design and evaluation of a stiffness ... - tu delft

Design and evaluation of a stiffnesscompensating ankle-foot orthosis

Freek VerbakelMarch 18, 2013

Mas

tero

fScie

nce

Thes

is

Design and evaluation of astiffness compensating

ankle-foot orthosis

MASTER OF SCIENCE THESIS

For the degree of Master of Science in MechanicalEngineering at Delft University of Technology

Freek Verbakel

March 18, 2013

Examination commiteeDr. Ir. E. de Vlugt (BMechE, 3ME)Dr. Ir. J.H. de Groot (LUMC)Dr. J.F.M. Molenbroek (Applied Ergonomics, IDE)

Faculty of Mechanical, Maritime and Materials Engineering (3mE)Delft University of Technology

PrefaceI have always been interested in both technology and medicine. And the logicresult for me was subscribing for the track Biomechanical Design within the Mas-ter of Science program in Mechanical Engineering. I took a long time thinkingabout a suitable thesis subject. Thinking about the System Identification & Pa-rameter Estimation (SIPE) classes I decided to contact Erwin de Vlugt for doinga thesis within his research field. There were plenty of opportunities and I choseto do my thesis on a project that was completely new: designing an ankle-footorthosis with a negative stiffness mechanism. The assignment, containing bothdesign aspects as well as performing clinical experiments, was really suitable forme because it gave me the opportunity to bring both my interests together.

Before really starting with the thesis a literature study and an internship had tobe done. For the literature study I wrote a report in which I described differentankle-foot orthosis that are currently used to treat patients with gait disorders.From reading all the literature I formed a good image in my mind about howa wearable ankle-foot orthosis should be designed. For my internship I workedthree months at the Northwestern University Prosthetics and Orthotics Center(NUPOC) in Chicago on a data analysis project about the effects of ankle-footorthosis on the gait of stroke survivors. During my internship I learned morethan I could have imagined about normal and pathological gait and how ankle-foot orthosis influence gait.

Thereafter I really started with my thesis. The first part was the design andconstruction of the negative stiffness orthosis (NSO) in which I worked closelywith a fellow graduation student, Leon Derks. Together we worked on transfer-ring the concept of a negative stiffness mechanism to a design of an ankle-footorthosis. My main focus was on the mechanism, where Leon Derks’ focus was onthe patient interaction. After the construction of the NSO I thought of ways toevaluate the behavior of the NSO. I decided to start with doing measurementson able-bodied subjects, in which a patient is simulated artificially, using the‘Achilles’ measurement setup.

I would like to thank my supervisors Erwin de Vlugt and Jurriaan de Grootfor their time having interesting discussions and helping me throughout theentire project. I would also like to thank the staff of the Delft University ofTechnology for their help with the construction of the NSO and the staff ofthe Leiden University Medical Center for their help in getting familiar with theexperiment setup. I really enjoyed the company of all the other students thathave been working in the lab. Thank you all!

Freek VerbakelMarch 2013

I

Table of contents

I Scientific paper 11 Introduction . . . . . . . . . . . . . . . . . . . . 5

2 Methods . . . . . . . . . . . . . . . . . . . . . . 12

3 Results . . . . . . . . . . . . . . . . . . . . . . 18

4 Discussion . . . . . . . . . . . . . . . . . . . . . 28

5 Conclusion . . . . . . . . . . . . . . . . . . . . . 33

References . . . . . . . . . . . . . . . . . . . . . 34

II Appendices 37A Experiment setup . . . . . . . . . . . . . . . . . . . 39

B Experiment protocol . . . . . . . . . . . . . . . . . 42

C Checklist experiments . . . . . . . . . . . . . . . . . 44

D Data processing . . . . . . . . . . . . . . . . . . . 46

III

Part IScientific paper

1

Design and evaluation of a stiffnesscompensating ankle-foot orthosis

Freek Verbakel1, Erwin de Vlugt1, Jurriaan H. de Groot2

1 Laboratory of Neuromuscular Control, Department of Biomechanical Engineering,

Faculty of Mechanical, Maritime and Materials Engineering, Delft University of Tech-

nology, Mekelweg 2, 2628 CD Delft, The Netherlands2 Laboratory for Kinematics and Neuromechanics, Department of Rehabilitation Medicine,

Leiden University Medical Center, Albinusdreef 2, 2333 ZA Leiden, The Netherlands

Abstract

Introduction: Upper motor neuron diseases (UMND’s) primarily cause pare-sis with secondarily an increase in viscoelasticity of passive tissues and musclesand results in decreased Range of Motion (RoM) and may obstruct relearningof movement patterns. Existing ankle-foot orthoses (AFO’s) ‘take over’ jointcontrol by adding positive stiffness, restricting movement for stabilization andgenerate external work. A need exists for an AFO that promotes self-control ofthe patient. This study proposes an AFO based on a novel concept to compen-sate increased stiffness by adding negative stiffness to the joint. The negativestiffness working principle is a mechanism consisting of a spring generating forcearound a rotation axis, with a variable moment arm. The spring force is decreas-ing, when rotating the axis, less than the moment arm is increasing and torquethus increases. The main goals of this study are to adjust the aforementionedtheoretical mechanism for use in an AFO, using a simulation, and to build aphysical functional model of an AFO with the negative stiffness mechanism, inwhich subjects are able to experience negative stiffness. A secondary goal is tovalidate the resulting functional model, the negative stiffness orthosis (NSO).Methods: A mathematical model has been used to dimension different designparameters. Ten able-bodied subjects with an artificially increased ankle jointstiffness performed passive and active RoM tasks while wearing the NSO. Theimportant outcome parameters of the study are the maximal dorsiflexion angle(active and passive) and EMG-activity as a function of the ankle angle duringthe active RoM task, as an indication of user effort during active movements.Results: The negative stiffness working principle has been realized with thefirst NSO. The NSO increases ankle angles for constant torque input. Maximalactive dorsiflexion angle increased as well, while dorsiflexor muscle EMG-activitydecreased for equal ankle angles.Conclusion: The results show that the negative stiffness mechanism is ableto partly compensate stiffness and support a better dorsiflexion angle to forceratio, which is especially beneficial in UMND patients with weaker muscles thannormal. A prototype, suitable for walking, is to be developed.

3

List of abbreviations

Active RoM Range of motion during active movementsAFO Ankle-foot orthosisCAD Computer aided designCNS Central nervous systemDF DorsiflexionEMG ElectromyographyLSD Least significant differenceNSO Negative stiffness orthosisPassive RoM Range of motion during passive movementsPF PlantarflexionRoM Range of MotionTA Tibialis anterior muscleTS Triceps surrae muscle groupUMND Upper motor neuron disease

4

1 Introduction

Problem

Upper motor neuron diseases (UMND’s, stroke a.o.) are deficits of the centralnervous system (CNS) that affects the control of upper motor neurons in thebrain. Three fundamental mechanisms of impairment rise due to UMND [1].The first mechanism is paresis, a neurological deficit, which leads to a reductionof voluntary motor unit recruitment and concomitant muscle force. The secondmechanism is an increased gain in reflexive activity, i.e. spasticity. The thirdmechanism is immobilization of the musculoskeletal system, which results in apermanent stiffening and eventually shortening of muscles and contracture ofthe joint. These mechanisms of impairment lead to a patient induced disuseof the musculoskeletal system. Gracies (2005a,b) reported a vicious cycle ofparesis-disuse-paresis and thus a progressive development of the paresis. It isalso suggested that disuse, in combination with shorter muscles, leads to softtissue rearrangements [1, 2]. The increase in viscoelasticity of passive tissuesand muscles hamper joint rotation, in addition to the paresis [3, 4]. Increasedviscoelasticity of passive tissues and muscles may eventually change the neutraljoint position. UMND patients have limited movement ability, due to hamperedjoint rotation. Hampered joint rotation results in decreased joint Range of Mo-tion (RoM), promotes disuse and obstructs relearning of movement patterns,such as walking.Figure 1 shows a schematic force-length characteristic for active and passive tis-sues under constant muscle activation in the normal situation and the assumedeffects in UMND being paresis, shorter muscles and increased stiffness of passivetissues.

5

Figure 1: Schematic presentation of the force-length characteristics of active and pas-sive tissues of the ankle joint that act in parallel under constant muscle ac-tivation. The length where active and passive forces are equal determinesthe maximal dorsiflexion angle. Dashed lines represent the normal situa-tion and the solid lines show the assumed effects in UMND being paresis,shorter muscles and increased stiffness of passive tissues (from top to bot-tom panel respectively). The difference between the green (normal) and thered dot (impaired) represents the assumed effects in UMND patients on themaximal dorsiflexion angle.

