You are on page 1of 6

2014 American Control Conference (ACC)

June 4-6, 2014. Portland, Oregon, USA

Optimal Design of a Transfemoral Prosthesis with


Energy Storage and Regeneration
Rick Rarick, Hanz Richter, Antonie van den Bogert,
Dan Simon, Holly Warner, Taylor Barto

Abstract – We describe the preliminary optimal design of an Energy regeneration is a phrase used to describe the
electromechanical above-knee active prosthesis with energy storage and reuse of energy in a system. Batteries have been
storage and regeneration. A DC motor-generator applies a the traditional means for energy storage since regeneration
positive or negative torque to the knee. The control system became common in electric vehicles. When batteries and
regulates the exchange of energy between the motor-generator associated electronics are used, storage efficiencies are
and a supercapacitor. The central idea of the design is typically close to 20% [10]. Manufacturers still implement
motivated by the mechanics, energy management, and sensor- regeneration despite this low efficiency. This is due to the
based control that constitute human movement. We use large levels of kinetic energy involved in electric vehicles.
biogeography-based optimization, which is an evolutionary
algorithm, to optimize the system parameters, and we evaluate Despite improvements in battery energy density and
its performance with Simulink® models. We optimize three internal resistance, regeneration based on batteries alone a
alternative prosthesis designs. Simulation results indicate that less practical value as power levels decrease. The rate at
the prosthesis can be optimized to achieve knee angle tracking which mechanical energy must be transferred for effective
with an RMS error on the order of 0.2 degrees. deceleration is higher than the rate at which batteries can
I. INTRODUCTION store energy; that is, batteries have insufficient power storage
densities. Power conversion is traditionally handled by
Transfemoral (above-knee) amputees consume up to 50% electronics that dissipate energy as heat. At small power
more metabolic energy than able-bodied individuals [1], levels, the net energy that can be stored may be too small to
despite the fact that amputees have fewer muscles. This justify the complexity and cost of regeneration.
implies large differences in function between prosthetic and
natural limbs, requiring human adaptation that is On the other hand, supercapacitors have power transfer
energetically costly. Also, mechanical stress is increased in characteristics opposite that of batteries: they store small
amputees, leading to overuse injuries and osteoarthritis [2]. amounts of energy at fast rates. They have small energy
densities but large power densities. Also, supercapacitors are
Microprocessor knees such as the C-Leg and Rheo Knee
light weight, low cost, and have an almost indefinite lifespan.
reduce energy cost by less than 5% compared to passive
Energy regeneration in our research is inspired by
knees [3]. We suggest that these modern knees do not solve
regeneration involving batteries and supercapacitors in
the energy problem because they are essentially controlled
electric vehicles, where regeneration efficiencies greater than
dampers that cannot generate positive work. Also, these
80% have been shown [11], [12].
knees are typically used with a spring-like ankle that does not
provide active push-off. Fully active knees and ankles with The few reported attempts at regeneration in a
torque motors are becoming available [4], [5], [6], but such transfemoral prosthesis were confined to traditional battery
devices consume substantial amounts of power, requiring a and pulse-width converter configuration [13], [14]. Small
trade-off between battery mass and operating time. efficiencies (18% for the former, 30% for the latter) were
achieved in practice. Based on the much higher efficiencies
A solution to the energy problem is suggested by
obtained in the automobile industry using supercapacitors, we
considering power generation and absorption during
have applied biogeography-based optimization (BBO), a type
walking [7], [8]. The knee and ankle both have phases of
of evolutionary algorithm, to the design of a prosthesis which
both positive and negative work during gait. During level
implements energy regeneration using a supercapacitor.
walking and slow running, the knee absorbs an amount of
energy greater than that required by the ankle [9]. Therefore, The goal of our research is to use BBO to optimize design
a well-designed integrated prosthetic limb should be able to parameters to achieve knee angle tracking (compared to able-
operate without external energy, if it has the ability to bodied reference data) when the knee torque is applied to the
redistribute energy and store energy during periods of generator to charge the capacitor; that is, when the knee
negative work for later use. performs negative work. We describe three approaches to this
design goal: (1) the knee torque is coupled to the generator
via a gearset, and the generator is coupled to the capacitor via
an ideal DC transformer; (2) the DC transformer is replaced
This work was supported by Grant 0826124 in the CMMI Division of
the Engineering Directorate of the National Science Foundation, and by the by an H-bridge voltage source converter; (3) the gearset is
Wright Center for Sensor Systems Engineering through the state of Ohio. replaced by a ball screw and ball nut coupled to a linkage.
The authors are with Cleveland State University, Cleveland, OH 44115, Note that other design approaches are possible and may be
USA. Email addresses [r.rarick, h.richter, a.vandenbogert, d.j.simon, considered in future work, but our work in this paper is
h.warner, t.a.barto]@csuohio.edu. Rick Rarick is the corresponding author. limited to these three designs.

