A Preliminary Study of Ankle Variable Hybrid Above-knee

Prostheses

Su-Hong Eom

, Sun-Jong Na

, Sang-Hyun Lee

, Se-Hoon Park

and Eung-Hyuk Lee

Department of Electronic Engineering, Korea Polytechnic University, Siheung City, Gyeonggi-do, Korea

Korea Orthopedics & Rehabilitation Engineering Center, Incheon, Korea

Keywords: Ankle Variable Prostheses, Hybrid Above-knee Prosthesis, Active and Semiactive Prostheses.

Abstract: This study is a preliminary study to solve problems in gait imbalance at slope ways and low ramps with ankle

variable hybrid above-knee prostheses. For the purpose of implementing ankle variable control, the stance

phase in gait was determined as a step-by-step manner and the threshold values were derived through the

decision tree learning method based on inertial sensor data in verifying the swing phase. It can be used to

perform the ankle variable control. The control of the hybrid above-knee prosthesis was demonstrated by

measuring butterfly diagrams on a low ramp for verifying the gait balance in the test ramp.

1 INTRODUCTION

Prostheses are an aid used by people who have

amputated legs due to natural causes or accidents

(Dongfang Xu, et al., 2018). According to a WHO

report in 2017, there are an estimated 30 million

people who have lower limb amputation disorders,

and 60 million people around 2025 throughout the

world (WHO, 2018). Therefore, more convenient

and more natural functional consideration are being

paid to their lost body restoration devices (Steven

Culver, et al., 2018; Matthew L. Handford and

Manoj Srinivasan, 2018).

Prostheses are divided into two different types,

above-knee prostheses and below-knee prostheses,

depending on the area of the amputation level. The

above-knee prostheses are those used by the

amputee below the knee, and the below-knee

prostheses are those used by the lower ankle

amputee. Due to this location of amputation, the

below-knee prosthesis users are generally less

uncomfortable with walking by using it than the

above knee prosthesis users (Jinying Zhu, et al.,

2014; Kyle J. Kaveny, et al., 2018). Therefore, the

objective of this study is to develop prostheses with

more comfortable and natural functions for the

subjects of above-knee prosthesis users who feel

pains in their gait.

https://orcid.org/0000-0001-8493-1432

https://orcid.org/0000-0002-0601-9058

Recently, the research trend of above-knee

prostheses is developing from passive to intelligent

type prostheses, and the intelligent type prostheses

are divided into powered types, which are directly

involved in the movement of the knee joint, and

passive types (Elissa D. Ledoux and Michael

Goldfarb, 2017; D. Quintero, D. J. Villarreal, D.

Lambert, S. Kapp, and R. D. Gregg, 2018). In the

case of powered prostheses, it can produce a similar

reproduction of a person's gait trajectory because it

drives the knee joint using a motor. However, there

is a risk that the prostheses may stop while walking

due to the limitations of weight and application time

of batteries. In order to overcome the limitations,

recently, hybrid type prostheses, which combine

passive and powered types, are being researched and

developed (M. Bellmann, et al., 2010; B. Lawson, et

al., 2013; B. E. Lawson, et al., 2014).

The current representative hybrid above-knee

prostheses are the MIT Knee and RIC's hybrid knee

(T. Lenzi, et al., 2015). The MIT Knee uses motor

power only in the swing phase and is powered as a

passive type in the stance phase. Unlike the MIT

Knee, the RIC's hybrid knee operates in a passive

type in the swing phase and uses motor power

selectively when optional active force is required in

the stance phase. Therefore, although this control

method is not a big problem with level walking, it

Eom, S., Na, S., Lee, S., Park, S. and Lee, E.

A Preliminary Study of Ankle Variable Hybrid Above-knee Prostheses.

DOI: 10.5220/0007952807890794

In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 789-794

ISBN: 978-989-758-380-3

789

causes gait imbalance due to short widths of the

affected area caused by passive type prostheses

because the bounce trajectory of the knees generated

by the initial contact (I.C.) phase, loading response

(L.R.), phase, and intermediate stance (M.St., Mid-

Stance) phase in the gait of the general stance phase

is the same as that of the walkway when entering

slopes or on low ramps.