This study focuses on the human ankle as a crucial joint for movement (posture,walking) and because often the distal joints are affected most in UMND. Nextto the reported increase in stiffness, the ankle joint is also largely maintainedin a plantarflexed position (i.e. a changed neutral position, pointing down-wards), due to the triceps surrae muscle group shortening relatively more thanthe tibialis anterior muscle in UMND [5]. A lack of ankle joint dorsiflexion (i.e.pointing upwards) may be the result of the increased stiffness and the changedneutral ankle position. Ankle joint dorsiflexion is, however, crucial for effectivegait [6, 7]. In the stance phase of gait a lack of ankle dorsiflexion prevents thebody from rotating forward around the ankle, and is thus not able to generatea normal step length. In the swing phase of gait a lack of ankle dorsiflexion pre-vents sufficient foot clearance with the risk of hitting the ground, inhibits therisk of tripping and prevents an appropriate initial ground contact, i.e. landingon heel and forefoot.

Rehabilitation often prescribes mechanical support of the joint using ankle-foot orthoses (AFO’s). Most are for permanent use. Rehabilitation AFO’sare designed using active and passive actuation principles. The active AFO’sare generating additional torque and are thus able to describe the movement

6

pattern the user should follow. Torque is generated in active AFO’s using pneu-matic muscles (e.g. the Michigan AFO [8]) or with series elastics actuatorsthat change impedance during the gait cycle (e.g. the MIT AFO [9] and therobotic tendon design [10, 11]). Disadvantage is that it places the user out ofthe sensory-motor loop, i.e. moving the user beyond his own control, the userhas to do little on himself and will not be able to fully ‘take back’ joint control.Another disadvantage is the complexity of the actuation as well as the massand volume of all the components (batteries, processors a.o.) that are needed.Figure 2 shows a schematic force-length characteristic for active and passivetissues under constant muscle activation in the case of an unassisted UMNDpatient and the assumed effect of active actuated devices, stiffness increasingpassive devices and stiffness compensating passive devices. Note that the as-sumed UMND patient in Figure 2 (solid line) experiences effects from all threetypes of impairments as were presented in Figure 1. Every type of assistanceimproves the force-length characteristic by partly remedying the effect of onespecific impairment.

Figure 2: Schematic presentation of the force-length characteristics of active and pas-sive tissues of the ankle joint that act in parallel under constant muscleactivation. The length where active and passive forces are equal deter-mines the maximal dorsiflexion angle. Dashed lines represent the case ofan unassisted UMND patient and the solid lines show the assumed effectsof active actuated devices, stiffness increasing passive devices and stiffnesscompensating passive devices (from top to bottom panel respectively). Thedifference between the red (impaired) and the green dot (assisted) representsthe assumed effects of the assisting devices on the maximal dorsiflexion an-gle.

7

Passive actuated AFO’s are mainly designed to prevent drop foot gait. A car-bon fiber spring or a dorsiflexion assist controlled by spring (DACS) AFO arepassive actuated AFO’s that prevent plantarflexion movement (i.e. generat-ing additional stiffness) and generate a continuous torque assisting in dorsi-flexion [12, 13]. Limitation of passive actuated AFO’s is that the actuator isa (linear) spring, adding positive stiffness and restricting movements for stabi-lization. Figure 2 (middle) shows the effect of stiffness increasing devices onthe force-length characteristic of muscles. Passive actuated devices have advan-tages. First, passive devices need fewer components and are less complex ingeneral than active devices. Second, passive devices are, when accommodat-ing movements, assisting the user to train his own muscles and maintain theself-controllability.

Motivation

Several studies have shown that passive devices are able to promote relearn-ing of typical complex movement patterns, by accommodating movements, andenhance necessary plastic rearrangements in the CNS, i.e. the breaking andforming of neural connections [14, 15]. Other studies suggested that a relationexists between muscle usage and plastic rearrangements in the CNS [2,16]. Us-ing the plasticity of the CNS it becomes possible to use the remaining neuralcapacity to relearn complex movement patterns [17]. To increase movement ca-pacity by optimal usage of remaining neural control, a need exists for an AFOthat promotes self-control. This study proposes a passive AFO based on a novelconcept to compensate increased joint stiffness by adding negative stiffness tothe joint, i.e. (partial) compensation of the joint such that for the same muscleforce an increase in RoM is accomplished. Figure 2 (bottom) shows the expectedeffect of a stiffness compensating passive AFO on the force-length characteristicof active and passive tissues by lowering the passive tissue characteristic andincrease the maximal joint angle and thus RoM.

Working principle

The main function of the proposed negative stiffness mechanism is to generatea specific torque-angle characteristic. A spring under pre-tension generates anincreasing torque (T) around a rotation axis. The spring force (F) is decreasingless than the moment arm (r) is increasing. Therefore the product (T = F·r)increases. Figure 3 shows the behavior of the spring force and the moment armrelative to each other, to generate a negative stiffness behavior. Figure 3 alsoshows the resulting mechanism torque. Schmit and Okada (2011) proposed amechanism for a negative stiffness hinge joint [18]. The mechanism consists of alinear tensile spring connected to a non-circular curved cable spool mechanism.Figure 4 shows a schematic drawing of the negative stiffness mechanism. Thespool rotates around the center of rotation (point O). The spool contour isdefined in the spool coordinate system (Rspool), which rotates around point Oat an angle θs with respect to the global coordinate system (Rref ). The cableis running from the fixed pulley at point P (at a distance R from the centerof rotation) to the point T on the spool, where the cable runs tangent to thespool contour. The radius r is the varying distance, from the center of rotationto point T, θr is the rotation angle of r with respect to Rspool. The cable

8

winds over the spool contour and the position of point T and the correspondingmoment arm, i.e. distance r at point T, changes. The torque delivered is a non-linear function of the spool’s rotation angle. The mechanism can generate anytorque-angle characteristic by dimensioning of the four key design parameters.The key design parameters are: 1) the spool contour, defining the relation ofthe moment arm as a function of the rotation angle, 2) the distance from pointO to point P (distance R, the location of the fixed pulley), defining the ‘zeromoment arm’ (i.e. the orientation) of the spool, 3) the spring pre-tension and 4)the spring stiffness. The last two parameters are determining the force appliedby the spring during the joint rotation. Figure 5 shows two conditions of anexample mechanism, with a simpler spool contour compared to the mechanismin Figure 4, and demonstrates how the changing moment arm influences thetorque around the center of rotation.

Figure 3: Relative behavior of the spring force, the moment arm and the correspond-ing mechanism torque as a function of the rotation angle to generate a neg-ative stiffness behavior. The spring force (red line) is decreasing less thanthe moment arm (blue line) is increasing, the mechanism torque (greenline) is thus increasing for increasing rotation angle.

9

Figure 4: Schematic drawing of the negative stiffness mechanism as proposed bySchmit and Okada (2011). The spool rotates around the fixed center ofrotation (point O) and changes the tension in the cable, running over thefixed pulley (P) and attached to the spring. The key design parameters arethe contour of the spool (r and θr), the distance R, the spring stiffness (k)and the spring pre-tension (q0) [18].

Figure 5: Effects of an example negative stiffness mechanism. The spring force (k =25 N/m) is decreasing (0.33 N to 0.25 N, left to right graphic) less thanthe moment arm (2.3 mm to 12.3 mm) is increasing. The torque increases(0.8·10−3 Nm to 3.1·10−3 Nm) for a spool rotating from -9 ◦ to 13 ◦.

10

Goals

The main goals of this study are to adjust the aforementioned theoretical mech-anism design from Schmit and Okada (2011) for use in an AFO, using a simu-lation, and to build a physical functional model of an AFO with the negativestiffness mechanism, in which subjects are able to experience negative stiffness.A secondary goal is to validate the resulting functional model, the negative stiff-ness orthosis (NSO), on able-bodied subjects at a single joint level. Figure 6gives a schematic overview of the different phases in this study.

Mechanism from literature:Schmit and Okada (2011)

Computer simulation to optimizethe mechanism behavior by

adjusting key design parameters

Design and construction ofthe mechanism in an AFO:a physical functional model

Validation of the result-ing physical model us-

ing able-bodied subjects

Application: passive AFOto compensate increased

ankle joint stiffness

Physical design constraints

Figure 6: Schematic overview of this study. The first step is to adjust the mechanismfrom Schmit and Okada (2011), using a simulation, for use in an AFO.The second step is to design and construct a physical functional model ofan AFO with the proposed mechanism. The last step is to validate theresulting functional model using able-bodied subjects.

11

2 Methods

Negative stiffness mechanism

The proposed mechanism by Schmit and Okada (2011) will be applied on thehuman ankle joint. The mechanism will be generating torque around the joint:at the neutral ankle position torque is zero and potential energy is stored in thespring; during dorsiflexion movements the energy is released and dorsiflexiontorque increases. The work, T·θs, is increasing for a decreasing spring force, F,by an increasing moment arm, r(θs).

It would be beneficial if the mechanism could be tuned for different demands(e.g. different patients, improved gait). Three key design parameters (momentarm of the spool, distance R, and the spring stiffness k) can be changed by creat-ing new components. Creating new components may be inefficient from an eco-nomic perspective. The simplest way to accommodate for different demands isby changing the spring characteristics. An adjustable spring pre-tension shouldenable the mechanism to shift the torque-angle characteristic.

The main function of the negative stiffness mechanism is to generate a spe-cific torque-angle characteristic. The negative stiffness mechanism should bedesigned based on a desired torque-angle characteristic. The desired torque-angle characteristic is approximated by an exponential function [3], see Eq. 1.

