Anda di halaman 1dari 4

Concurrent Multibody and Finite Element Analysis of the

Lower-Limb During Amputee Running *

Stacey M. Rigney, Anne Simmons, and Lauren Kark, Member, IEEE

therefore raises doubts over the validity and accuracy of

1. 

Abstract—Lower-limb amputee athletes use Carbon fiber

Energy Storage and Return (ESAR) prostheses during high existing amputee running research that relies upon classical
impact activities such as running. The advantage provided to rigid-body mechanics models [5].
amputee athletes due to the energy-storing properties of ESAR
prostheses is as yet uncertain. Conventional energy analysis As an alternative to rigid-body mechanics, continuum-
methods for prostheses rely upon multibody models with mechanics describes not only the rotation and translation of
articulating joints. Alternatively, Finite Element (FE) analysis solids, but also their deformation [6]; continuum-mechanics
treats bodies as a deforming continuum and can therefore is much more suitable for the dynamic analysis of ESAR
calculate the energy stored without using these rigid-body prostheses. Finite Element (FE) analysis is a form of
mechanics assumptions. This paper presents a concurrent computational modeling based on continuum-mechanics
multibody and FE model of the femur, tibia, socket and ESAR which can be driven by prescribed displacements to calculate
prosthesis of a transtibial amputee athlete during sprinting. resulting forces and moments. We propose a subject-specific
Gait analysis spatial data was used to conduct an offline multibody model that combines a rigid-body model of the
simulation of the affected leg’s stance phase in COMSOL skeletal elements of the amputated limb with a FE model of
Multiphysics. The calculated peak elastic strain energy of the the prosthetic component. Ultimately, the model proposed
prosthesis was 80J, with an overall RMSE of simulated marker here will be expanded into a full dynamic (time-dependent)
displacement of 4.19mm. This concurrent model presents a simulation with seamless integration between FE and motion
novel method for analyzing in vivo ESAR prosthesis behavior. capture software, the results of which can be compared to
conventional rigid-body mechanics methods to determine the
I. INTRODUCTION accuracy improvement. The aim of this study is to conduct a
The proportion of athletes and highly active people within preliminary quasi-static investigation of concurrent
the amputee population is increasing [1], which has resulted multibody and FE analysis to determine the energy storage
in the development of carbon-fiber Energy Storage and capacity of an ESAR prosthesis during amputee running. We
Return (ESAR) prostheses for high impact activities. These hypothesize that this new model of the lower-limb will
prostheses are designed to behave as springs, storing and provide information that would otherwise be unavailable
returning energy during the stance phase of gait. Recently it using rigid-body mechanics, and will therefore be a useful
has been asked whether lower-limb amputees sprinting with tool for future gait analysis of amputee running.
ESAR prostheses have an unfair advantage over their able-
bodied counterparts [2]. In order to ascertain whether II. METHODS
amputee athletes are indeed super-abled versus disabled, first Model development and validation encompassed two
we must develop accurate mechanical models for simulated stages: experimental and numerical analysis. Anthropometric
and experimental data acquisition. and gait analysis data was first measured in order to develop
The methods commonly used to analyze amputee gait are a subject-specific lower-limb computational model, and the
based on anatomical models of the lower-limb despite material properties of the subject’s prosthesis were calculated
inherent inaccuracies [3], namely the violation of the rigid- according to a previously described procedure [7]. The
body mechanics assumptions of conventional inverse numerical analysis utilized COMSOL Multiphysics 4.4
dynamics calculations [4]. Conversely, the material and (Stockholm, Sweden) Multi-body Dynamics (MBD) module
geometric construction of ESAR prostheses are purposely to simulate the skeletal and prosthesis components during
designed to emulate a leaf spring; the function of the stance phase of one of the experimental trials.
traditional articulating ‘ankle’ joint is replaced by the A. Experimental data collection
inherent compression of the prosthesis. This incongruence
Gait analysis was performed on a healthy 16-year-old
female right-sided transtibial amputee of mass 53.18kg and
*Research supported provided by the Commonwealth Scientific and height 171.0cm. The subject was accustomed to sprinting on
Industrial Research Organisation (CSIRO) and Australian Institute of Sport her Category 2 1E90 Sprinter by Ottobock® (Duderstadt,
(AIS) in memory of Dr Tony Collings.
S. M. Rigney is with the Graduate School of Biomedical Engineering, Germany), and volunteered for the study in accordance with
UNSW Australia, Sydney, NSW 2052 AUS (email: UNSW Australia Human Research Ethics Committee approval HC13054. For five trials, the subject sprinted at
A. Simmons is with the School of Mechanical and Manufacturing maximum speed along an indoor 100m straight running track