In this study, therefore, the gait imbalance will be

compensated by changing the ankle joint for the

missing in the knee joint using the below-knee

prosthesis, which is implemented by a variable

hydraulic cylinder applied to the ankle joint in order

to solve the problem.

The method proposed in this study is to attach a

variable below-knee with a variable hydraulic

cylinder to the above-knee prosthesis that uses power

only in the swing phase to give the ankle a flexible

change. However, the flexible state of the ankle shall

be fixed before the toe-off point at which the foot falls

off the ground. This is because the toe-off is the time

to gain the gait momentum. Thus, the pressure of the

hydraulic cylinder was set to a maximum level before

the toe-off by estimating the stance phase of the

prosthesis step-by-step.

In order to determine whether this method of gait

can solve the imbalance, the butterfly diagrams

between the affected and the unaffected sides were

measured and evaluated.

2 METHOD

2.1 Ankle Variable Hybrid Prosthesis

The prostheses used in this study are those that were

co-studied with the Korea Orthopedics &

Rehabilitation Engineering Center, as shown in Fig.1.

The range of motion of the knee joint is 0º～90º and

the range of motion of the ankle joint is presented by

16° of planar flexion and 18° of dorsiflexion. The

total weight is 4Kg; knee 2.1Kg, ankle 1.4Kg, battery

0.35Kg, and controller 0.15Kg. In the below-knee

prosthesis, there is only a motor to control the

hydraulic nozzle and no motor for the powered assist

function.

Figure 1: Hybrid above-knee prosthesis of KOREC.

2.2 Gait Phase Detection Algorithm

Since the hybrid prosthesis applied in this paper does

not have power assist in the stance phase, the gait

trajectory in the stance phase resembles the passive

prosthesis as shown in Fig. 2. Thus, it is not possible

to determine the gait phase based on the gait

trajectory. In this study, therefore, the initial contact

phase, intermediate stance phase, terminal stance

phase, and swing phase were determined by the hip

angles, knee angles, progressive angular velocity of

the prosthesis, and the angular velocity of the Yaw-

axis of gait based on inertial sensors. The threshold

values for the determination were derived using an IF

～THEN format based on the decision tree learning

method in as shown in Fig. 2. Here, the decision tree

learning method was applied to the learner because

the sequence of gait is interpreted as the State

Machine behavior, which is the specification for the

sequential occurrence.

Figure 2: Gait phase detection using a learner.

2.2.1 Decision Tree Learning Method

The decision tree learning method is a learner that

classifies or predicts dependent variables by changing

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

790

the rule of decision making into a tree structure as one

of the instructional learning that is taught with a label

on the training data. It explores the characteristics,

patterns, and rules of the target based on structural or

unstructured data and analyzes various factors that

show relevance. It is possible to see the process of the

analysis and is easy to interpret and understand. Also,

it has the advantage of being stable and not having to

process data. The decision tree learning method is a

learner that can be both categorization and regression

and can predict both categories and continuous

numbers. Recently, a research has been conducted to

classify images using the decision tree learning

method as well as numerical data (Han Liu, et al.,

2017).



x, Y











,



,



,⋯,



,



(1)

The decision tree learning method is represented by

Eq. (1) and vector Y consists of labels and vector x,

which is a set of input variables 



, 



, 



, and so on.

In the decision tree, the starting node is called a

root node and it consists of a branch that connects

between the child nodes. The last node that ends the

classification is called the terminal node, and the

number of branches between the root node and the

end node is called depth.

The decision tree learning method proceeds with

the separation of data in a way that reduces the

impurity as much. The Gini Index, used as an

indicator of the impurity, refers to the probability that

an item will be randomly selected from a set and then

misplaced when estimating a label. The equation for

the Gini Index is as shown in Eq. (2).

. 







1

∑











w (2)

here, k∈1,2,3,⋯,m and p is the data in a set

labeled i and the Gini Index as a value between 0

and 1.