τ(θ) = c1ec2(θ−c3) (1)

The exponential function in Eq. 1 only estimates torque accurately in dorsi-flexion. The exponential function has an upward slope and torque level doesnot change in plantarflexion. The dorsiflexion area is the area of rotation thatwould ideally be affected by the negative stiffness mechanism. The variablesc1, c2 and c3 are derived by fitting Eq. 1 on measured characteristics found inthe literature. The desired torque-angle characteristic for the negative stiffnessmechanism was based on an average normal (τnormal) and pathological (τpatient)torque-angle characteristics as reported by De Vlugt (2012) [3], according to Eq.2.

τnso = τnormal − τpatient (2)

The four key design parameters are interdependent, all are affecting the shape ofthe torque-angle characteristic. Based on a desired torque-angle characteristicτ(θs) with a corresponding RoM vector θs, it is possible to determine all four keydesign parameters uniquely. The RoM vector determines the working area of themechanism. The length of the working area is chosen to be 45 ◦ in this study.Eq. 3 represents the translation from the desired torque-angle characteristicconstituting the negative stiffness to the required geometry. J(θs) defines thekinematic input/output characteristic that the spool must achieve to synthesizethe torque-angle characteristic τ(θs). Eq. 3 (from [18]) is taking spring pre-tension q0 and stiffness k, into account.

J(θs) =τ(θs)√

2 · k∫ θs0τ(u)du+ (k · q0)2

(3)

12

Eq. 4 (from [18]) provides the explicit solution for the radius, a vector with thesame size as θs from the kinematic input/output characteristic, as defined inEq. 3, and the parameter R.

r =

√J2(θs) +

J ′2(θs) · (R2 − J2(θs))

(J ′(θs) +√R2 − J2(θs))2

(4)

Eq. 3 and Eq. 4 can be used in a mathematical model to simulate the effectsof different key design parameters on the resulting torque-angle characteristic,τ(θs).

Functional design

Two important geometrical design constraints exist in the physical design of theNSO. One is the maximum spool radius (rmax), i.e. the size of the spool, beingsmaller than the distance from the center of rotation to the ground, to preventthe spool from hitting the ground. The value of rmax, is the constant distancefrom the center of rotation to the end of the spool and influences the value ofthe mechanism moment arm. The distance R is along the shank and the springshould be mounted above distance R. The distance R plus the spring, whenfully elongated, should not exceed the length of the shank. Figure 7 shows aninitial sketch of the NSO. The geometrical design constraints can be deductedfrom the sketch in Figure 7. The spool moment arm cannot be exactly zero,since a thickness of the spool is necessary over the entire contour. Next to theaforementioned constraints, a few extra design criteria are selected. A linearoff the shelve spring should be selected that has a favorable energy to size ra-tio. The NSO is intended for experimental research, and therefore it shouldbe designed such that it can easily be assembled and disassembled. The fourkey design parameters need to be adjusted easily (e.g. replacing or rotatingspool, replacing spring or increasing spring pre-tension) in the design of theNSO. For the validation in this study will the spring pre-tension be the onlykey design parameter that is adjusted. The NSO should be able to compen-sate the ankle stiffness of an average patient (Ashworth score 2 [3]) and adjustfor other patients by increasing or decreasing the level of stiffness compensation.

The validation study was intended to evaluate the NSO at a single joint level.The NSO connected to the shank does not need to move, as was mentioned inthe introduction. The NSO can therefore be designed rigid enough regardlessof mass and volume.

13

Figure 7: Sketch of the NSO worn by a potential user to show the geometrical designconstraints. The size of the mechanism is limited by the length of the shankand the height of the ankle center of rotation.

Validation of the NSO

The validation is intended to verify the simulated increase in RoM, by adjust-ing the spring pre-tension. Adjustment of the spring pre-tension scales thecompensation torque-angle characteristic. Another important aspect for assess-ing proper application of the NSO is the ability of the muscles to generate forcewhile stretched to lengths that were not usual before application of the NSO, seethe effects on the force-length characteristic in Figure 2 (bottom). Thus, testsneed to be conducted that include determination of passive RoM (validation ofstiffness compensation) and active RoM (prime benefits of stiffness compensa-tion).It is hypothesized that the NSO increases the maximal dorsiflexion angle, passiveand active, when stiffness compensation is increased. Secondly, it is hypothe-sized that EMG-activity decreases for dorsiflexion movements.

Experimental conditions

Ten able-bodied subjects (5 male/5 female) were recruited for participation inthe validation study. Patients are simulated in this study by increasing theankle stiffness in dorsiflexion of able-bodied subjects. The ankle stiffness inable-bodied subjects was increased in all conditions, with the exception of thecontrol (NORM) condition, by applying sports tape, see Figure 8(a). The tapingtechnique is one that is commonly used to relieve achilles tendinitis patients,i.e. restricting the achilles tendon movements. The ankle stiffness increase indorsiflexion is applied to simulate a patient condition in able-bodied subjects.Measurements are done in five different conditions:

14

• NORM: The subject will undergo the experiment with his bare foot strappedto the footplate of the manipulator (no NSO, no footwear and no tape).

• INERT: The subject will undergo the experiment with a taped ankle wear-ing the NSO without spring tension over the entire RoM.

• LOW: The subject will undergo the experiment with a taped ankle wearingthe NSO with a spring pre-tension of exactly 0 mm in the neutral ankleposition. The spring will therefore only tense when the ankle rotates inplantarflexion.

• MEDIUM: The subject will undergo the experiment with a taped anklewearing the NSO with a 4 mm extra spring pre-tension. The ankle angleat which the spring relaxes moves more into dorsiflexion.

• LARGE: The subject will undergo the experiment with a taped anklewearing the NSO with a 4 mm extra spring pre-tension.

The four conditions involving the NSO are performed in the presented order forhalf of the subjects. For the other half of the subjects are the four conditionsinvolving the NSO performed in the reversed order, to exclude degeneratingeffects of the tape. The NORM condition is always measured first. The subjectneeded to stand up between the measurements of the NORM condition andthe other conditions to mount the NSO. It is likely that the seating position isslightly changed between the NORM condition and the other four conditions.In order to minimize the change in seating position is the subject asked to sittightly against the back of the seat and the subject is fixated to the seat by abelt around the waist. The rotation of the subject’s knee is maintained at aconstant level and is checked before measuring the NORM condition and beforemeasuring the other four conditions.

(a) (b)

Figure 8: (a) Taping technique used in the experiment to increase stiffness in dorsi-flexion by restricting achilles tendon movements. (b) Achilles robotic anklemanipulator.

15

Equipment

Subjects are in a seated position in single joint experiments, moving the anklejoint around a fixed axis and having all other joints at rest, similar to control-ling the throttle while driving a car. The shank, and the NSO connected toit, does not need to move. A robotic ankle manipulator (‘Achilles’, MOOGFCS Inc, Nieuw Vennep, The Netherlands, see Figure 8(b)) was used in theexperiments. The NSO can be connected to the ankle manipulator. The leg isfully supported while the subjects experience the negative stiffness of the NSOaround their ankle joint. The subjects are able to perform active and passiveankle movement tasks in the ankle manipulator, with all other joints at rest.The ankle manipulator is recording torque, rotation angle, rotational velocityand acceleration of the ankle at a rate of 1024 Hz.Surface EMG-activity is measured using a Porti7 interface with unipolar elec-trodes (TMSI, Oldenzaal, The Netherlands) on the tibialis anterior, soleus, gas-trocnemius medialis and gastrocnemius lateralis muscles. Two electrodes wereplaced on the muscle belly in the direction of the muscle fibers for bipolardeduction and placement was in accordance with the guidelines developed bySENIAM.org (inter electrode distance is 2 cm). The EMG-activity signal hasbeen sampled at a rate of 2000 Hz. The EMG-activity signal is amplified 20xand low-pass filtered at 540 Hz, online.Post processing: The EMG-activity signal is normalized and initially high-passfiltered at 20 Hz using a third order recurrent Butterworth filter [19], offline.The EMG-activity signal of all four individual muscles is derived by calculatingthe difference between the two unipolar electrodes on each muscle. The EMG-activity signal is rectified and low-pass filtered at 15 Hz using a third orderrecurrent Butterworth filter [19], offline.

Tasks

Passive RoM task: The first part of the experiment is a passive RoM task, i.e.a task to measure the subject’s passive RoM (i.e. RoM while moving passively).In the passive RoM task the ankle manipulator moves the subject’s relaxedankle by supplying an 80 seconds long triangular torque input signal, with amaximum of 7.5 Nm and a minimum of -7.5 Nm.Active RoM task: The second part is an active RoM task, i.e. a task to measurethe subject’s active RoM (i.e. RoM while moving actively). In the active RoMtask the subject is asked to move the ankle manipulator into the maximal dorsi-flexion and plantarflexion angle and repeat during 30 seconds with a self-chosenvelocity. Direct feedback of the ankle angle is presented on a screen in front ofthe measurement setup.

