Proceedings of the International Conference on Communication and Electronics Systems (ICCES 2018)

IEEE Xplore Part Number:CFP18AWO-ART; ISBN:978-1-5386-4765-3

Development of Electromyogram Based

Rehabilitation Device for Upper Limb
Amputation using Neural Network
Gayathri Priyadarshini.R Suryarajan.R Prasad J
Assistant Professor: Electronics Assistant Professor: Biomedical Assistant Professor (Sr.Gr):
and Instrumentation Engineering Engineering Biomedical Engineering
Bannari Amman Institute of Sri Ramakrishna Engineering Sri Ramakrishna Engineering
Technology, Erode, India College College
gayathripriyadarshini@bitsathy. Coimbatore, India Coimbatore, India
ac.in [email protected] [email protected]

Abstract—Implementation of technology in integrating While considering earlier studies, one technology

humans and machines provides a variety of that provides support for upper extremity can be
opportunities in the medical industry, in rehabilitation. classified based on human interaction with the device:
One of the major concerns is to provide a better 1. Handle the end-effector, 2. Cable driven support.
solution for people in providing post-operative support But the use of more number of cables along with the
for physical rehabilitation. This could be done either motors used for actuation could prove costly. Devices
using mechanical, electrical or myo-electric functions. of the first category include MIT Manus (G. Kwakkel
The objective of this paper is to indicate EMG based
framework control connected to mechanized gadget
et al 1997), (D. Reinkensmeyer et al1993). These
which copies the upper appendage exercises. Through designs add inertia to a subject’s hand during motion.
two biometrical sensors it captures biceps EMG signal, The second set of device includes Armin (C. Carignan
which is then filtered and processed. Finally an et al2005) and MGA (A. Frisoli et al2005). These
output/control signal is sent to the actuators, which will designs are usually, anthropomorphic and made of
then perform the movement of upper limb. The goal is rigid links attached to the human arm. Other study
to predict and restore the lower arm movement through suggests exoskeleton model for upper hand, especially
the EMG signal analysis. in finger with 16 actuation joints for each finger [3]
(Christian Fleischer et al 2006). Quantitative analysis
Keywords: Physical Rehabilitation, upper limb, Myo-
electric signal (EMG), Actuators.
of Electromyogram signals which shows an important
source of information for the diagnosis of disorders in
neuromuscular regions. Now with the advancements
of computer aided EMG equipment, time and
I. INTRODUCTION frequency domain quantitative analysis is possible.
In this present world, human-machine interface is (Abdul hamit Subasi et al 2006), gives an examination
providing with multiple opportunities in all fields, between two classifiers to be specific feed forward
especially in rehabilitation technology. Here the mistake back proliferation fake neural systems and
technology deals with the physical exercise to restore wavelet neural system which gives the connection to
the lost functions in the body [1]. Studies have shown their precision in grouping of EMG signals [4].
that intensive movement oriented training will affect
the recovery of neuromuscular functions of the human In this paper, we use a model that mimics the
arm. Basically, it is based on manual training for the original arm, which is driven based on the EMG
patients which requires manpower. But mostly non- signals. So the working of the model arm supports in
availability of therapists and more hospital stay could restoring the lost functions of the hand. Moreover,
prove costly. So as to overcome these difficulties, we while replacing the subject has to be ensured that the
provide a solution by providing an artificial support to original function being restored, basically dealt with
the patients by restoring the lost functions based on prosthesis. Prosthesis is a condition was the dis-
the movement of their other arm, assisting them in functional organ or part is being completely replaced
recovering from their lost functions [2]. Here the with another part [1]. Mainly rehabilitation is a boon in
device potentially offers uniform performance and the biomedical industry, were more researches are
provide quantitative measures. carried out to mimic the functions of the original body
part and help restore the lost functions. Thus, here we

978-1-5386-4765-3/18/$31.00 ©2018 IEEE 826

Proceedings of the International Conference on Communication and Electronics Systems (ICCES 2018)
IEEE Xplore Part Number:CFP18AWO-ART; ISBN:978-1-5386-4765-3