978-1-4799-3271-9/$31.00 ©2014 AACC 4108


II. PROBLEM STATEMENT substituted for the ideal DC transformer in Figure 3 to
Figure 1 shows human knee angle data obtained from human provide a simulation that is similar to the electronics that will
motion studies [15]. The goal of the prosthesis design is to be used in a hardware prototype. The gearset driving the
replicate this knee angle trajectory as closely as possible. generator remains the same as in Figure 3.
Figure 2 shows the profile of the knee power for the same set
of data. BBO was used to optimize design parameters to
minimize the RMS tracking error of the knee angle, but only
in the K1 generating region. In the following subsections, we
summarize the three prosthesis designs mentioned above.

Figure 3: Electromechanical prosthetic system schematic. The input is


Figure 1: Human knee angle data for one stride (fast walk). The heel- the knee angle torque M K which we obtained from clinical data.
strike (HS) and toe-off (TO) points are indicated, as well as the flexion
(F) and extension (E) regions. JG : gear train inertia (kg-m 2 )
300 JM : motor train inertia (kg-m 2 )
TS : spring torque (N-m)
200
TA : armature shaft torque (N-m)
Knee Power (W)

100 e: DC machine back-emf (V)


0 a: motor constant (N-m/A)
RA : armature resistance (Ω)
-100
C: capacitance (F)
-200 LA : armature inductance (H)
-300
K0 K1 K2 K3 K4 K5 K6 K7 K8 iA : armature current (A)
0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 iC : capacitor current (A)
Time (s)
Figure 2: Knee Power – The negative power regions (K1, K3, K5, K7) u: ideal transformer ratio
are regions where power is delivered to the generator (generating
mode). The positive regions are those in which power is produced by Figure 4: The parameters in the prosthetic system of Figure 3
the motor (motoring mode).
 M K − TS − nTA = ( J G + n 2 J M ) φK
A. Ideal DC Transformer with gearset transmission. Mechanical: 
The first design that we consider uses a simplified ideal  TS = KφK
DC transformer (power converter) as an initial feasibility  e − RAi A − LA ( di A / dt ) − vT =
0
study for our research. A schematic for this design is shown Electrical: 
in Figure 3, the definitions of the physical parameters are  iC = C ( diC / dt )
listed in Figure 4, and the dynamic equations in Figure 5.  v = uvC
DC Transformer:  T
A torsion spring connected to the shaft stores or releases  iC = ui A
mechanical energy. The output shaft of the transmission
applies a torque TA to the armature of the generator, and the  e = anφK
Electromechanical: 
generator produces a terminal voltage vT across the input of TA = ai A
the transformer, which causes the capacitor to charge. BBO Figure 5: The dynamic equations for the system in Figure 3. BBO is
used to optimize n, C, K, u, and q c0 (the initial charge of the capacitor.)
optimizes n, C, K, u, and q c0 (initial charge). We assume u is
constant for simplicity, but in practice there is no need for
The H-bridge operates in four modes: motoring
such an assumption; future work will consider time-varying
plus/minus and generating plus/minus, where the plus and
transformer ratios. Note that initial charge at the beginning of
minus indicate the direction of machine rotation. When the
each stride can be set by an auxiliary controller
bridge is operating in the motoring mode, a pulse-width
B. H-bridge voltage source converter modulation (PWM) signal switches the bridge so that either
We now replace the ideal DC transformer of Figure 3 the motor current is flowing from the capacitor into the motor
with a more realistic H-bridge voltage source converter. or the motor terminals are shorted to provide damping.
Figure 6 shows a schematic for the converter design, which is
based on [14]. The H-bridge (inside the dashed block) can be