Outcome parameters

The important outcomes of the study are the maximal dorsiflexion angles, activeand passive, and motoric efficiency, i.e. EMG-activity as a function of the ankleangle during the active RoM task. In order to compare torque-angle character-istics between the different conditions simple graphs were generated by fitting afifth order polynomial on the measured ankle angle as a function of the torqueinput signal. The outcome parameters are the ankle angles at two specific torquelevels. The first torque level is 0 Nm, which is at rest. The second torque levelis 7.5 Nm, the highest measured in the experiment and resembles the maximal

16

passive dorsiflexion angle.In order to analyze the active RoM task, the recorded dorsiflexion movementswere separated from the plantarflexion movements. In order to make the ac-tive movements comparable between subjects, all performed dorsiflexion and allperformed plantarflexion movements are normalized in time. The measured ro-tational velocity is plotted as a function of the measured ankle angle (averagedover all performed movements made during the 30 seconds measurements). Theimportant outcome parameter of the active RoM task is the largest measureddorsiflexion angle and resembles the maximal active dorsiflexion angle.In order to control for the muscular input, the motoric efficiency is analyzedby plotting the EMG-activity signal as a function of the corresponding ankleangle in the active RoM task. The EMG-activity of the dorsiflexor muscle, i.e.tibialis anterior (TA), is plotted as a function of the ankle angle in the dorsiflex-ion movements (averaged over all the movements made during the 30 secondsmeasurements). The EMG-activity of the plantarflexor muscle, i.e. the tricepssurrae (TS) which is a summation of the soleus, gastrocnemius medialis andgastrocnemius lateralis, is plotted as a function of the ankle angle of the plan-tarflexion movements. The EMG-activity of both muscle groups is amplitudenormalized, after the summation of the TS, by the maximum and the minimumvalue. The outcome parameters of the motoric efficiency are the values of theaverage normalized EMG-activity at specific ankle angles: -10 ◦, -5 ◦, 0 ◦, 5 ◦

and 10 ◦.

Statistics

The differences of the means between conditions of the defined outcome param-eters are analyzed using a least significant difference (LSD) post-hoc one-wayANOVA test for significance (α = 0.05). The following hypotheses are testedusing IBM SPSS version 19. 1) The measured ankle angle at 0 Nm and 7.5 Nmin the passive RoM task is different for the different experimental conditions. 2)The maximal reached dorsiflexion angle during the active RoM task is differentfor the different experimental conditions. 3) The EMG-activity at -10 ◦, -5 ◦,0 ◦, 5 ◦ and 10 ◦ in the active RoM task is different for the different experimentalconditions.

17

3 Results


Figure 9 (left) shows the optimized contour of the spool (thick lines), illustratedfor three different positions (rotations of the ankle joint) and the correspondingcable paths (thin lines), that results in an exponential torque-angle character-istic. Figure 9 (right) shows the corresponding moment arm as a function ofthe ankle angle (top), spring force as a function of the ankle angle (middle) andtorque-angle characteristics (bottom) of the mechanism (black) and the optimalaccording to Eq. 4 (red). The characteristics of the moment arm and the springforce can now be scaled to fit within the geometrical design constraints and togenerate an optimal torque-angle characteristic.

Figure 9: Schematic overview of the optimized negative stiffness mechanism, thicklines represent the spool (from left to right: in 33 ◦ plantarflexion, in neu-tral and in 12 ◦ dorsiflexion), the thin lines represent the correspondingcable paths from the fixed pulley in the top (left), the corresponding mo-ment arm as a function of the ankle angle (top right), spring force as afunction of the ankle angle (middle right) and torque-angle characteristics,black represents the optimized mechanism and red represents the optimalcharacteristic according to Eq. 4 (bottom right).

Figure 10 shows the simulated torque-angle characteristics (approximated by anexponential function) that can be covered by the negative stiffness mechanism,for a specific patient. The torque of the negative stiffness mechanism is opposingthe torque of the patient ankle joint. The range of possible torque compensation,by varying the spring pre-tension, lies in the dark gray area in Figure 10. Thecompensation characteristic adds to the patient torque-angle characteristic, apatient with Ashworth score 2 (AS2) in this situation, to obtain the torque-angle

18

characteristic of the total system (light gray area). Torque-angle characteristicsof other patients and an able-bodied subject are shown as well for comparison.

Figure 10: Simulated torque-angle characteristics, i.e. measured torque as a func-tion of the ankle angle. The solid lines represent specific patient anklesaccording to the Ashworth scale and an able-bodied subject, the dashed linerepresents the theoretical available compensation of the negative stiffnessmechanism, the dark gray area represents the area in which the compensa-tion characteristic can be found and the light gray area represents the areawhere the patient (AS2) + compensation can be found, the more springpre-tension the lower the characteristic lies. The range of compensationcould as well be applied on patients with a different Ashworth score. Thecharacteristics are approximated by an exponential function [3].

Functional design

Overcoming human tissue stiffness generally requires a large amount of force,larger than is common in spring balanced mechanisms of this size. As a conse-quence, a regular linear tensile spring having an appropriate stiffness and thatobeys the geometrical design constraints was not available, i.e. such a springwould be too large. The NSO was therefore built with two springs at both sidesof the NSO. The NSO was equipped with two 41.35 N/mm springs, 87.2 mmlong at rest and 113.7 mm at 1.25 kN. Figure 11 shows the level of torque com-pensation at 10 ◦ dorsiflexion as a function of rmax (influencing moment arm,see Figure 9 (top right)) and spring stiffness (influencing spring force, see Figure9 (middle right)). The two black lines show the selected values for the NSO. Armax of 70 mm was selected, to meet the first geometrical design constraint.

19

Figure 11: Level of torque compensation as a function of maximum spool radius, i.e.the constant distance from the center of rotation to the end of the spool,and spring stiffness, the two black lines represent the selected values forthe NSO. The spring stiffness is accomplished by two parallel springs.

Figure 12 shows details of the mechanism design. The mechanism design istranslated from the simulated design in Figure 9. A simple stainless steel cable(2 mm diameter) is clamped at the lower end of the spool, which is directlymounted to the frame of the AFO. The upper end of the spool is mountedin the center of rotation. The center of rotation is a special designed hingejoint between the frames connected medially and laterally to the shaft throughball bearings. The spring is mounted in the shaft, with a simple turning screwthrough the top, to vary the spring pre-tension. The design is made out ofaluminum and stainless steel, where necessary, and carbon fiber in the userinterfaces.

Figure 12: Details of the spool design, with details of dimensions, attachment point(clamping) and center of rotation.

20

Using the optimized mechanism and taking into account the geometrical designconstraints a detailed CAD design was made (Figure 13(a)) and was subse-quently built, being the first NSO (Figure 13(b)). The two springs are mountedin the shafts on both sides of the NSO.

(a) (b)

Figure 13: (a) CAD design and (b) physical design of the NSO.


Figure 14 shows one of the subjects wearing the NSO during the validationstudy. Figure 15 shows the results for a typical subject for all five conditionsin the passive RoM task. The maximal passive dorsiflexion angle is restrictedin the INERT condition, due to the applied taping technique. The maximaldorsiflexion angle is (partly) restored and the torque-angle characteristics areshifted into dorsiflexion as an effect of increasing spring pre-tension.

21

Figure 14: Subject wearing the NSO connected to the ankle manipulator during thevalidation.

Figure 15: Torque-angle graphs (i.e. torque-angle characteristics) measured in thepassive RoM task of a typical subject for all five conditions. The anklemanipulator applied a triangular torque input signal, with a maximum of7.5 Nm and a minimum of -7.5 Nm, and measured the ankle angle.

22

Figure 16 shows the outcome parameters for all subjects for all five conditionsin the passive RoM task. The average ankle angles are more in dorsiflexion forlarger spring pre-tension, similar as in the individual subject in Figure 15. Themean of the ankle angle in every condition is significantly different from all otherconditions at 0 Nm, with the exception of the means of the NORM and INERTconditions. The mean ankle angle in the LARGE condition is significantlydifferent from the mean ankle angle in the INERT and LOW conditions at 7.5Nm, and not significantly different from the mean ankle angle in the NORMcondition. The mean of the ankle angle at 7.5 Nm is 7.25 ◦ larger in the LARGEcondition compared to the INERT condition.

Figure 16: Measured ankle angle at specific torque levels: 0 Nm and 7.5 Nm, averagedover all ten subjects (mean ± standard deviation) for all five conditionsduring the passive RoM task. The ankle manipulator applied the torqueand measured the ankle angle. The asterix denotes significant differences.

23

Figure 17 shows the rotational velocity as a function of the ankle angle in theactive RoM task averaged over all movements for a typical subject, positive ve-locity represents dorsiflexion movements and negative velocity represents plan-tarflexion movements.

Figure 17: Averaged rotational velocity (mean ± standard deviation) as a functionof the ankle angle, dorsiflexion movements (positive velocity) and plan-tarflexion movements (negative velocity), in the active RoM task of atypical subject for all five conditions.