mainly discuss on the physical rehabilitation of upper D. Mean Absolute Value

extremities of the arm. Moving window is the major tool for calculating the
mean absolute value (MAV). It is calculated for each
II. METHODS
window of data according to the equation,
Once the signal is obtained from the biceps of the ... (4)
subject with the use of surface electrodes, signal
amplification and filtering plays a critical role in Where, MAV- Mean Absolute Value
EMG practical application systems [5]. The process is S- Window Length
expected to give as much information from the E. Integrated EMG
physiological EMG signals as possible while it
minimizes the noise components. The maximization Incorporated EMG (IEMG) is ascertained as the
summation of the supreme estimations of the sEMG
of the signal to noise ratio and the minimization of the
flag sufficiency. By and large, IEMG is utilized as a
distortion of the signal are the two major functions of
beginning record to distinguish the muscle
amplification and filtering process [6]. After signal
movement that used to approaching the control order
amplification and filtering, to process the signal
of assistive control gadget. It is identified with the
feature extraction of the signal is required.
sEMG flag succession terminating point,
A. FeatureExtraction
... (5)
Feature extraction involves the process of reducing
the large set of data from the resources inorder to
obtain certain information. In general term, while Where, N- Length of the Signal
describing the data with sufficient accuracy various Xn- sEMG signal in Segment
methods of constructing combinations of the variables
to get around with the processing [7]. F. Amplitude
A measure of its change over a solitary period is
B. Wavelet Multiresolution Analysis thought to be the sufficiency of an intermittent
Information content of a signal are analyzed using variable. The elements of the greatness of the
multiresolution representations [4]. By decomposing distinction between the variable's outtrageous
the signal on a wavelet orthonormal basis of L2 (Rn), qualities,
here the difference of information between the ... (6)
approximation of a signal at the resolution 2j+1 and 2j Where, A--Peak amplitude of the wave
(j-integer) can be extracted, the vector space of x-- Oscillating variable
measurable, square-integral n-dimensional functions. -- Angular frequency
This decomposition defines an orthogonal t-- time
multiresolution representations called wavelet K and b are arbitrary constants
representation which is given as,
For every wavelet Ψ, III. EXPERMENTAL SETUP
Vj L2 (R), j Z ... (1) A model of an upper limb was developed to support
Vj=……+Wj-2+Wj-1 ... (2) or assist the upper limb post traumatic patients to
Where, Vj -- Large Scale regain their movements at a faster rate and to reduce
Wj --Small Scale the frequent hospital visits. The model is made up of
L--Details aluminum rods which are less in weight which is also
portable and the design is less complex in nature,
C. Root Mean Square
when compared with the cable driven arm exoskeleton
RMS can be defined as the statistical measure of based rehabilitation device [8] (Ying Mao et al 2012).
the magnitude of a varying quantity, here the The patient needs to rest the arm in the device, which
EMG signals. It is the square root of the arithmetic contains the upper arm cuff and the forearm cuff.
mean of the squares of the set of values. This will Once being rested, the EMG electrodes which are
help to find the digital values required for the placed on the other arm (biceps) is used for
control of the exoskeleton arm. Let Xn represent controlling the device [9]. The exoskeleton arm moves
the samples then RMS is given by, according to the movement of the other arm for
2n ... (3) lexicon and extension activities. There is a control
console so that the patient can stop and start the setup
Where, Vrms--Root mean square value
whenever discomfort occurs [10]. A 12V DC motor is
Xn-- Value of samples
used alongside of the forearm cuff to drive the setup
n-- No of samples

978-1-5386-4765-3/18/$31.00 ©2018 IEEE 827

Proceedings of the International Conference on Communication and Electronics Systems (ICCES 2018)
IEEE Xplore Part Number:CFP18AWO-ART; ISBN:978-1-5386-4765-3

and a SPDT switch (Relay) is used for the flexion and

extension [11][12]. Model of the system is shown in
Figure1a and 1b.

Figure1a. Experimental setup Figure1b. Model of the Exoskeleton Arm

Parts of the Exoskeleton model:
1.Upper arm cuff 2. Motor 3. Lower arm cuff
4. Support Stand 5. Base Plate
IV. RESULTS
The raw EMG signal is obtained using unwanted and line interference signals. Then the
the surface electrodes taken from the biceps of the signal is being de-noised using the wavelet de-
subject is shown in the figure 2. In figure 3, The noising to eliminate high frequency noise in the
EMG signals were filtered using the Band-pass figure 4.
filter and Band-stop filter to eliminate the

Figure 2. Raw EMG Signal

Figure 3. Filtered EMG Signal

978-1-5386-4765-3/18/$31.00 ©2018 IEEE 828

Proceedings of the International Conference on Communication and Electronics Systems (ICCES 2018)
IEEE Xplore Part Number:CFP18AWO-ART; ISBN:978-1-5386-4765-3

Figure 4. Wavelet De-Noised Signal

Figure 5. Feature Extracted Values

Figure 6. Front Panel of LabVIEW with NN classification-Extension

Figure 7. Front Panel of LabVIEW with NN classification-Flexion

After Wavelet de-noising, RMS, MAV, neural network was given as test signal to find the
Amplitude and IEMG are calculated termed as error percentage between the training data and test
feature extraction and is shown in figure 5. After data. Once the test data are similar to the training
the feature extraction the values are used for pattern, it is confirmed that the network was
classification in neural network for identifying the trained for the best results.
different EMG patterns (Flexion and Extension). After the classification in neural network,
The Figure 8 shows the performance which the program is called in the LabVIEW as control
represents the training pattern along with the test signals for the motor to run in forward and reverse
signal given by the training algorithm itself. direction represents flexion and extension. Figure
Now once the training is over, the 8 shows the operation of matlab in LabVIEW. The
remaining data that are not used for training the