Engineering, UNSW Australia, Sydney, NSW 2052 AUS (email: instrumented with eight 9287B 1000Hz Kistler®
L. Kark is with the Graduate School of Biomedical Engineering, UNSW
(Winterthur, Switzerland) force plates and a MX T40-S
Australia, Sydney, NSW 2052 AUS (phone: +61 2 9385 0560, email: 250Hz Vicon® (Denver, Colorado, USA) motion capture system composed of 20 cameras. The residual limb length,
978-1-4244-9270-1/15/$31.00 ©2015 Crown 2434
procedure described previously [7,9,10]. Briefly, the ESAR
RPB1 RPL1 prosthesis was held at angles α = 12°, 22° and the free distal
RPB2 RPL2 end was compressed with constant strain-rate loading and
unloading of ż = ±0.001ms-1 up to a maximum load of 1000N
RPB3 RPL3 using a uniaxial displacement-controlled crosshead fitted
RTIB with a 5000N load cell. Vertical reaction force was sampled
at 1000Hz.
B. Numerical model
The anthropometric measurements of the right femur and
RPL6 tibia were obtained from the marker locations during the in
RPL7 vivo static trial using the subject-specific scaling function
within OpenSim 3.2 (Stanford, California, USA)
RHEE RTOE musculoskeletal modeling software [11-15]. Scaled 3D
geometry files of the femur and tibia were then exported from
Figure 1. Markers placed on the prosthesis during gait analysis.
OpenSim as a *.VTP file, converted into a *.STL file using
the open-source software ParaView 4.2 (Clifton Park, New
York, USA), converted into a CAD file (*.SLDPRT) using
SOLIDWORKS® 2014 (Waltham, Massachusetts, USA)
z’ then imported into COMSOL Multiphysics. Prosthetic
componentry was drawn in SOLIDWORKS using the
x’ y’ measurements taken of the prosthesis and socket. The socket
was approximated as a cylinder and the shoe of the prosthesis
was simplified for modeling purposes. The geometry file of
the right tibia was amputated according to the level of the
z subject’s amputation, measured during experimental gait
z’ The skeletal structures and prosthetic socket were
α modeled as rigid-bodies using the COMSOL Mutiphysics
y MBD module; a rigid-body is treated as a homogenous
continuum with zero internal stress [6]. The prosthesis was
modeled as a linearly elastic isotropic homogenous solid with
y’ Poisson’s ratio of 0.3 and Young’s Modulus of 3.5×104 MPa
z for the carbon fiber component, and Poisson’s ratio of 0.3
and Young’s Modulus of 1.0×103 MPa for the rubber shoe.
y Young’s Modulus of the carbon fiber was estimated from the
experimental in vitro mechanical testing data using inverse
Figure 2. Translation between coordinate systems and the resulant FE analysis according to the procedure described previously
angle α made by the prosthesis longitudinal axis z’ with the laboratory [7], the result of which was consistent with values provided
vertical axis z. in literature for a carbon fiber ESAR prosthesis [16]. All
inertial effects were ignored due to the quasi-static nature of
socket dimensions and prosthesis measurements were
the simulation.
recorded using Mitutoyo® (Kawasaki-shi, Kanagawa, Japan)
digital vernier calipers. Passive-reflective markers were The residual limb skeletal model was connected to the
placed on both sides of the body according to the full-body socket and prosthetic foot models via fixed attachments with
Plug-in-gait marker set [8], where locations on the prosthetic zero degrees of freedom, whilst the femur and tibia were
limb were approximated from the height of the anatomical connected via a one-degree of freedom (1DOF) hinge joint
locations. Ten additional markers were placed on the about the global x-axis to simulate the knee joint. A 1DOF
prosthesis – three along its longitudinal axis in the sagittal rigid-body prescribed angular displacement θx (in the sagittal
plane (right prosthesis lateral 1-3, RPL1-3), and three along plane only) was applied to the femur at the location
its longitudinal axis in the frontal plane (right prosthesis back corresponding to the RTHI marker, calculated using the
1-3, RPB1-3), and four markers along its lateral profile experimental translational displacement of markers RTHI and
(RPL4-7) (Figure 1). RKNE. A 3DOF translational (x,y,z) prescribed displacement
was applied at the location of the RPL1 marker via a rigid
Three-dimensional marker position data was exported
connector at the boundary on the prosthesis domain
from Vicon Nexus® 1.8 in order to determine the
corresponding to the socket attachment site. Similarly, a
translational and rotational displacement of the lower-limb
rigid connector was also applied at the distal boundary of the
anatomical landmarks as well as the range of angles α that the
prosthetic shoe to prescribe a 3DOF translational
longitudinal axis of the prosthesis z’ makes with the
displacement at the RTOE marker. The domains were free to
laboratory vertical axis z during stance (Figure 2). These
move in all other degrees of freedom.
angles were used as boundary conditions for mechanical
testing of the prosthesis, conducted in an Instron® (Norwood, A tetrahedral discretization scheme was utilized for all
Massachusetts, USA) testing machine according to the domains, with a minimum element size of 0.0055m for the
prosthesis geometry as determined via a mesh convergence
study, and 0.143m for the remaining rigid bodies; a minimum 1800 90