24

Figure 18 shows the maximal active dorsiflexion angle averaged over all subjects.The mean of the maximal active dorsiflexion angle in the NORM condition issignificantly different from the mean of the maximal active dorsiflexion angle inthe INERT, LOW and MEDIUM conditions, but not significantly different fromthe mean of the maximal active dorsiflexion angle in the LARGE condition. Themeans of the maximal active dorsiflexion angle in the INERT, LOW, MEDIUMand LARGE conditions are not significantly different. The mean of the maximalactive dorsiflexion angle is 4.8 ◦ higher in the LARGE condition compared tothe INERT condition.

Figure 18: Maximal active dorsiflexion angle averaged over all ten subjects (mean ±standard deviation) for all five conditions in the active RoM task. Theasterix denotes significant differences.

25

Figure 19 shows the averaged EMG-activity for a typical subject of the TAmuscle during active dorsiflexion movements (left column) and for the TS musclegroup during active plantarflexion movements (right column), as a function ofankle angle. The left column in Figure 19 shows that an increase in springpre-tension lowers the EMG-activity during dorsiflexion movements, for equalankle angles. The subject needs less muscle activation than in the NORMcondition. A range of rotations in dorsiflexion exists where EMG-activity is zero.The increase of spring pre-tension increases the range of ankle angles where nomuscle activity is required. The right column in Figure 19 shows an inverse effectduring plantarflexion movements. The effect of the increased spring pre-tensionincreases the EMG-activity. The EMG-activity is higher because the elongationof the spring requires energy input and is thus less efficient, and limited. Storedenergy is conserved until the spring relaxes again during dorsiflexion movements.

Figure 19: Averaged EMG-activity, filtered and normalized, (mean ± standard devi-ation) as a function of the ankle angle, for dorsiflexion movements dueto TA muscle activity (left) and for plantarflexion movements due to TSmuscle activity (right), in the active RoM task of a typical subject for allfive conditions.

26

Figure 20 shows the outcome parameters of the motoric efficiency averagedover all ten subjects. Measurements of TA EMG-activity are displayed at fivedifferent ankle angles: -10 ◦, -5 ◦, 0 ◦, 5 ◦ and 10 ◦. Not all subjects reached 10 ◦

in all conditions. The 10 ◦ measurements are left out for subjects that did notreach 10 ◦ in all conditions. The mean EMG-activity in the LARGE conditionis 49.6% lower than the mean EMG-activity in the INERT condition at -10 ◦,48.9% at -5 ◦, 48.1% at 0 ◦ and 40.8% at 5 ◦. The mean EMG-activity in theLARGE condition is 29.4% lower than the mean EMG-activity in the INERTcondition at 10 ◦, it must be noted that the population is smaller at 10 ◦ (n =8). The mean EMG-activity in the INERT condition is significantly differentfrom the mean EMG-activity in the MEDIUM and LARGE conditions at -10 ◦,-5 ◦, 0 ◦ and 5 ◦. The mean EMG-activity in the LOW condition is significantlydifferent from the mean EMG-activity in the LARGE condition at 0 ◦ and 5 ◦.The mean EMG-activity in the INERT condition is significantly different fromthe mean EMG-activity in the LARGE condition at 10 ◦.

Figure 20: Filtered and normalized EMG-activity of the TA muscle averaged over allten subjects (mean ± standard deviation) for all five conditions duringthe active RoM task at different ankle angles: -10 ◦ (PF), -5 ◦ (PF), 0 ◦,5 ◦ (DF) and 10 ◦ (DF). Note that the population that reached 10 ◦ (DF)is smaller than the total population (n = 8).

27

4 Discussion

A physical functional model of an AFO with negative stiffness has been built,based on a theoretical concept, in which subjects are able to experience negativestiffness. The effect of different levels of stiffness compensation supplied by theNSO on dorsiflexion movements and control has been tested on able-bodied sub-jects with artificially stiffened ankles. The NSO is able to increase the maximaldorsiflexion angle and reduces muscle activation during movement.


A negative stiffness mechanism has been designed based on a cable spool mech-anism as proposed by Schmit and Okada (2011). The mechanism consisted of aspring attached to a hinge joint, where the combination of decreasing spring forceand a proportionally larger increase of the moment arm around the joint resultsin a specific torque-angle characteristic. The shape of the resulting torque-anglecharacteristic is determined by four key design parameters. There is a trade-offbetween the two most important key design parameters, the spring stiffness andthe maximum spool radius (rmax). Figure 11 illustrated this trade-off. Thermax must be larger when the spring stiffness is lower to generate the sametorque-angle characteristic. Multiple combinations of spring stiffness and rmaxcould be selected. Optimal would be to select a spring with a lower stiffness,than the spring that has been selected, and build a spool with a larger rmax,resulting in less force generated by the spring, which makes it possible to designthe mechanism less rigid. However rmax is limited by the height of the anklecenter of rotation, because otherwise the spool will be hitting the ground. Thespring stiffness thus needs to be as large as was selected (82.7 N/mm dividedover two parallel springs).When designing the desired torque-angle characteristic to compensate humantissue stiffness, it should be observed that there is probably a trade-off betweenmovement ability and joint stability. The trade-off between movement abilityand joint stability can be seen in the two passive AFO’s in Figure 2 (middle andbottom). The stiffness increasing AFO’s emphasized on increasing joint stabil-ity, while the stiffness compensating AFO’s emphasized on increasing movementability. The level of dorsiflexion torque potentially generated by the mechanismis higher than the resisting torque in able-bodied subjects and patients withan Ashworth score 0, see Figure 10. Therefore overcompensation (i.e. jointstability is insufficient) of the joint stiffness is possible when using the negativestiffness mechanism to compensate human joints. Overcompensation may givethe user the feeling of moving too fast and the user may potentially react toslow down. Loss of control and moving in undesired positions may be the resultof overcompensation and must therefore be averted, by carefully selecting thelevel of stiffness compensation. On the other hand, undercompensation (i.e.movement ability is insufficient) would mean less effective usage of the negativestiffness mechanism.The results of this study showed that partial or inaccurate compensation, basedon the chosen exponential torque-angle characteristic, supports an increasedmaximal dorsiflexion angle and increased control, which can contribute to moreeffective ankle movements. The findings are similar to the theory of stiffnesscompensating passive AFO’s, as was presented in Figure 2. The shape of the

28

mechanism torque-angle characteristic was derived from the average patientcharacteristic and taken as an exponential function, because passive joint resis-tance in general follows exponential behavior [20]. Ideal would be to design aspecific torque-angle characteristic for individual patients, determined by a spe-cialist. The patient specific torque-angle characteristic should generate a level oftorque compensation that gives the patient the best combination of movementability and joint stability.

Functional design

A wearable NSO, with the negative stiffness mechanism, was built in whichthe subject-AFO interaction could be tested in vivo on able-bodied subjects.In order to build a functional model that demonstrates the functioning of thenegative stiffness mechanism, that is easily assembled and disassembled and iscertain to withstand the concomitant high forces, the NSO is built with heavierand thicker material than is normal in AFO’s. Therefore the NSO was designedfor mass and volume not being design criteria and the current functional modelwas thus not feasible to be applied during walking. The NSO shows that theeffects are beneficial in dorsiflexion, but adverse in plantarflexion. Currently,RoM is decreased because the maximal plantarflexion angle decreases more thanthe maximal dorsiflexion angle increases. The spool was constructed with athickness larger than 0 mm, like the design in Figure 9, to make the rotationaxis solid enough to withstand high forces. The moment arm should have been 0mm when rotated in plantarflexion in the simulation, but was 7 mm in the NSO.The moment arm generated an undesired resisting torque, which increases forlarger spring pre-tension, in plantarflexion and caused the increase of stiffness inplantarflexion. The increase of stiffness in plantarflexion influences the resultingtorque-angle characteristic, and the effect should therefore have been minimized.


The design and construction process proceeded quickly enough to allow time totest the NSO-subject interaction in an experiment setup, but the NSO was notspecifically intended for experiments which gave two noticeable problems in thevalidation study. The first problem was the shape of the foot interface. It provedto be difficult to restrict foot movements within the NSO, because no subjectspecific shells were available, i.e. the foot position was not constant with respectto the NSO. The movement of the foot could lead to a slight misalignmentbetween the ankle joint and the rotation axis of the ankle manipulator. Thesecond problem was the lack of alignment adjustability of the rotation axis of theNSO with the rotation axis of the ankle manipulator. Misalignment of rotationaxes generated a resisting torque and affected RoM negatively.The NSO could not have been validated using UMND patients, because it wouldbe unethical to expose patients to a model that is still in the experimentalphase. Therefore has the NSO been validated using able-bodied subjects withan artificially increased stiffness in dorsiflexion by applying sports tape. Theeffect of taping the ankle in able-bodied subjects resulted in larger measuredtorques for the same ankle angle and a decreased maximal dorsiflexion angle.The shape of the torque-angle characteristic was still the same after applyingthe tape.