4109
(MBSN) linkage assembly is more compact, and such
designs are typically used in commercial prostheses. For
clarity, the MSBN assembly is shown displaced from its true
position; in the actual device it is inserted in place of the b-
link. The length of the b-link changes as the screw rotates. In
the generating mode, knee torque back-drives the ball screw
and hence the generator.
The dynamics of this system can be represented by a
bond graph [16] as shown in Figure 8. This figure completely
replaces Figure 3 when the screw assembly is used. The mass
of the motor is m, and l is the lead (pitch) of the screw. The
symbol TF represents a generalized transformer, which is
analogous to an electrical AC transformer. The first TF in the
bond graph represents the ball/screw/nut assembly with ratio
l. The second TF represents the DC transformer of Figure 3
Figure 6: The prosthetic system of Figure 3 with the ideal DC with ratio u. The symbol MTF represents a modulated
transformer replaced by an H-bridge voltage source converter. The transformer with variable ratio. In the bond graph MTF
gearset coupled to the DC machine is not shown here.
represents the linkage which couples the angle θ 2 to the
When the converter is operating in the generating mode, length of the b-link though the variable ratio G (b) =
the PWM signal is used to control the voltage across the dθ 2 / db, which is nonlinear. The symbol GY represents a
capacitor by modulating the pulse width. The knee angle generalized gyrator, which is the motor/generator in the
tracking error e(t ) is input to a proportional-integral- prosthesis with generator constant a.
derivative (PID) controller which controls the duty cycle
u (t ) through the following equation:

∆u(t )= k p e(t ) + ki ∫ e(t )dt + kd e(t )

(t ) φKref − φK , φKref (t ) is the reference knee angle,


where e=
φK (t ) is the knee angle, and k p , ki , and kd are the PID gains.
If the converter is in the motoring plus/minus mode,
1) u( k ) ± ∆u( k ), where k is the discrete sample time
u( k +=
in Simulink. The formula is the same in the generating plus-
minus mode, except that the ± is replaced by  .
The H-bridge design uses the same mechanical and
electromechanical equations as the ideal DC transformer
configuration in Section II(A), except that in place of the
electrical and DC transformer equations, there are two
electrical equations that depend on which mode the converter
is in. When the prosthesis is in the generating-plus mode, one
of the following voltage loop equations is used depending on
the state of the PWM signal:
Charging: e − vLa − vRa − vD1 − vC − vQ 4 =
0 Figure 7: Assembly for coupling the knee torque to the motor-
Shorted: e − vLa − vRa − vQ 2 − vQ 4 =
0 generator which is used instead of the rotary actuator in Figure 3. The
motor, ball screw, and ball nut assembly (MBSN) is shown displaced
from its true position in place of the b-link. Dimensions c, d, e, γ, and � L
The loop equations for the generating-minus mode are are constant, and the diagram is not to scale.
determined similarly. The generating plus/minus modes
determine which MOSFETS have current flowing through
them, and the path of current flow required to maintain
generation. BBO is used to optimize n, C, K, q c0 , R Q2 , R Q4 ,
k p , k i , and k d , where R Q2 and R Q4 are the resistances of the
Figure 8: A bond graph of the prosthetic system using the MBSN
transistors. linkage assembly.
C. Ball screw and nut coupled to a linkage mechanism.
Knee torque M K enters the prosthetic system at the left
An alternative design to the rotary actuator (gearset and
DC machine) in Figure 3 is the assembly shown in Figure 7. of Figure 8 and is converted via the first transformer (the
The rotary actuator is replaced by a motor, ball screw, ball nonlinear linkage) to a force which is applied to the second
nut, and connecting linkage, which might be more easily transformer (the ball screw). The force applied to the ball
integrated in a prosthetic leg. A motor/ball/screw/nut screw is converted to a generator torque which supplies