Ground Reaction Force (N)

of one element is required for accurate simulation of rigid

Elastic Strain Energy (J)

bodies within COMSOL Multiphysics MBD. A quasi-static 1200 60
parametric solver with intervals of Δt = 0.004s was used to
solve the stance phase of the amputated leg, with initial 900
contact at t = 0.000s and toe-off at t = 0.150s. 600 30

Validation of the in vivo multibody model was achieved 300

through two methods: ground reaction force and marker data. 0 0
The simulated ground reaction force was calculated via a -300
surface integration of the stress along the distal boundary of
-600 -30
the prosthetic shoe, and then compared to the experimental 0 20 40 60 80 100
ground reaction force data. Likewise the displacement of
Stance Phase (%)
points corresponding to markers RPL4, RPL5, RPL6, RPL7,
RANK and RHEE was calculated and compared to the Exp RFx Exp RFy Exp RFz FEA RFx
experimental results. As the experimental markers had a FEA RFy FEA RFz FEA Energy
diameter of 10mm yet were modeled numerically as a single Figure 3. Experimental and simulated ground reaction force
point, the validation acceptance criterion was a RMSE of less
than 10mm. Once validated, the total elastic strain energy MPa
was calculated for each parametric quasi-static time step,
which is the total work done to produce the elastic strain in
the material.