29

Figure 17 showed that the subjects had, for the same level of torque input (0Nm and 7.5 Nm are shown) a larger ankle dorsiflexion rotation as spring pre-tension increases, indicating that subjects experienced a larger dorsiflexion angleto force ratio. A larger dorsiflexion angle to force ratio enabled the subjects tomove the ankle more into dorsiflexion while requiring less muscle force to do so.The maximal active dorsiflexion angle is increased as an effect of increasedspring pre-tension and EMG-activity for equal ankle angles is decreased, aswas hypothesized. Subjects with increased ankle stiffness using the NSO maypotentially move the ankle from 10 ◦ plantarflexion to 10 ◦ dorsiflexion with lesseffort, because of decreased EMG-activity for equal ankle angles as indicated byFigure 20.

Safety

Overcompensation implies the risk of joint instability (total stiffness negative).Maximal ankle joint rotation and rotational velocity should be imposed on theNSO by physical barriers. In dorsiflexion the rotation is limited to the pointwhere the spring is relaxed (the spring force is decreasing when the ankle isrotating in dorsiflexion). More potential hazards rise due to the large forcesin the NSO. The force is transferred from the spring to the mechanism bya simple stainless steel cable, clamped by a standard wire clamp. The usedstandard parts are prone to wear, and therefore not suited for long term usage.Long term usage was not required in this study, but the NSO would be lessmaintenance demanding if the spring to mechanism interface was made out ofone custom made part.

Clinical relevance

The used taping technique on able-bodied subjects was not able to fully sim-ulate an UMND patient, because it does not account for the decreased neuralcapacity. Therefore it cannot be stated whether sufficient joint stability re-mains when using the NSO, and that walking is not comprehended. Researchhas shown that increased ankle joint stiffness in UMND patients is a positiveadaption mechanism of the paresis to accommodate functional movements on asimpler level of organization [21,22].The NSO in this study is an AFO that promotes self-control to increase move-ment capacity of patients that suffer from increased ankle joint stiffness. Self-control is promoted by building the NSO with a passive negative stiffness mech-anism. The increased stiffness of the ankle joint can partly be compensated bythe NSO, such that for the same muscle force an increase in the maximal dorsi-flexion angle is accomplished. The NSO supports a larger dorsiflexion angle toforce ratio, which is especially beneficial in UMND patients that have weakermuscles than normal. The current findings support the idea that an AFO withthe negative stiffness mechanism may be able to assist UMND patients to per-form complex movements, such as walking.The decreased effort in moving the ankle from 10 ◦ plantarflexion to 10 ◦ dor-siflexion may benefit in the stance phase and the swing phase of gait. Thedecreased effort may make dorsiflexion movements possible and allow startingto rotate the body forward over the ankle during the stance phase of gait. Thesame decreased effort may enable the patients to rotate the ankle enough into

30

dorsiflexion to not hit the ground during the swing phase of gait. Otherwise theNSO will have the option of changing the spool orientation, e.g. rotating thespool and thus the area of decreased effort more into dorsiflexion.The subjects were able to move with decreased effort. But it is remarkable thatthe able-bodied subjects in this study did not use the assistance supplied by theNSO to increase the rotational velocity of the ankle joint (the subjects were al-lowed to move with their own self-selected velocity after all) instead of loweringthe EMG-activity of the TA muscle, as was shown in Figure 20. It may be thecase that the able-bodied subjects subconsciously select a maximal rotationalvelocity that results in an optimal ankle rotation and that faster movementsmay result in an increased reflexive activity.The proposed NSO changes the neutral ankle position, which is not in accor-dance with the design criteria. A changed neutral ankle position generates thesame effect as existing AFO’s preventing drop foot gait. Because the effect of theparesis is asymmetric around the joint, i.e. the plantarflexor muscles are pro-portionally contracting more than the dorsiflexor muscles [1, 23]. Asymmetriccontraction of ankle muscles causes drop foot gait, i.e. excessive plantarflexion.As mentioned earlier the majority of existing AFO’s are focusing on preventingdrop foot gait [12, 24–26]. Based on this study it cannot be stated if and howthe NSO should have influenced the neutral ankle position.

Future prototype

Suggestions for an improved NSO prototype can be made. Emphasis shouldbe put on aspects that where not covered by the design criteria in this study:mass- and volume reduction, fabricating a custom made spring and creatingmore adjustability of the device to create a new prototype. Unlike the NSO inthis study it is advised that the new prototype will be suited for walking experi-ments, i.e. multi joint evaluation, and clinical experiments. A prototype designsuitable for walking can be achieved by especially critical looking at the volumeat het medial side of the shank, where too much volume may hamper oppositeshank movements. Mass and volume can be reduced by creating comfortablyfitting shells for the foot and shank, which must be designed patient specific.The shells can best be made out of a lightweight, though stiff, material, likecarbon fiber.The enormous spring force (650 N), that is needed to overcome human tissuestiffness with a relatively small moment arm (from 0 mm to 10 mm), only comesin very large springs (there is a trade-off between the spring stiffness and elon-gation). It is worthwhile to create a design that grants the mechanism moreavailable space for the spool and thus a larger available moment arm and a cor-respondingly smaller spring. It can be thought of a system in which the spoolcenter of rotation is not at the ankle center of rotation and that the torque istransferred trough an alternative mechanism between the centers of rotation.Another possible solution is to mount the spool in the direction of the foot’slong axis and adjust the cable paths such that the mechanism can generate thesame torque. It is also worthwhile to evaluate more advanced spring mecha-nisms to save mass and volume, e.g. a spring with a better energy to size ratio.But it remains a challenge to generate enough torque (up to 10 Nm) using asmall mechanism. Two important advantages rise when a lower spring stiffnessis required. The first advantage is that thinner and lighter materials can be

31

used. The second advantage is that the NSO can suffice with only one springand only one negative stiffness mechanism, placed on the lateral side of theshank. Both lead to a reduction is mass and volume of the NSO.To prevent injuries the spring and mechanism must be completely shielded fromthe user and the construction must be build rigid enough. Ultimately, a FiniteElement Analysis can be used to find regions in the construction that are proneto deform or snap, and to find regions that are designed with too much mate-rial. Regular check-ups of the load bearing parts for wear and deformations arerequired.

Future research

Future research should point out to what extent stiffness compensation is pos-sible when performing multi joint movement tasks, e.g. walking, and whetherRoM and motoric efficiency can be increased as well. Interesting will be to verifywhether the NSO is able to assist UMND patients to (re)learn movement pat-terns, based on optimal usage of remaining neural capacity. It should be verifiedwhether usage of the NSO leads to plastic rearrangements of the CNS, whichwill give the UMND patient the ability to perform the (re)learned movementpatterns without external mechanical support.

32

5 Conclusion

In this study a new passive AFO based on a novel concept to compensate anklejoint stiffness was proposed. A physical functional model of an AFO has beendeveloped consisting of a spring balanced negative stiffness mechanism, suit-able for evaluation at a single joint level. The negative stiffness mechanism wasequipped with a spring, with adjustable spring pre-tension, the chosen key de-sign parameter to change the level of stiffness compensation. The experimentsshowed that the NSO compensated more stiffness for a higher spring pre-tension.And with the compensated stiffness comes a better dorsiflexion angle to forceratio. The active RoM task was intended to identify the prime benefits of theNSO usage. The experiments showed that the maximal dorsiflexion angle canbe increased, with less dorsiflexor muscle EMG-activity. This study suggeststhat a negative stiffness orthosis may potentially benefit UMND patients withincreased passive tissue viscoelasticity around the ankle joint during functionaltasks, such as walking. A prototype, suitable for walking, is to be developed.

33

References

[1] Gracies, J., Pathophysiology of spastic paresis. I: Paresis and soft tissuechanges. Muscle & Nerve, 2005. 31: p. 531-551.

[2] Gracies, J., Pathophysiology of spastic paresis. II: Emergence of muscleoveractivity. Muscle & Nerve, 2005. 31: p. 552-571.

[3] De Vlugt, E., et al, Clonus is explained from increased reflex gain andenlarged tissue viscoelasticity, Journal of Biomechanics, 2012. 45(1): p.148-155.

[4] Lieber, R.L., et al, Structural and functional changes in spastic skeletalmuscles. Muscle & Nerve, 2004. 29(5): p. 615-627.

[5] McDonald, M.F., Garisson, M.K., Schmit, B.D., Length-tension propertiesof ankle muscles in chronic human spinal cord injury. Journal of Biome-chanics, 2010. 38(12): p. 2344-2353.

[6] Ballaz, L., Plamondon, S., Lemay, M., Ankle range of motion is key to gaitefficiency in adolescents with cerebral palsy. Clinical Biomechanics, 2010.25(9): p. 944-948.

[7] Bregman, D.J.J., The Optimal Ankle Foot Orthosis: The influence of me-chanical properties of Ankle Foot Orthoses on the walking ability of patientswith central neurological disorder. 2011.

[8] Ferris, D.P., et al, An improved powered ankle-foot orthosis using propor-tional myoelectric control. Gait and Posture, 2006. 23: p. 425-428.

[9] Blaya, J.A., Herr, H., Adaptive Control of a Variable-Impedance Ankle-FootOrthosis to Assist Drop-Foot Gait. IEEE Transactions on Neural Systemsand Rehabilitation Engineering, 2004. 12(1): p. 24-31.

