Design and Preliminary Evaluation of A Passive Spine Exoskeleton
Design and Preliminary Evaluation of A Passive Spine Exoskeleton
Haohan Zhang
Department of Mechanical Engineering,
Columbia University,
Evaluation of a Passive
New York, NY 10027
e-mail: [email protected] Spine Exoskeleton
Abhijit Kadrolkar This paper presents a design of a passive spine exoskeleton which implements a
Department of Mechanical “push–pull” external assistive strategy. The spine exoskeleton was designed for reducing
and Industrial Engineering, the risk of back injury. It applies a pulling force on thoracic region and a pushing force
University of Massachusetts Amherst, on lumbar region during spine flexion/extension. The design was inspired by previous
160 Governor’s Drive, simulation work, where the results highly supported benefits of the push–pull strategy on
Amherst, MA 01003 reducing the back muscular efforts and bending moment for the sagittal spine flexion/
e-mail: [email protected] extension. A passive physical prototype was designed and constructed to test the
push–pull strategy on human subjects. Three subjects were able to repeat the identical
Frank C. Sup, IV1 dynamic spine flexion and extension tasks with the spine exoskeleton prototype. The sur-
Department of Mechanical face electromyography showed a reduction of up to 24% at lumbar and 54% at thoracic
and Industrial Engineering, level muscle for the human subjects wearing the exoskeleton suit to accomplish the same
University of Massachusetts Amherst, static tasks without any external assistance. The muscle force and intervertebral bending
160 Governor’s Drive, moment were estimated to be reduced by up to 479 N and 36 N m, respectively.
Amherst, MA 01003 [DOI: 10.1115/1.4031798]
e-mail: [email protected]
Keywords: push–pull, spine flexion/extension, exoskeleton
identical presumably the inertial effects of the cam can be displacements, DlL and Dlh , can be obtained from the model
neglected, which deduces geometry used in the simulation. Therefore, the spring constant of
the spring applying pushing force is determined. Since there are
T r ¼ FL rL (1) two extension springs on either side of the spine, the spring con-
stant for each spring should be half of kh . The final springs
where r and rL are the geometric parameters of the cam that can selected for the physical prototype are shown in Table 1.
be measured from the physical model. In addition, both the virtual The components of the physical prototype were 3D printed
spring and the physical spring force obey the Hooke’s Law such using acrylonitrile butadiene styrene plastic (Stratasys, uPrint SE).
that T ¼ kh Dlh and FL ¼ kL DlL , where kh is the spring constant The posts for holding rotational joints of the cams (shown in pur-
of the lumbar spring and kL is the spring constant provided by ple in Fig. 3(c)) have a hollow core that allows steel rods to be
the optimization result from OPENSIM, which is the set of spring inserted to strengthen the overall structure. The rotational joints
constants that deliver the best assistance. Additionally, the were supported by ball bearings. Buckles were used to adjust the
Rest length (m) Spring constant (N/m) Outer diameter (m) Extended Lg. (m)
3 Evaluation
The performance of the prototype was evaluated experimentally
Fig. 4 Model of static analysis for subject flexion with sEMG and biomechanics data from human subjects performing specific
sensor attached upper body motions were collected. The data included surface
electromyogram (sEMG) readings from selected muscle groups
prototype and to tension the cables. The fabricated prototype and inertial data from wearable accelerometers mounted on the
including all components weighs 3.0 kg which was less than 5% body of the subjects. The control data for the protocol were data
of the total body weight for the subjects tested in this paper. A ma- from the subject performing the same motions while not wearing
jority of the weight is located on the pelvis while the parts carried the exoskeleton prototype. From the experimental data, the reduc-
by the human torso are light weight. Since the prototype was pri- tions of back muscular effort and bending moment were estimated
marily designed to evaluate movement in the sagittal plane, it is for each subject wearing the spine exoskeleton. The study was
passively compliant with the movement of the human body in approved by the Institutional Review Board at the University of
other planes through the flexibility of the prototype structures. Massachusetts Amherst.
2.2 Estimating Muscle Force and Bending Moment. A 3.1 Protocol. The experiment consisted of two tasks:
model for the static test in vivo was applied in the sagittal plane to dynamic and static. Subjects completed both tasks without and
estimate the muscular force and intervertebral bending moment with the exoskeleton for comparison. Control data for each subject
reduction, as shown in Fig. 4. The free body of interest was the were collected without wearing the exoskeleton. Subsequent to
torso above the dashed line indicated the L5–S1 joint. The coordi- this, test data for each subject were collected after donning the
nate frame was constructed according to the accelerometer shown exoskeleton. The goal was to evaluate the effect of the exoskele-
with the green block. The muscle force FM is assumed parallel- ton suit on a person in different situations. It relied on kinematics
ing to the sensor axis at the attaching point [2]. The weight of the data, sEMG magnitudes from the erector spinae muscles, and esti-
exoskeleton suit is negligible since it is much lighter than the torso mates of expected reductions in muscular forces and intervertebral
masses of the subjects. When the participant wore the exoskeleton bending moments. Figures 5(a) and 5(b) show a photograph of a
suit, the two spring forces applied to their torso are the pulling subject wearing the prototype. The subjects were first tested with-
force on thoracic region FTS and the pushing force on lumbar area out the exoskeleton suit and then with in order to not positively
FLS . During the static test, the readings ax and az from the accel- favor the results with the exoskeleton due to fatigue.
