinforced annuloplasty band for
Han Y, Park C and Choi S (2003) End-point position control minimally invasive mitral valve repair. The Annals of Thor-
of a single-link arm using shape memory alloy actuators. acic Surgery 88(4): 1312–1316.
Journal of Mechanical Engineering Science 217(8): 871– Sato A, Chishima E, Yamaji Y, et al. (1984) Orientation and
882. composition dependencies of shape memory effect in
Heintze O (2004) A computationally efficient free energy model Fe–Mn–Si alloys. Acta Metallurgica 32(4): 539–547.
for shape memory alloys—experiments and theory. Doctoral Seelecke S (2002) Modeling the dynamic behavior of shape
Dissertation, North Carolina State University, Raleigh, memory alloys. Journal of Non-Linear Mechanics 37:
NC. 1363–1374.
Heintze O and Seelecke S (2005) A free energy model for the Seelecke S and Muller I (2004) Shape memory alloy actuators
inner loop behavior of pseudoelastic shape memory alloys. in smart structures: modeling and simulation. Applied
In: Proceedings of materials research society symposium, Mechanics Reviews 57(1): 23–46.
vol. 881, pp. CC1.6.1–12. Cambridge University Press. Shaw K, Nortcliffe AL, Thompson M, et al. (1999) Assessing the
Heintze O, Seelecke S and Buskens C (2003) Modeling and performance of multiobjective genetic algorithms for optimi-
optimal control of microscale SMA actuators. In: Proceed- zation of a batch process scheduling problem. In: Proceedings
ings of the SPIE, smart structures and materials 2003: mod- of the congress on evolutionary computation, Washington, DC,
eling, signal processing, and control, San Diego, CA, 3 6 July 1999 to 09 July 1999, vol. 1, pp.37–45.
March 2003, vol. 5049, pp.495–505. Shi K-S and Lee Y-J (2002) A genetic algorithm application
Huo Y (1989) A mathematical model for the hysteresis in in bankruptcy prediction modeling. Expert Systems with
shape memory alloys. Continuum Mechanics and Thermo- Applications 23(3): 321–328.
dynamics 1: 283–303. Shinjo N and Swain G (2004) Use of a shape memory alloy
Ikuta K (1990) Micro/miniature shape memory alloy actua- for the design of an oscillatory propulsion system. IEEE
tor. In: Proceedings of the IEEE international conference on Journal of Ocean Engineering 29(3): 750–755.
robotics and automation, Cincinnati, OH, 13–18 May 1990, Smith R (2005) Smart Material Systems: Model Development.
vol. 3, pp.2156–2161. Philadelphia, PA: SIAM.
Lan C-C and Yang Y-N (2009) A computational design Smith R and Braun T (2006) Efficient implementation algo-
method for a shape memory alloy actuated compliant fin- rithms for homogenized energy models. Continuum
ger. Journal of Mechanical Design 131: 21009-1–21009-9. Mechanics and Thermodynamics 18(3): 137–155.
Li Q (2006) Modeling and finite element analysis of smart Smith R and Massad J (2005) A homogenized free energy
materials. Doctoral Dissertation, North Carolina State model for hysteresis in thin-film shape memory alloys.
University, Raleigh, NC. Thin Solid Films 489(1–2): 266–290.
Srinivas M and Patnaik L (1994) Genetic algorithms: a Veeramani A, Crews J and Buckner G (2008b) Design and
survey. Computer 27(6): 17–26. control of a shape memory alloy actuated robotic catheter.
Vanderplaats G (2007) Multidiscipline Design Optimization. In: Proceedings of the ASME 2008 dynamic systems and
Monterey, CA: Vanderplaats Research and Development, control conference, Ann Arbor, MI, 20–22 October 2008,
Inc. pp.1131–1138.
Veeramani A (2009) A transformative tool for minimally inva- Yang K and Gu C (2002) A novel robot hand with embedded
sive procedures: design, modeling and real-time control of a shape memory alloy actuators. Proceedings of the Institu-
polycrystalline shape memory alloy actuated robotic cathe- tion of Mechanical Engineers, Part C: Journal of Mechani-
ter. Doctoral Dissertation, North Carolina State Univer- cal Engineering Science 216(7): 737–745.
sity, Raleigh, NC. Zhou G and Lloyd P (2009) Design, manufacture and evalua-
Veeramani A, Buckner GD, Owen SB, et al. (2008a) Model- tion of bending behaviour of composite beams embedded
ing the dynamic behavior of a shape memory alloy actu- with SMA wires. Composites Science and Technology
ated catheter. Smart Materials and Structures 19(1): 1–14. 69(13): 2034–2041.