4110
power to the electrical system. The electrical system includes suffices to assume a linear relationship between an island's
resistance RA , the DC transformer, and capacitor C. HSI and its migration rates, and to assume that these rates are
the same for all islands under consideration (the BBO
The parameters in Figure 8 are either physical constants population).
or known inputs except for G, which is a varying transformer
ratio governed by the crank-slider geometry. To calculate G, In order to apply BBO to an optimization problem, the n-
a kinematic analysis of the mechanism was performed by tuple ( SIV1 ,  , SIVn ) associated with the features of an
defining the associated loop equations and a geometric island is viewed as a candidate solution to the optimization
constraint equation to yield the following system of problem. In other words, the set of all such n-tuples is the
equations: search space from which an optimal solution will be
determined. The value of HSI for a particular island is viewed
e cos θ 2 − b cos θ 3 − c cos θ 4 − d = 0
as the objective function value associated with that candidate
e sin θ 2 −  sin
b θ 3 −  sin
c θ4 =
0 solution. The goal of BBO, then, is to determine the solutions
θ3 − θ4 − γ =0 which maximize HSI over the entire search space.
We use the migration rates of each candidate solution to
Solving these three equations for 𝜃2 in terms of 𝑏 and
probabilistically share features between islands. For each SIV
differentiating with respect to 𝑏 yields an equation expressing
(feature) in each island (candidate solution), we
the transformer ratio G between the rotational motion of the
probabilistically decide whether or not to immigrate. If
knee and the linear extension of the ball screw.
immigration is selected for a given SIV, then the emigrating
1 island is selected probabilistically using roulette wheel
 b + c cos (γ )  2
selection [19] based on µ. After migration, mutation is
dθ 2  
G = 
= 
d −
db  e  1
( 2 2
−e + b + 2b   cos
c ( γ )+c −d ) 
2 2 2 

applied to increase diversity in the population. Mutation is
probabilistically applied to each feature in each island for
  4 e 2 d 2 
every generation. Figure 9 summarizes this algorithm. We
  
use yk ( s ) to denote the s-th feature of the k-th island in
The knee angle 𝜑𝐾 in Figure 7 is directly related to θ 2 population y. Migration and mutation of the entire population
through a constant design angle 𝜑𝐿 , defined by the geometry take place before any of the candidate solutions are replaced,
of the link which attaches to the thigh. This relationship is which requires the use of temporary population z in Figure 9.
described by the equation ϕ K =π − ϕ L − θ 2 . BBO is used to
Initialize a population { yk } of candidate solutions
optimize C, K, u, e, c, d, γ , and 𝜑𝐿 for the ball screw design. While not (termination criterion)
{zk } ← { y k }
III. BIOGEOGRAPHY-BASED OPTIMIZATION For each island zk
For each SIV s
Biogeography-based optimization is an evolutionary Use λ k to decide whether to immigrate to zk ( s )
algorithm that is based on the mathematics of biogeography – If immigrating to zk ( s ) then
the study of the migration, extinction, and distribution of Use µ to select the emigrating island y j
species [17], [18]. In BBO, candidate problem solutions are zk ( s ) ← y j ( s )
represented as islands (also called habitats), and the sharing End if
of features (independent variables) between candidate Probabilistically decide whether to mutate zk ( s )
solutions is represented as migration between islands. Next SIV : s ← s + 1
Next island: k ← k + 1
Species of plants and animals migrate between islands by { y k } ← {zk }
Next generation
various means, such as being carried along by driftwood,
fish, birds, and the wind. Over long periods of time, some Figure 9: Biogeography-based optimization algorithm.
islands tend to accumulate more species than others because
they possess certain features that are more suitable. This The best individuals from each generation are called
ability to sustain larger numbers of species can be associated elites, and a certain number of them are saved and inserted
with a fitness measure that we can quantify with a habitat into the population in the next generation. The number of
suitability index (HSI). The value of HSI depends on many elites is a BBO tuning parameter, as are the mutation rate, the
features of the island. If a value is assigned to each feature, size of the population, and the number of generations.
then the HSI is a function of these values. Each of these We calculate the cost of the i-th candidate prosthesis
values is represented by a suitability index variable (SIV). design in the BBO population with the following equation:
These mappings are summarized as follows:
2