eration sensors (without gravity elimination) indicated compo- In the dynamic test, the subjects were asked to perform repeated
nents of the reaction forces FRX and FRZ at the lumbar joint, spine flexion/extension in the sagittal plane (three second flexion
which were decoupled along the frame axes. sR denotes the reac- and three second extension) for 120 s with a constant speed pro-
tion moment about that joint. The equilibrium equations for this vided by an audible tone from a metronome. The subjects started
model are as follows: from standing, flexed their torso approximately 90 deg, and
Without the exoskeleton suit extended it back to the original standing position to finish a com-
8 plete flexion/extension motion. All motion was limited to the sag-
< FRX ¼ max ¼ G cos h þ FM
> ittal plane and lateral or rotational motion was avoided. At all
FRZ ¼ maz ¼ G sin h (2) times during the test, arms were crossed across the chest, and
>
:P hands were placed on opposite shoulders. The subjects fully rested
Mo ¼ 0 ¼ GD FM d sR
before approaching into the static tests.
The static tests required the subjects to hold three different flex-
With the exoskeleton suit (denoted with 0 ) ion positions (small, medium, and full-range flexion) for between
8 0 90 and 120 s. The flexion angle in each position was determined
0
0 0
< FRX ¼ max ¼ G cos h þ FM þ FTS cos a
> by a reference height measuring from subject0 s left shoulder to the
0 0 0
FRZ ¼ maz ¼ G sin h þ FLS FTS sin a (3) ground. These reference heights were measured before the experi-
>
:P ments for each individual subject and varied as anticipated due to
Mo ¼ 0 ¼ GD F0M d FTS dTS FLS dLS s0R the differences of their total height and lengths of body segments.
The subjects were instructed to maintain their flexion angle by
Subtracting Eq. (2) from Eq. (3) gives the difference equation in aligning their left shoulders with a specific reference height during
two equilibrium situations each trial. The subjects rested between every two trials. After the
Table 2 sEMG reductions of all subjects. The greatest reductions at each position were bolded among all subjects at lumbar and
thoracic levels, respectively.
static tests without wearing with exoskeleton suit, the subjects electrode placement. The electrodes that physically interfered
fully rested and then completed the identical tests wearing the with the prototype were removed. In addition, the removed sen-
exoskeleton suit. sors in device tests varied among subjects due to their size differ-
ences. Even with removal of some electrodes, both lumbar and
thoracic level muscular activities on both sides were recorded for
3.2 Participants. Three males participated in this study
each subject and are shown in Table 2.
(height ¼ 172 6 12 cm, weight ¼ 68:3 6 6:7 kg, and age The skin was lightly abraded and swabbed with alcohol prior to
¼ 26:7 6 3:3 years). Subject 1 was the target subject that the affixing the sensors along the muscles with double-sided tape.
spine exoskeleton was designed for. He participated in the biome-
Electrodes were separated by a 10 mm distance and electrically
chanics motion capture experiment where the data were utilized to arranged in a bipolar configuration to obtain sEMG data from
construct the simulation model and perform the computational
each location. The analog sEMG signals were sampled at 2 kHz
evaluations. Subjects 2 and 3 participated for result comparison, and band-pass filtered (20–450 Hz) by the wireless sEMG system.
where subject 2 was similar in size to subject 1 while subject 3 Triaxial accelerometers embedded in the sEMG sensors were used
was about 15 cm taller.
to obtain the inertial data during the evaluations. The inertial data
were sampled at 148.1 Hz and low-pass filtered at 50 Hz. Flexion
3.3 Instrumentation. The sEMG data were captured using a angles were integrated from the acceleration data using the left
commercial measurement system (DelsysTM–TrignoTM Wireless). side accelerometers and linear accelerations were obtained from
Eight electrodes were placed on the torso to capture muscle acti- the right side by assuming the subjects moved symmetrically with
vations during spine flexion/extension and placed as shown in respect to the sagittal plane.
Fig. 5(c). The muscular activities of lumbar erector spinae (LES)
at L3 and L1 levels and thoracic erector spinae (TES) at T11 and
T10 levels were captured on both sides. It needs to be pointed out 3.4 Data Processing. The sEMG signals were processed
that although eight-channel sEMG was collected for all subjects according to Refs. [16–18] prior to result comparison. A second
during control tests, only a subset was used during the device test- order high-pass Butterworth filter with a cut-off frequency of 2 Hz
ing. This is because some components (i.e., springs and cams) of was used to remove DC offsets and motion artifacts. The signals
the wearable device physically interfered with some attached elec- were then normalized by the largest value of the magnitude from
trodes. The control tests were performed first and the sensor loca- each sensing channel for each subject, separately for the control
tion was not adjusted for the exoskeleton tests to avoid altering and test data. Finally, the signals were full-wave rectified and
Fig. 7 Representative sEMG data comparison from a single participant undertaking static
tests: (a)–(c) comparison of normalized muscle sEMG at lumbar level at positions 1 (barely
flexion) to 3 (full-range flexion) and (d)–(f) comparison of normalized muscle sEMG at thorax
level at positions 1–3