[10] Hollander, K.W., et al, An efficient Robotic Tendon for Gait Assistance.Journal of Biomechanical Engineering, 2006 128: p. 788-791.

[11] Boehler, A.W., et al, Design, Implementation and Test Results of a RobustControl Method for a Powered Ankle Foot Orthosis (AFO), in IEEE In-ternational Conference on Robotics and Automation. 2008: Pasadena, CA,USA. p. 2025-2030.

[12] Bartonek, A., Erikson, M., Gutierez-Farewik, E.M., A new carbon fibrespring orthosis for children with plantarflexor weakness. Gait and Posture,2007. 25: p. 652-656.

[13] Yamamoto, S., et al, Development of an Ankle-Foot Orthosis with Dorsi-flexion Assist, Part 2: Structure and Evaluation. Journal of Prostheticsand Orthotics, 1999. 11(2): p. 24-27.

[14] Banala, S.K., et al, Gravity-Balancing Leg Orthosis and Its PerformanceEvaluation. IEEE Transactions on Robotics, 2006. 22(6): p. 1228-1239.

[15] Rahman, T., Design and Testing of WREX. Advances in RehabilitationRobotics, 2004 306: p.243-250.

34

[16] Diserens, K., et al, The effect of repetitive arm cycling on post stroke spas-ticity and motor control. Journal of the Neurological Sciences, 2007. 253(1):p. 18-24.

[17] Kwakkel, G., et al, Impact of Early Applied Upper Limb Stimulation: TheEXPLICIT-Stroke Programme Design. BMC Neurology, 2008 8(49).

[18] Schmit, N., Okada, M., Synthesis of a non-circular cable spool to realizea nonlinear rotational spring, in International Conference on IntelligentRobots and Systems. 2011. p. 762-767.

[19] Konrad, P., The ABC of EMG: A Practical Introduction to KinesiologicalElectromyography. 2005.

[20] Esteki, A., Mansour, J.M., An experimentally based nonlinear viscoelasticmodel of joint passive moment. Journal of Biomechanics, 1996. 29(4): p.443-450.

[21] Dietz, V., Proprioception and locomotor disorders. Nature Reviews: Neu-roscience, 2002. 3(10): p. 781-790.

[22] Dietz, V., Sinkjaer, T., Spastic movement disorder: impairment reflex func-tion and altered muscle mechanics. The Lancet Neurology, 2007. 6(8): p.725-733.

[23] Pelletier, C.A., Hicks, A.L., The length-tension relationship of human dor-siflexor and plantarflexor muscles after spinal cord injury. Spinal Cord,2010. 48(3): p. 202-206.

[24] Sumiya, T., Suzuki, Y., Kasahara, T., Stiffness control in posterior-typeplastic ankle-foot orthoses: effect of ankle trimline. Part 2: orthosis char-acteristics and orthosis/patient matching. Prosthetics and Orthotics Inter-national, 1996. 20: p. 132-137.

[25] Churchill, A.J.G., Halligan, P.W., Wade, D.T., Relative contribution offootwear to the efficacy of ankle-foot orthoses. Clinical Rehabilitation, 2003.17: p. 553-557.

[26] Chin, R., et al, A pneumatic power harvesting ankle-foot orthosis to preventfoot-drop. Journal of NeuroEngineering and Rehabilitation, 2009. 6(19).

35

Part IIAppendices

37

Appendix A

Experiment setup

This appendix describes the measurement setup of the validation part in thisstudy. The experiments are done in the motion analysis laboratory in the LeidenUniversity Medical Center. The setup consists of the following devices:

• Achilles robotic ankle manipulator: The Achilles robotic ankle manipula-tor (MOOG FCS Inc, Nieuw Vennep, The Netherlands) is a robot that iscapable of perturbing the human ankle joint. The Achilles is intended forresearch in a single joint setup, meaning that the subject is in a seatedposition and will only rotate his ankle. Active and passive tasks can beperformed on the Achilles and the ankle rotation, rotational velocity andtorque (i.e. kinematic signals) are measured at a rate of 2048 Hz maxi-mally. Figure A.1(a) shows the Achilles robotic ankle manipulator withthe feedback screen above.

• Porti7 interface: The Porti7 interface (TMSI, Oldenzaal, The Nether-lands) is a device recording EMG-activity signals through maximal 32unipolar channels in real-time. Only eight channels are required in thisstudy in order to measure four muscles bipolar. Bipolar deduction needs tobe done in the post processing phase. EMG-activity signals are recordedat a rate of 2000 Hz maximally. The Porti7 communicates with the com-puter through a FUSBI interface (TMSI, Oldenzaal, The Netherlands)that receives signals from the Porti7 through a glass fiber cable and sendsthe signals to the computer through a USB cable. Figure A.1(b) showsthe Porti7 and the FUSBI interfaces with the corresponding cables.

• RobinGUI: The RobinGUI is a guided user interface programmed in Mat-lab version R2010b (Mathworks Inc, Natick, MA, USA) on a 32bit Win-dows XP computer connected to the measurement devices. The RobinGUIloads and runs custom made files to govern the Achilles robotic ankle ma-nipulator (e.g. no force input for active tasks) such that the requiredexperimental tasks can be executed as intended. The tasks are describedin Appendix B. The RobinGUI is collecting data from the Achilles andfrom the Porti7 interface, this is described in more detail below.

• NSO: The negative stiffness orthosis is an AFO with the negative stiffnessmechanism. The purpose of the NSO is to let subjects experience negativestiffness. The NSO is designed and manufactured at the Delft Universityof Technology. A stainless steel footplate is rigidly attached to the NSO.The footplate on the NSO is designed such that it can be used to replacethe standard footplate on the Achilles robotic ankle manipulator and isused to quickly switch between measurements with and without the NSO.

39

(a) (b)

Figure A.1: (a) Achilles robotic ankle manipulator and the feedback screen. Thesubject is seated in front of the setup. The feedback screen providesfeedback about the ankle rotation during the measurements. (b) ThePorti7 (below) and FUSBI (top) interfaces. The EMG-activity signalsare the input in the Porti7 through the eight yellow cables (plus the cyancable as the ground signal). The orange cable sends all the signals to theFUSBI through a glass fiber cable and the black cable sends the signalsto the computer through a USB cable.

The measured data from both the Achilles and the Porti7 interface must besynchronized with the perturbing signal (the Achilles drivefile) and with thefeedback information that is presented to the subject. The measurement signalfrom the Achilles is stored using a dataloggerthread that makes the Achillesmeasure the signals during the measurement trial and stores the data in a .matfile on the computer afterwards. There is also a dataloggerflush signal fromthe Achilles that continuously sends signals to the computer for the purpose ofdirect feedback, but this signal is not stored on the computer. The measure-ment signal from the Porti7 interface must be stored on the computer whilethe measurement trial is still running, due to a too small buffer size. FigureA.2 describes schematically how the guided user interface collects signals fromboth devices in parallel. The first step is to start (initialize) all devices. Thesecond step is to start the Achilles drivefile (i.e. the programmed movements ofthe perturbator for the duration of the whole trial) and the dataloggerthread.While measuring, the Porti7 buffer is read once in every 2.5 seconds. The rest ofthe measurements are the Porti7 commands bypassed and the feedback screenis updated based on the dataloggerflush signal. At the end of each trial is themeasurement loop aborted, is all the required data stored and are all devicesstopped. There is a time delay between the start of the measurement of theEMG-activity signal and the start of the measurement of the kinematic sig-nals. A special experiment has been designed and performed to identify thetime delay between both signals. An EMG electrode is rigidly placed at a smalldistance above the stainless steel footplate of the Achilles. The Achilles drivefileis programmed to generate a small rotational movement in the direction of theelectrode that will give a signal at the moment the electrode and the stainlesssteel plate make contact. Ultimately the delay between both signals is identified

40

to be 70 samples at the Achilles sampling rate. For proper analysis of the datawill the first 70 samples of the Achilles data be removed. More about the datapost processing can be found in Appendix D.

Init Porti7 bufferInit Achilles virtual environment

Start Achilles drivefileStart Achilles dataloggerthreadStart Achilles dataloggerflush

tread≥ 2.5s

Read Porti7 buffer

Store Porti7 data of the last2.5 seconds in a matrixRestart tread (set to 0)

Refresh feedback monitor basedon dataloggerflush signals

Wait 0.01 seconds

t ≥ T

Store Achilles dataloggerthreadStore matrix from

Porti7 measurements

NO

NO

Y ES

Y ES

Figure A.2: Schematic overview of the data collection from the Achilles and thePorti7 interface. The Achilles drivefile is governing the Achilles move-ments and the Achilles dataloggerthread is storing all measurement sig-nals from the Achilles robotic ankle manipulator. The data from thePorti7 buffer is read every 2.5 seconds (tread) in a measurement loop,the feedback monitor is updated in the same loop. T is the programmedmeasurement duration and t is the time since the start of the trial.