Island  ( feature1 ,  ,featuren )  ( SIV1 ,  , SIVn )  HSI


( BBO cost )i = ∫K 1 φKref (t ) − φKi (t )  dt

An island with a large number of species (a large HSI) φKi (t ) is the knee angle trajectory that results from simulating
has an abundance of species which can emigrate to other BBO’s i-th candidate prosthesis design. We re-order the N
islands, so its rate of emigration, µ , is correspondingly large. prosthesis designs from highest cost ( i = 1) to lowest cost
The island is also less likely to be able to sustain further (i = N ), and calculate the migration rate of each design from
immigration because of the growing demand on its resources,
so its immigration rate, λ , is small. For many applications it the design’s HSI. The independent variable set optimized by

4111
BBO is different for each of the designs, as discussed in the that the H-bridge design starts with the highest cost at the
following section. 0-th generation. This indicates that a realistic converter
results in a system that is more difficult to optimize. Figure
IV. SIMULATIONS AND RESULTS 13 also shows that the ball screw design took the longest time
The BBO algorithm was applied to the three design to converge, which is another indication that the ball screw
variations discussed in Section II. The physical constants design is more difficult to optimize.
used for the simulations (Figure 10) are taken from a
commercially available Maxon motor [20] except for the ball
screw lead which was chosen to be a typical value.
a = 0.054 (N-m) motor constant
=RA 0.0821 (Ω) armature resistance
LA = 3.08 × 10−5 (H) armature inductance
J m = 1.29 × 10−4 (kg-m 2 ) motor train inertia
J g = 0.088 × 10−4 (kg-m 2 ) gear train inertia
m = 2.1 (kg) mass of motor
Figure 11: Tracking results for the ideal DC transformer design (A) in
l = 2.0 (mm/rev) screw lead the K1 generating region. The plots for the other designs are similar.

Figure 10: Physical constants used for the simulations.

All of the BBO simulations used a mutation probability of


0.02, and two elites. The BBO population sizes are 50 for the
DC transformer design, 25 for the H-bridge design, and 50
for the ball screw design. A summary of the BBO results is
given in Table 1.
Table 1: BBO Simulation Results Figure 12: The capacitor charge is increasing during the K1 generating
region for the ball screw design (C) indicating energy storage.
Electrical Mechanical Optimized Parameter
Design Design Parameter Value
A Ideal DC Rotary Gearset n 72
Transformer C 193 (F)
K 51 (N-m)
(RMS Tracking U 68
Error = 0.14 deg) q C0 55 (C)
B H-Bridge Voltage Rotary Gearset n 98
Source Converter C 409 (F)
K 1.0 (N-m)
(RMS Tracking q C0 35 (C)
Error = 0.19 deg) R Q2 1.0 (Ω)
R Q4 7.0 (Ω)
kp 662
ki 934
kd 349
C C. Ideal DC Ball Screw & C 6.3 (F)
Transformer Linkage K 20.1 (N-m)
u 198
(RMS Tracking e 0.4139 (m)
Error = 0.30 deg) c 0.4712 (m)
d 0.4080 (m)
γ 2.782
φL 2.088
Figure 13: BBO cost results for the three designs in the K1 region.
Table 1 shows that all three designs provided good knee
angle tracking, but the ball screw design resulted in an RMS V. CONCLUSION AND FUTURE WORK
error that was twice as large as the other two designs. The We have proposed and simulated an electromechanical,
nonlinearity of the ball screw system might make it more regenerative, above-knee prosthesis. We have considered
difficult to optimize. three approaches to the prosthesis design. The first approach
The tracking results for the ideal DC transformer design uses a rotary gear to drive the knee joint, and an ideal DC
(A) are displayed in Figure 11. The tracking plots for the transformer to charge a super-capacitor; this approach
other designs are similar. Energy storage during the K1 includes five design parameters. The second approach uses a
generating region for the ball screw design (C) is indicated by rotary gear and a realistic voltage source converter to charge
the increasing capacitor charge in Figure 12. The the super-capacitor; this approach includes nine design
convergence of BBO is shown in Figure 13, which shows parameters. The third approach uses a ball screw and linkage