2.3 Hybrid Prosthesis Control

The control of the ankle variable above-knee

prosthesis proposed in this study is as follows. The

control is largely divided into two stages. The first

control is to detect the intermediate stance phase in

the stance phase so that the ankle variable can be fixed

before the final stance phase in order to obtain gait

momentum in the terminal stance phase. The second

control is to release the lock on the ankle from the

swing phase in order to move the ankle during the

initial contact and intermediate stance phases by

inertia. Fig. 3 shows this control sequence.

Figure 3: Control block of the ankle variable hybrid

prosthesis.

3 IMPLEMENTATION AND

EVALUATION OF THE

PRELIMINARY STUDY

3.1 Training and Learning in the Gait

Phase

In this study, a decision tree learning method was

proposed to determine the gait phase using inertial

sensors in the passive prosthesis-based gait.

Therefore, to obtain training data for learning, we

used the 3R60 prosthesis, Ottobock, to the socket that

can be worn by normal people, as shown in Figure 4,

in order to obtain training data for learning. Then, a

precision small encoder was attached to make certain

changes in the joint, together with inertial sensor data.

The small rotary encoder was a product of Autonics

and the specification is 360 pulses/revolution. The

inertial sensor was NGIMU from x-io Technologies

and acquired all data through WiFi communication.

The reason that we chose the 3R60 prosthesis as a

passive type is to fit into the same experimental

condition as the user of above-knee prostheses.

The training data were obtained from two healthy

70Kg and three 80Kg men in their 20s, and a passive

type above-knee prosthesis user with 72Kg weight on

the right side of the Stump Length 44cm.

In the experiment, the gait data was obtained by a

total of 500 times with a width of 30cm and five

continuous gait steps in which 80% were used as

training data, and 20% were used as verification data.

A Preliminary Study of Ankle Variable Hybrid Above-knee Prostheses

791

Figure 4: Passive type prosthesis adaptor and sensor

attachment locations, experimental setting image and

coordinate system used in this experiment.

The values used as an element of the decision tree

in the acquired data are the angle of the change in the

knee angles obtained by the small encoder, the change

in the hip angles of the prosthesis leg acquired by the

inertia sensor, the angular velocity of the gait

direction of the prosthesis, and the angular velocity of

the Yaw axis issued during gait. Fig. 5 represents the

acquired data.

In Fig. 5, the knee angle increases sharply at the

beginning of the swing phase, and the hip angle

increases gradually from the stance phase and begins

to decrease at the beginning of the swing phase. The

Gyro Z is a Yaw axis movement in the direction of the

prosthesis in the swing phase. Data changes on the

Yaw axis can be interpreted as a reflection of the

amount of movement that occurs when the pelvis

moves during gait. Therefore, it is possible to verify

that the value of the change is significant in the swing

phase.

However, the decision tree was learned with the

emphasis on the relation between the knee angle, hip

angle, and Gyro Z mentioned previously because the

Giro X value, the angular velocity in the gait progress,

did not change much of the acquired data.

Figure 5: Single gait phase for the machine learning.

3.1.1 Evaluation of the Threshold Values

Determined by the Decision Tree

Learning Method

The threshold values derived from the decision tree

learning method performed to identify the gait phase

of the passive type prosthesis proposed in this study

are shown in Table 1. The threshold values are

optimized for the right-side above-knee prosthesis

wearer for the final evaluation in this study.

For the evaluation of the derived threshold values,

an evaluation program was produced as shown in Fig.

6 and a simulation was performed by applying 20%

of the acquired data randomly. As a result, the

accuracy of 98.6% of the initial contact phase, 97.2%

of the intermediate stance phase, 98.9% of the

terminal stance phase, and 99.8% of the swing phase

was derived. Also, the trigger signal to release the

lock on the ankle was 99% accuracy

In addition, it showed that there is room for

improving the accuracy of the trigger signal because

the trigger signal for setting the lock on the ankle was

detected with an average of 97.2% accuracy.

Table 1: Threshold values of the gait phase derived from

the learning.