41

Appendix B

Experiment protocol

This appendix gives additional information to the experimental protocol usedfor the validation study. The experiments will be performed on able-bodiedsubjects. To emulate an increased ankle joint stiffness as in stroke patients,the ankle will be artificially stiffened. The increased ankle joint stiffness shouldresult in larger necessary torques in dorsiflexion and a decreased maximal dor-siflexion angle. Different methods are tested for applicability for this specificgoal. Restraining the achilles tendon by non-elastic sports tape is found to bethe most appropriate.The experiments will be conducted on the Achilles ankle perturbator. The studyconsists of two experimental tasks performed in five conditions. See Part 1 fordescriptions of the measurement conditions.The level of pre-tension of the NSO, needed to create the different experimentalconditions, is verified by measuring the length of the screw thread that standsout at the top of the spring shafts, inside the shaft the thread is connected tothe spring. The distance the thread stands out of the shaft can be changed byturning the nut. The more the thread stands out the higher the pre-tensionthat is imposed on the spring. Preliminary tests were done to determine thedistance that the thread should stand out in every experimental condition. Itwas found that zero pre-tension occurs when the thread stands out 5.6 cm. Thedistance in the other conditions are: 6 cm (LOW), 6.4 cm (MEDIUM) and 6.8cm (LARGE). Figure B.1 shows the thread when it stands out 6.8 cm (for themeasurements in the LARGE conditions).

Figure B.1: The nut in top of the NSO can be turned to adjust the distance the threadstands out of the spring shafts and thus the pre-tension of the springs.The current configuration corresponds with the LARGE condition.

42

Passive RoM task: The subject is asked to fully relax while the ankle is movedby the ankle manipulator. The Achilles drivefile is programmed to generate atriangular torque profile (moving from 0 Nm to 7.5 Nm to -7.5 Nm and back to 0Nm). The duration of this measurement is 80 seconds. The important outcomesof the passive RoM task are the torque-angle characteristic (i.e. torque as afunction of the measured angle) and the maximal passive dorsiflexion angle (i.e.the largest measured ankle angle).Active RoM task: The Subject is asked to rotate the ankle in dorsiflexionas far as possible and in plantarflexion thereafter. The angle between maximaldorsiflexion and plantarflexion is defined as the active RoM. The subject willfeel no resistance from the ankle manipulator because the Achilles drivefile isprogrammed to generate a torque profile that is always zero. The duration ofthe measurement is 30 seconds. The subject is asked to move in dorsiflexionand in plantarflexion continuously at his own comfortable velocity until theend of the measurement. To motivate the subject the real-time rotation of hisankle is presented on a screen in front of the setup. Figure B.2 shows how theinformation is presented.

Figure B.2: Feedback information presented to the subject. The red bar representsthe actual rotation of the ankle, the large blue box represents the neutralorientation of the ankle and the small blue bars represent the minimaland maximal reached angles (i.e. the RoM). On the left of the screenare the values presented that correspond to the red and the blue bars.

43

Appendix C

Checklist experiments

Preparation and measurement for the NORM condition

1. Introduce subject to the research and explain about the functioning of theAchilles robotic ankle manipulator.

2. Make left shank accessible and prepare with EMG preparation gel andalcohol.

3. Attach eight EMG markers on the appropriate locations (derived fromSENIAM.org).

4. Mount the correct footplate to the Achilles.

5. Check the position of the ankle with respect to the center of rotation andadjust if needed.

6. Check positions of workspace extremities (hardware safety stops).

7. Initialise Achilles with the new workspace extremities (software safetystops).

8. Strap foot onto the footplate.

9. Connect EMG electrodes to the corresponding EMG markers.

10. Check seat position and adjust if necessary.

11. Ask the subject to take and maintain a relaxed seating position and movethe ankle into a 90 degree position.

12. Start the active RoM trial (used to measure absolute rotations) and writedown the angle of the 90 degree position, check the ankle position using aruler.

13. Ask subject to move the ankle up and down slowly after starting activeROM trial for the second time.

14. Check resulting EMG-activity signals and relocate markers if needed, re-peat previous and current step until clear EMG-activity signals are mea-sured.

15. Check whether correct name and 90 degree angle are set in the GUI.

16. Check whether all tasks are added into the GUI, enough for all experi-mental conditions.

44

17. Start experiment.

18. Execute both trials in NORM condition.

Preparation and measurement for the other (NSO) conditions

1. Decouple EMG electrodes.

2. Remove the subject’s foot from the footplate.

3. Tape the subject’s achilles tendon, according to the achilles tendinitistechnique.

4. Determine the order the subject will perform the different conditions (i.e.from INERT to LARGE or from LARGE to INERT).

5. Check the spring pre-tension of the NSO according to the condition thatis measured first.

6. Remove footplate and mount NSO functional model and strap the subjectinto the NSO functional model.

7. Adjust seat position such that the subject’s knee is in the same orientationas before.

8. Let subject retake approximately the same relaxed seating position.

9. Connect EMG electrodes to the corresponding EMG markers.

10. Execute both trials in the first condition (INERT or LARGE).

11. Adjust spring pre-tension force on NSO, subject remains seated.

12. Execute both trials in the next condition and repeat previous step untilall conditions are measured (INERT, LOW, MEDIUM and LARGE).

13. End experiment.

After measurements

1. Remove EMG electrodes, release subject from NSO and remove EMGmarkers.

2. Copy measurement data to a folder containing name of subject in the title.

3. Check whether all data is presentable.

45

Appendix D

Data processing

In the following overview are the Matlab commands described that are used inpost processing the measured signals from the Achilles robotic ankle manipula-tor and the Porti7 interface.

The following commands are used to process the data measured in the passiveRoM task:

% Loading Achilles data, i.e. Session.Datadat = cell2mat(Session.Data(1));pr1 = dat(:,8); % Position measured [rad]Tr1 = dat(:,10); % Torque measured [Nm]pr1 = (pr1*180)/pi;% Adjusting for subject specific neutral ankle position, measured manuallypr1 = pr1 − horpos;% Dorsiflexion torque is negative in the Achilles coordinate systemTr1 = (−1)*Tr1;% Fitting a fifth order polynomial on the torque−angle characteristicp1 = polyfit(pr1,Tr1,5);y1 = polyval(p1,pr1);

The following commands are used to process the data measured in the activeRoM task:

% Loading Achilles data, i.e. Session.Datadat = cell2mat(Session.Data(2));tr2 = dat(:,1); % Time [s]pr2 = dat(:,8); % Position measured [rad]vr5 = dat(:,9); % Velocity measured [rad/s]vr5 = vr5*180/pi;% Adjusting for subject specific neutral ankle position, measured manuallypr2 = pr2*180/pi − horpos;% Removing first 70 samples to synchronize with EMG−measurementsvr5 = vr5(71:end);pr2 = pr2(71:end);tr2 = tr2(1:end−70);% Filtering velocity signalfSample = 1024; % [Hz]fLP = 5; % [Hz]butterOrder = 3;[Blo,Alo] = butter(butterOrder, fLP/(fSample/2), 'low');vr5 = filtfilt(Blo, Alo, vr5);

46

% Creating EMG−activity filtersfSample = 2000; % [Hz]fHP = 20; % [Hz]fLP = 4; % [Hz]butterOrder = 3;% Initial EMG filter[Bhi,Ahi] = butter(butterOrder, fHP/(fSample/2), 'high');% Post−processing filter[Blo,Alo] = butter(butterOrder, fLP/(fSample/2), 'low');

% Loading Porti7 data, i.e. Channels.Samplesfor i = 1:8 % Eight channels, two for each muscle

Channels{i}.Samples(1) = []; % Removing first samplemeanCh(i) = mean(Channels{i}.Samples);avgCh(i,:) = Channels{i}.Samples − meanCh(i); % NormalizationavgCh(i,:) = filtfilt(Bhi, Ahi, avgCh(i,:)); % Initial filter

end% Bipolar deductionmusc12 = avgCh(1,:) − avgCh(2,:); % Tibialis Anterior EMG−activitymusc22 = avgCh(3,:) − avgCh(4,:); % Soleus EMG−activitymusc32 = avgCh(5,:) − avgCh(6,:); % Gastrocnemius Medialis EMG−activitymusc42 = avgCh(7,:) − avgCh(8,:); % Gastrocnemius Lateralis EMG−activity% Rectificationmusc12 = abs(musc12);musc22 = abs(musc22);musc32 = abs(musc32);musc42 = abs(musc42);% Post processing filtermusc12 = filtfilt(Blo, Alo, musc12);musc22 = filtfilt(Blo, Alo, musc22);musc32 = filtfilt(Blo, Alo, musc32);musc42 = filtfilt(Blo, Alo, musc42);musc2 = musc22 + musc32 + musc42; % Plantarflexor muscle EMG−activity% Creating time vectort = 1/2000:1/2000:81;t2 = t(1:length(musc12));

% Matching the Achilles data with the EMG−activity datapr2 = pr2';pr2 = pr2(1:t2(end)*1024);pr2 = pr2(1:length(pr2)/length(musc12):end);musc12 = musc12(1:end−1);musc2 = musc2(1:end−1);

47

design and evaluation of a stiffness ... - tu delft

Documents