978-1-5386-4765-3/18/$31.00 ©2018 IEEE 829

Proceedings of the International Conference on Communication and Electronics Systems (ICCES 2018)
IEEE Xplore Part Number:CFP18AWO-ART; ISBN:978-1-5386-4765-3

matlab program is called in LabVIEW using effective than the existing system. In addition to the
Mathscript function. current application, the device can be developed for
prosthetic applications like artificial limb, finger
prosthesis. More likely to add, lower limb
rehabilitation can also be developed. The most
prominent system lies with the human machine
interface for rehabilitation. As to say EMG provides
with one best solution in controlling the muscle
action.
REFERENCES
[1] Deepak Joshi, SonalAtreya, A. S. Arora, SnehAnand 2009,
“Trends in EMG based Prosthetic Hand Development: A
Review”, Indian Journal of Biomechanics: Special Issue
NCBM 7-8, pp.228-232
[2] Carignan C, M. Liszka, and S. Roderick 2005, “Design of an
arm exoskeleton with scapula motion for shoulder
rehabilitation”, in Proc. Int. Conf. Adv. Robot, pp. 524–531.
[3] Christian Fleischer, Andreas Wege, Konstantin Kondak and
Gunter Hommel 2006, “Application of EMG signals for
controlling exoskeleton robots”, Biomed Tech 51, pp.314–
Figure 8. Performance plot of the training 319
data-NN [4] AbdulhamitSubasi, Mustafa Yilmaz, HasanRizaOzcalik
2006, “Classification of EMG Signals Using Wavelet Neural
After the above steps, the feature extracted Network”, Journal of Neuroscience Methods (156), pp.360-
367
values are given to the Mathscript node. Here the
[5] Jacob Rosen, Moshe Brand, Moshe B. Fuchs, and
trained values and feature extracted values are MirceaArcan 2001, “A Myosignal-Based Powered
compared for similarity and the corresponding output Exoskeleton System”, IEEE Transaction on Systems, Man,
is given out from the script node. The output is And Cybernetics—Part A: Systems And Humans, vol.
31,no.3,pp.1-15
considered as the control signals for driving the
[6] Carlo J. De Luca, L. Donald Gilmore, Mikhail Kuznetsov,
motor. Figure 6 & 7 explains the front panel of the Serge H. Roy 2010, “Filtering the surface EMG signal:
LabVIEW for Extension and Flexion after the Movement artifact and baseline noise contamination”,
execution of the neural network. Journal of Biomechanics 43, pp.1573-1579 (Science Direct)
In this manner the outcomes from the Neural [7] Panagiotis K. Artemiadis and Kostas J. Kyriakopoulos
Network gives the yield (named as control signals) to (2010), “An EMG-Based Robot Control Scheme Robust to
Time-Varying EMG Signal Features”, IEEE Transactions On
drive the engine in forward and switch bearing (i.e. Information Technology In Biomedicine, vol. 14, no.3,
flexion and expansion) of the exoskeleton arm. pp.582-588
[8] Ying Mao and Sunil Kumar Agrawal 2012, “Design of a
V. CONCLUSION AND FUTURE WORK Cable-Driven Arm Exoskeleton (CAREX) for Neural
Rehabilitation”, IEEE Transactions on Robotics, vol. 28,
The paper describes the working model of an no.4, pp.922-931
exoskeleton arm which could be used as an assistive [9] Nicola Vitiello¸ TommasoLenzi, Stefano Roccella, Stefano
device during rehabilitation for the upper limb. The Marco Maria De Rossi, EmanueleCattin, Francesco
device is controlled by the subjects own EMG signals Giovacchini, FabrizioVecchi, Maria Chiara Carrozza 2012,
from their other arm (placement in biceps). When “NEUROExos: A Powered Elbow Exoskeleton for Physical
Rehabilitation”, IEEE Transactions on Robotics, pp.1-15
compared with the existing technology the model that
[10] Nan Bu, Masaru Okamoto, and Toshio Tsuji 2009, “A
is suggested here is less in weight and portable. The Hybrid Motion Classification Approach for EMG-Based
system design is based on aluminum, which is Human–Robot Interfaces Using Bayesian and Neural
naturally less weight. With this system the Networks”, IEEE Transactions on Robotics, vol.25 no.3, pp.
rehabilitation time is reduced, faster treatment is 502-511
provided and very effective for the patients. [11] Mohan C, Vinod Kumar Giri 2011, “DC Motor Control
using EMG Signal for Prosthesis”, International Journal of
The work done is for two degree of freedom i.e. Electronics & Communication Technology, vol. 2, no.2,
flexion (upward movement) and extension (downward pp.163-166
movement), but in future the degree of freedom can [12] Shobhitha Ann Job, R.Jegan, Melwin Abraham C (2013),
be increased. To obtain more degrees of freedom, “OWI-535 EDGE Robotic Arm Control Using
Electromyogram (EMG) Signals”,International Journal of
multichannel electrodes has to be used. Using Innovative Technology and Exploring Engineering (IJITEE)
multichannel electrodes will provide best results and ISSN: 2278-3075, vol.2, no.6,pp.282-286.
the classification of arm movements will be more

978-1-5386-4765-3/18/$31.00 ©2018 IEEE 830