4112
to drive the knee joint, and an ideal DC transformer; this REFERENCES
approach includes eight design parameters. In each case we [1] A. Gitter, J. Czerniecki, and K. Weaver, “A Reassessment of Center-
used biogeography-based optimization (BBO) to optimize the of-Mass Dynamics as a Determinate of the Metabolic Inefficiency of
system design for the first power generation region of the Above-Knee Amputee Ambulation,” American Journal of Physical
stride, which occurs near the beginning of stance phase. BBO Medicine and Rehabilitation, vol. 74, pp. 332–338, 1995.
successfully optimizes the prosthesis parameters to provide [2] P. Struyf, C. van Heugten, M. Hitters, and R. Smeets, “The Prevalence
of Osteoarthritis of the Intact Hip and Knee among Traumatic Leg
RMS tracking errors on the order of 0.2 degrees. Amputees,” Archives of Physical Medicine and Rehabilitation,
This research has opened the way to many possibilities vol. 90, pp. 440–446, 2009.
[3] J. Johansson, D. Sherrill, P. Riley, P. Bonato, and H. Herr, “A clinical
for future work. First, we need to optimize the system design comparison of variable-damping and mechanically passive prosthetic
over the remaining regions of gait. Some system parameters knee devices,” American Journal of Physical Medicine and
are constant over the entire stride, but other parameters could Rehabilitation, vol. 84, pp. 563–575, 2005.
vary from one region to the next. For instance, the spring [4] M. Highsmith, J. Kahle, D. Bongiorni, B. Sutton, S. Groer, and K.
constant cannot change during the amputee’s stride, but we Kaufman, “Safety, energy efficiency, and cost efficacy of the C-Leg
for transfemoral amputees: A review of the literature,” Prosthetics and
could vary the total capacitance during the amputee’s stride Orthotics International, vol. 34, pp. 362–377, 2010.
by switching capacitors in and out of the circuit. [5] F. Sup, H. Varol, J. Mitchell, T. Withrow, and M. Goldfarb, “Design
and Control of an Active Electrical Knee and Ankle Prosthesis,” IEEE
Another direction for future research is to design the International Conference on Biomedical Robotics and
electronics and the control algorithms for the motoring Biomechatronics, Scottsdale, AZ, pp. 523–528, 2008.
regions of the stride. After that task is done, the system will [6] M. Eilenberg, H. Geyer, and H. Herr, “Control of a Powered Ankle–
need to be optimized again for the generating regions because Foot Prosthesis Based on a Neuromuscular Model,” IEEE
the system’s operation during the motoring regions affects Transactions on Neural Systems and Rehabilitation Engineering,
vol. 18, pp. 164–173, 2010.
the initial conditions of the generating regions, and vice [7] D. Winter, “Energy generation and absorption at the ankle and knee
versa. during fast, natural, and slow cadences,” Clinical Orthopedics and
Related Research, vol. 175, pp. 147–154, 1983.
Another direction for future research is to optimize the [8] D. Winter, Biomechanics and Motor Control of Human Movement
prosthesis design using multi-objective optimization (MOO) (4th edition), John Wiley & Sons, 2009.
algorithms. The multiple objectives could include not only [9] D. J. Farris and G. S. Sawicki, “The mechanics and energetics of
knee angle tracking, but also objectives such as energy human walking and running: a joint level perspective,” Journal of the
efficiency, system cost, system complexity, compatibility Royal Society Interface, vol. 9, no. 66, pp. 110–118, 2011.
[10] J. Moreno, M. Ortuzar, and J. Dixon, “Energy-Management System
with off-the-shelf components, robustness, and others. Many for a Hybrid Electric Vehicle, Using Ultracapacitors and Neural