The simulated ground reaction force followed the same
pattern as the experimental ground reaction force data,
reaching a peak vertical reaction force of approximately
1700N at t = 0.077s or at 52% of stance phase (Figure 3). The
experimental ground reaction force presented an oscillating
response during initial contact to mid-stance; the quasi-static
simulation was not able to mimic these results. From mi-
stance onwards, the simulated vertical reaction force showed
good agreement with the experimental results. The
longitudinal reaction force, which presents primarily as Figure 4. Von mises stress during mid-stance at t = 0.02s.
friction between the ground and the shoe, changed from
posterior to anterior 0.018s or 12% later in the simulation in
comparison to the experiment. During late stance, the RMSE (mm) x-axis y-axis z-axis Mean
simulated longitudinal reaction force reached a peak that was RPL1 0.00 0.12 0.08 0.07
110N lower than the experimental value. The lateral ground RPL4 2.47 6.51 2.81 3.93
reaction force was over-predicted by approximately 60N RPL5 5.57 6.32 3.60 5.16
RPL6 5.27 8.58 4.59 6.15
during the simulation. The elastic strain energy peaked at 80J RPL7 5.13 8.51 3.29 5.64
at mid-stance, and followed a similar trajectory to the vertical RANK 3.96 6.49 3.25 4.57
reaction force (Figure 3). The peak stress occurred in the RHEE 5.13 4.33 4.93 4.80
prosthesis at the sharpest radii (Figure 4), and reached a RTOE 0.12 7.33 2.09 3.18
maximum of 3.83×102 MPa at t = 0.077s. The femur, tibia Mean 3.46 6.02 3.08 4.19
and socket showed zero stress due to their rigid nature. The
simulated marker displacements showed good agreement which is a reasonable value in comparison to previously
with the experimental data (Figure 5), with a RMSE less than calculated values for walking self-selected and fast paces
the validation criteria of 10mm for all markers and an overall where the peak strain energy were 15J and 18J respectively
RMSE of 4.19mm (Table I). [16]. Stored strain energy is an important characteristic as
prosthesis overall efficiency is defined by the ratio of this
IV. DISCUSSION stored strain energy to dissipated energy. The model
therefore presents an improvement on the current practice of
A concurrent multibody and FE model of the femur, tibia, calculating energy storage from in vitro mechanical testing
socket and prosthetic foot of a transtibial amputee sprinter only, as this type of testing has been shown to be in variance
was simulated using experimental spatial data. This to the true in vivo behavior [9]. Additionally, combining FE
preliminary study is the first to conduct FE analysis of an with a multibody model of the skeletal elements provided a
ESAR prosthesis during the stance phase of running. The simulation more indicative of the true experimental behavior
numerical analysis allowed calculation of variables not in comparison to traditional direct dynamics FE methods,
available using conventional gait analysis alone, such as the during which the experimental ground reaction forces is
strain energy stored in the prosthesis. In this study, the stored applied as a boundary condition to the FE model [17]. This
strain energy peaked during mid-stance with a value of 80J,

Marker Displacement (mm)

(a) 400 (b) (c)
500 300 500 500
Marker Displacement (mm)

Marker Displacement (mm)

400 400 400
300 0 300 300

Axis Title
-100 0 0.04 0.08 0.12
200 200 200
Time (s)
100 100 100

0 0 0
0 20 40 60 80 100 0 20 40 60 80 100 0 20 40 60 80 100
-100 -100 -100
Stance Phase (%) Stance Phase (%) Stance Phase (%)

Figure 5. Experimental and simulated marker displacement in the laboratory (a) lateral or x-axis, (b) longitudinal or y-axis, and (c) vertical or z-axis.