Phases

Boundary Value

Initial

Contact

Gyro Z < -10.392, Hip angle < 1.09251,

Knee angle <= 0.432182, and the other

three thresholds

Mid

Stance

Gyro Z < -10.392, Hip angle >= 1.09251,

Hip angle < -11.2027, and the other one

threshold

Terminal

Stance

Gyro Z < -10.392, Hip angle >= 1.09251,

Hip angle < -11.2027, and Gyro X >=

7.72385

Swing

Gyro Z >= -10.392, Knee angle >= 1.09251

Hip angle >= -3.47349, and the other

three thresholds

Figure 6: Evaluation program of the learner.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

792

3.2 Evaluation of the Gait Balance in

the Ankle Variable Control

An experiment that determines whether the variable

control was properly performed was implemented in

order to evaluate the gait balance as the ankle variable

control was applied based on the gait phase of the

classified stance phase. The experiment confirmed

the change of the ankle with a three-dimensional

motion analyzer based on the control trigger signal

from the classifier. A three-dimensional motion

analysis system (Cortex 6.02, Motion Analysis Corp.,

USA) was used to confirm the results. The system

consists of 12 infrared cameras, with 120 Hz of

sampling and 12.5Cm diameter reflective markers.

The experiment for the evaluation measured changes

in the ankle by attaching markers on both ankles.

In the first experiment, the angle changes of the

planar flexion and dorsiflexion of the ankle were

observed. The measured graph is shown in Fig. 7. It

was confirmed that proper changes in the ankle alone

represent gait trajectories, which are similar to the

actual healthy angle. Thus, it has been confirmed that

the ankle variable control can help the gait balance of

the above-knee prosthesis users.

Figure 7: Gait trajectory of Helthy Ankle, Variable (Non-

Powered) Ankle and Fixed Ankle.

In the second experiment, seen the differences of

unaffected stride in general ankles fixed type

prosthesis whether can be solved. The experiment

utilizes a load cell sensor mounted on the prosthesis.

The load cell is tied to the joint of the transfemoral

prosthesis and lower limb prosthesis. The experiment

was conducted under three conditions: the ankle was

fixed from the beginning, the variable state from the

swing phase to the intermediate stance phase and

fixed after the intermediate stance phase, and the

variable conditions between all gait cycles. The

results are presented in Fig. 8.

Figure 8: Gait stance according to the Variation of Ankle

Control.

When walking with the ankle in a fixed state, the

toe off point tends to be shortened, and in contrast, the

toe off point could be confirmed to be longer if the

ankle is set as a flexible state. Therefore, it has been

confirmed that the appropriate toe off timing can help

with gait balance.

Finally, the dynamic foot pressure examination

was checked to verify the actual gait balance between

the affected and the unaffected sides. The evaluation

was implemented using the Gait analysis treadmill

from Zebris. The gait speed was 1.5km/h, and the

gradient was 4°. In the evaluation method, the reason

for the low slope rather than level ground is the

arbitrary setting for measuring the gait imbalance on

the ramp entry section as a problem mentioned in the

introduction. Fig. 9 shows the evaluation graph. As

shown in Fig. 9, it was verified that the ankle variable

hybrid prosthesis proposed in this study represents the

gait balance in comparison to the 3R80 passive type

above-knee prosthesis with the clearly fixed ankle.

Figure 9: Butterfly diagram of Existing prosthesis and

Prosthesis with proposed techniques.

A Preliminary Study of Ankle Variable Hybrid Above-knee Prostheses

793

This is because proper changes in the ankle joint

during gait have partially compensated for the bounce

role of the load reactor in the stance phase gait

trajectory in the gait of normal people. Also, it can be

attributed to the fact that the swing phase assisted

power to the movement of the knee joint.

4 CONCLUSIONS

In this study, a preliminary study was performed to

solve the gait imbalance of passive type prostheses

that occur on the entrance or low slope of the ramp,

through applying the ankle variable hybrid prosthesis

developed by the Korea Orthopedics & Rehabilitation

Engineering Center. As a result of the performance, it

was confirmed that only the variation in the ankle

showed the gait balance. The results confirmed that

even though the powered below-knee prosthesis was

not used, the gait balance was achieved in the hybrid

above-knee prosthesis. Thus, it has been confirmed

that this gait control can improve the convenience of

the users of above-knee prostheses. However, this is

the early stage of the study and is a lack of suggestion

of quantitative ankle joint resistances. Also, it still

requires further experimentations in a gait

environment with high slopes.