evolutionary algorithms, including BBO, have already been Networks,” IEEE Transactions on Industrial Electronics, vol. 53,
applied to MOO [18], so the extension of our research to pp. 614–623, 2006.
MOO is a natural direction for future work. [11] J. Dixon and M. Ortuzar, “Ultracapacitors + DC-DC converters in
regenerative braking system,” IEEE Aerospace and Electronic
Another direction for future research is to add Systems Magazine, vol. 17, pp. 16–21, 2002.
components to the model in this paper to make it more [12] P. Weissler, “Mazda Introduces Supercapacitor-Type Regenerative
flexible and realistic. These components include one or more Braking,” Automotive Engineering
battery storage elements, switched capacitor banks, and Online, http://www.sae.org/mags/aei/11845. Accessed March 10,
2014.
realistic feedback sensors. After completing system
[13] M. Tucker and K. Fite, “Mechanical Damping with Electrical
optimization for the prosthetic knee, our next research goal Regeneration for a Powered Transfemoral Prosthesis,” IEEE/ASME
will be to integrate the prosthesis with human kinetic and International Conference on Advanced Intelligent Mechatronics,
kinematic models, and repeat the system optimization process Montreal, Canada, pp. 13–18, 2010.
with those additional constraints. The next goal will be to [14] K. Tabor, The Real-Time Digital Control of a Regenerative Above-
include an electromechanical model for a prosthetic ankle. Knee Prosthesis, MSc. Thesis, Electrical Engineering and Computer
Since the knee is a net absorber of energy and the ankle is a Science, Massachusetts Institute of Technology, 1988.
net consumer of energy [7], our integrated knee/ankle [15] A. van den Bogert, S. Samorezov, B. Davis, and W. Smith,
“Modeling and Optimal Control of an Energy-Storing Prosthetic
prosthesis will be designed to achieve net energy transfer Knee,” Journal of Biomechanical Engineering, vol. 134, no. 5,
from the knee to the ankle while emulating able-bodied gait pp. 0510071–0510078, 2012.
kinematics. [16] D. Karnopp, D. Margolis, and R. Rosenberg, System Dynamics, 5th
edition, Wiley, 2012.
Finally, the prosthesis design that results from this
[17] D. Simon, “Biogeography-based optimization,” IEEE Transactions on
ongoing research will be manufactured as a prototype and Evolutionary Computation, vol. 12, no. 6, pp. 702–713, Dec. 2008.
tested with Cleveland State University’s hip robot [21] to [18] D. Simon, Evolutionary Optimization Algorithms, Wiley, 2013.
confirm its feasibility as a commercializable system. The
[19] D. Goldberg, Genetic Algorithms in Search, Optimization, and
testing will use a semi-active virtual control method with Machine Learning, Addison Wesley, 1989.
regenerative energy-storage joints [22]. The control approach [20] Datasheet, Maxon RE65 Motor, Part No.
consists of three steps. First, a virtual control design is 353295, www.maxonmotorusa.com, Accessed March 10, 2014.
conducted by any suitable means, assuming a fully actuated [21] H. Richter and D. Simon, “Robust Tracking Control of a Prosthesis
system. Second, virtual control inputs are matched by a Test Robot,” ASME Journal of Dyamic Systems, Measurements and
parameter modulation law. Third, the storage dynamics are Control, v. 136, no. 3, doi:10.1115/1.4026342, 2014.
shaped using design parameters. The storage dynamics [22] H. Richter, D. Simon, A. van den Bogert, “Semiactive Virtual Control
Method for Robots with Regenerative Energy-Storing Joints,” World
coincide with the system’s internal dynamics under exact Congress of the International Federation of Automatic Control, Cape
virtual control matching. Town, South Africa, August 2014.

4113

You might also like