preliminary quasi-static study demonstrates the potential of [2] G. P. Bruggeman, A. Arampatzis, F. Emrich and W. Potthast,
subject-specific multibody models for amputee gait analysis. “Biomechanics of double transtibial amputee sprinting using dedicated
sprint prostheses,” Sports Technol., vol. 4-5, no. 1, pp. 220-227, 2008.
A number of areas have been identified where the model [3] A.B. Sawers and M.E. Hahn, “The potential for error with use of
could be further refined. The simulated reaction force did not inverse dynamic calculations in gait analysis of individuals with lower
exhibit the oscillations of the experiments due to the quasi- limb loss: A review of model selection and assumptions,” J. Prosthet.
Orthot., vol. 22, no. 1, pp. 56-61, 2010.
static nature of the simulation. The use of a time-dependent
[4] D.A. Winter, Biomechanics and motor control of human movement.
solver that includes inertial effects may improve the accuracy Hoboken, New Jersey: John Wiley & Sons, Inc., 2009.
of the oscillatory phenomena at initial contact. The model [5] M. D. Geil, M. Parnianpour, P. Quesada, N. Berme and S. Simon,
utilized a 1DOF prescribed displacement of the femur and “Comparison of methods for the calculation of energy storage and
1DOF hinge joint at the knee; this simplification considerably return in a dynamic elastic response prosthesis,” J. Biomech., vol. 33,
reduced computational effort, however this affected the no. 12, pp. 1745-1750, 2000.
accuracy of the simulated prosthesis motion in the lateral [6] C. A. Truesdell, Essays in the history of mechanics. New York:
Springer, 1968.
direction (x-axis). The reaction force and moment in this
[7] S. M. Rigney, A. Simmons and L. Kark, “Finite element analysis of a
plane of movement is very low, therefore it is unlikely to be lower-limb running-specific prosthesis,” in 8th Aust. Cong. App.
of great consequence when calculating joint torque and Mech., Barton, A.C.T, Australia, 2014, pp. 297-305.
power in future time-dependent studies. The shape of the [8] M.P. Kadaba, H. Ramakrishnan and M. Wootten, “Measurement of
longitudinal (y-axis) reaction force was offset from the lower extremity kinematics during level walking,” J. Orthop. Res.,
experimental results in the time domain, which is likely to be vol. 8, no. 3, pp. 383-392, 1990.
improved by refining the shoe geometry. Finally, improved [9] S. M. Rigney, A. Simmons and L. Kark, “Stance phase mechanical
characterisation of a running-specific lower-limb prosthesis,” in Proc.
accuracy could be obtained by modeling the carbon fiber as a 2014 Aust. Biomed. Eng. Conf., Canberra, A.C.T., Australia, 2014, to
heterogeneous orthotropic material, as its very nature as a be published.
composite is to have varying stiffness depending on [10] S. M. Rigney, A. Simmons and L. Kark, “Mechanical Efficiency of a
thickness, composition and layering. Running-Specific Energy Storage and Return Prosthesis,” in Bk
Abstracts 9th Aust. Biomech. Conf., Wollongong, NSW, Australia,
Extension of this project will include the addition of 2014, pp. 43.
viscoelastic damping parameters to enable a time-dependent [11] S. L. Delp, F. C. Anderson, A. S. Arnold, J. P. Loan, A. Habib, C. T.
simulation, which will result in the ability to calculate: joint John, E. Guendelman and D. G. Thelen, “OpenSim: open-source
torque and power; energy absorbed, dissipated and returned software to create and analyze dynamic simulations of movement,”
IEEE Trans. Biomed. Eng., vol. 54. No. 11, pp. 1940-1950, 2007.
by the prosthesis; prosthesis work and efficiency; and the
[12] S. L. Delp, J. P. Loan, M. G. Hoy, F. E. Zajac, E. L. Topp and J. M.
comparison of these values with results obtained through Rosen, “An interactive graphics-based model of the lower extremity to
conventional link-segment rigid-body mechanics models. study orthopaedic surgical procedures,” IEEE Trans. Biomed. Eng.,
Seamless integration with motion capture software, such as vol. 37, no. 8, pp. 757-767, 1990.
Vicon Nexus, will facilitate the integration of FE analysis [13] F. C. Anderson and M.G. Pandy, “A dynamic optimization solution
with existing clinical practices. Additionally, the presented for vertical jumping in three dimensions,” Comput. Meth. Biomech.
skeletal and prosthesis model can be combined with other Biomed. Eng., vol. 2, no. 3, pp. 201-231, 1999.
[14] F. C. Anderson and M.G. Pandy, “Dynamic optimization of human
clinical measures, such as MRI, to enable patient-specific walking,” J. Biomech. Eng., vol. 123, no. 5, pp. 381-390, 2001.
hard and soft tissue modeling, such as the menisci of the knee [15] G.T. Yamaguchi and F.E. Zajac, “A planar model of the knee joint to
and/or residual limb tissue mechanics. Concurrent multibody characterize the knee extensor mechanism,” J. Biomech., vol. 22, no.
and FE models present an exciting new avenue for patient- 1, pp. 1-10, 1989.
specific modeling. [16] X. Bonnet, H. Pillet, P. Fodé, F. Lavaste and W. Skalli, “Finite
element modelling of an energy-storing prosthetic foot during the
stance phase of transtibial amputee gait,” Proc. Inst. Mech. Eng. Part
REFERENCES H: J. Eng. Med., vol. 226, no. 1, pp. 70-75, 2012.
[1] L. G. Stansbury, S. J. Lalliess, J. G. Branstetter, M. R. Bagg and J. B. [17] M. Omasta, D. Paloušek, T. Návrat and J. Rosický, “Finite element
Holcomb, “Amputations in US military personnel in the current analysis for the evaluation of the structural behaviour of a prosthesis
conflicts in Afghanistan and Iraq,” J. Orthop. Trauma, vol. 22, no. 1, for trans-tibial amputees,” Med. Eng. Phys., vol. 34, no. 1, pp. 38-45,
pp. 43-46, 2008. 2012.