ACKNOWLEDGEMENTS

This research was supported by Basic Science

Research Program through the National Research

Foundation of Korea (NRF) funded by the Ministry

of Education (NRF-2017R1A2B2006958).

This research was supported by the Bio & Medical

Technology Development Program of the NRF

funded by the Korean government, MSIP (NRF-

2017M3A9E2063260).

This research was supported by the MSIT

(Ministry of Science and ICT), Korea, under the

ITRC (Information Technology Research Center)

support program (IITP-2018-0-01426) supervised by

the IITP.

REFERENCES

Dongfang Xu, Yanggang Feng, Jingeng Mai, and Qining

Wang, 2018. Real-Time On-Board Recognition of

Continuous Locomotion Modes for Amputees with

Robotic Transtibial Prostheses. IEEE Transactions on

Neural Systems and Rehabilitation Engineering,

Volume 26, Issue 10, pp 2015-2025.

WHO. 2018. World Report on Disability.

Steven Culver, Harrison Bartlett, Amanda Shultz, Michael

Goldfarb. 2018. A Stair Ascent and Descent Controller

for a Powered Ankle Prosthesis. IEEE Transactions on

Neural Systems and Rehabilitation Engineering, pp

993-1002.

Matthew L. Handford, Manoj Srinivasan, 2018. Energy-

Optimal Human Walking With Feedback-Controlled

Robotic Prostheses: A Computational Study. IEEE

Transactions on Neural Systems and Rehabilitation

Engineering, VOL 26, NO 9, pp 1773-1782.

Jinying Zhu, Qining Wang, Long Wang, 2014. On the

Design of a Powered Transtibial Prosthesis with

Stiffness Adaptable Ankle and Toe Joints. IEEE

Transactions on Industrial Electronics, Volume 61,

Issue 9, pp 4797-4807.

Kyle J. Kaveny, Ann M. Simon, Tommaso Lenzi, Suzanne

B. Finucane, Emily A. Seyforth, Graci Finco, Kasen L.

Culler, Levi J. Hargrove, 2018. Initial Results of a

Variable Speed Knee Controller for Walking with a

Powered Knee and Ankle Prosthesis. 2018 7th IEEE

International Conference on Biomedical Robotics and

Biomechatronics. Biorob, pp 764-769.

Elissa D. Ledoux and Michael Goldfarb, 2017. Control and

Evaluation of a Powered Transfemoral Prosthesis for

Stair Ascent. IEEE Transactions on Neural Systems

and Rehabilitation Engineering, vol 25, NO 7, pp 917-

924.

D. Quintero, D. J. Villarreal, D. Lambert, S. Kapp, and R.

D. Gregg, 2018. Continuous-phase control of a

powered knee-ankle prosthesis: Amputee experiments

across speeds and inclines. IEEE Transactions on

Robotics, vol. 34, no. 3, pp 686-701.

M. Bellmann, T. Schmalz, S. Blumentritt, 2010.

Comparative biomechanical analysis of current

microprocessor-controlled prosthetic knee joints. Arch.

Phys. Med. Rehabil., vol 91, no 4, pp 644–652.

B. Lawson, H. Varol, A. Huff, E. Erdemir, and M.

Goldfarb, 2013. Control of stair ascent and descent with

a powered transfemoral prosthesis. IEEE Trans. Neural

Syst. Rehabil. Eng., vol 21, no 3, pp 466–473.

B. E. Lawson, J. Mitchell, D. Truex, A. Shultz, E. Ledoux,

and M. Goldfarb, 2014. A robotic leg prosthesis:

Design, control, and implementation. IEEE Robotics &

Automation Magazine, vol 21, no 4, pp 70–81.

T. Lenzi, J. Sensinger, J. Lipsey, L. Hargrove, T. Kuiken,

2015. Design and preliminary testing of the RIC hybrid

knee prosthesis. 2015 37th Annual International

Conference of the IEEE Engineering in Medicine and

Biology Society, pp 1683-1686.

Han Liu, Mihaela Cocea, Weili Ding, 2017. Decision tree

learning based feature evaluation and selection for

image classification. International Conference on

Machine Learning and Cybernetics, pp569-574.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

794